linux/arch/s390/boot/startup.c

649 lines
22 KiB

s390/smp: rework absolute lowcore access Temporary unsetting of the prefix page in memcpy_absolute() routine poses a risk of executing code path with unexpectedly disabled prefix page. This rework avoids the prefix page uninstalling and disabling of normal and machine check interrupts when accessing the absolute zero memory. Although memcpy_absolute() routine can access the whole memory, it is only used to update the absolute zero lowcore. This rework therefore introduces a new mechanism for the absolute zero lowcore access and scraps memcpy_absolute() routine for good. Instead, an area is reserved in the virtual memory that is used for the absolute lowcore access only. That area holds an array of 8KB virtual mappings - one per CPU. Whenever a CPU is brought online, the corresponding item is mapped to the real address of the previously installed prefix page. The absolute zero lowcore access works like this: a CPU calls the new primitive get_abs_lowcore() to obtain its 8KB mapping as a pointer to the struct lowcore. Virtual address references to that pointer get translated to the real addresses of the prefix page, which in turn gets swapped with the absolute zero memory addresses due to prefixing. Once the pointer is not needed it must be released with put_abs_lowcore() primitive: struct lowcore *abs_lc; unsigned long flags; abs_lc = get_abs_lowcore(&flags); abs_lc->... = ...; put_abs_lowcore(abs_lc, flags); To ensure the described mechanism works large segment- and region- table entries must be avoided for the 8KB mappings. Failure to do so results in usage of Region-Frame Absolute Address (RFAA) or Segment-Frame Absolute Address (SFAA) large page fields. In that case absolute addresses would be used to address the prefix page instead of the real ones and the prefixing would get bypassed. Reviewed-by: Heiko Carstens <hca@linux.ibm.com> Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com> Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
2022-07-20 08:22:01 +02:00
s390: Add stackprotector support Stackprotector support was previously unavailable on s390 because by default compilers generate code which is not suitable for the kernel: the canary value is accessed via thread local storage, where the address of thread local storage is within access registers 0 and 1. Using those registers also for the kernel would come with a significant performance impact and more complicated kernel entry/exit code, since access registers contents would have to be exchanged on every kernel entry and exit. With the upcoming gcc 16 release new compiler options will become available which allow to generate code suitable for the kernel. [1] Compiler option -mstack-protector-guard=global instructs gcc to generate stackprotector code that refers to a global stackprotector canary value via symbol __stack_chk_guard. Access to this value is guaranteed to occur via larl and lgrl instructions. Furthermore, compiler option -mstack-protector-guard-record generates a section containing all code addresses that reference the canary value. To allow for per task canary values the instructions which load the address of __stack_chk_guard are patched so they access a lowcore field instead: a per task canary value is available within the task_struct of each task, and is written to the per-cpu lowcore location on each context switch. Also add sanity checks and debugging option to be consistent with other kernel code patching mechanisms. Full debugging output can be enabled with the following kernel command line options: debug_stackprotector bootdebug ignore_loglevel earlyprintk dyndbg="file stackprotector.c +p" Example debug output: stackprot: 0000021e402d4eda: c010005a9ae3 -> c01f00070240 where "<insn address>: <old insn> -> <new insn>". [1] gcc commit 0cd1f03939d5 ("s390: Support global stack protector") Reviewed-by: Sven Schnelle <svens@linux.ibm.com> Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2025-11-17 15:09:53 +01:00
s390/smp: rework absolute lowcore access Temporary unsetting of the prefix page in memcpy_absolute() routine poses a risk of executing code path with unexpectedly disabled prefix page. This rework avoids the prefix page uninstalling and disabling of normal and machine check interrupts when accessing the absolute zero memory. Although memcpy_absolute() routine can access the whole memory, it is only used to update the absolute zero lowcore. This rework therefore introduces a new mechanism for the absolute zero lowcore access and scraps memcpy_absolute() routine for good. Instead, an area is reserved in the virtual memory that is used for the absolute lowcore access only. That area holds an array of 8KB virtual mappings - one per CPU. Whenever a CPU is brought online, the corresponding item is mapped to the real address of the previously installed prefix page. The absolute zero lowcore access works like this: a CPU calls the new primitive get_abs_lowcore() to obtain its 8KB mapping as a pointer to the struct lowcore. Virtual address references to that pointer get translated to the real addresses of the prefix page, which in turn gets swapped with the absolute zero memory addresses due to prefixing. Once the pointer is not needed it must be released with put_abs_lowcore() primitive: struct lowcore *abs_lc; unsigned long flags; abs_lc = get_abs_lowcore(&flags); abs_lc->... = ...; put_abs_lowcore(abs_lc, flags); To ensure the described mechanism works large segment- and region- table entries must be avoided for the 8KB mappings. Failure to do so results in usage of Region-Frame Absolute Address (RFAA) or Segment-Frame Absolute Address (SFAA) large page fields. In that case absolute addresses would be used to address the prefix page instead of the real ones and the prefixing would get bypassed. Reviewed-by: Heiko Carstens <hca@linux.ibm.com> Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com> Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
2022-07-20 08:22:01 +02:00
s390/mm: rework arch_get_mappable_range() callback As per description in mm/memory_hotplug.c platforms should define arch_get_mappable_range() that provides maximum possible addressable physical memory range for which the linear mapping could be created. The current implementation uses VMEM_MAX_PHYS macro as the maximum mappable physical address and it is simply a cast to vmemmap. Since the address is in physical address space the natural upper limit of MAX_PHYSMEM_BITS is honoured: vmemmap_start = min(vmemmap_start, 1UL << MAX_PHYSMEM_BITS); Further, to make sure the identity mapping would not overlay with vmemmap, the size of identity mapping could be stripped like this: ident_map_size = min(ident_map_size, vmemmap_start); Similarily, any other memory that could be added (e.g DCSS segment) should not overlay with vmemmap as well and that is prevented by using vmemmap (VMEM_MAX_PHYS macro) as the upper limit. However, while the use of VMEM_MAX_PHYS brings the desired result it actually poses two issues: 1. As described, vmemmap is handled as a physical address, although it is actually a pointer to struct page in virtual address space. 2. As vmemmap is a virtual address it could have been located anywhere in the virtual address space. However, the desired necessity to honour MAX_PHYSMEM_BITS limit prevents that. Rework arch_get_mappable_range() callback in a way it does not use VMEM_MAX_PHYS macro and does not confuse the notion of virtual vs physical address spacees as result. That paves the way for moving vmemmap elsewhere and optimizing the virtual address space layout. Introduce max_mappable preserved boot variable and let function setup_kernel_memory_layout() set it up. As result, the rest of the code is does not need to know the virtual memory layout specifics. Reviewed-by: Heiko Carstens <hca@linux.ibm.com> Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com> Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2023-07-02 23:11:18 +02:00
s390/boot: Rework deployment of the kernel image Rework deployment of kernel image for both compressed and uncompressed variants as defined by CONFIG_KERNEL_UNCOMPRESSED kernel configuration variable. In case CONFIG_KERNEL_UNCOMPRESSED is disabled avoid uncompressing the kernel to a temporary buffer and copying it to the target address. Instead, uncompress it directly to the target destination. In case CONFIG_KERNEL_UNCOMPRESSED is enabled avoid moving the kernel to default 0x100000 location when KASLR is disabled or failed. Instead, use the uncompressed kernel image directly. In case KASLR is disabled or failed .amode31 section location in memory is not randomized and precedes the kernel image. In case CONFIG_KERNEL_UNCOMPRESSED is disabled that location overlaps the area used by the decompression algorithm. That is fine, since that area is not used after the decompression finished and the size of .amode31 section is not expected to exceed BOOT_HEAP_SIZE ever. There is no decompression in case CONFIG_KERNEL_UNCOMPRESSED is enabled. Therefore, rename decompress_kernel() to deploy_kernel(), which better describes both uncompressed and compressed cases. Introduce AMODE31_SIZE macro to avoid immediate value of 0x3000 (the size of .amode31 section) in the decompressor linker script. Modify the vmlinux linker script to force the size of .amode31 section to AMODE31_SIZE (the value of (_eamode31 - _samode31) could otherwise differ as result of compiler options used). Introduce __START_KERNEL macro that defines the kernel ELF image entry point and set it to the currrent value of 0x100000. Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com>
2024-03-22 14:39:57 +01:00
s390/boot: Rework deployment of the kernel image Rework deployment of kernel image for both compressed and uncompressed variants as defined by CONFIG_KERNEL_UNCOMPRESSED kernel configuration variable. In case CONFIG_KERNEL_UNCOMPRESSED is disabled avoid uncompressing the kernel to a temporary buffer and copying it to the target address. Instead, uncompress it directly to the target destination. In case CONFIG_KERNEL_UNCOMPRESSED is enabled avoid moving the kernel to default 0x100000 location when KASLR is disabled or failed. Instead, use the uncompressed kernel image directly. In case KASLR is disabled or failed .amode31 section location in memory is not randomized and precedes the kernel image. In case CONFIG_KERNEL_UNCOMPRESSED is disabled that location overlaps the area used by the decompression algorithm. That is fine, since that area is not used after the decompression finished and the size of .amode31 section is not expected to exceed BOOT_HEAP_SIZE ever. There is no decompression in case CONFIG_KERNEL_UNCOMPRESSED is enabled. Therefore, rename decompress_kernel() to deploy_kernel(), which better describes both uncompressed and compressed cases. Introduce AMODE31_SIZE macro to avoid immediate value of 0x3000 (the size of .amode31 section) in the decompressor linker script. Modify the vmlinux linker script to force the size of .amode31 section to AMODE31_SIZE (the value of (_eamode31 - _samode31) could otherwise differ as result of compiler options used). Introduce __START_KERNEL macro that defines the kernel ELF image entry point and set it to the currrent value of 0x100000. Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com>
2024-03-22 14:39:57 +01:00
s390/boot: rework decompressor reserved tracking Currently several approaches for finding unused memory in decompressor are utilized. While "safe_addr" grows towards higher addresses, vmem code allocates paging structures top down. The former requires careful ordering. In addition to that ipl report handling code verifies potential intersections with secure boot certificates on its own. Neither of two approaches are memory holes aware and consistent with each other in low memory conditions. To solve that, existing approaches are generalized and combined together, as well as online memory ranges are now taken into consideration. physmem_info has been extended to contain reserved memory ranges. New set of functions allow to handle reserves and find unused memory. All reserves and memory allocations are "typed". In case of out of memory condition decompressor fails with detailed info on current reserved ranges and usable online memory. Linux version 6.2.0 ... Kernel command line: ... mem=100M Our of memory allocating 100000 bytes 100000 aligned in range 0:5800000 Reserved memory ranges: 0000000000000000 0000000003e33000 DECOMPRESSOR 0000000003f00000 00000000057648a3 INITRD 00000000063e0000 00000000063e8000 VMEM 00000000063eb000 00000000063f4000 VMEM 00000000063f7800 0000000006400000 VMEM 0000000005800000 0000000006300000 KASAN Usable online memory ranges (info source: sclp read info [3]): 0000000000000000 0000000006400000 Usable online memory total: 6400000 Reserved: 61b10a3 Free: 24ef5d Call Trace: (sp:000000000002bd58 [<0000000000012a70>] physmem_alloc_top_down+0x60/0x14c) sp:000000000002bdc8 [<0000000000013756>] _pa+0x56/0x6a sp:000000000002bdf0 [<0000000000013bcc>] pgtable_populate+0x45c/0x65e sp:000000000002be90 [<00000000000140aa>] setup_vmem+0x2da/0x424 sp:000000000002bec8 [<0000000000011c20>] startup_kernel+0x428/0x8b4 sp:000000000002bf60 [<00000000000100f4>] startup_normal+0xd4/0xd4 physmem_alloc_range allows to find free memory in specified range. It should be used for one time allocations only like finding position for amode31 and vmlinux. physmem_alloc_top_down can be used just like physmem_alloc_range, but it also allows multiple allocations per type and tries to merge sequential allocations together. Which is useful for paging structures allocations. If sequential allocations cannot be merged together they are "chained", allowing easy per type reserved ranges enumeration and migration to memblock later. Extra "struct reserved_range" allocated for chaining are not tracked or reserved but rely on the fact that both physmem_alloc_range and physmem_alloc_top_down search for free memory only below current top down allocator position. All reserved ranges should be transferred to memblock before memblock allocations are enabled. The startup code has been reordered to delay any memory allocations until online memory ranges are detected and occupied memory ranges are marked as reserved to be excluded from follow-up allocations. Ipl report certificates are a special case, ipl report certificates list is checked together with other memory reserves until certificates are saved elsewhere. KASAN required memory for shadow memory allocation and mapping is reserved as 1 large chunk which is later passed to KASAN early initialization code. Acked-by: Heiko Carstens <hca@linux.ibm.com> Reviewed-by: Alexander Gordeev <agordeev@linux.ibm.com> Signed-off-by: Vasily Gorbik <gor@linux.ibm.com> Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2023-02-02 13:59:36 +01:00
s390/boot: rework decompressor reserved tracking Currently several approaches for finding unused memory in decompressor are utilized. While "safe_addr" grows towards higher addresses, vmem code allocates paging structures top down. The former requires careful ordering. In addition to that ipl report handling code verifies potential intersections with secure boot certificates on its own. Neither of two approaches are memory holes aware and consistent with each other in low memory conditions. To solve that, existing approaches are generalized and combined together, as well as online memory ranges are now taken into consideration. physmem_info has been extended to contain reserved memory ranges. New set of functions allow to handle reserves and find unused memory. All reserves and memory allocations are "typed". In case of out of memory condition decompressor fails with detailed info on current reserved ranges and usable online memory. Linux version 6.2.0 ... Kernel command line: ... mem=100M Our of memory allocating 100000 bytes 100000 aligned in range 0:5800000 Reserved memory ranges: 0000000000000000 0000000003e33000 DECOMPRESSOR 0000000003f00000 00000000057648a3 INITRD 00000000063e0000 00000000063e8000 VMEM 00000000063eb000 00000000063f4000 VMEM 00000000063f7800 0000000006400000 VMEM 0000000005800000 0000000006300000 KASAN Usable online memory ranges (info source: sclp read info [3]): 0000000000000000 0000000006400000 Usable online memory total: 6400000 Reserved: 61b10a3 Free: 24ef5d Call Trace: (sp:000000000002bd58 [<0000000000012a70>] physmem_alloc_top_down+0x60/0x14c) sp:000000000002bdc8 [<0000000000013756>] _pa+0x56/0x6a sp:000000000002bdf0 [<0000000000013bcc>] pgtable_populate+0x45c/0x65e sp:000000000002be90 [<00000000000140aa>] setup_vmem+0x2da/0x424 sp:000000000002bec8 [<0000000000011c20>] startup_kernel+0x428/0x8b4 sp:000000000002bf60 [<00000000000100f4>] startup_normal+0xd4/0xd4 physmem_alloc_range allows to find free memory in specified range. It should be used for one time allocations only like finding position for amode31 and vmlinux. physmem_alloc_top_down can be used just like physmem_alloc_range, but it also allows multiple allocations per type and tries to merge sequential allocations together. Which is useful for paging structures allocations. If sequential allocations cannot be merged together they are "chained", allowing easy per type reserved ranges enumeration and migration to memblock later. Extra "struct reserved_range" allocated for chaining are not tracked or reserved but rely on the fact that both physmem_alloc_range and physmem_alloc_top_down search for free memory only below current top down allocator position. All reserved ranges should be transferred to memblock before memblock allocations are enabled. The startup code has been reordered to delay any memory allocations until online memory ranges are detected and occupied memory ranges are marked as reserved to be excluded from follow-up allocations. Ipl report certificates are a special case, ipl report certificates list is checked together with other memory reserves until certificates are saved elsewhere. KASAN required memory for shadow memory allocation and mapping is reserved as 1 large chunk which is later passed to KASAN early initialization code. Acked-by: Heiko Carstens <hca@linux.ibm.com> Reviewed-by: Alexander Gordeev <agordeev@linux.ibm.com> Signed-off-by: Vasily Gorbik <gor@linux.ibm.com> Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2023-02-02 13:59:36 +01:00
s390/boot: rework decompressor reserved tracking Currently several approaches for finding unused memory in decompressor are utilized. While "safe_addr" grows towards higher addresses, vmem code allocates paging structures top down. The former requires careful ordering. In addition to that ipl report handling code verifies potential intersections with secure boot certificates on its own. Neither of two approaches are memory holes aware and consistent with each other in low memory conditions. To solve that, existing approaches are generalized and combined together, as well as online memory ranges are now taken into consideration. physmem_info has been extended to contain reserved memory ranges. New set of functions allow to handle reserves and find unused memory. All reserves and memory allocations are "typed". In case of out of memory condition decompressor fails with detailed info on current reserved ranges and usable online memory. Linux version 6.2.0 ... Kernel command line: ... mem=100M Our of memory allocating 100000 bytes 100000 aligned in range 0:5800000 Reserved memory ranges: 0000000000000000 0000000003e33000 DECOMPRESSOR 0000000003f00000 00000000057648a3 INITRD 00000000063e0000 00000000063e8000 VMEM 00000000063eb000 00000000063f4000 VMEM 00000000063f7800 0000000006400000 VMEM 0000000005800000 0000000006300000 KASAN Usable online memory ranges (info source: sclp read info [3]): 0000000000000000 0000000006400000 Usable online memory total: 6400000 Reserved: 61b10a3 Free: 24ef5d Call Trace: (sp:000000000002bd58 [<0000000000012a70>] physmem_alloc_top_down+0x60/0x14c) sp:000000000002bdc8 [<0000000000013756>] _pa+0x56/0x6a sp:000000000002bdf0 [<0000000000013bcc>] pgtable_populate+0x45c/0x65e sp:000000000002be90 [<00000000000140aa>] setup_vmem+0x2da/0x424 sp:000000000002bec8 [<0000000000011c20>] startup_kernel+0x428/0x8b4 sp:000000000002bf60 [<00000000000100f4>] startup_normal+0xd4/0xd4 physmem_alloc_range allows to find free memory in specified range. It should be used for one time allocations only like finding position for amode31 and vmlinux. physmem_alloc_top_down can be used just like physmem_alloc_range, but it also allows multiple allocations per type and tries to merge sequential allocations together. Which is useful for paging structures allocations. If sequential allocations cannot be merged together they are "chained", allowing easy per type reserved ranges enumeration and migration to memblock later. Extra "struct reserved_range" allocated for chaining are not tracked or reserved but rely on the fact that both physmem_alloc_range and physmem_alloc_top_down search for free memory only below current top down allocator position. All reserved ranges should be transferred to memblock before memblock allocations are enabled. The startup code has been reordered to delay any memory allocations until online memory ranges are detected and occupied memory ranges are marked as reserved to be excluded from follow-up allocations. Ipl report certificates are a special case, ipl report certificates list is checked together with other memory reserves until certificates are saved elsewhere. KASAN required memory for shadow memory allocation and mapping is reserved as 1 large chunk which is later passed to KASAN early initialization code. Acked-by: Heiko Carstens <hca@linux.ibm.com> Reviewed-by: Alexander Gordeev <agordeev@linux.ibm.com> Signed-off-by: Vasily Gorbik <gor@linux.ibm.com> Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2023-02-02 13:59:36 +01:00
s390/boot: rework decompressor reserved tracking Currently several approaches for finding unused memory in decompressor are utilized. While "safe_addr" grows towards higher addresses, vmem code allocates paging structures top down. The former requires careful ordering. In addition to that ipl report handling code verifies potential intersections with secure boot certificates on its own. Neither of two approaches are memory holes aware and consistent with each other in low memory conditions. To solve that, existing approaches are generalized and combined together, as well as online memory ranges are now taken into consideration. physmem_info has been extended to contain reserved memory ranges. New set of functions allow to handle reserves and find unused memory. All reserves and memory allocations are "typed". In case of out of memory condition decompressor fails with detailed info on current reserved ranges and usable online memory. Linux version 6.2.0 ... Kernel command line: ... mem=100M Our of memory allocating 100000 bytes 100000 aligned in range 0:5800000 Reserved memory ranges: 0000000000000000 0000000003e33000 DECOMPRESSOR 0000000003f00000 00000000057648a3 INITRD 00000000063e0000 00000000063e8000 VMEM 00000000063eb000 00000000063f4000 VMEM 00000000063f7800 0000000006400000 VMEM 0000000005800000 0000000006300000 KASAN Usable online memory ranges (info source: sclp read info [3]): 0000000000000000 0000000006400000 Usable online memory total: 6400000 Reserved: 61b10a3 Free: 24ef5d Call Trace: (sp:000000000002bd58 [<0000000000012a70>] physmem_alloc_top_down+0x60/0x14c) sp:000000000002bdc8 [<0000000000013756>] _pa+0x56/0x6a sp:000000000002bdf0 [<0000000000013bcc>] pgtable_populate+0x45c/0x65e sp:000000000002be90 [<00000000000140aa>] setup_vmem+0x2da/0x424 sp:000000000002bec8 [<0000000000011c20>] startup_kernel+0x428/0x8b4 sp:000000000002bf60 [<00000000000100f4>] startup_normal+0xd4/0xd4 physmem_alloc_range allows to find free memory in specified range. It should be used for one time allocations only like finding position for amode31 and vmlinux. physmem_alloc_top_down can be used just like physmem_alloc_range, but it also allows multiple allocations per type and tries to merge sequential allocations together. Which is useful for paging structures allocations. If sequential allocations cannot be merged together they are "chained", allowing easy per type reserved ranges enumeration and migration to memblock later. Extra "struct reserved_range" allocated for chaining are not tracked or reserved but rely on the fact that both physmem_alloc_range and physmem_alloc_top_down search for free memory only below current top down allocator position. All reserved ranges should be transferred to memblock before memblock allocations are enabled. The startup code has been reordered to delay any memory allocations until online memory ranges are detected and occupied memory ranges are marked as reserved to be excluded from follow-up allocations. Ipl report certificates are a special case, ipl report certificates list is checked together with other memory reserves until certificates are saved elsewhere. KASAN required memory for shadow memory allocation and mapping is reserved as 1 large chunk which is later passed to KASAN early initialization code. Acked-by: Heiko Carstens <hca@linux.ibm.com> Reviewed-by: Alexander Gordeev <agordeev@linux.ibm.com> Signed-off-by: Vasily Gorbik <gor@linux.ibm.com> Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2023-02-02 13:59:36 +01:00
s390: compile relocatable kernel without -fPIE On s390, currently kernel uses the '-fPIE' compiler flag for compiling vmlinux. This has a few problems: - It uses dynamic symbols (.dynsym), for which the linker refuses to allow more than 64k sections. This can break features which use '-ffunction-sections' and '-fdata-sections', including kpatch-build [1] and Function Granular KASLR. - It unnecessarily uses GOT relocations, adding an extra layer of indirection for many memory accesses. Instead of using '-fPIE', resolve all the relocations at link time and then manually adjust any absolute relocations (R_390_64) during boot. This is done by first telling the linker to preserve all relocations during the vmlinux link. (Note this is harmless: they are later stripped in the vmlinux.bin link.) Then use the 'relocs' tool to find all absolute relocations (R_390_64) which apply to allocatable sections. The offsets of those relocations are saved in a special section which is then used to adjust the relocations during boot. (Note: For some reason, Clang occasionally creates a GOT reference, even without '-fPIE'. So Clang-compiled kernels have a GOT, which needs to be adjusted.) On my mostly-defconfig kernel, this reduces kernel text size by ~1.3%. [1] https://github.com/dynup/kpatch/issues/1284 [2] https://gcc.gnu.org/pipermail/gcc-patches/2023-June/622872.html [3] https://gcc.gnu.org/pipermail/gcc-patches/2023-August/625986.html Compiler consideration: Gcc recently implemented an optimization [2] for loading symbols without explicit alignment, aligning with the IBM Z ELF ABI. This ABI mandates symbols to reside on a 2-byte boundary, enabling the use of the larl instruction. However, kernel linker scripts may still generate unaligned symbols. To address this, a new -munaligned-symbols option has been introduced [3] in recent gcc versions. This option has to be used with future gcc versions. Older Clang lacks support for handling unaligned symbols generated by kernel linker scripts when the kernel is built without -fPIE. However, future versions of Clang will include support for the -munaligned-symbols option. When the support is unavailable, compile the kernel with -fPIE to maintain the existing behavior. In addition to it: move vmlinux.relocs to safe relocation When the kernel is built with CONFIG_KERNEL_UNCOMPRESSED, the entire uncompressed vmlinux.bin is positioned in the bzImage decompressor image at the default kernel LMA of 0x100000, enabling it to be executed in-place. However, the size of .vmlinux.relocs could be large enough to cause an overlap with the uncompressed kernel at the address 0x100000. To address this issue, .vmlinux.relocs is positioned after the .rodata.compressed in the bzImage. Nevertheless, in this configuration, vmlinux.relocs will overlap with the .bss section of vmlinux.bin. To overcome that, move vmlinux.relocs to a safe location before clearing .bss and handling relocs. Compile warning fix from Sumanth Korikkar: When kernel is built with CONFIG_LD_ORPHAN_WARN and -fno-PIE, there are several warnings: ld: warning: orphan section `.rela.iplt' from `arch/s390/kernel/head64.o' being placed in section `.rela.dyn' ld: warning: orphan section `.rela.head.text' from `arch/s390/kernel/head64.o' being placed in section `.rela.dyn' ld: warning: orphan section `.rela.init.text' from `arch/s390/kernel/head64.o' being placed in section `.rela.dyn' ld: warning: orphan section `.rela.rodata.cst8' from `arch/s390/kernel/head64.o' being placed in section `.rela.dyn' Orphan sections are sections that exist in an object file but don't have a corresponding output section in the final executable. ld raises a warning when it identifies such sections. Eliminate the warning by placing all .rela orphan sections in .rela.dyn and raise an error when size of .rela.dyn is greater than zero. i.e. Dont just neglect orphan sections. This is similar to adjustment performed in x86, where kernel is built with -fno-PIE. commit 5354e84598f2 ("x86/build: Add asserts for unwanted sections") [sumanthk@linux.ibm.com: rebased Josh Poimboeuf patches and move vmlinux.relocs to safe location] [hca@linux.ibm.com: merged compile warning fix from Sumanth] Tested-by: Sumanth Korikkar <sumanthk@linux.ibm.com> Acked-by: Vasily Gorbik <gor@linux.ibm.com> Signed-off-by: Josh Poimboeuf <jpoimboe@kernel.org> Signed-off-by: Sumanth Korikkar <sumanthk@linux.ibm.com> Link: https://lore.kernel.org/r/20240219132734.22881-4-sumanthk@linux.ibm.com Link: https://lore.kernel.org/r/20240219132734.22881-5-sumanthk@linux.ibm.com Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2024-02-19 14:27:33 +01:00
s390/mm: Uncouple physical vs virtual address spaces The uncoupling physical vs virtual address spaces brings the following benefits to s390: - virtual memory layout flexibility; - closes the address gap between kernel and modules, it caused s390-only problems in the past (e.g. 'perf' bugs); - allows getting rid of trampolines used for module calls into kernel; - allows simplifying BPF trampoline; - minor performance improvement in branch prediction; - kernel randomization entropy is magnitude bigger, as it is derived from the amount of available virtual, not physical memory; The whole change could be described in two pictures below: before and after the change. Some aspects of the virtual memory layout setup are not clarified (number of page levels, alignment, DMA memory), since these are not a part of this change or secondary with regard to how the uncoupling itself is implemented. The focus of the pictures is to explain why __va() and __pa() macros are implemented the way they are. Memory layout in V==R mode: | Physical | Virtual | +- 0 --------------+- 0 --------------+ identity mapping start | | S390_lowcore | Low-address memory | +- 8 KB -----------+ | | | | | identity | phys == virt | | mapping | virt == phys | | | +- AMODE31_START --+- AMODE31_START --+ .amode31 rand. phys/virt start |.amode31 text/data|.amode31 text/data| +- AMODE31_END ----+- AMODE31_END ----+ .amode31 rand. phys/virt start | | | | | | +- __kaslr_offset, __kaslr_offset_phys| kernel rand. phys/virt start | | | | kernel text/data | kernel text/data | phys == kvirt | | | +------------------+------------------+ kernel phys/virt end | | | | | | | | | | | | +- ident_map_size -+- ident_map_size -+ identity mapping end | | | ... unused gap | | | +---- vmemmap -----+ 'struct page' array start | | | virtually mapped | | memory map | | | +- __abs_lowcore --+ | | | Absolute Lowcore | | | +- __memcpy_real_area | | | Real Memory Copy| | | +- VMALLOC_START --+ vmalloc area start | | | vmalloc area | | | +- MODULES_VADDR --+ modules area start | | | modules area | | | +------------------+ UltraVisor Secure Storage limit | | | ... unused gap | | | +KASAN_SHADOW_START+ KASAN shadow memory start | | | KASAN shadow | | | +------------------+ ASCE limit Memory layout in V!=R mode: | Physical | Virtual | +- 0 --------------+- 0 --------------+ | | S390_lowcore | Low-address memory | +- 8 KB -----------+ | | | | | | | | ... unused gap | | | | +- AMODE31_START --+- AMODE31_START --+ .amode31 rand. phys/virt start |.amode31 text/data|.amode31 text/data| +- AMODE31_END ----+- AMODE31_END ----+ .amode31 rand. phys/virt end (<2GB) | | | | | | +- __kaslr_offset_phys | kernel rand. phys start | | | | kernel text/data | | | | | +------------------+ | kernel phys end | | | | | | | | | | | | +- ident_map_size -+ | | | | ... unused gap | | | +- __identity_base + identity mapping start (>= 2GB) | | | identity | phys == virt - __identity_base | mapping | virt == phys + __identity_base | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | +---- vmemmap -----+ 'struct page' array start | | | virtually mapped | | memory map | | | +- __abs_lowcore --+ | | | Absolute Lowcore | | | +- __memcpy_real_area | | | Real Memory Copy| | | +- VMALLOC_START --+ vmalloc area start | | | vmalloc area | | | +- MODULES_VADDR --+ modules area start | | | modules area | | | +- __kaslr_offset -+ kernel rand. virt start | | | kernel text/data | phys == (kvirt - __kaslr_offset) + | | __kaslr_offset_phys +- kernel .bss end + kernel rand. virt end | | | ... unused gap | | | +------------------+ UltraVisor Secure Storage limit | | | ... unused gap | | | +KASAN_SHADOW_START+ KASAN shadow memory start | | | KASAN shadow | | | +------------------+ ASCE limit Unused gaps in the virtual memory layout could be present or not - depending on how partucular system is configured. No page tables are created for the unused gaps. The relative order of vmalloc, modules and kernel image in virtual memory is defined by following considerations: - start of the modules area and end of the kernel should reside within 4GB to accommodate relative 32-bit jumps. The best way to achieve that is to place kernel next to modules; - vmalloc and module areas should locate next to each other to prevent failures and extra reworks in user level tools (makedumpfile, crash, etc.) which treat vmalloc and module addresses similarily; - kernel needs to be the last area in the virtual memory layout to easily distinguish between kernel and non-kernel virtual addresses. That is needed to (again) simplify handling of addresses in user level tools and make __pa() macro faster (see below); Concluding the above, the relative order of the considered virtual areas in memory is: vmalloc - modules - kernel. Therefore, the only change to the current memory layout is moving kernel to the end of virtual address space. With that approach the implementation of __pa() macro is straightforward - all linear virtual addresses less than kernel base are considered identity mapping: phys == virt - __identity_base All addresses greater than kernel base are kernel ones: phys == (kvirt - __kaslr_offset) + __kaslr_offset_phys By contrast, __va() macro deals only with identity mapping addresses: virt == phys + __identity_base .amode31 section is mapped separately and is not covered by __pa() macro. In fact, it could have been handled easily by checking whether a virtual address is within the section or not, but there is no need for that. Thus, let __pa() code do as little machine cycles as possible. The KASAN shadow memory is located at the very end of the virtual memory layout, at addresses higher than the kernel. However, that is not a linear mapping and no code other than KASAN instrumentation or API is expected to access it. When KASLR mode is enabled the kernel base address randomized within a memory window that spans whole unused virtual address space. The size of that window depends from the amount of physical memory available to the system, the limit imposed by UltraVisor (if present) and the vmalloc area size as provided by vmalloc= kernel command line parameter. In case the virtual memory is exhausted the minimum size of the randomization window is forcefully set to 2GB, which amounts to in 15 bits of entropy if KASAN is enabled or 17 bits of entropy in default configuration. The default kernel offset 0x100000 is used as a magic value both in the decompressor code and vmlinux linker script, but it will be removed with a follow-up change. Acked-by: Heiko Carstens <hca@linux.ibm.com> Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com>
2024-03-01 07:15:22 +01:00
s390: compile relocatable kernel without -fPIE On s390, currently kernel uses the '-fPIE' compiler flag for compiling vmlinux. This has a few problems: - It uses dynamic symbols (.dynsym), for which the linker refuses to allow more than 64k sections. This can break features which use '-ffunction-sections' and '-fdata-sections', including kpatch-build [1] and Function Granular KASLR. - It unnecessarily uses GOT relocations, adding an extra layer of indirection for many memory accesses. Instead of using '-fPIE', resolve all the relocations at link time and then manually adjust any absolute relocations (R_390_64) during boot. This is done by first telling the linker to preserve all relocations during the vmlinux link. (Note this is harmless: they are later stripped in the vmlinux.bin link.) Then use the 'relocs' tool to find all absolute relocations (R_390_64) which apply to allocatable sections. The offsets of those relocations are saved in a special section which is then used to adjust the relocations during boot. (Note: For some reason, Clang occasionally creates a GOT reference, even without '-fPIE'. So Clang-compiled kernels have a GOT, which needs to be adjusted.) On my mostly-defconfig kernel, this reduces kernel text size by ~1.3%. [1] https://github.com/dynup/kpatch/issues/1284 [2] https://gcc.gnu.org/pipermail/gcc-patches/2023-June/622872.html [3] https://gcc.gnu.org/pipermail/gcc-patches/2023-August/625986.html Compiler consideration: Gcc recently implemented an optimization [2] for loading symbols without explicit alignment, aligning with the IBM Z ELF ABI. This ABI mandates symbols to reside on a 2-byte boundary, enabling the use of the larl instruction. However, kernel linker scripts may still generate unaligned symbols. To address this, a new -munaligned-symbols option has been introduced [3] in recent gcc versions. This option has to be used with future gcc versions. Older Clang lacks support for handling unaligned symbols generated by kernel linker scripts when the kernel is built without -fPIE. However, future versions of Clang will include support for the -munaligned-symbols option. When the support is unavailable, compile the kernel with -fPIE to maintain the existing behavior. In addition to it: move vmlinux.relocs to safe relocation When the kernel is built with CONFIG_KERNEL_UNCOMPRESSED, the entire uncompressed vmlinux.bin is positioned in the bzImage decompressor image at the default kernel LMA of 0x100000, enabling it to be executed in-place. However, the size of .vmlinux.relocs could be large enough to cause an overlap with the uncompressed kernel at the address 0x100000. To address this issue, .vmlinux.relocs is positioned after the .rodata.compressed in the bzImage. Nevertheless, in this configuration, vmlinux.relocs will overlap with the .bss section of vmlinux.bin. To overcome that, move vmlinux.relocs to a safe location before clearing .bss and handling relocs. Compile warning fix from Sumanth Korikkar: When kernel is built with CONFIG_LD_ORPHAN_WARN and -fno-PIE, there are several warnings: ld: warning: orphan section `.rela.iplt' from `arch/s390/kernel/head64.o' being placed in section `.rela.dyn' ld: warning: orphan section `.rela.head.text' from `arch/s390/kernel/head64.o' being placed in section `.rela.dyn' ld: warning: orphan section `.rela.init.text' from `arch/s390/kernel/head64.o' being placed in section `.rela.dyn' ld: warning: orphan section `.rela.rodata.cst8' from `arch/s390/kernel/head64.o' being placed in section `.rela.dyn' Orphan sections are sections that exist in an object file but don't have a corresponding output section in the final executable. ld raises a warning when it identifies such sections. Eliminate the warning by placing all .rela orphan sections in .rela.dyn and raise an error when size of .rela.dyn is greater than zero. i.e. Dont just neglect orphan sections. This is similar to adjustment performed in x86, where kernel is built with -fno-PIE. commit 5354e84598f2 ("x86/build: Add asserts for unwanted sections") [sumanthk@linux.ibm.com: rebased Josh Poimboeuf patches and move vmlinux.relocs to safe location] [hca@linux.ibm.com: merged compile warning fix from Sumanth] Tested-by: Sumanth Korikkar <sumanthk@linux.ibm.com> Acked-by: Vasily Gorbik <gor@linux.ibm.com> Signed-off-by: Josh Poimboeuf <jpoimboe@kernel.org> Signed-off-by: Sumanth Korikkar <sumanthk@linux.ibm.com> Link: https://lore.kernel.org/r/20240219132734.22881-4-sumanthk@linux.ibm.com Link: https://lore.kernel.org/r/20240219132734.22881-5-sumanthk@linux.ibm.com Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2024-02-19 14:27:33 +01:00
s390: compile relocatable kernel without -fPIE On s390, currently kernel uses the '-fPIE' compiler flag for compiling vmlinux. This has a few problems: - It uses dynamic symbols (.dynsym), for which the linker refuses to allow more than 64k sections. This can break features which use '-ffunction-sections' and '-fdata-sections', including kpatch-build [1] and Function Granular KASLR. - It unnecessarily uses GOT relocations, adding an extra layer of indirection for many memory accesses. Instead of using '-fPIE', resolve all the relocations at link time and then manually adjust any absolute relocations (R_390_64) during boot. This is done by first telling the linker to preserve all relocations during the vmlinux link. (Note this is harmless: they are later stripped in the vmlinux.bin link.) Then use the 'relocs' tool to find all absolute relocations (R_390_64) which apply to allocatable sections. The offsets of those relocations are saved in a special section which is then used to adjust the relocations during boot. (Note: For some reason, Clang occasionally creates a GOT reference, even without '-fPIE'. So Clang-compiled kernels have a GOT, which needs to be adjusted.) On my mostly-defconfig kernel, this reduces kernel text size by ~1.3%. [1] https://github.com/dynup/kpatch/issues/1284 [2] https://gcc.gnu.org/pipermail/gcc-patches/2023-June/622872.html [3] https://gcc.gnu.org/pipermail/gcc-patches/2023-August/625986.html Compiler consideration: Gcc recently implemented an optimization [2] for loading symbols without explicit alignment, aligning with the IBM Z ELF ABI. This ABI mandates symbols to reside on a 2-byte boundary, enabling the use of the larl instruction. However, kernel linker scripts may still generate unaligned symbols. To address this, a new -munaligned-symbols option has been introduced [3] in recent gcc versions. This option has to be used with future gcc versions. Older Clang lacks support for handling unaligned symbols generated by kernel linker scripts when the kernel is built without -fPIE. However, future versions of Clang will include support for the -munaligned-symbols option. When the support is unavailable, compile the kernel with -fPIE to maintain the existing behavior. In addition to it: move vmlinux.relocs to safe relocation When the kernel is built with CONFIG_KERNEL_UNCOMPRESSED, the entire uncompressed vmlinux.bin is positioned in the bzImage decompressor image at the default kernel LMA of 0x100000, enabling it to be executed in-place. However, the size of .vmlinux.relocs could be large enough to cause an overlap with the uncompressed kernel at the address 0x100000. To address this issue, .vmlinux.relocs is positioned after the .rodata.compressed in the bzImage. Nevertheless, in this configuration, vmlinux.relocs will overlap with the .bss section of vmlinux.bin. To overcome that, move vmlinux.relocs to a safe location before clearing .bss and handling relocs. Compile warning fix from Sumanth Korikkar: When kernel is built with CONFIG_LD_ORPHAN_WARN and -fno-PIE, there are several warnings: ld: warning: orphan section `.rela.iplt' from `arch/s390/kernel/head64.o' being placed in section `.rela.dyn' ld: warning: orphan section `.rela.head.text' from `arch/s390/kernel/head64.o' being placed in section `.rela.dyn' ld: warning: orphan section `.rela.init.text' from `arch/s390/kernel/head64.o' being placed in section `.rela.dyn' ld: warning: orphan section `.rela.rodata.cst8' from `arch/s390/kernel/head64.o' being placed in section `.rela.dyn' Orphan sections are sections that exist in an object file but don't have a corresponding output section in the final executable. ld raises a warning when it identifies such sections. Eliminate the warning by placing all .rela orphan sections in .rela.dyn and raise an error when size of .rela.dyn is greater than zero. i.e. Dont just neglect orphan sections. This is similar to adjustment performed in x86, where kernel is built with -fno-PIE. commit 5354e84598f2 ("x86/build: Add asserts for unwanted sections") [sumanthk@linux.ibm.com: rebased Josh Poimboeuf patches and move vmlinux.relocs to safe location] [hca@linux.ibm.com: merged compile warning fix from Sumanth] Tested-by: Sumanth Korikkar <sumanthk@linux.ibm.com> Acked-by: Vasily Gorbik <gor@linux.ibm.com> Signed-off-by: Josh Poimboeuf <jpoimboe@kernel.org> Signed-off-by: Sumanth Korikkar <sumanthk@linux.ibm.com> Link: https://lore.kernel.org/r/20240219132734.22881-4-sumanthk@linux.ibm.com Link: https://lore.kernel.org/r/20240219132734.22881-5-sumanthk@linux.ibm.com Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2024-02-19 14:27:33 +01:00
s390/boot: Do not adjust GOT entries for undef weak sym Since commit 778666df60f0 ("s390: compile relocatable kernel without -fPIE") and commit 00cda11d3b2e ("s390: Compile kernel with -fPIC and link with -no-pie") the kernel on s390x may have a Global Offset Table (GOT) whose entries are adjusted for KASLR in kaslr_adjust_got(). The GOT may contain entries for undefined weak symbols that resolved to zero. That is the resulting GOT entry value is zero. Adjusting those entries unconditionally in kaslr_adjust_got() is wrong. Otherwise the following sample code would erroneously assume foo to be defined, due to the adjustment changing the zero-value to a non-zero one: extern int foo __attribute__((weak)); if (*foo) /* foo is defined [or undefined and erroneously adjusted] */ The vmlinux build at commit 00cda11d3b2e ("s390: Compile kernel with -fPIC and link with -no-pie") with defconfig actually had two GOT entries for the undefined weak symbols __start_BTF and __stop_BTF: $ objdump -tw vmlinux | grep -F "*UND*" 0000000000000000 w *UND* 0000000000000000 __stop_BTF 0000000000000000 w *UND* 0000000000000000 __start_BTF $ readelf -rw vmlinux | grep -E "R_390_GOTENT +0{16}" 000000345760 2776a0000001a R_390_GOTENT 0000000000000000 __stop_BTF + 2 000000345766 2d5480000001a R_390_GOTENT 0000000000000000 __start_BTF + 2 The s390-specific vmlinux linker script sets the section start to __START_KERNEL, which is currently defined as 0x100000 on s390x. Access to lowcore is performed via a pointer of 0 and not a symbol in a section starting at 0. The first 64K are reserved for the loader on s390x. Thus it is safe to assume that __START_KERNEL will never be 0. As a result there cannot be any defined symbols resolving to zero in the kernel. Note that the first three GOT entries are reserved for the dynamic loader on s390x. [1] In the kernel they are zero. Therefore no extra handling is required to skip these. Skip adjusting GOT entries with a value of zero in kaslr_adjust_got(). While at it update the comment when a GOT exists on s390x. Since commit 00cda11d3b2e ("s390: Compile kernel with -fPIC and link with -no-pie") it no longer only exists when compiling with Clang, but also with GCC. [1]: s390x ELF ABI, section "Global Offset Table", https://github.com/IBM/s390x-abi/releases Fixes: 778666df60f0 ("s390: compile relocatable kernel without -fPIE") Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com> Acked-by: Sumanth Korikkar <sumanthk@linux.ibm.com> Acked-by: Alexander Gordeev <agordeev@linux.ibm.com> Signed-off-by: Jens Remus <jremus@linux.ibm.com> Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com>
2024-06-20 11:12:49 +02:00
s390: compile relocatable kernel without -fPIE On s390, currently kernel uses the '-fPIE' compiler flag for compiling vmlinux. This has a few problems: - It uses dynamic symbols (.dynsym), for which the linker refuses to allow more than 64k sections. This can break features which use '-ffunction-sections' and '-fdata-sections', including kpatch-build [1] and Function Granular KASLR. - It unnecessarily uses GOT relocations, adding an extra layer of indirection for many memory accesses. Instead of using '-fPIE', resolve all the relocations at link time and then manually adjust any absolute relocations (R_390_64) during boot. This is done by first telling the linker to preserve all relocations during the vmlinux link. (Note this is harmless: they are later stripped in the vmlinux.bin link.) Then use the 'relocs' tool to find all absolute relocations (R_390_64) which apply to allocatable sections. The offsets of those relocations are saved in a special section which is then used to adjust the relocations during boot. (Note: For some reason, Clang occasionally creates a GOT reference, even without '-fPIE'. So Clang-compiled kernels have a GOT, which needs to be adjusted.) On my mostly-defconfig kernel, this reduces kernel text size by ~1.3%. [1] https://github.com/dynup/kpatch/issues/1284 [2] https://gcc.gnu.org/pipermail/gcc-patches/2023-June/622872.html [3] https://gcc.gnu.org/pipermail/gcc-patches/2023-August/625986.html Compiler consideration: Gcc recently implemented an optimization [2] for loading symbols without explicit alignment, aligning with the IBM Z ELF ABI. This ABI mandates symbols to reside on a 2-byte boundary, enabling the use of the larl instruction. However, kernel linker scripts may still generate unaligned symbols. To address this, a new -munaligned-symbols option has been introduced [3] in recent gcc versions. This option has to be used with future gcc versions. Older Clang lacks support for handling unaligned symbols generated by kernel linker scripts when the kernel is built without -fPIE. However, future versions of Clang will include support for the -munaligned-symbols option. When the support is unavailable, compile the kernel with -fPIE to maintain the existing behavior. In addition to it: move vmlinux.relocs to safe relocation When the kernel is built with CONFIG_KERNEL_UNCOMPRESSED, the entire uncompressed vmlinux.bin is positioned in the bzImage decompressor image at the default kernel LMA of 0x100000, enabling it to be executed in-place. However, the size of .vmlinux.relocs could be large enough to cause an overlap with the uncompressed kernel at the address 0x100000. To address this issue, .vmlinux.relocs is positioned after the .rodata.compressed in the bzImage. Nevertheless, in this configuration, vmlinux.relocs will overlap with the .bss section of vmlinux.bin. To overcome that, move vmlinux.relocs to a safe location before clearing .bss and handling relocs. Compile warning fix from Sumanth Korikkar: When kernel is built with CONFIG_LD_ORPHAN_WARN and -fno-PIE, there are several warnings: ld: warning: orphan section `.rela.iplt' from `arch/s390/kernel/head64.o' being placed in section `.rela.dyn' ld: warning: orphan section `.rela.head.text' from `arch/s390/kernel/head64.o' being placed in section `.rela.dyn' ld: warning: orphan section `.rela.init.text' from `arch/s390/kernel/head64.o' being placed in section `.rela.dyn' ld: warning: orphan section `.rela.rodata.cst8' from `arch/s390/kernel/head64.o' being placed in section `.rela.dyn' Orphan sections are sections that exist in an object file but don't have a corresponding output section in the final executable. ld raises a warning when it identifies such sections. Eliminate the warning by placing all .rela orphan sections in .rela.dyn and raise an error when size of .rela.dyn is greater than zero. i.e. Dont just neglect orphan sections. This is similar to adjustment performed in x86, where kernel is built with -fno-PIE. commit 5354e84598f2 ("x86/build: Add asserts for unwanted sections") [sumanthk@linux.ibm.com: rebased Josh Poimboeuf patches and move vmlinux.relocs to safe location] [hca@linux.ibm.com: merged compile warning fix from Sumanth] Tested-by: Sumanth Korikkar <sumanthk@linux.ibm.com> Acked-by: Vasily Gorbik <gor@linux.ibm.com> Signed-off-by: Josh Poimboeuf <jpoimboe@kernel.org> Signed-off-by: Sumanth Korikkar <sumanthk@linux.ibm.com> Link: https://lore.kernel.org/r/20240219132734.22881-4-sumanthk@linux.ibm.com Link: https://lore.kernel.org/r/20240219132734.22881-5-sumanthk@linux.ibm.com Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2024-02-19 14:27:33 +01:00
s390/boot: Do not adjust GOT entries for undef weak sym Since commit 778666df60f0 ("s390: compile relocatable kernel without -fPIE") and commit 00cda11d3b2e ("s390: Compile kernel with -fPIC and link with -no-pie") the kernel on s390x may have a Global Offset Table (GOT) whose entries are adjusted for KASLR in kaslr_adjust_got(). The GOT may contain entries for undefined weak symbols that resolved to zero. That is the resulting GOT entry value is zero. Adjusting those entries unconditionally in kaslr_adjust_got() is wrong. Otherwise the following sample code would erroneously assume foo to be defined, due to the adjustment changing the zero-value to a non-zero one: extern int foo __attribute__((weak)); if (*foo) /* foo is defined [or undefined and erroneously adjusted] */ The vmlinux build at commit 00cda11d3b2e ("s390: Compile kernel with -fPIC and link with -no-pie") with defconfig actually had two GOT entries for the undefined weak symbols __start_BTF and __stop_BTF: $ objdump -tw vmlinux | grep -F "*UND*" 0000000000000000 w *UND* 0000000000000000 __stop_BTF 0000000000000000 w *UND* 0000000000000000 __start_BTF $ readelf -rw vmlinux | grep -E "R_390_GOTENT +0{16}" 000000345760 2776a0000001a R_390_GOTENT 0000000000000000 __stop_BTF + 2 000000345766 2d5480000001a R_390_GOTENT 0000000000000000 __start_BTF + 2 The s390-specific vmlinux linker script sets the section start to __START_KERNEL, which is currently defined as 0x100000 on s390x. Access to lowcore is performed via a pointer of 0 and not a symbol in a section starting at 0. The first 64K are reserved for the loader on s390x. Thus it is safe to assume that __START_KERNEL will never be 0. As a result there cannot be any defined symbols resolving to zero in the kernel. Note that the first three GOT entries are reserved for the dynamic loader on s390x. [1] In the kernel they are zero. Therefore no extra handling is required to skip these. Skip adjusting GOT entries with a value of zero in kaslr_adjust_got(). While at it update the comment when a GOT exists on s390x. Since commit 00cda11d3b2e ("s390: Compile kernel with -fPIC and link with -no-pie") it no longer only exists when compiling with Clang, but also with GCC. [1]: s390x ELF ABI, section "Global Offset Table", https://github.com/IBM/s390x-abi/releases Fixes: 778666df60f0 ("s390: compile relocatable kernel without -fPIE") Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com> Acked-by: Sumanth Korikkar <sumanthk@linux.ibm.com> Acked-by: Alexander Gordeev <agordeev@linux.ibm.com> Signed-off-by: Jens Remus <jremus@linux.ibm.com> Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com>
2024-06-20 11:12:49 +02:00
s390/boot: Do not adjust GOT entries for undef weak sym Since commit 778666df60f0 ("s390: compile relocatable kernel without -fPIE") and commit 00cda11d3b2e ("s390: Compile kernel with -fPIC and link with -no-pie") the kernel on s390x may have a Global Offset Table (GOT) whose entries are adjusted for KASLR in kaslr_adjust_got(). The GOT may contain entries for undefined weak symbols that resolved to zero. That is the resulting GOT entry value is zero. Adjusting those entries unconditionally in kaslr_adjust_got() is wrong. Otherwise the following sample code would erroneously assume foo to be defined, due to the adjustment changing the zero-value to a non-zero one: extern int foo __attribute__((weak)); if (*foo) /* foo is defined [or undefined and erroneously adjusted] */ The vmlinux build at commit 00cda11d3b2e ("s390: Compile kernel with -fPIC and link with -no-pie") with defconfig actually had two GOT entries for the undefined weak symbols __start_BTF and __stop_BTF: $ objdump -tw vmlinux | grep -F "*UND*" 0000000000000000 w *UND* 0000000000000000 __stop_BTF 0000000000000000 w *UND* 0000000000000000 __start_BTF $ readelf -rw vmlinux | grep -E "R_390_GOTENT +0{16}" 000000345760 2776a0000001a R_390_GOTENT 0000000000000000 __stop_BTF + 2 000000345766 2d5480000001a R_390_GOTENT 0000000000000000 __start_BTF + 2 The s390-specific vmlinux linker script sets the section start to __START_KERNEL, which is currently defined as 0x100000 on s390x. Access to lowcore is performed via a pointer of 0 and not a symbol in a section starting at 0. The first 64K are reserved for the loader on s390x. Thus it is safe to assume that __START_KERNEL will never be 0. As a result there cannot be any defined symbols resolving to zero in the kernel. Note that the first three GOT entries are reserved for the dynamic loader on s390x. [1] In the kernel they are zero. Therefore no extra handling is required to skip these. Skip adjusting GOT entries with a value of zero in kaslr_adjust_got(). While at it update the comment when a GOT exists on s390x. Since commit 00cda11d3b2e ("s390: Compile kernel with -fPIC and link with -no-pie") it no longer only exists when compiling with Clang, but also with GCC. [1]: s390x ELF ABI, section "Global Offset Table", https://github.com/IBM/s390x-abi/releases Fixes: 778666df60f0 ("s390: compile relocatable kernel without -fPIE") Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com> Acked-by: Sumanth Korikkar <sumanthk@linux.ibm.com> Acked-by: Alexander Gordeev <agordeev@linux.ibm.com> Signed-off-by: Jens Remus <jremus@linux.ibm.com> Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com>
2024-06-20 11:12:49 +02:00
s390: compile relocatable kernel without -fPIE On s390, currently kernel uses the '-fPIE' compiler flag for compiling vmlinux. This has a few problems: - It uses dynamic symbols (.dynsym), for which the linker refuses to allow more than 64k sections. This can break features which use '-ffunction-sections' and '-fdata-sections', including kpatch-build [1] and Function Granular KASLR. - It unnecessarily uses GOT relocations, adding an extra layer of indirection for many memory accesses. Instead of using '-fPIE', resolve all the relocations at link time and then manually adjust any absolute relocations (R_390_64) during boot. This is done by first telling the linker to preserve all relocations during the vmlinux link. (Note this is harmless: they are later stripped in the vmlinux.bin link.) Then use the 'relocs' tool to find all absolute relocations (R_390_64) which apply to allocatable sections. The offsets of those relocations are saved in a special section which is then used to adjust the relocations during boot. (Note: For some reason, Clang occasionally creates a GOT reference, even without '-fPIE'. So Clang-compiled kernels have a GOT, which needs to be adjusted.) On my mostly-defconfig kernel, this reduces kernel text size by ~1.3%. [1] https://github.com/dynup/kpatch/issues/1284 [2] https://gcc.gnu.org/pipermail/gcc-patches/2023-June/622872.html [3] https://gcc.gnu.org/pipermail/gcc-patches/2023-August/625986.html Compiler consideration: Gcc recently implemented an optimization [2] for loading symbols without explicit alignment, aligning with the IBM Z ELF ABI. This ABI mandates symbols to reside on a 2-byte boundary, enabling the use of the larl instruction. However, kernel linker scripts may still generate unaligned symbols. To address this, a new -munaligned-symbols option has been introduced [3] in recent gcc versions. This option has to be used with future gcc versions. Older Clang lacks support for handling unaligned symbols generated by kernel linker scripts when the kernel is built without -fPIE. However, future versions of Clang will include support for the -munaligned-symbols option. When the support is unavailable, compile the kernel with -fPIE to maintain the existing behavior. In addition to it: move vmlinux.relocs to safe relocation When the kernel is built with CONFIG_KERNEL_UNCOMPRESSED, the entire uncompressed vmlinux.bin is positioned in the bzImage decompressor image at the default kernel LMA of 0x100000, enabling it to be executed in-place. However, the size of .vmlinux.relocs could be large enough to cause an overlap with the uncompressed kernel at the address 0x100000. To address this issue, .vmlinux.relocs is positioned after the .rodata.compressed in the bzImage. Nevertheless, in this configuration, vmlinux.relocs will overlap with the .bss section of vmlinux.bin. To overcome that, move vmlinux.relocs to a safe location before clearing .bss and handling relocs. Compile warning fix from Sumanth Korikkar: When kernel is built with CONFIG_LD_ORPHAN_WARN and -fno-PIE, there are several warnings: ld: warning: orphan section `.rela.iplt' from `arch/s390/kernel/head64.o' being placed in section `.rela.dyn' ld: warning: orphan section `.rela.head.text' from `arch/s390/kernel/head64.o' being placed in section `.rela.dyn' ld: warning: orphan section `.rela.init.text' from `arch/s390/kernel/head64.o' being placed in section `.rela.dyn' ld: warning: orphan section `.rela.rodata.cst8' from `arch/s390/kernel/head64.o' being placed in section `.rela.dyn' Orphan sections are sections that exist in an object file but don't have a corresponding output section in the final executable. ld raises a warning when it identifies such sections. Eliminate the warning by placing all .rela orphan sections in .rela.dyn and raise an error when size of .rela.dyn is greater than zero. i.e. Dont just neglect orphan sections. This is similar to adjustment performed in x86, where kernel is built with -fno-PIE. commit 5354e84598f2 ("x86/build: Add asserts for unwanted sections") [sumanthk@linux.ibm.com: rebased Josh Poimboeuf patches and move vmlinux.relocs to safe location] [hca@linux.ibm.com: merged compile warning fix from Sumanth] Tested-by: Sumanth Korikkar <sumanthk@linux.ibm.com> Acked-by: Vasily Gorbik <gor@linux.ibm.com> Signed-off-by: Josh Poimboeuf <jpoimboe@kernel.org> Signed-off-by: Sumanth Korikkar <sumanthk@linux.ibm.com> Link: https://lore.kernel.org/r/20240219132734.22881-4-sumanthk@linux.ibm.com Link: https://lore.kernel.org/r/20240219132734.22881-5-sumanthk@linux.ibm.com Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2024-02-19 14:27:33 +01:00
s390/physmem_info: Query diag500(STORAGE LIMIT) to support QEMU/KVM memory devices To support memory devices under QEMU/KVM, such as virtio-mem, we have to prepare our kernel virtual address space accordingly and have to know the highest possible physical memory address we might see later: the storage limit. The good old SCLP interface is not suitable for this use case. In particular, memory owned by memory devices has no relationship to storage increments, it is always detected using the device driver, and unaware OSes (no driver) must never try making use of that memory. Consequently this memory is located outside of the "maximum storage increment"-indicated memory range. Let's use our new diag500 STORAGE_LIMIT subcode to query this storage limit that can exceed the "maximum storage increment", and use the existing interfaces (i.e., SCLP) to obtain information about the initial memory that is not owned+managed by memory devices. If a hypervisor does not support such memory devices, the address exposed through diag500 STORAGE_LIMIT will correspond to the maximum storage increment exposed through SCLP. To teach kdump on s390 to include memory owned by memory devices, there will be ways to query the relevant memory ranges from the device via a driver running in special kdump mode (like virtio-mem already implements to filter /proc/vmcore access so we don't end up reading from unplugged device blocks). Update setup_ident_map_size(), to clarify that there can be more than just online and standby memory. Tested-by: Mario Casquero <mcasquer@redhat.com> Signed-off-by: David Hildenbrand <david@redhat.com> Acked-by: Heiko Carstens <hca@linux.ibm.com> Reviewed-by: Alexander Gordeev <agordeev@linux.ibm.com> Tested-by: Sumanth Korikkar <sumanthk@linux.ibm.com> Acked-by: Christian Borntraeger <borntraeger@linux.ibm.com> Link: https://lore.kernel.org/r/20241025141453.1210600-4-david@redhat.com Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2024-10-25 16:14:48 +02:00
s390/physmem_info: Query diag500(STORAGE LIMIT) to support QEMU/KVM memory devices To support memory devices under QEMU/KVM, such as virtio-mem, we have to prepare our kernel virtual address space accordingly and have to know the highest possible physical memory address we might see later: the storage limit. The good old SCLP interface is not suitable for this use case. In particular, memory owned by memory devices has no relationship to storage increments, it is always detected using the device driver, and unaware OSes (no driver) must never try making use of that memory. Consequently this memory is located outside of the "maximum storage increment"-indicated memory range. Let's use our new diag500 STORAGE_LIMIT subcode to query this storage limit that can exceed the "maximum storage increment", and use the existing interfaces (i.e., SCLP) to obtain information about the initial memory that is not owned+managed by memory devices. If a hypervisor does not support such memory devices, the address exposed through diag500 STORAGE_LIMIT will correspond to the maximum storage increment exposed through SCLP. To teach kdump on s390 to include memory owned by memory devices, there will be ways to query the relevant memory ranges from the device via a driver running in special kdump mode (like virtio-mem already implements to filter /proc/vmcore access so we don't end up reading from unplugged device blocks). Update setup_ident_map_size(), to clarify that there can be more than just online and standby memory. Tested-by: Mario Casquero <mcasquer@redhat.com> Signed-off-by: David Hildenbrand <david@redhat.com> Acked-by: Heiko Carstens <hca@linux.ibm.com> Reviewed-by: Alexander Gordeev <agordeev@linux.ibm.com> Tested-by: Sumanth Korikkar <sumanthk@linux.ibm.com> Acked-by: Christian Borntraeger <borntraeger@linux.ibm.com> Link: https://lore.kernel.org/r/20241025141453.1210600-4-david@redhat.com Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2024-10-25 16:14:48 +02:00
s390/mm: Uncouple physical vs virtual address spaces The uncoupling physical vs virtual address spaces brings the following benefits to s390: - virtual memory layout flexibility; - closes the address gap between kernel and modules, it caused s390-only problems in the past (e.g. 'perf' bugs); - allows getting rid of trampolines used for module calls into kernel; - allows simplifying BPF trampoline; - minor performance improvement in branch prediction; - kernel randomization entropy is magnitude bigger, as it is derived from the amount of available virtual, not physical memory; The whole change could be described in two pictures below: before and after the change. Some aspects of the virtual memory layout setup are not clarified (number of page levels, alignment, DMA memory), since these are not a part of this change or secondary with regard to how the uncoupling itself is implemented. The focus of the pictures is to explain why __va() and __pa() macros are implemented the way they are. Memory layout in V==R mode: | Physical | Virtual | +- 0 --------------+- 0 --------------+ identity mapping start | | S390_lowcore | Low-address memory | +- 8 KB -----------+ | | | | | identity | phys == virt | | mapping | virt == phys | | | +- AMODE31_START --+- AMODE31_START --+ .amode31 rand. phys/virt start |.amode31 text/data|.amode31 text/data| +- AMODE31_END ----+- AMODE31_END ----+ .amode31 rand. phys/virt start | | | | | | +- __kaslr_offset, __kaslr_offset_phys| kernel rand. phys/virt start | | | | kernel text/data | kernel text/data | phys == kvirt | | | +------------------+------------------+ kernel phys/virt end | | | | | | | | | | | | +- ident_map_size -+- ident_map_size -+ identity mapping end | | | ... unused gap | | | +---- vmemmap -----+ 'struct page' array start | | | virtually mapped | | memory map | | | +- __abs_lowcore --+ | | | Absolute Lowcore | | | +- __memcpy_real_area | | | Real Memory Copy| | | +- VMALLOC_START --+ vmalloc area start | | | vmalloc area | | | +- MODULES_VADDR --+ modules area start | | | modules area | | | +------------------+ UltraVisor Secure Storage limit | | | ... unused gap | | | +KASAN_SHADOW_START+ KASAN shadow memory start | | | KASAN shadow | | | +------------------+ ASCE limit Memory layout in V!=R mode: | Physical | Virtual | +- 0 --------------+- 0 --------------+ | | S390_lowcore | Low-address memory | +- 8 KB -----------+ | | | | | | | | ... unused gap | | | | +- AMODE31_START --+- AMODE31_START --+ .amode31 rand. phys/virt start |.amode31 text/data|.amode31 text/data| +- AMODE31_END ----+- AMODE31_END ----+ .amode31 rand. phys/virt end (<2GB) | | | | | | +- __kaslr_offset_phys | kernel rand. phys start | | | | kernel text/data | | | | | +------------------+ | kernel phys end | | | | | | | | | | | | +- ident_map_size -+ | | | | ... unused gap | | | +- __identity_base + identity mapping start (>= 2GB) | | | identity | phys == virt - __identity_base | mapping | virt == phys + __identity_base | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | +---- vmemmap -----+ 'struct page' array start | | | virtually mapped | | memory map | | | +- __abs_lowcore --+ | | | Absolute Lowcore | | | +- __memcpy_real_area | | | Real Memory Copy| | | +- VMALLOC_START --+ vmalloc area start | | | vmalloc area | | | +- MODULES_VADDR --+ modules area start | | | modules area | | | +- __kaslr_offset -+ kernel rand. virt start | | | kernel text/data | phys == (kvirt - __kaslr_offset) + | | __kaslr_offset_phys +- kernel .bss end + kernel rand. virt end | | | ... unused gap | | | +------------------+ UltraVisor Secure Storage limit | | | ... unused gap | | | +KASAN_SHADOW_START+ KASAN shadow memory start | | | KASAN shadow | | | +------------------+ ASCE limit Unused gaps in the virtual memory layout could be present or not - depending on how partucular system is configured. No page tables are created for the unused gaps. The relative order of vmalloc, modules and kernel image in virtual memory is defined by following considerations: - start of the modules area and end of the kernel should reside within 4GB to accommodate relative 32-bit jumps. The best way to achieve that is to place kernel next to modules; - vmalloc and module areas should locate next to each other to prevent failures and extra reworks in user level tools (makedumpfile, crash, etc.) which treat vmalloc and module addresses similarily; - kernel needs to be the last area in the virtual memory layout to easily distinguish between kernel and non-kernel virtual addresses. That is needed to (again) simplify handling of addresses in user level tools and make __pa() macro faster (see below); Concluding the above, the relative order of the considered virtual areas in memory is: vmalloc - modules - kernel. Therefore, the only change to the current memory layout is moving kernel to the end of virtual address space. With that approach the implementation of __pa() macro is straightforward - all linear virtual addresses less than kernel base are considered identity mapping: phys == virt - __identity_base All addresses greater than kernel base are kernel ones: phys == (kvirt - __kaslr_offset) + __kaslr_offset_phys By contrast, __va() macro deals only with identity mapping addresses: virt == phys + __identity_base .amode31 section is mapped separately and is not covered by __pa() macro. In fact, it could have been handled easily by checking whether a virtual address is within the section or not, but there is no need for that. Thus, let __pa() code do as little machine cycles as possible. The KASAN shadow memory is located at the very end of the virtual memory layout, at addresses higher than the kernel. However, that is not a linear mapping and no code other than KASAN instrumentation or API is expected to access it. When KASLR mode is enabled the kernel base address randomized within a memory window that spans whole unused virtual address space. The size of that window depends from the amount of physical memory available to the system, the limit imposed by UltraVisor (if present) and the vmalloc area size as provided by vmalloc= kernel command line parameter. In case the virtual memory is exhausted the minimum size of the randomization window is forcefully set to 2GB, which amounts to in 15 bits of entropy if KASAN is enabled or 17 bits of entropy in default configuration. The default kernel offset 0x100000 is used as a magic value both in the decompressor code and vmlinux linker script, but it will be removed with a follow-up change. Acked-by: Heiko Carstens <hca@linux.ibm.com> Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com>
2024-03-01 07:15:22 +01:00
s390/mm: Uncouple physical vs virtual address spaces The uncoupling physical vs virtual address spaces brings the following benefits to s390: - virtual memory layout flexibility; - closes the address gap between kernel and modules, it caused s390-only problems in the past (e.g. 'perf' bugs); - allows getting rid of trampolines used for module calls into kernel; - allows simplifying BPF trampoline; - minor performance improvement in branch prediction; - kernel randomization entropy is magnitude bigger, as it is derived from the amount of available virtual, not physical memory; The whole change could be described in two pictures below: before and after the change. Some aspects of the virtual memory layout setup are not clarified (number of page levels, alignment, DMA memory), since these are not a part of this change or secondary with regard to how the uncoupling itself is implemented. The focus of the pictures is to explain why __va() and __pa() macros are implemented the way they are. Memory layout in V==R mode: | Physical | Virtual | +- 0 --------------+- 0 --------------+ identity mapping start | | S390_lowcore | Low-address memory | +- 8 KB -----------+ | | | | | identity | phys == virt | | mapping | virt == phys | | | +- AMODE31_START --+- AMODE31_START --+ .amode31 rand. phys/virt start |.amode31 text/data|.amode31 text/data| +- AMODE31_END ----+- AMODE31_END ----+ .amode31 rand. phys/virt start | | | | | | +- __kaslr_offset, __kaslr_offset_phys| kernel rand. phys/virt start | | | | kernel text/data | kernel text/data | phys == kvirt | | | +------------------+------------------+ kernel phys/virt end | | | | | | | | | | | | +- ident_map_size -+- ident_map_size -+ identity mapping end | | | ... unused gap | | | +---- vmemmap -----+ 'struct page' array start | | | virtually mapped | | memory map | | | +- __abs_lowcore --+ | | | Absolute Lowcore | | | +- __memcpy_real_area | | | Real Memory Copy| | | +- VMALLOC_START --+ vmalloc area start | | | vmalloc area | | | +- MODULES_VADDR --+ modules area start | | | modules area | | | +------------------+ UltraVisor Secure Storage limit | | | ... unused gap | | | +KASAN_SHADOW_START+ KASAN shadow memory start | | | KASAN shadow | | | +------------------+ ASCE limit Memory layout in V!=R mode: | Physical | Virtual | +- 0 --------------+- 0 --------------+ | | S390_lowcore | Low-address memory | +- 8 KB -----------+ | | | | | | | | ... unused gap | | | | +- AMODE31_START --+- AMODE31_START --+ .amode31 rand. phys/virt start |.amode31 text/data|.amode31 text/data| +- AMODE31_END ----+- AMODE31_END ----+ .amode31 rand. phys/virt end (<2GB) | | | | | | +- __kaslr_offset_phys | kernel rand. phys start | | | | kernel text/data | | | | | +------------------+ | kernel phys end | | | | | | | | | | | | +- ident_map_size -+ | | | | ... unused gap | | | +- __identity_base + identity mapping start (>= 2GB) | | | identity | phys == virt - __identity_base | mapping | virt == phys + __identity_base | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | +---- vmemmap -----+ 'struct page' array start | | | virtually mapped | | memory map | | | +- __abs_lowcore --+ | | | Absolute Lowcore | | | +- __memcpy_real_area | | | Real Memory Copy| | | +- VMALLOC_START --+ vmalloc area start | | | vmalloc area | | | +- MODULES_VADDR --+ modules area start | | | modules area | | | +- __kaslr_offset -+ kernel rand. virt start | | | kernel text/data | phys == (kvirt - __kaslr_offset) + | | __kaslr_offset_phys +- kernel .bss end + kernel rand. virt end | | | ... unused gap | | | +------------------+ UltraVisor Secure Storage limit | | | ... unused gap | | | +KASAN_SHADOW_START+ KASAN shadow memory start | | | KASAN shadow | | | +------------------+ ASCE limit Unused gaps in the virtual memory layout could be present or not - depending on how partucular system is configured. No page tables are created for the unused gaps. The relative order of vmalloc, modules and kernel image in virtual memory is defined by following considerations: - start of the modules area and end of the kernel should reside within 4GB to accommodate relative 32-bit jumps. The best way to achieve that is to place kernel next to modules; - vmalloc and module areas should locate next to each other to prevent failures and extra reworks in user level tools (makedumpfile, crash, etc.) which treat vmalloc and module addresses similarily; - kernel needs to be the last area in the virtual memory layout to easily distinguish between kernel and non-kernel virtual addresses. That is needed to (again) simplify handling of addresses in user level tools and make __pa() macro faster (see below); Concluding the above, the relative order of the considered virtual areas in memory is: vmalloc - modules - kernel. Therefore, the only change to the current memory layout is moving kernel to the end of virtual address space. With that approach the implementation of __pa() macro is straightforward - all linear virtual addresses less than kernel base are considered identity mapping: phys == virt - __identity_base All addresses greater than kernel base are kernel ones: phys == (kvirt - __kaslr_offset) + __kaslr_offset_phys By contrast, __va() macro deals only with identity mapping addresses: virt == phys + __identity_base .amode31 section is mapped separately and is not covered by __pa() macro. In fact, it could have been handled easily by checking whether a virtual address is within the section or not, but there is no need for that. Thus, let __pa() code do as little machine cycles as possible. The KASAN shadow memory is located at the very end of the virtual memory layout, at addresses higher than the kernel. However, that is not a linear mapping and no code other than KASAN instrumentation or API is expected to access it. When KASLR mode is enabled the kernel base address randomized within a memory window that spans whole unused virtual address space. The size of that window depends from the amount of physical memory available to the system, the limit imposed by UltraVisor (if present) and the vmalloc area size as provided by vmalloc= kernel command line parameter. In case the virtual memory is exhausted the minimum size of the randomization window is forcefully set to 2GB, which amounts to in 15 bits of entropy if KASAN is enabled or 17 bits of entropy in default configuration. The default kernel offset 0x100000 is used as a magic value both in the decompressor code and vmlinux linker script, but it will be removed with a follow-up change. Acked-by: Heiko Carstens <hca@linux.ibm.com> Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com>
2024-03-01 07:15:22 +01:00
s390/mm: Uncouple physical vs virtual address spaces The uncoupling physical vs virtual address spaces brings the following benefits to s390: - virtual memory layout flexibility; - closes the address gap between kernel and modules, it caused s390-only problems in the past (e.g. 'perf' bugs); - allows getting rid of trampolines used for module calls into kernel; - allows simplifying BPF trampoline; - minor performance improvement in branch prediction; - kernel randomization entropy is magnitude bigger, as it is derived from the amount of available virtual, not physical memory; The whole change could be described in two pictures below: before and after the change. Some aspects of the virtual memory layout setup are not clarified (number of page levels, alignment, DMA memory), since these are not a part of this change or secondary with regard to how the uncoupling itself is implemented. The focus of the pictures is to explain why __va() and __pa() macros are implemented the way they are. Memory layout in V==R mode: | Physical | Virtual | +- 0 --------------+- 0 --------------+ identity mapping start | | S390_lowcore | Low-address memory | +- 8 KB -----------+ | | | | | identity | phys == virt | | mapping | virt == phys | | | +- AMODE31_START --+- AMODE31_START --+ .amode31 rand. phys/virt start |.amode31 text/data|.amode31 text/data| +- AMODE31_END ----+- AMODE31_END ----+ .amode31 rand. phys/virt start | | | | | | +- __kaslr_offset, __kaslr_offset_phys| kernel rand. phys/virt start | | | | kernel text/data | kernel text/data | phys == kvirt | | | +------------------+------------------+ kernel phys/virt end | | | | | | | | | | | | +- ident_map_size -+- ident_map_size -+ identity mapping end | | | ... unused gap | | | +---- vmemmap -----+ 'struct page' array start | | | virtually mapped | | memory map | | | +- __abs_lowcore --+ | | | Absolute Lowcore | | | +- __memcpy_real_area | | | Real Memory Copy| | | +- VMALLOC_START --+ vmalloc area start | | | vmalloc area | | | +- MODULES_VADDR --+ modules area start | | | modules area | | | +------------------+ UltraVisor Secure Storage limit | | | ... unused gap | | | +KASAN_SHADOW_START+ KASAN shadow memory start | | | KASAN shadow | | | +------------------+ ASCE limit Memory layout in V!=R mode: | Physical | Virtual | +- 0 --------------+- 0 --------------+ | | S390_lowcore | Low-address memory | +- 8 KB -----------+ | | | | | | | | ... unused gap | | | | +- AMODE31_START --+- AMODE31_START --+ .amode31 rand. phys/virt start |.amode31 text/data|.amode31 text/data| +- AMODE31_END ----+- AMODE31_END ----+ .amode31 rand. phys/virt end (<2GB) | | | | | | +- __kaslr_offset_phys | kernel rand. phys start | | | | kernel text/data | | | | | +------------------+ | kernel phys end | | | | | | | | | | | | +- ident_map_size -+ | | | | ... unused gap | | | +- __identity_base + identity mapping start (>= 2GB) | | | identity | phys == virt - __identity_base | mapping | virt == phys + __identity_base | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | +---- vmemmap -----+ 'struct page' array start | | | virtually mapped | | memory map | | | +- __abs_lowcore --+ | | | Absolute Lowcore | | | +- __memcpy_real_area | | | Real Memory Copy| | | +- VMALLOC_START --+ vmalloc area start | | | vmalloc area | | | +- MODULES_VADDR --+ modules area start | | | modules area | | | +- __kaslr_offset -+ kernel rand. virt start | | | kernel text/data | phys == (kvirt - __kaslr_offset) + | | __kaslr_offset_phys +- kernel .bss end + kernel rand. virt end | | | ... unused gap | | | +------------------+ UltraVisor Secure Storage limit | | | ... unused gap | | | +KASAN_SHADOW_START+ KASAN shadow memory start | | | KASAN shadow | | | +------------------+ ASCE limit Unused gaps in the virtual memory layout could be present or not - depending on how partucular system is configured. No page tables are created for the unused gaps. The relative order of vmalloc, modules and kernel image in virtual memory is defined by following considerations: - start of the modules area and end of the kernel should reside within 4GB to accommodate relative 32-bit jumps. The best way to achieve that is to place kernel next to modules; - vmalloc and module areas should locate next to each other to prevent failures and extra reworks in user level tools (makedumpfile, crash, etc.) which treat vmalloc and module addresses similarily; - kernel needs to be the last area in the virtual memory layout to easily distinguish between kernel and non-kernel virtual addresses. That is needed to (again) simplify handling of addresses in user level tools and make __pa() macro faster (see below); Concluding the above, the relative order of the considered virtual areas in memory is: vmalloc - modules - kernel. Therefore, the only change to the current memory layout is moving kernel to the end of virtual address space. With that approach the implementation of __pa() macro is straightforward - all linear virtual addresses less than kernel base are considered identity mapping: phys == virt - __identity_base All addresses greater than kernel base are kernel ones: phys == (kvirt - __kaslr_offset) + __kaslr_offset_phys By contrast, __va() macro deals only with identity mapping addresses: virt == phys + __identity_base .amode31 section is mapped separately and is not covered by __pa() macro. In fact, it could have been handled easily by checking whether a virtual address is within the section or not, but there is no need for that. Thus, let __pa() code do as little machine cycles as possible. The KASAN shadow memory is located at the very end of the virtual memory layout, at addresses higher than the kernel. However, that is not a linear mapping and no code other than KASAN instrumentation or API is expected to access it. When KASLR mode is enabled the kernel base address randomized within a memory window that spans whole unused virtual address space. The size of that window depends from the amount of physical memory available to the system, the limit imposed by UltraVisor (if present) and the vmalloc area size as provided by vmalloc= kernel command line parameter. In case the virtual memory is exhausted the minimum size of the randomization window is forcefully set to 2GB, which amounts to in 15 bits of entropy if KASAN is enabled or 17 bits of entropy in default configuration. The default kernel offset 0x100000 is used as a magic value both in the decompressor code and vmlinux linker script, but it will be removed with a follow-up change. Acked-by: Heiko Carstens <hca@linux.ibm.com> Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com>
2024-03-01 07:15:22 +01:00
s390/mm: start kernel with DAT enabled The setup of the kernel virtual address space is spread throughout the sources, boot stages and config options like this: 1. The available physical memory regions are queried and stored as mem_detect information for later use in the decompressor. 2. Based on the physical memory availability the virtual memory layout is established in the decompressor; 3. If CONFIG_KASAN is disabled the kernel paging setup code populates kernel pgtables and turns DAT mode on. It uses the information stored at step [1]. 4. If CONFIG_KASAN is enabled the kernel early boot kasan setup populates kernel pgtables and turns DAT mode on. It uses the information stored at step [1]. The kasan setup creates early_pg_dir directory and directly overwrites swapper_pg_dir entries to make shadow memory pages available. Move the kernel virtual memory setup to the decompressor and start the kernel with DAT turned on right from the very first istruction. That completely eliminates the boot phase when the kernel runs in DAT-off mode, simplies the overall design and consolidates pgtables setup. The identity mapping is created in the decompressor, while kasan shadow mappings are still created by the early boot kernel code. Share with decompressor the existing kasan memory allocator. It decreases the size of a newly requested memory block from pgalloc_pos and ensures that kernel image is not overwritten. pgalloc_low and pgalloc_pos pointers are made preserved boot variables for that. Use the bootdata infrastructure to setup swapper_pg_dir and invalid_pg_dir directories used by the kernel later. The interim early_pg_dir directory established by the kasan initialization code gets eliminated as result. As the kernel runs in DAT-on mode only the PSW_KERNEL_BITS define gets PSW_MASK_DAT bit by default. Additionally, the setup_lowcore_dat_off() and setup_lowcore_dat_on() routines get merged, since there is no DAT-off mode stage anymore. The memory mappings are created with RW+X protection that allows the early boot code setting up all necessary data and services for the kernel being booted. Just before the paging is enabled the memory protection is changed to RO+X for text, RO+NX for read-only data and RW+NX for kernel data and the identity mapping. Reviewed-by: Heiko Carstens <hca@linux.ibm.com> Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com> Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2022-12-13 11:35:11 +01:00
s390/mm: Uncouple physical vs virtual address spaces The uncoupling physical vs virtual address spaces brings the following benefits to s390: - virtual memory layout flexibility; - closes the address gap between kernel and modules, it caused s390-only problems in the past (e.g. 'perf' bugs); - allows getting rid of trampolines used for module calls into kernel; - allows simplifying BPF trampoline; - minor performance improvement in branch prediction; - kernel randomization entropy is magnitude bigger, as it is derived from the amount of available virtual, not physical memory; The whole change could be described in two pictures below: before and after the change. Some aspects of the virtual memory layout setup are not clarified (number of page levels, alignment, DMA memory), since these are not a part of this change or secondary with regard to how the uncoupling itself is implemented. The focus of the pictures is to explain why __va() and __pa() macros are implemented the way they are. Memory layout in V==R mode: | Physical | Virtual | +- 0 --------------+- 0 --------------+ identity mapping start | | S390_lowcore | Low-address memory | +- 8 KB -----------+ | | | | | identity | phys == virt | | mapping | virt == phys | | | +- AMODE31_START --+- AMODE31_START --+ .amode31 rand. phys/virt start |.amode31 text/data|.amode31 text/data| +- AMODE31_END ----+- AMODE31_END ----+ .amode31 rand. phys/virt start | | | | | | +- __kaslr_offset, __kaslr_offset_phys| kernel rand. phys/virt start | | | | kernel text/data | kernel text/data | phys == kvirt | | | +------------------+------------------+ kernel phys/virt end | | | | | | | | | | | | +- ident_map_size -+- ident_map_size -+ identity mapping end | | | ... unused gap | | | +---- vmemmap -----+ 'struct page' array start | | | virtually mapped | | memory map | | | +- __abs_lowcore --+ | | | Absolute Lowcore | | | +- __memcpy_real_area | | | Real Memory Copy| | | +- VMALLOC_START --+ vmalloc area start | | | vmalloc area | | | +- MODULES_VADDR --+ modules area start | | | modules area | | | +------------------+ UltraVisor Secure Storage limit | | | ... unused gap | | | +KASAN_SHADOW_START+ KASAN shadow memory start | | | KASAN shadow | | | +------------------+ ASCE limit Memory layout in V!=R mode: | Physical | Virtual | +- 0 --------------+- 0 --------------+ | | S390_lowcore | Low-address memory | +- 8 KB -----------+ | | | | | | | | ... unused gap | | | | +- AMODE31_START --+- AMODE31_START --+ .amode31 rand. phys/virt start |.amode31 text/data|.amode31 text/data| +- AMODE31_END ----+- AMODE31_END ----+ .amode31 rand. phys/virt end (<2GB) | | | | | | +- __kaslr_offset_phys | kernel rand. phys start | | | | kernel text/data | | | | | +------------------+ | kernel phys end | | | | | | | | | | | | +- ident_map_size -+ | | | | ... unused gap | | | +- __identity_base + identity mapping start (>= 2GB) | | | identity | phys == virt - __identity_base | mapping | virt == phys + __identity_base | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | +---- vmemmap -----+ 'struct page' array start | | | virtually mapped | | memory map | | | +- __abs_lowcore --+ | | | Absolute Lowcore | | | +- __memcpy_real_area | | | Real Memory Copy| | | +- VMALLOC_START --+ vmalloc area start | | | vmalloc area | | | +- MODULES_VADDR --+ modules area start | | | modules area | | | +- __kaslr_offset -+ kernel rand. virt start | | | kernel text/data | phys == (kvirt - __kaslr_offset) + | | __kaslr_offset_phys +- kernel .bss end + kernel rand. virt end | | | ... unused gap | | | +------------------+ UltraVisor Secure Storage limit | | | ... unused gap | | | +KASAN_SHADOW_START+ KASAN shadow memory start | | | KASAN shadow | | | +------------------+ ASCE limit Unused gaps in the virtual memory layout could be present or not - depending on how partucular system is configured. No page tables are created for the unused gaps. The relative order of vmalloc, modules and kernel image in virtual memory is defined by following considerations: - start of the modules area and end of the kernel should reside within 4GB to accommodate relative 32-bit jumps. The best way to achieve that is to place kernel next to modules; - vmalloc and module areas should locate next to each other to prevent failures and extra reworks in user level tools (makedumpfile, crash, etc.) which treat vmalloc and module addresses similarily; - kernel needs to be the last area in the virtual memory layout to easily distinguish between kernel and non-kernel virtual addresses. That is needed to (again) simplify handling of addresses in user level tools and make __pa() macro faster (see below); Concluding the above, the relative order of the considered virtual areas in memory is: vmalloc - modules - kernel. Therefore, the only change to the current memory layout is moving kernel to the end of virtual address space. With that approach the implementation of __pa() macro is straightforward - all linear virtual addresses less than kernel base are considered identity mapping: phys == virt - __identity_base All addresses greater than kernel base are kernel ones: phys == (kvirt - __kaslr_offset) + __kaslr_offset_phys By contrast, __va() macro deals only with identity mapping addresses: virt == phys + __identity_base .amode31 section is mapped separately and is not covered by __pa() macro. In fact, it could have been handled easily by checking whether a virtual address is within the section or not, but there is no need for that. Thus, let __pa() code do as little machine cycles as possible. The KASAN shadow memory is located at the very end of the virtual memory layout, at addresses higher than the kernel. However, that is not a linear mapping and no code other than KASAN instrumentation or API is expected to access it. When KASLR mode is enabled the kernel base address randomized within a memory window that spans whole unused virtual address space. The size of that window depends from the amount of physical memory available to the system, the limit imposed by UltraVisor (if present) and the vmalloc area size as provided by vmalloc= kernel command line parameter. In case the virtual memory is exhausted the minimum size of the randomization window is forcefully set to 2GB, which amounts to in 15 bits of entropy if KASAN is enabled or 17 bits of entropy in default configuration. The default kernel offset 0x100000 is used as a magic value both in the decompressor code and vmlinux linker script, but it will be removed with a follow-up change. Acked-by: Heiko Carstens <hca@linux.ibm.com> Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com>
2024-03-01 07:15:22 +01:00
s390/mm: start kernel with DAT enabled The setup of the kernel virtual address space is spread throughout the sources, boot stages and config options like this: 1. The available physical memory regions are queried and stored as mem_detect information for later use in the decompressor. 2. Based on the physical memory availability the virtual memory layout is established in the decompressor; 3. If CONFIG_KASAN is disabled the kernel paging setup code populates kernel pgtables and turns DAT mode on. It uses the information stored at step [1]. 4. If CONFIG_KASAN is enabled the kernel early boot kasan setup populates kernel pgtables and turns DAT mode on. It uses the information stored at step [1]. The kasan setup creates early_pg_dir directory and directly overwrites swapper_pg_dir entries to make shadow memory pages available. Move the kernel virtual memory setup to the decompressor and start the kernel with DAT turned on right from the very first istruction. That completely eliminates the boot phase when the kernel runs in DAT-off mode, simplies the overall design and consolidates pgtables setup. The identity mapping is created in the decompressor, while kasan shadow mappings are still created by the early boot kernel code. Share with decompressor the existing kasan memory allocator. It decreases the size of a newly requested memory block from pgalloc_pos and ensures that kernel image is not overwritten. pgalloc_low and pgalloc_pos pointers are made preserved boot variables for that. Use the bootdata infrastructure to setup swapper_pg_dir and invalid_pg_dir directories used by the kernel later. The interim early_pg_dir directory established by the kasan initialization code gets eliminated as result. As the kernel runs in DAT-on mode only the PSW_KERNEL_BITS define gets PSW_MASK_DAT bit by default. Additionally, the setup_lowcore_dat_off() and setup_lowcore_dat_on() routines get merged, since there is no DAT-off mode stage anymore. The memory mappings are created with RW+X protection that allows the early boot code setting up all necessary data and services for the kernel being booted. Just before the paging is enabled the memory protection is changed to RO+X for text, RO+NX for read-only data and RW+NX for kernel data and the identity mapping. Reviewed-by: Heiko Carstens <hca@linux.ibm.com> Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com> Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2022-12-13 11:35:11 +01:00
s390/mm: start kernel with DAT enabled The setup of the kernel virtual address space is spread throughout the sources, boot stages and config options like this: 1. The available physical memory regions are queried and stored as mem_detect information for later use in the decompressor. 2. Based on the physical memory availability the virtual memory layout is established in the decompressor; 3. If CONFIG_KASAN is disabled the kernel paging setup code populates kernel pgtables and turns DAT mode on. It uses the information stored at step [1]. 4. If CONFIG_KASAN is enabled the kernel early boot kasan setup populates kernel pgtables and turns DAT mode on. It uses the information stored at step [1]. The kasan setup creates early_pg_dir directory and directly overwrites swapper_pg_dir entries to make shadow memory pages available. Move the kernel virtual memory setup to the decompressor and start the kernel with DAT turned on right from the very first istruction. That completely eliminates the boot phase when the kernel runs in DAT-off mode, simplies the overall design and consolidates pgtables setup. The identity mapping is created in the decompressor, while kasan shadow mappings are still created by the early boot kernel code. Share with decompressor the existing kasan memory allocator. It decreases the size of a newly requested memory block from pgalloc_pos and ensures that kernel image is not overwritten. pgalloc_low and pgalloc_pos pointers are made preserved boot variables for that. Use the bootdata infrastructure to setup swapper_pg_dir and invalid_pg_dir directories used by the kernel later. The interim early_pg_dir directory established by the kasan initialization code gets eliminated as result. As the kernel runs in DAT-on mode only the PSW_KERNEL_BITS define gets PSW_MASK_DAT bit by default. Additionally, the setup_lowcore_dat_off() and setup_lowcore_dat_on() routines get merged, since there is no DAT-off mode stage anymore. The memory mappings are created with RW+X protection that allows the early boot code setting up all necessary data and services for the kernel being booted. Just before the paging is enabled the memory protection is changed to RO+X for text, RO+NX for read-only data and RW+NX for kernel data and the identity mapping. Reviewed-by: Heiko Carstens <hca@linux.ibm.com> Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com> Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2022-12-13 11:35:11 +01:00
s390/mm: start kernel with DAT enabled The setup of the kernel virtual address space is spread throughout the sources, boot stages and config options like this: 1. The available physical memory regions are queried and stored as mem_detect information for later use in the decompressor. 2. Based on the physical memory availability the virtual memory layout is established in the decompressor; 3. If CONFIG_KASAN is disabled the kernel paging setup code populates kernel pgtables and turns DAT mode on. It uses the information stored at step [1]. 4. If CONFIG_KASAN is enabled the kernel early boot kasan setup populates kernel pgtables and turns DAT mode on. It uses the information stored at step [1]. The kasan setup creates early_pg_dir directory and directly overwrites swapper_pg_dir entries to make shadow memory pages available. Move the kernel virtual memory setup to the decompressor and start the kernel with DAT turned on right from the very first istruction. That completely eliminates the boot phase when the kernel runs in DAT-off mode, simplies the overall design and consolidates pgtables setup. The identity mapping is created in the decompressor, while kasan shadow mappings are still created by the early boot kernel code. Share with decompressor the existing kasan memory allocator. It decreases the size of a newly requested memory block from pgalloc_pos and ensures that kernel image is not overwritten. pgalloc_low and pgalloc_pos pointers are made preserved boot variables for that. Use the bootdata infrastructure to setup swapper_pg_dir and invalid_pg_dir directories used by the kernel later. The interim early_pg_dir directory established by the kasan initialization code gets eliminated as result. As the kernel runs in DAT-on mode only the PSW_KERNEL_BITS define gets PSW_MASK_DAT bit by default. Additionally, the setup_lowcore_dat_off() and setup_lowcore_dat_on() routines get merged, since there is no DAT-off mode stage anymore. The memory mappings are created with RW+X protection that allows the early boot code setting up all necessary data and services for the kernel being booted. Just before the paging is enabled the memory protection is changed to RO+X for text, RO+NX for read-only data and RW+NX for kernel data and the identity mapping. Reviewed-by: Heiko Carstens <hca@linux.ibm.com> Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com> Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2022-12-13 11:35:11 +01:00
s390/mm: Uncouple physical vs virtual address spaces The uncoupling physical vs virtual address spaces brings the following benefits to s390: - virtual memory layout flexibility; - closes the address gap between kernel and modules, it caused s390-only problems in the past (e.g. 'perf' bugs); - allows getting rid of trampolines used for module calls into kernel; - allows simplifying BPF trampoline; - minor performance improvement in branch prediction; - kernel randomization entropy is magnitude bigger, as it is derived from the amount of available virtual, not physical memory; The whole change could be described in two pictures below: before and after the change. Some aspects of the virtual memory layout setup are not clarified (number of page levels, alignment, DMA memory), since these are not a part of this change or secondary with regard to how the uncoupling itself is implemented. The focus of the pictures is to explain why __va() and __pa() macros are implemented the way they are. Memory layout in V==R mode: | Physical | Virtual | +- 0 --------------+- 0 --------------+ identity mapping start | | S390_lowcore | Low-address memory | +- 8 KB -----------+ | | | | | identity | phys == virt | | mapping | virt == phys | | | +- AMODE31_START --+- AMODE31_START --+ .amode31 rand. phys/virt start |.amode31 text/data|.amode31 text/data| +- AMODE31_END ----+- AMODE31_END ----+ .amode31 rand. phys/virt start | | | | | | +- __kaslr_offset, __kaslr_offset_phys| kernel rand. phys/virt start | | | | kernel text/data | kernel text/data | phys == kvirt | | | +------------------+------------------+ kernel phys/virt end | | | | | | | | | | | | +- ident_map_size -+- ident_map_size -+ identity mapping end | | | ... unused gap | | | +---- vmemmap -----+ 'struct page' array start | | | virtually mapped | | memory map | | | +- __abs_lowcore --+ | | | Absolute Lowcore | | | +- __memcpy_real_area | | | Real Memory Copy| | | +- VMALLOC_START --+ vmalloc area start | | | vmalloc area | | | +- MODULES_VADDR --+ modules area start | | | modules area | | | +------------------+ UltraVisor Secure Storage limit | | | ... unused gap | | | +KASAN_SHADOW_START+ KASAN shadow memory start | | | KASAN shadow | | | +------------------+ ASCE limit Memory layout in V!=R mode: | Physical | Virtual | +- 0 --------------+- 0 --------------+ | | S390_lowcore | Low-address memory | +- 8 KB -----------+ | | | | | | | | ... unused gap | | | | +- AMODE31_START --+- AMODE31_START --+ .amode31 rand. phys/virt start |.amode31 text/data|.amode31 text/data| +- AMODE31_END ----+- AMODE31_END ----+ .amode31 rand. phys/virt end (<2GB) | | | | | | +- __kaslr_offset_phys | kernel rand. phys start | | | | kernel text/data | | | | | +------------------+ | kernel phys end | | | | | | | | | | | | +- ident_map_size -+ | | | | ... unused gap | | | +- __identity_base + identity mapping start (>= 2GB) | | | identity | phys == virt - __identity_base | mapping | virt == phys + __identity_base | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | +---- vmemmap -----+ 'struct page' array start | | | virtually mapped | | memory map | | | +- __abs_lowcore --+ | | | Absolute Lowcore | | | +- __memcpy_real_area | | | Real Memory Copy| | | +- VMALLOC_START --+ vmalloc area start | | | vmalloc area | | | +- MODULES_VADDR --+ modules area start | | | modules area | | | +- __kaslr_offset -+ kernel rand. virt start | | | kernel text/data | phys == (kvirt - __kaslr_offset) + | | __kaslr_offset_phys +- kernel .bss end + kernel rand. virt end | | | ... unused gap | | | +------------------+ UltraVisor Secure Storage limit | | | ... unused gap | | | +KASAN_SHADOW_START+ KASAN shadow memory start | | | KASAN shadow | | | +------------------+ ASCE limit Unused gaps in the virtual memory layout could be present or not - depending on how partucular system is configured. No page tables are created for the unused gaps. The relative order of vmalloc, modules and kernel image in virtual memory is defined by following considerations: - start of the modules area and end of the kernel should reside within 4GB to accommodate relative 32-bit jumps. The best way to achieve that is to place kernel next to modules; - vmalloc and module areas should locate next to each other to prevent failures and extra reworks in user level tools (makedumpfile, crash, etc.) which treat vmalloc and module addresses similarily; - kernel needs to be the last area in the virtual memory layout to easily distinguish between kernel and non-kernel virtual addresses. That is needed to (again) simplify handling of addresses in user level tools and make __pa() macro faster (see below); Concluding the above, the relative order of the considered virtual areas in memory is: vmalloc - modules - kernel. Therefore, the only change to the current memory layout is moving kernel to the end of virtual address space. With that approach the implementation of __pa() macro is straightforward - all linear virtual addresses less than kernel base are considered identity mapping: phys == virt - __identity_base All addresses greater than kernel base are kernel ones: phys == (kvirt - __kaslr_offset) + __kaslr_offset_phys By contrast, __va() macro deals only with identity mapping addresses: virt == phys + __identity_base .amode31 section is mapped separately and is not covered by __pa() macro. In fact, it could have been handled easily by checking whether a virtual address is within the section or not, but there is no need for that. Thus, let __pa() code do as little machine cycles as possible. The KASAN shadow memory is located at the very end of the virtual memory layout, at addresses higher than the kernel. However, that is not a linear mapping and no code other than KASAN instrumentation or API is expected to access it. When KASLR mode is enabled the kernel base address randomized within a memory window that spans whole unused virtual address space. The size of that window depends from the amount of physical memory available to the system, the limit imposed by UltraVisor (if present) and the vmalloc area size as provided by vmalloc= kernel command line parameter. In case the virtual memory is exhausted the minimum size of the randomization window is forcefully set to 2GB, which amounts to in 15 bits of entropy if KASAN is enabled or 17 bits of entropy in default configuration. The default kernel offset 0x100000 is used as a magic value both in the decompressor code and vmlinux linker script, but it will be removed with a follow-up change. Acked-by: Heiko Carstens <hca@linux.ibm.com> Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com>
2024-03-01 07:15:22 +01:00
s390/mm: Uncouple physical vs virtual address spaces The uncoupling physical vs virtual address spaces brings the following benefits to s390: - virtual memory layout flexibility; - closes the address gap between kernel and modules, it caused s390-only problems in the past (e.g. 'perf' bugs); - allows getting rid of trampolines used for module calls into kernel; - allows simplifying BPF trampoline; - minor performance improvement in branch prediction; - kernel randomization entropy is magnitude bigger, as it is derived from the amount of available virtual, not physical memory; The whole change could be described in two pictures below: before and after the change. Some aspects of the virtual memory layout setup are not clarified (number of page levels, alignment, DMA memory), since these are not a part of this change or secondary with regard to how the uncoupling itself is implemented. The focus of the pictures is to explain why __va() and __pa() macros are implemented the way they are. Memory layout in V==R mode: | Physical | Virtual | +- 0 --------------+- 0 --------------+ identity mapping start | | S390_lowcore | Low-address memory | +- 8 KB -----------+ | | | | | identity | phys == virt | | mapping | virt == phys | | | +- AMODE31_START --+- AMODE31_START --+ .amode31 rand. phys/virt start |.amode31 text/data|.amode31 text/data| +- AMODE31_END ----+- AMODE31_END ----+ .amode31 rand. phys/virt start | | | | | | +- __kaslr_offset, __kaslr_offset_phys| kernel rand. phys/virt start | | | | kernel text/data | kernel text/data | phys == kvirt | | | +------------------+------------------+ kernel phys/virt end | | | | | | | | | | | | +- ident_map_size -+- ident_map_size -+ identity mapping end | | | ... unused gap | | | +---- vmemmap -----+ 'struct page' array start | | | virtually mapped | | memory map | | | +- __abs_lowcore --+ | | | Absolute Lowcore | | | +- __memcpy_real_area | | | Real Memory Copy| | | +- VMALLOC_START --+ vmalloc area start | | | vmalloc area | | | +- MODULES_VADDR --+ modules area start | | | modules area | | | +------------------+ UltraVisor Secure Storage limit | | | ... unused gap | | | +KASAN_SHADOW_START+ KASAN shadow memory start | | | KASAN shadow | | | +------------------+ ASCE limit Memory layout in V!=R mode: | Physical | Virtual | +- 0 --------------+- 0 --------------+ | | S390_lowcore | Low-address memory | +- 8 KB -----------+ | | | | | | | | ... unused gap | | | | +- AMODE31_START --+- AMODE31_START --+ .amode31 rand. phys/virt start |.amode31 text/data|.amode31 text/data| +- AMODE31_END ----+- AMODE31_END ----+ .amode31 rand. phys/virt end (<2GB) | | | | | | +- __kaslr_offset_phys | kernel rand. phys start | | | | kernel text/data | | | | | +------------------+ | kernel phys end | | | | | | | | | | | | +- ident_map_size -+ | | | | ... unused gap | | | +- __identity_base + identity mapping start (>= 2GB) | | | identity | phys == virt - __identity_base | mapping | virt == phys + __identity_base | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | +---- vmemmap -----+ 'struct page' array start | | | virtually mapped | | memory map | | | +- __abs_lowcore --+ | | | Absolute Lowcore | | | +- __memcpy_real_area | | | Real Memory Copy| | | +- VMALLOC_START --+ vmalloc area start | | | vmalloc area | | | +- MODULES_VADDR --+ modules area start | | | modules area | | | +- __kaslr_offset -+ kernel rand. virt start | | | kernel text/data | phys == (kvirt - __kaslr_offset) + | | __kaslr_offset_phys +- kernel .bss end + kernel rand. virt end | | | ... unused gap | | | +------------------+ UltraVisor Secure Storage limit | | | ... unused gap | | | +KASAN_SHADOW_START+ KASAN shadow memory start | | | KASAN shadow | | | +------------------+ ASCE limit Unused gaps in the virtual memory layout could be present or not - depending on how partucular system is configured. No page tables are created for the unused gaps. The relative order of vmalloc, modules and kernel image in virtual memory is defined by following considerations: - start of the modules area and end of the kernel should reside within 4GB to accommodate relative 32-bit jumps. The best way to achieve that is to place kernel next to modules; - vmalloc and module areas should locate next to each other to prevent failures and extra reworks in user level tools (makedumpfile, crash, etc.) which treat vmalloc and module addresses similarily; - kernel needs to be the last area in the virtual memory layout to easily distinguish between kernel and non-kernel virtual addresses. That is needed to (again) simplify handling of addresses in user level tools and make __pa() macro faster (see below); Concluding the above, the relative order of the considered virtual areas in memory is: vmalloc - modules - kernel. Therefore, the only change to the current memory layout is moving kernel to the end of virtual address space. With that approach the implementation of __pa() macro is straightforward - all linear virtual addresses less than kernel base are considered identity mapping: phys == virt - __identity_base All addresses greater than kernel base are kernel ones: phys == (kvirt - __kaslr_offset) + __kaslr_offset_phys By contrast, __va() macro deals only with identity mapping addresses: virt == phys + __identity_base .amode31 section is mapped separately and is not covered by __pa() macro. In fact, it could have been handled easily by checking whether a virtual address is within the section or not, but there is no need for that. Thus, let __pa() code do as little machine cycles as possible. The KASAN shadow memory is located at the very end of the virtual memory layout, at addresses higher than the kernel. However, that is not a linear mapping and no code other than KASAN instrumentation or API is expected to access it. When KASLR mode is enabled the kernel base address randomized within a memory window that spans whole unused virtual address space. The size of that window depends from the amount of physical memory available to the system, the limit imposed by UltraVisor (if present) and the vmalloc area size as provided by vmalloc= kernel command line parameter. In case the virtual memory is exhausted the minimum size of the randomization window is forcefully set to 2GB, which amounts to in 15 bits of entropy if KASAN is enabled or 17 bits of entropy in default configuration. The default kernel offset 0x100000 is used as a magic value both in the decompressor code and vmlinux linker script, but it will be removed with a follow-up change. Acked-by: Heiko Carstens <hca@linux.ibm.com> Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com>
2024-03-01 07:15:22 +01:00
s390/mm: Uncouple physical vs virtual address spaces The uncoupling physical vs virtual address spaces brings the following benefits to s390: - virtual memory layout flexibility; - closes the address gap between kernel and modules, it caused s390-only problems in the past (e.g. 'perf' bugs); - allows getting rid of trampolines used for module calls into kernel; - allows simplifying BPF trampoline; - minor performance improvement in branch prediction; - kernel randomization entropy is magnitude bigger, as it is derived from the amount of available virtual, not physical memory; The whole change could be described in two pictures below: before and after the change. Some aspects of the virtual memory layout setup are not clarified (number of page levels, alignment, DMA memory), since these are not a part of this change or secondary with regard to how the uncoupling itself is implemented. The focus of the pictures is to explain why __va() and __pa() macros are implemented the way they are. Memory layout in V==R mode: | Physical | Virtual | +- 0 --------------+- 0 --------------+ identity mapping start | | S390_lowcore | Low-address memory | +- 8 KB -----------+ | | | | | identity | phys == virt | | mapping | virt == phys | | | +- AMODE31_START --+- AMODE31_START --+ .amode31 rand. phys/virt start |.amode31 text/data|.amode31 text/data| +- AMODE31_END ----+- AMODE31_END ----+ .amode31 rand. phys/virt start | | | | | | +- __kaslr_offset, __kaslr_offset_phys| kernel rand. phys/virt start | | | | kernel text/data | kernel text/data | phys == kvirt | | | +------------------+------------------+ kernel phys/virt end | | | | | | | | | | | | +- ident_map_size -+- ident_map_size -+ identity mapping end | | | ... unused gap | | | +---- vmemmap -----+ 'struct page' array start | | | virtually mapped | | memory map | | | +- __abs_lowcore --+ | | | Absolute Lowcore | | | +- __memcpy_real_area | | | Real Memory Copy| | | +- VMALLOC_START --+ vmalloc area start | | | vmalloc area | | | +- MODULES_VADDR --+ modules area start | | | modules area | | | +------------------+ UltraVisor Secure Storage limit | | | ... unused gap | | | +KASAN_SHADOW_START+ KASAN shadow memory start | | | KASAN shadow | | | +------------------+ ASCE limit Memory layout in V!=R mode: | Physical | Virtual | +- 0 --------------+- 0 --------------+ | | S390_lowcore | Low-address memory | +- 8 KB -----------+ | | | | | | | | ... unused gap | | | | +- AMODE31_START --+- AMODE31_START --+ .amode31 rand. phys/virt start |.amode31 text/data|.amode31 text/data| +- AMODE31_END ----+- AMODE31_END ----+ .amode31 rand. phys/virt end (<2GB) | | | | | | +- __kaslr_offset_phys | kernel rand. phys start | | | | kernel text/data | | | | | +------------------+ | kernel phys end | | | | | | | | | | | | +- ident_map_size -+ | | | | ... unused gap | | | +- __identity_base + identity mapping start (>= 2GB) | | | identity | phys == virt - __identity_base | mapping | virt == phys + __identity_base | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | +---- vmemmap -----+ 'struct page' array start | | | virtually mapped | | memory map | | | +- __abs_lowcore --+ | | | Absolute Lowcore | | | +- __memcpy_real_area | | | Real Memory Copy| | | +- VMALLOC_START --+ vmalloc area start | | | vmalloc area | | | +- MODULES_VADDR --+ modules area start | | | modules area | | | +- __kaslr_offset -+ kernel rand. virt start | | | kernel text/data | phys == (kvirt - __kaslr_offset) + | | __kaslr_offset_phys +- kernel .bss end + kernel rand. virt end | | | ... unused gap | | | +------------------+ UltraVisor Secure Storage limit | | | ... unused gap | | | +KASAN_SHADOW_START+ KASAN shadow memory start | | | KASAN shadow | | | +------------------+ ASCE limit Unused gaps in the virtual memory layout could be present or not - depending on how partucular system is configured. No page tables are created for the unused gaps. The relative order of vmalloc, modules and kernel image in virtual memory is defined by following considerations: - start of the modules area and end of the kernel should reside within 4GB to accommodate relative 32-bit jumps. The best way to achieve that is to place kernel next to modules; - vmalloc and module areas should locate next to each other to prevent failures and extra reworks in user level tools (makedumpfile, crash, etc.) which treat vmalloc and module addresses similarily; - kernel needs to be the last area in the virtual memory layout to easily distinguish between kernel and non-kernel virtual addresses. That is needed to (again) simplify handling of addresses in user level tools and make __pa() macro faster (see below); Concluding the above, the relative order of the considered virtual areas in memory is: vmalloc - modules - kernel. Therefore, the only change to the current memory layout is moving kernel to the end of virtual address space. With that approach the implementation of __pa() macro is straightforward - all linear virtual addresses less than kernel base are considered identity mapping: phys == virt - __identity_base All addresses greater than kernel base are kernel ones: phys == (kvirt - __kaslr_offset) + __kaslr_offset_phys By contrast, __va() macro deals only with identity mapping addresses: virt == phys + __identity_base .amode31 section is mapped separately and is not covered by __pa() macro. In fact, it could have been handled easily by checking whether a virtual address is within the section or not, but there is no need for that. Thus, let __pa() code do as little machine cycles as possible. The KASAN shadow memory is located at the very end of the virtual memory layout, at addresses higher than the kernel. However, that is not a linear mapping and no code other than KASAN instrumentation or API is expected to access it. When KASLR mode is enabled the kernel base address randomized within a memory window that spans whole unused virtual address space. The size of that window depends from the amount of physical memory available to the system, the limit imposed by UltraVisor (if present) and the vmalloc area size as provided by vmalloc= kernel command line parameter. In case the virtual memory is exhausted the minimum size of the randomization window is forcefully set to 2GB, which amounts to in 15 bits of entropy if KASAN is enabled or 17 bits of entropy in default configuration. The default kernel offset 0x100000 is used as a magic value both in the decompressor code and vmlinux linker script, but it will be removed with a follow-up change. Acked-by: Heiko Carstens <hca@linux.ibm.com> Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com>
2024-03-01 07:15:22 +01:00
s390/mm: Uncouple physical vs virtual address spaces The uncoupling physical vs virtual address spaces brings the following benefits to s390: - virtual memory layout flexibility; - closes the address gap between kernel and modules, it caused s390-only problems in the past (e.g. 'perf' bugs); - allows getting rid of trampolines used for module calls into kernel; - allows simplifying BPF trampoline; - minor performance improvement in branch prediction; - kernel randomization entropy is magnitude bigger, as it is derived from the amount of available virtual, not physical memory; The whole change could be described in two pictures below: before and after the change. Some aspects of the virtual memory layout setup are not clarified (number of page levels, alignment, DMA memory), since these are not a part of this change or secondary with regard to how the uncoupling itself is implemented. The focus of the pictures is to explain why __va() and __pa() macros are implemented the way they are. Memory layout in V==R mode: | Physical | Virtual | +- 0 --------------+- 0 --------------+ identity mapping start | | S390_lowcore | Low-address memory | +- 8 KB -----------+ | | | | | identity | phys == virt | | mapping | virt == phys | | | +- AMODE31_START --+- AMODE31_START --+ .amode31 rand. phys/virt start |.amode31 text/data|.amode31 text/data| +- AMODE31_END ----+- AMODE31_END ----+ .amode31 rand. phys/virt start | | | | | | +- __kaslr_offset, __kaslr_offset_phys| kernel rand. phys/virt start | | | | kernel text/data | kernel text/data | phys == kvirt | | | +------------------+------------------+ kernel phys/virt end | | | | | | | | | | | | +- ident_map_size -+- ident_map_size -+ identity mapping end | | | ... unused gap | | | +---- vmemmap -----+ 'struct page' array start | | | virtually mapped | | memory map | | | +- __abs_lowcore --+ | | | Absolute Lowcore | | | +- __memcpy_real_area | | | Real Memory Copy| | | +- VMALLOC_START --+ vmalloc area start | | | vmalloc area | | | +- MODULES_VADDR --+ modules area start | | | modules area | | | +------------------+ UltraVisor Secure Storage limit | | | ... unused gap | | | +KASAN_SHADOW_START+ KASAN shadow memory start | | | KASAN shadow | | | +------------------+ ASCE limit Memory layout in V!=R mode: | Physical | Virtual | +- 0 --------------+- 0 --------------+ | | S390_lowcore | Low-address memory | +- 8 KB -----------+ | | | | | | | | ... unused gap | | | | +- AMODE31_START --+- AMODE31_START --+ .amode31 rand. phys/virt start |.amode31 text/data|.amode31 text/data| +- AMODE31_END ----+- AMODE31_END ----+ .amode31 rand. phys/virt end (<2GB) | | | | | | +- __kaslr_offset_phys | kernel rand. phys start | | | | kernel text/data | | | | | +------------------+ | kernel phys end | | | | | | | | | | | | +- ident_map_size -+ | | | | ... unused gap | | | +- __identity_base + identity mapping start (>= 2GB) | | | identity | phys == virt - __identity_base | mapping | virt == phys + __identity_base | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | +---- vmemmap -----+ 'struct page' array start | | | virtually mapped | | memory map | | | +- __abs_lowcore --+ | | | Absolute Lowcore | | | +- __memcpy_real_area | | | Real Memory Copy| | | +- VMALLOC_START --+ vmalloc area start | | | vmalloc area | | | +- MODULES_VADDR --+ modules area start | | | modules area | | | +- __kaslr_offset -+ kernel rand. virt start | | | kernel text/data | phys == (kvirt - __kaslr_offset) + | | __kaslr_offset_phys +- kernel .bss end + kernel rand. virt end | | | ... unused gap | | | +------------------+ UltraVisor Secure Storage limit | | | ... unused gap | | | +KASAN_SHADOW_START+ KASAN shadow memory start | | | KASAN shadow | | | +------------------+ ASCE limit Unused gaps in the virtual memory layout could be present or not - depending on how partucular system is configured. No page tables are created for the unused gaps. The relative order of vmalloc, modules and kernel image in virtual memory is defined by following considerations: - start of the modules area and end of the kernel should reside within 4GB to accommodate relative 32-bit jumps. The best way to achieve that is to place kernel next to modules; - vmalloc and module areas should locate next to each other to prevent failures and extra reworks in user level tools (makedumpfile, crash, etc.) which treat vmalloc and module addresses similarily; - kernel needs to be the last area in the virtual memory layout to easily distinguish between kernel and non-kernel virtual addresses. That is needed to (again) simplify handling of addresses in user level tools and make __pa() macro faster (see below); Concluding the above, the relative order of the considered virtual areas in memory is: vmalloc - modules - kernel. Therefore, the only change to the current memory layout is moving kernel to the end of virtual address space. With that approach the implementation of __pa() macro is straightforward - all linear virtual addresses less than kernel base are considered identity mapping: phys == virt - __identity_base All addresses greater than kernel base are kernel ones: phys == (kvirt - __kaslr_offset) + __kaslr_offset_phys By contrast, __va() macro deals only with identity mapping addresses: virt == phys + __identity_base .amode31 section is mapped separately and is not covered by __pa() macro. In fact, it could have been handled easily by checking whether a virtual address is within the section or not, but there is no need for that. Thus, let __pa() code do as little machine cycles as possible. The KASAN shadow memory is located at the very end of the virtual memory layout, at addresses higher than the kernel. However, that is not a linear mapping and no code other than KASAN instrumentation or API is expected to access it. When KASLR mode is enabled the kernel base address randomized within a memory window that spans whole unused virtual address space. The size of that window depends from the amount of physical memory available to the system, the limit imposed by UltraVisor (if present) and the vmalloc area size as provided by vmalloc= kernel command line parameter. In case the virtual memory is exhausted the minimum size of the randomization window is forcefully set to 2GB, which amounts to in 15 bits of entropy if KASAN is enabled or 17 bits of entropy in default configuration. The default kernel offset 0x100000 is used as a magic value both in the decompressor code and vmlinux linker script, but it will be removed with a follow-up change. Acked-by: Heiko Carstens <hca@linux.ibm.com> Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com>
2024-03-01 07:15:22 +01:00
s390/mm: define KMSAN metadata for vmalloc and modules The pages for the KMSAN metadata associated with most kernel mappings are taken from memblock by the common code. However, vmalloc and module metadata needs to be defined by the architectures. Be a little bit more careful than x86: allocate exactly MODULES_LEN for the module shadow and origins, and then take 2/3 of vmalloc for the vmalloc shadow and origins. This ensures that users passing small vmalloc= values on the command line do not cause module metadata collisions. Link: https://lkml.kernel.org/r/20240621113706.315500-32-iii@linux.ibm.com Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com> Reviewed-by: Alexander Potapenko <glider@google.com> Acked-by: Alexander Gordeev <agordeev@linux.ibm.com> Acked-by: Heiko Carstens <hca@linux.ibm.com> Cc: Christian Borntraeger <borntraeger@linux.ibm.com> Cc: Christoph Lameter <cl@linux.com> Cc: David Rientjes <rientjes@google.com> Cc: Dmitry Vyukov <dvyukov@google.com> Cc: Hyeonggon Yoo <42.hyeyoo@gmail.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: <kasan-dev@googlegroups.com> Cc: Marco Elver <elver@google.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Masami Hiramatsu (Google) <mhiramat@kernel.org> Cc: Pekka Enberg <penberg@kernel.org> Cc: Roman Gushchin <roman.gushchin@linux.dev> Cc: Steven Rostedt (Google) <rostedt@goodmis.org> Cc: Sven Schnelle <svens@linux.ibm.com> Cc: Vasily Gorbik <gor@linux.ibm.com> Cc: Vlastimil Babka <vbabka@suse.cz> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2024-06-21 13:35:15 +02:00
s390/mm: Uncouple physical vs virtual address spaces The uncoupling physical vs virtual address spaces brings the following benefits to s390: - virtual memory layout flexibility; - closes the address gap between kernel and modules, it caused s390-only problems in the past (e.g. 'perf' bugs); - allows getting rid of trampolines used for module calls into kernel; - allows simplifying BPF trampoline; - minor performance improvement in branch prediction; - kernel randomization entropy is magnitude bigger, as it is derived from the amount of available virtual, not physical memory; The whole change could be described in two pictures below: before and after the change. Some aspects of the virtual memory layout setup are not clarified (number of page levels, alignment, DMA memory), since these are not a part of this change or secondary with regard to how the uncoupling itself is implemented. The focus of the pictures is to explain why __va() and __pa() macros are implemented the way they are. Memory layout in V==R mode: | Physical | Virtual | +- 0 --------------+- 0 --------------+ identity mapping start | | S390_lowcore | Low-address memory | +- 8 KB -----------+ | | | | | identity | phys == virt | | mapping | virt == phys | | | +- AMODE31_START --+- AMODE31_START --+ .amode31 rand. phys/virt start |.amode31 text/data|.amode31 text/data| +- AMODE31_END ----+- AMODE31_END ----+ .amode31 rand. phys/virt start | | | | | | +- __kaslr_offset, __kaslr_offset_phys| kernel rand. phys/virt start | | | | kernel text/data | kernel text/data | phys == kvirt | | | +------------------+------------------+ kernel phys/virt end | | | | | | | | | | | | +- ident_map_size -+- ident_map_size -+ identity mapping end | | | ... unused gap | | | +---- vmemmap -----+ 'struct page' array start | | | virtually mapped | | memory map | | | +- __abs_lowcore --+ | | | Absolute Lowcore | | | +- __memcpy_real_area | | | Real Memory Copy| | | +- VMALLOC_START --+ vmalloc area start | | | vmalloc area | | | +- MODULES_VADDR --+ modules area start | | | modules area | | | +------------------+ UltraVisor Secure Storage limit | | | ... unused gap | | | +KASAN_SHADOW_START+ KASAN shadow memory start | | | KASAN shadow | | | +------------------+ ASCE limit Memory layout in V!=R mode: | Physical | Virtual | +- 0 --------------+- 0 --------------+ | | S390_lowcore | Low-address memory | +- 8 KB -----------+ | | | | | | | | ... unused gap | | | | +- AMODE31_START --+- AMODE31_START --+ .amode31 rand. phys/virt start |.amode31 text/data|.amode31 text/data| +- AMODE31_END ----+- AMODE31_END ----+ .amode31 rand. phys/virt end (<2GB) | | | | | | +- __kaslr_offset_phys | kernel rand. phys start | | | | kernel text/data | | | | | +------------------+ | kernel phys end | | | | | | | | | | | | +- ident_map_size -+ | | | | ... unused gap | | | +- __identity_base + identity mapping start (>= 2GB) | | | identity | phys == virt - __identity_base | mapping | virt == phys + __identity_base | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | +---- vmemmap -----+ 'struct page' array start | | | virtually mapped | | memory map | | | +- __abs_lowcore --+ | | | Absolute Lowcore | | | +- __memcpy_real_area | | | Real Memory Copy| | | +- VMALLOC_START --+ vmalloc area start | | | vmalloc area | | | +- MODULES_VADDR --+ modules area start | | | modules area | | | +- __kaslr_offset -+ kernel rand. virt start | | | kernel text/data | phys == (kvirt - __kaslr_offset) + | | __kaslr_offset_phys +- kernel .bss end + kernel rand. virt end | | | ... unused gap | | | +------------------+ UltraVisor Secure Storage limit | | | ... unused gap | | | +KASAN_SHADOW_START+ KASAN shadow memory start | | | KASAN shadow | | | +------------------+ ASCE limit Unused gaps in the virtual memory layout could be present or not - depending on how partucular system is configured. No page tables are created for the unused gaps. The relative order of vmalloc, modules and kernel image in virtual memory is defined by following considerations: - start of the modules area and end of the kernel should reside within 4GB to accommodate relative 32-bit jumps. The best way to achieve that is to place kernel next to modules; - vmalloc and module areas should locate next to each other to prevent failures and extra reworks in user level tools (makedumpfile, crash, etc.) which treat vmalloc and module addresses similarily; - kernel needs to be the last area in the virtual memory layout to easily distinguish between kernel and non-kernel virtual addresses. That is needed to (again) simplify handling of addresses in user level tools and make __pa() macro faster (see below); Concluding the above, the relative order of the considered virtual areas in memory is: vmalloc - modules - kernel. Therefore, the only change to the current memory layout is moving kernel to the end of virtual address space. With that approach the implementation of __pa() macro is straightforward - all linear virtual addresses less than kernel base are considered identity mapping: phys == virt - __identity_base All addresses greater than kernel base are kernel ones: phys == (kvirt - __kaslr_offset) + __kaslr_offset_phys By contrast, __va() macro deals only with identity mapping addresses: virt == phys + __identity_base .amode31 section is mapped separately and is not covered by __pa() macro. In fact, it could have been handled easily by checking whether a virtual address is within the section or not, but there is no need for that. Thus, let __pa() code do as little machine cycles as possible. The KASAN shadow memory is located at the very end of the virtual memory layout, at addresses higher than the kernel. However, that is not a linear mapping and no code other than KASAN instrumentation or API is expected to access it. When KASLR mode is enabled the kernel base address randomized within a memory window that spans whole unused virtual address space. The size of that window depends from the amount of physical memory available to the system, the limit imposed by UltraVisor (if present) and the vmalloc area size as provided by vmalloc= kernel command line parameter. In case the virtual memory is exhausted the minimum size of the randomization window is forcefully set to 2GB, which amounts to in 15 bits of entropy if KASAN is enabled or 17 bits of entropy in default configuration. The default kernel offset 0x100000 is used as a magic value both in the decompressor code and vmlinux linker script, but it will be removed with a follow-up change. Acked-by: Heiko Carstens <hca@linux.ibm.com> Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com>
2024-03-01 07:15:22 +01:00
s390/mm: define KMSAN metadata for vmalloc and modules The pages for the KMSAN metadata associated with most kernel mappings are taken from memblock by the common code. However, vmalloc and module metadata needs to be defined by the architectures. Be a little bit more careful than x86: allocate exactly MODULES_LEN for the module shadow and origins, and then take 2/3 of vmalloc for the vmalloc shadow and origins. This ensures that users passing small vmalloc= values on the command line do not cause module metadata collisions. Link: https://lkml.kernel.org/r/20240621113706.315500-32-iii@linux.ibm.com Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com> Reviewed-by: Alexander Potapenko <glider@google.com> Acked-by: Alexander Gordeev <agordeev@linux.ibm.com> Acked-by: Heiko Carstens <hca@linux.ibm.com> Cc: Christian Borntraeger <borntraeger@linux.ibm.com> Cc: Christoph Lameter <cl@linux.com> Cc: David Rientjes <rientjes@google.com> Cc: Dmitry Vyukov <dvyukov@google.com> Cc: Hyeonggon Yoo <42.hyeyoo@gmail.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: <kasan-dev@googlegroups.com> Cc: Marco Elver <elver@google.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Masami Hiramatsu (Google) <mhiramat@kernel.org> Cc: Pekka Enberg <penberg@kernel.org> Cc: Roman Gushchin <roman.gushchin@linux.dev> Cc: Steven Rostedt (Google) <rostedt@goodmis.org> Cc: Sven Schnelle <svens@linux.ibm.com> Cc: Vasily Gorbik <gor@linux.ibm.com> Cc: Vlastimil Babka <vbabka@suse.cz> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2024-06-21 13:35:15 +02:00
s390/mm: start kernel with DAT enabled The setup of the kernel virtual address space is spread throughout the sources, boot stages and config options like this: 1. The available physical memory regions are queried and stored as mem_detect information for later use in the decompressor. 2. Based on the physical memory availability the virtual memory layout is established in the decompressor; 3. If CONFIG_KASAN is disabled the kernel paging setup code populates kernel pgtables and turns DAT mode on. It uses the information stored at step [1]. 4. If CONFIG_KASAN is enabled the kernel early boot kasan setup populates kernel pgtables and turns DAT mode on. It uses the information stored at step [1]. The kasan setup creates early_pg_dir directory and directly overwrites swapper_pg_dir entries to make shadow memory pages available. Move the kernel virtual memory setup to the decompressor and start the kernel with DAT turned on right from the very first istruction. That completely eliminates the boot phase when the kernel runs in DAT-off mode, simplies the overall design and consolidates pgtables setup. The identity mapping is created in the decompressor, while kasan shadow mappings are still created by the early boot kernel code. Share with decompressor the existing kasan memory allocator. It decreases the size of a newly requested memory block from pgalloc_pos and ensures that kernel image is not overwritten. pgalloc_low and pgalloc_pos pointers are made preserved boot variables for that. Use the bootdata infrastructure to setup swapper_pg_dir and invalid_pg_dir directories used by the kernel later. The interim early_pg_dir directory established by the kasan initialization code gets eliminated as result. As the kernel runs in DAT-on mode only the PSW_KERNEL_BITS define gets PSW_MASK_DAT bit by default. Additionally, the setup_lowcore_dat_off() and setup_lowcore_dat_on() routines get merged, since there is no DAT-off mode stage anymore. The memory mappings are created with RW+X protection that allows the early boot code setting up all necessary data and services for the kernel being booted. Just before the paging is enabled the memory protection is changed to RO+X for text, RO+NX for read-only data and RW+NX for kernel data and the identity mapping. Reviewed-by: Heiko Carstens <hca@linux.ibm.com> Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com> Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2022-12-13 11:35:11 +01:00
s390/boot: Rework deployment of the kernel image Rework deployment of kernel image for both compressed and uncompressed variants as defined by CONFIG_KERNEL_UNCOMPRESSED kernel configuration variable. In case CONFIG_KERNEL_UNCOMPRESSED is disabled avoid uncompressing the kernel to a temporary buffer and copying it to the target address. Instead, uncompress it directly to the target destination. In case CONFIG_KERNEL_UNCOMPRESSED is enabled avoid moving the kernel to default 0x100000 location when KASLR is disabled or failed. Instead, use the uncompressed kernel image directly. In case KASLR is disabled or failed .amode31 section location in memory is not randomized and precedes the kernel image. In case CONFIG_KERNEL_UNCOMPRESSED is disabled that location overlaps the area used by the decompression algorithm. That is fine, since that area is not used after the decompression finished and the size of .amode31 section is not expected to exceed BOOT_HEAP_SIZE ever. There is no decompression in case CONFIG_KERNEL_UNCOMPRESSED is enabled. Therefore, rename decompress_kernel() to deploy_kernel(), which better describes both uncompressed and compressed cases. Introduce AMODE31_SIZE macro to avoid immediate value of 0x3000 (the size of .amode31 section) in the decompressor linker script. Modify the vmlinux linker script to force the size of .amode31 section to AMODE31_SIZE (the value of (_eamode31 - _samode31) could otherwise differ as result of compiler options used). Introduce __START_KERNEL macro that defines the kernel ELF image entry point and set it to the currrent value of 0x100000. Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com>
2024-03-22 14:39:57 +01:00
s390/boot: Rework deployment of the kernel image Rework deployment of kernel image for both compressed and uncompressed variants as defined by CONFIG_KERNEL_UNCOMPRESSED kernel configuration variable. In case CONFIG_KERNEL_UNCOMPRESSED is disabled avoid uncompressing the kernel to a temporary buffer and copying it to the target address. Instead, uncompress it directly to the target destination. In case CONFIG_KERNEL_UNCOMPRESSED is enabled avoid moving the kernel to default 0x100000 location when KASLR is disabled or failed. Instead, use the uncompressed kernel image directly. In case KASLR is disabled or failed .amode31 section location in memory is not randomized and precedes the kernel image. In case CONFIG_KERNEL_UNCOMPRESSED is disabled that location overlaps the area used by the decompression algorithm. That is fine, since that area is not used after the decompression finished and the size of .amode31 section is not expected to exceed BOOT_HEAP_SIZE ever. There is no decompression in case CONFIG_KERNEL_UNCOMPRESSED is enabled. Therefore, rename decompress_kernel() to deploy_kernel(), which better describes both uncompressed and compressed cases. Introduce AMODE31_SIZE macro to avoid immediate value of 0x3000 (the size of .amode31 section) in the decompressor linker script. Modify the vmlinux linker script to force the size of .amode31 section to AMODE31_SIZE (the value of (_eamode31 - _samode31) could otherwise differ as result of compiler options used). Introduce __START_KERNEL macro that defines the kernel ELF image entry point and set it to the currrent value of 0x100000. Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com>
2024-03-22 14:39:57 +01:00
s390/boot: Rework deployment of the kernel image Rework deployment of kernel image for both compressed and uncompressed variants as defined by CONFIG_KERNEL_UNCOMPRESSED kernel configuration variable. In case CONFIG_KERNEL_UNCOMPRESSED is disabled avoid uncompressing the kernel to a temporary buffer and copying it to the target address. Instead, uncompress it directly to the target destination. In case CONFIG_KERNEL_UNCOMPRESSED is enabled avoid moving the kernel to default 0x100000 location when KASLR is disabled or failed. Instead, use the uncompressed kernel image directly. In case KASLR is disabled or failed .amode31 section location in memory is not randomized and precedes the kernel image. In case CONFIG_KERNEL_UNCOMPRESSED is disabled that location overlaps the area used by the decompression algorithm. That is fine, since that area is not used after the decompression finished and the size of .amode31 section is not expected to exceed BOOT_HEAP_SIZE ever. There is no decompression in case CONFIG_KERNEL_UNCOMPRESSED is enabled. Therefore, rename decompress_kernel() to deploy_kernel(), which better describes both uncompressed and compressed cases. Introduce AMODE31_SIZE macro to avoid immediate value of 0x3000 (the size of .amode31 section) in the decompressor linker script. Modify the vmlinux linker script to force the size of .amode31 section to AMODE31_SIZE (the value of (_eamode31 - _samode31) could otherwise differ as result of compiler options used). Introduce __START_KERNEL macro that defines the kernel ELF image entry point and set it to the currrent value of 0x100000. Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com>
2024-03-22 14:39:57 +01:00
s390/mm: start kernel with DAT enabled The setup of the kernel virtual address space is spread throughout the sources, boot stages and config options like this: 1. The available physical memory regions are queried and stored as mem_detect information for later use in the decompressor. 2. Based on the physical memory availability the virtual memory layout is established in the decompressor; 3. If CONFIG_KASAN is disabled the kernel paging setup code populates kernel pgtables and turns DAT mode on. It uses the information stored at step [1]. 4. If CONFIG_KASAN is enabled the kernel early boot kasan setup populates kernel pgtables and turns DAT mode on. It uses the information stored at step [1]. The kasan setup creates early_pg_dir directory and directly overwrites swapper_pg_dir entries to make shadow memory pages available. Move the kernel virtual memory setup to the decompressor and start the kernel with DAT turned on right from the very first istruction. That completely eliminates the boot phase when the kernel runs in DAT-off mode, simplies the overall design and consolidates pgtables setup. The identity mapping is created in the decompressor, while kasan shadow mappings are still created by the early boot kernel code. Share with decompressor the existing kasan memory allocator. It decreases the size of a newly requested memory block from pgalloc_pos and ensures that kernel image is not overwritten. pgalloc_low and pgalloc_pos pointers are made preserved boot variables for that. Use the bootdata infrastructure to setup swapper_pg_dir and invalid_pg_dir directories used by the kernel later. The interim early_pg_dir directory established by the kasan initialization code gets eliminated as result. As the kernel runs in DAT-on mode only the PSW_KERNEL_BITS define gets PSW_MASK_DAT bit by default. Additionally, the setup_lowcore_dat_off() and setup_lowcore_dat_on() routines get merged, since there is no DAT-off mode stage anymore. The memory mappings are created with RW+X protection that allows the early boot code setting up all necessary data and services for the kernel being booted. Just before the paging is enabled the memory protection is changed to RO+X for text, RO+NX for read-only data and RW+NX for kernel data and the identity mapping. Reviewed-by: Heiko Carstens <hca@linux.ibm.com> Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com> Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2022-12-13 11:35:11 +01:00
s390: Add stackprotector support Stackprotector support was previously unavailable on s390 because by default compilers generate code which is not suitable for the kernel: the canary value is accessed via thread local storage, where the address of thread local storage is within access registers 0 and 1. Using those registers also for the kernel would come with a significant performance impact and more complicated kernel entry/exit code, since access registers contents would have to be exchanged on every kernel entry and exit. With the upcoming gcc 16 release new compiler options will become available which allow to generate code suitable for the kernel. [1] Compiler option -mstack-protector-guard=global instructs gcc to generate stackprotector code that refers to a global stackprotector canary value via symbol __stack_chk_guard. Access to this value is guaranteed to occur via larl and lgrl instructions. Furthermore, compiler option -mstack-protector-guard-record generates a section containing all code addresses that reference the canary value. To allow for per task canary values the instructions which load the address of __stack_chk_guard are patched so they access a lowcore field instead: a per task canary value is available within the task_struct of each task, and is written to the per-cpu lowcore location on each context switch. Also add sanity checks and debugging option to be consistent with other kernel code patching mechanisms. Full debugging output can be enabled with the following kernel command line options: debug_stackprotector bootdebug ignore_loglevel earlyprintk dyndbg="file stackprotector.c +p" Example debug output: stackprot: 0000021e402d4eda: c010005a9ae3 -> c01f00070240 where "<insn address>: <old insn> -> <new insn>". [1] gcc commit 0cd1f03939d5 ("s390: Support global stack protector") Reviewed-by: Sven Schnelle <svens@linux.ibm.com> Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2025-11-17 15:09:53 +01:00
s390/boot: Rework deployment of the kernel image Rework deployment of kernel image for both compressed and uncompressed variants as defined by CONFIG_KERNEL_UNCOMPRESSED kernel configuration variable. In case CONFIG_KERNEL_UNCOMPRESSED is disabled avoid uncompressing the kernel to a temporary buffer and copying it to the target address. Instead, uncompress it directly to the target destination. In case CONFIG_KERNEL_UNCOMPRESSED is enabled avoid moving the kernel to default 0x100000 location when KASLR is disabled or failed. Instead, use the uncompressed kernel image directly. In case KASLR is disabled or failed .amode31 section location in memory is not randomized and precedes the kernel image. In case CONFIG_KERNEL_UNCOMPRESSED is disabled that location overlaps the area used by the decompression algorithm. That is fine, since that area is not used after the decompression finished and the size of .amode31 section is not expected to exceed BOOT_HEAP_SIZE ever. There is no decompression in case CONFIG_KERNEL_UNCOMPRESSED is enabled. Therefore, rename decompress_kernel() to deploy_kernel(), which better describes both uncompressed and compressed cases. Introduce AMODE31_SIZE macro to avoid immediate value of 0x3000 (the size of .amode31 section) in the decompressor linker script. Modify the vmlinux linker script to force the size of .amode31 section to AMODE31_SIZE (the value of (_eamode31 - _samode31) could otherwise differ as result of compiler options used). Introduce __START_KERNEL macro that defines the kernel ELF image entry point and set it to the currrent value of 0x100000. Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com>
2024-03-22 14:39:57 +01:00
s390/mm: start kernel with DAT enabled The setup of the kernel virtual address space is spread throughout the sources, boot stages and config options like this: 1. The available physical memory regions are queried and stored as mem_detect information for later use in the decompressor. 2. Based on the physical memory availability the virtual memory layout is established in the decompressor; 3. If CONFIG_KASAN is disabled the kernel paging setup code populates kernel pgtables and turns DAT mode on. It uses the information stored at step [1]. 4. If CONFIG_KASAN is enabled the kernel early boot kasan setup populates kernel pgtables and turns DAT mode on. It uses the information stored at step [1]. The kasan setup creates early_pg_dir directory and directly overwrites swapper_pg_dir entries to make shadow memory pages available. Move the kernel virtual memory setup to the decompressor and start the kernel with DAT turned on right from the very first istruction. That completely eliminates the boot phase when the kernel runs in DAT-off mode, simplies the overall design and consolidates pgtables setup. The identity mapping is created in the decompressor, while kasan shadow mappings are still created by the early boot kernel code. Share with decompressor the existing kasan memory allocator. It decreases the size of a newly requested memory block from pgalloc_pos and ensures that kernel image is not overwritten. pgalloc_low and pgalloc_pos pointers are made preserved boot variables for that. Use the bootdata infrastructure to setup swapper_pg_dir and invalid_pg_dir directories used by the kernel later. The interim early_pg_dir directory established by the kasan initialization code gets eliminated as result. As the kernel runs in DAT-on mode only the PSW_KERNEL_BITS define gets PSW_MASK_DAT bit by default. Additionally, the setup_lowcore_dat_off() and setup_lowcore_dat_on() routines get merged, since there is no DAT-off mode stage anymore. The memory mappings are created with RW+X protection that allows the early boot code setting up all necessary data and services for the kernel being booted. Just before the paging is enabled the memory protection is changed to RO+X for text, RO+NX for read-only data and RW+NX for kernel data and the identity mapping. Reviewed-by: Heiko Carstens <hca@linux.ibm.com> Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com> Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2022-12-13 11:35:11 +01:00
s390/boot: rework decompressor reserved tracking Currently several approaches for finding unused memory in decompressor are utilized. While "safe_addr" grows towards higher addresses, vmem code allocates paging structures top down. The former requires careful ordering. In addition to that ipl report handling code verifies potential intersections with secure boot certificates on its own. Neither of two approaches are memory holes aware and consistent with each other in low memory conditions. To solve that, existing approaches are generalized and combined together, as well as online memory ranges are now taken into consideration. physmem_info has been extended to contain reserved memory ranges. New set of functions allow to handle reserves and find unused memory. All reserves and memory allocations are "typed". In case of out of memory condition decompressor fails with detailed info on current reserved ranges and usable online memory. Linux version 6.2.0 ... Kernel command line: ... mem=100M Our of memory allocating 100000 bytes 100000 aligned in range 0:5800000 Reserved memory ranges: 0000000000000000 0000000003e33000 DECOMPRESSOR 0000000003f00000 00000000057648a3 INITRD 00000000063e0000 00000000063e8000 VMEM 00000000063eb000 00000000063f4000 VMEM 00000000063f7800 0000000006400000 VMEM 0000000005800000 0000000006300000 KASAN Usable online memory ranges (info source: sclp read info [3]): 0000000000000000 0000000006400000 Usable online memory total: 6400000 Reserved: 61b10a3 Free: 24ef5d Call Trace: (sp:000000000002bd58 [<0000000000012a70>] physmem_alloc_top_down+0x60/0x14c) sp:000000000002bdc8 [<0000000000013756>] _pa+0x56/0x6a sp:000000000002bdf0 [<0000000000013bcc>] pgtable_populate+0x45c/0x65e sp:000000000002be90 [<00000000000140aa>] setup_vmem+0x2da/0x424 sp:000000000002bec8 [<0000000000011c20>] startup_kernel+0x428/0x8b4 sp:000000000002bf60 [<00000000000100f4>] startup_normal+0xd4/0xd4 physmem_alloc_range allows to find free memory in specified range. It should be used for one time allocations only like finding position for amode31 and vmlinux. physmem_alloc_top_down can be used just like physmem_alloc_range, but it also allows multiple allocations per type and tries to merge sequential allocations together. Which is useful for paging structures allocations. If sequential allocations cannot be merged together they are "chained", allowing easy per type reserved ranges enumeration and migration to memblock later. Extra "struct reserved_range" allocated for chaining are not tracked or reserved but rely on the fact that both physmem_alloc_range and physmem_alloc_top_down search for free memory only below current top down allocator position. All reserved ranges should be transferred to memblock before memblock allocations are enabled. The startup code has been reordered to delay any memory allocations until online memory ranges are detected and occupied memory ranges are marked as reserved to be excluded from follow-up allocations. Ipl report certificates are a special case, ipl report certificates list is checked together with other memory reserves until certificates are saved elsewhere. KASAN required memory for shadow memory allocation and mapping is reserved as 1 large chunk which is later passed to KASAN early initialization code. Acked-by: Heiko Carstens <hca@linux.ibm.com> Reviewed-by: Alexander Gordeev <agordeev@linux.ibm.com> Signed-off-by: Vasily Gorbik <gor@linux.ibm.com> Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2023-02-02 13:59:36 +01:00
s390/mm: start kernel with DAT enabled The setup of the kernel virtual address space is spread throughout the sources, boot stages and config options like this: 1. The available physical memory regions are queried and stored as mem_detect information for later use in the decompressor. 2. Based on the physical memory availability the virtual memory layout is established in the decompressor; 3. If CONFIG_KASAN is disabled the kernel paging setup code populates kernel pgtables and turns DAT mode on. It uses the information stored at step [1]. 4. If CONFIG_KASAN is enabled the kernel early boot kasan setup populates kernel pgtables and turns DAT mode on. It uses the information stored at step [1]. The kasan setup creates early_pg_dir directory and directly overwrites swapper_pg_dir entries to make shadow memory pages available. Move the kernel virtual memory setup to the decompressor and start the kernel with DAT turned on right from the very first istruction. That completely eliminates the boot phase when the kernel runs in DAT-off mode, simplies the overall design and consolidates pgtables setup. The identity mapping is created in the decompressor, while kasan shadow mappings are still created by the early boot kernel code. Share with decompressor the existing kasan memory allocator. It decreases the size of a newly requested memory block from pgalloc_pos and ensures that kernel image is not overwritten. pgalloc_low and pgalloc_pos pointers are made preserved boot variables for that. Use the bootdata infrastructure to setup swapper_pg_dir and invalid_pg_dir directories used by the kernel later. The interim early_pg_dir directory established by the kasan initialization code gets eliminated as result. As the kernel runs in DAT-on mode only the PSW_KERNEL_BITS define gets PSW_MASK_DAT bit by default. Additionally, the setup_lowcore_dat_off() and setup_lowcore_dat_on() routines get merged, since there is no DAT-off mode stage anymore. The memory mappings are created with RW+X protection that allows the early boot code setting up all necessary data and services for the kernel being booted. Just before the paging is enabled the memory protection is changed to RO+X for text, RO+NX for read-only data and RW+NX for kernel data and the identity mapping. Reviewed-by: Heiko Carstens <hca@linux.ibm.com> Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com> Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2022-12-13 11:35:11 +01:00
s390/boot: rework decompressor reserved tracking Currently several approaches for finding unused memory in decompressor are utilized. While "safe_addr" grows towards higher addresses, vmem code allocates paging structures top down. The former requires careful ordering. In addition to that ipl report handling code verifies potential intersections with secure boot certificates on its own. Neither of two approaches are memory holes aware and consistent with each other in low memory conditions. To solve that, existing approaches are generalized and combined together, as well as online memory ranges are now taken into consideration. physmem_info has been extended to contain reserved memory ranges. New set of functions allow to handle reserves and find unused memory. All reserves and memory allocations are "typed". In case of out of memory condition decompressor fails with detailed info on current reserved ranges and usable online memory. Linux version 6.2.0 ... Kernel command line: ... mem=100M Our of memory allocating 100000 bytes 100000 aligned in range 0:5800000 Reserved memory ranges: 0000000000000000 0000000003e33000 DECOMPRESSOR 0000000003f00000 00000000057648a3 INITRD 00000000063e0000 00000000063e8000 VMEM 00000000063eb000 00000000063f4000 VMEM 00000000063f7800 0000000006400000 VMEM 0000000005800000 0000000006300000 KASAN Usable online memory ranges (info source: sclp read info [3]): 0000000000000000 0000000006400000 Usable online memory total: 6400000 Reserved: 61b10a3 Free: 24ef5d Call Trace: (sp:000000000002bd58 [<0000000000012a70>] physmem_alloc_top_down+0x60/0x14c) sp:000000000002bdc8 [<0000000000013756>] _pa+0x56/0x6a sp:000000000002bdf0 [<0000000000013bcc>] pgtable_populate+0x45c/0x65e sp:000000000002be90 [<00000000000140aa>] setup_vmem+0x2da/0x424 sp:000000000002bec8 [<0000000000011c20>] startup_kernel+0x428/0x8b4 sp:000000000002bf60 [<00000000000100f4>] startup_normal+0xd4/0xd4 physmem_alloc_range allows to find free memory in specified range. It should be used for one time allocations only like finding position for amode31 and vmlinux. physmem_alloc_top_down can be used just like physmem_alloc_range, but it also allows multiple allocations per type and tries to merge sequential allocations together. Which is useful for paging structures allocations. If sequential allocations cannot be merged together they are "chained", allowing easy per type reserved ranges enumeration and migration to memblock later. Extra "struct reserved_range" allocated for chaining are not tracked or reserved but rely on the fact that both physmem_alloc_range and physmem_alloc_top_down search for free memory only below current top down allocator position. All reserved ranges should be transferred to memblock before memblock allocations are enabled. The startup code has been reordered to delay any memory allocations until online memory ranges are detected and occupied memory ranges are marked as reserved to be excluded from follow-up allocations. Ipl report certificates are a special case, ipl report certificates list is checked together with other memory reserves until certificates are saved elsewhere. KASAN required memory for shadow memory allocation and mapping is reserved as 1 large chunk which is later passed to KASAN early initialization code. Acked-by: Heiko Carstens <hca@linux.ibm.com> Reviewed-by: Alexander Gordeev <agordeev@linux.ibm.com> Signed-off-by: Vasily Gorbik <gor@linux.ibm.com> Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2023-02-02 13:59:36 +01:00
s390/boot: rework decompressor reserved tracking Currently several approaches for finding unused memory in decompressor are utilized. While "safe_addr" grows towards higher addresses, vmem code allocates paging structures top down. The former requires careful ordering. In addition to that ipl report handling code verifies potential intersections with secure boot certificates on its own. Neither of two approaches are memory holes aware and consistent with each other in low memory conditions. To solve that, existing approaches are generalized and combined together, as well as online memory ranges are now taken into consideration. physmem_info has been extended to contain reserved memory ranges. New set of functions allow to handle reserves and find unused memory. All reserves and memory allocations are "typed". In case of out of memory condition decompressor fails with detailed info on current reserved ranges and usable online memory. Linux version 6.2.0 ... Kernel command line: ... mem=100M Our of memory allocating 100000 bytes 100000 aligned in range 0:5800000 Reserved memory ranges: 0000000000000000 0000000003e33000 DECOMPRESSOR 0000000003f00000 00000000057648a3 INITRD 00000000063e0000 00000000063e8000 VMEM 00000000063eb000 00000000063f4000 VMEM 00000000063f7800 0000000006400000 VMEM 0000000005800000 0000000006300000 KASAN Usable online memory ranges (info source: sclp read info [3]): 0000000000000000 0000000006400000 Usable online memory total: 6400000 Reserved: 61b10a3 Free: 24ef5d Call Trace: (sp:000000000002bd58 [<0000000000012a70>] physmem_alloc_top_down+0x60/0x14c) sp:000000000002bdc8 [<0000000000013756>] _pa+0x56/0x6a sp:000000000002bdf0 [<0000000000013bcc>] pgtable_populate+0x45c/0x65e sp:000000000002be90 [<00000000000140aa>] setup_vmem+0x2da/0x424 sp:000000000002bec8 [<0000000000011c20>] startup_kernel+0x428/0x8b4 sp:000000000002bf60 [<00000000000100f4>] startup_normal+0xd4/0xd4 physmem_alloc_range allows to find free memory in specified range. It should be used for one time allocations only like finding position for amode31 and vmlinux. physmem_alloc_top_down can be used just like physmem_alloc_range, but it also allows multiple allocations per type and tries to merge sequential allocations together. Which is useful for paging structures allocations. If sequential allocations cannot be merged together they are "chained", allowing easy per type reserved ranges enumeration and migration to memblock later. Extra "struct reserved_range" allocated for chaining are not tracked or reserved but rely on the fact that both physmem_alloc_range and physmem_alloc_top_down search for free memory only below current top down allocator position. All reserved ranges should be transferred to memblock before memblock allocations are enabled. The startup code has been reordered to delay any memory allocations until online memory ranges are detected and occupied memory ranges are marked as reserved to be excluded from follow-up allocations. Ipl report certificates are a special case, ipl report certificates list is checked together with other memory reserves until certificates are saved elsewhere. KASAN required memory for shadow memory allocation and mapping is reserved as 1 large chunk which is later passed to KASAN early initialization code. Acked-by: Heiko Carstens <hca@linux.ibm.com> Reviewed-by: Alexander Gordeev <agordeev@linux.ibm.com> Signed-off-by: Vasily Gorbik <gor@linux.ibm.com> Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2023-02-02 13:59:36 +01:00
s390/boot: Rework deployment of the kernel image Rework deployment of kernel image for both compressed and uncompressed variants as defined by CONFIG_KERNEL_UNCOMPRESSED kernel configuration variable. In case CONFIG_KERNEL_UNCOMPRESSED is disabled avoid uncompressing the kernel to a temporary buffer and copying it to the target address. Instead, uncompress it directly to the target destination. In case CONFIG_KERNEL_UNCOMPRESSED is enabled avoid moving the kernel to default 0x100000 location when KASLR is disabled or failed. Instead, use the uncompressed kernel image directly. In case KASLR is disabled or failed .amode31 section location in memory is not randomized and precedes the kernel image. In case CONFIG_KERNEL_UNCOMPRESSED is disabled that location overlaps the area used by the decompression algorithm. That is fine, since that area is not used after the decompression finished and the size of .amode31 section is not expected to exceed BOOT_HEAP_SIZE ever. There is no decompression in case CONFIG_KERNEL_UNCOMPRESSED is enabled. Therefore, rename decompress_kernel() to deploy_kernel(), which better describes both uncompressed and compressed cases. Introduce AMODE31_SIZE macro to avoid immediate value of 0x3000 (the size of .amode31 section) in the decompressor linker script. Modify the vmlinux linker script to force the size of .amode31 section to AMODE31_SIZE (the value of (_eamode31 - _samode31) could otherwise differ as result of compiler options used). Introduce __START_KERNEL macro that defines the kernel ELF image entry point and set it to the currrent value of 0x100000. Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com>
2024-03-22 14:39:57 +01:00
s390/boot: rework decompressor reserved tracking Currently several approaches for finding unused memory in decompressor are utilized. While "safe_addr" grows towards higher addresses, vmem code allocates paging structures top down. The former requires careful ordering. In addition to that ipl report handling code verifies potential intersections with secure boot certificates on its own. Neither of two approaches are memory holes aware and consistent with each other in low memory conditions. To solve that, existing approaches are generalized and combined together, as well as online memory ranges are now taken into consideration. physmem_info has been extended to contain reserved memory ranges. New set of functions allow to handle reserves and find unused memory. All reserves and memory allocations are "typed". In case of out of memory condition decompressor fails with detailed info on current reserved ranges and usable online memory. Linux version 6.2.0 ... Kernel command line: ... mem=100M Our of memory allocating 100000 bytes 100000 aligned in range 0:5800000 Reserved memory ranges: 0000000000000000 0000000003e33000 DECOMPRESSOR 0000000003f00000 00000000057648a3 INITRD 00000000063e0000 00000000063e8000 VMEM 00000000063eb000 00000000063f4000 VMEM 00000000063f7800 0000000006400000 VMEM 0000000005800000 0000000006300000 KASAN Usable online memory ranges (info source: sclp read info [3]): 0000000000000000 0000000006400000 Usable online memory total: 6400000 Reserved: 61b10a3 Free: 24ef5d Call Trace: (sp:000000000002bd58 [<0000000000012a70>] physmem_alloc_top_down+0x60/0x14c) sp:000000000002bdc8 [<0000000000013756>] _pa+0x56/0x6a sp:000000000002bdf0 [<0000000000013bcc>] pgtable_populate+0x45c/0x65e sp:000000000002be90 [<00000000000140aa>] setup_vmem+0x2da/0x424 sp:000000000002bec8 [<0000000000011c20>] startup_kernel+0x428/0x8b4 sp:000000000002bf60 [<00000000000100f4>] startup_normal+0xd4/0xd4 physmem_alloc_range allows to find free memory in specified range. It should be used for one time allocations only like finding position for amode31 and vmlinux. physmem_alloc_top_down can be used just like physmem_alloc_range, but it also allows multiple allocations per type and tries to merge sequential allocations together. Which is useful for paging structures allocations. If sequential allocations cannot be merged together they are "chained", allowing easy per type reserved ranges enumeration and migration to memblock later. Extra "struct reserved_range" allocated for chaining are not tracked or reserved but rely on the fact that both physmem_alloc_range and physmem_alloc_top_down search for free memory only below current top down allocator position. All reserved ranges should be transferred to memblock before memblock allocations are enabled. The startup code has been reordered to delay any memory allocations until online memory ranges are detected and occupied memory ranges are marked as reserved to be excluded from follow-up allocations. Ipl report certificates are a special case, ipl report certificates list is checked together with other memory reserves until certificates are saved elsewhere. KASAN required memory for shadow memory allocation and mapping is reserved as 1 large chunk which is later passed to KASAN early initialization code. Acked-by: Heiko Carstens <hca@linux.ibm.com> Reviewed-by: Alexander Gordeev <agordeev@linux.ibm.com> Signed-off-by: Vasily Gorbik <gor@linux.ibm.com> Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2023-02-02 13:59:36 +01:00
s390/boot: rework decompressor reserved tracking Currently several approaches for finding unused memory in decompressor are utilized. While "safe_addr" grows towards higher addresses, vmem code allocates paging structures top down. The former requires careful ordering. In addition to that ipl report handling code verifies potential intersections with secure boot certificates on its own. Neither of two approaches are memory holes aware and consistent with each other in low memory conditions. To solve that, existing approaches are generalized and combined together, as well as online memory ranges are now taken into consideration. physmem_info has been extended to contain reserved memory ranges. New set of functions allow to handle reserves and find unused memory. All reserves and memory allocations are "typed". In case of out of memory condition decompressor fails with detailed info on current reserved ranges and usable online memory. Linux version 6.2.0 ... Kernel command line: ... mem=100M Our of memory allocating 100000 bytes 100000 aligned in range 0:5800000 Reserved memory ranges: 0000000000000000 0000000003e33000 DECOMPRESSOR 0000000003f00000 00000000057648a3 INITRD 00000000063e0000 00000000063e8000 VMEM 00000000063eb000 00000000063f4000 VMEM 00000000063f7800 0000000006400000 VMEM 0000000005800000 0000000006300000 KASAN Usable online memory ranges (info source: sclp read info [3]): 0000000000000000 0000000006400000 Usable online memory total: 6400000 Reserved: 61b10a3 Free: 24ef5d Call Trace: (sp:000000000002bd58 [<0000000000012a70>] physmem_alloc_top_down+0x60/0x14c) sp:000000000002bdc8 [<0000000000013756>] _pa+0x56/0x6a sp:000000000002bdf0 [<0000000000013bcc>] pgtable_populate+0x45c/0x65e sp:000000000002be90 [<00000000000140aa>] setup_vmem+0x2da/0x424 sp:000000000002bec8 [<0000000000011c20>] startup_kernel+0x428/0x8b4 sp:000000000002bf60 [<00000000000100f4>] startup_normal+0xd4/0xd4 physmem_alloc_range allows to find free memory in specified range. It should be used for one time allocations only like finding position for amode31 and vmlinux. physmem_alloc_top_down can be used just like physmem_alloc_range, but it also allows multiple allocations per type and tries to merge sequential allocations together. Which is useful for paging structures allocations. If sequential allocations cannot be merged together they are "chained", allowing easy per type reserved ranges enumeration and migration to memblock later. Extra "struct reserved_range" allocated for chaining are not tracked or reserved but rely on the fact that both physmem_alloc_range and physmem_alloc_top_down search for free memory only below current top down allocator position. All reserved ranges should be transferred to memblock before memblock allocations are enabled. The startup code has been reordered to delay any memory allocations until online memory ranges are detected and occupied memory ranges are marked as reserved to be excluded from follow-up allocations. Ipl report certificates are a special case, ipl report certificates list is checked together with other memory reserves until certificates are saved elsewhere. KASAN required memory for shadow memory allocation and mapping is reserved as 1 large chunk which is later passed to KASAN early initialization code. Acked-by: Heiko Carstens <hca@linux.ibm.com> Reviewed-by: Alexander Gordeev <agordeev@linux.ibm.com> Signed-off-by: Vasily Gorbik <gor@linux.ibm.com> Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2023-02-02 13:59:36 +01:00
s390/boot: rework decompressor reserved tracking Currently several approaches for finding unused memory in decompressor are utilized. While "safe_addr" grows towards higher addresses, vmem code allocates paging structures top down. The former requires careful ordering. In addition to that ipl report handling code verifies potential intersections with secure boot certificates on its own. Neither of two approaches are memory holes aware and consistent with each other in low memory conditions. To solve that, existing approaches are generalized and combined together, as well as online memory ranges are now taken into consideration. physmem_info has been extended to contain reserved memory ranges. New set of functions allow to handle reserves and find unused memory. All reserves and memory allocations are "typed". In case of out of memory condition decompressor fails with detailed info on current reserved ranges and usable online memory. Linux version 6.2.0 ... Kernel command line: ... mem=100M Our of memory allocating 100000 bytes 100000 aligned in range 0:5800000 Reserved memory ranges: 0000000000000000 0000000003e33000 DECOMPRESSOR 0000000003f00000 00000000057648a3 INITRD 00000000063e0000 00000000063e8000 VMEM 00000000063eb000 00000000063f4000 VMEM 00000000063f7800 0000000006400000 VMEM 0000000005800000 0000000006300000 KASAN Usable online memory ranges (info source: sclp read info [3]): 0000000000000000 0000000006400000 Usable online memory total: 6400000 Reserved: 61b10a3 Free: 24ef5d Call Trace: (sp:000000000002bd58 [<0000000000012a70>] physmem_alloc_top_down+0x60/0x14c) sp:000000000002bdc8 [<0000000000013756>] _pa+0x56/0x6a sp:000000000002bdf0 [<0000000000013bcc>] pgtable_populate+0x45c/0x65e sp:000000000002be90 [<00000000000140aa>] setup_vmem+0x2da/0x424 sp:000000000002bec8 [<0000000000011c20>] startup_kernel+0x428/0x8b4 sp:000000000002bf60 [<00000000000100f4>] startup_normal+0xd4/0xd4 physmem_alloc_range allows to find free memory in specified range. It should be used for one time allocations only like finding position for amode31 and vmlinux. physmem_alloc_top_down can be used just like physmem_alloc_range, but it also allows multiple allocations per type and tries to merge sequential allocations together. Which is useful for paging structures allocations. If sequential allocations cannot be merged together they are "chained", allowing easy per type reserved ranges enumeration and migration to memblock later. Extra "struct reserved_range" allocated for chaining are not tracked or reserved but rely on the fact that both physmem_alloc_range and physmem_alloc_top_down search for free memory only below current top down allocator position. All reserved ranges should be transferred to memblock before memblock allocations are enabled. The startup code has been reordered to delay any memory allocations until online memory ranges are detected and occupied memory ranges are marked as reserved to be excluded from follow-up allocations. Ipl report certificates are a special case, ipl report certificates list is checked together with other memory reserves until certificates are saved elsewhere. KASAN required memory for shadow memory allocation and mapping is reserved as 1 large chunk which is later passed to KASAN early initialization code. Acked-by: Heiko Carstens <hca@linux.ibm.com> Reviewed-by: Alexander Gordeev <agordeev@linux.ibm.com> Signed-off-by: Vasily Gorbik <gor@linux.ibm.com> Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2023-02-02 13:59:36 +01:00
s390/mm: Uncouple physical vs virtual address spaces The uncoupling physical vs virtual address spaces brings the following benefits to s390: - virtual memory layout flexibility; - closes the address gap between kernel and modules, it caused s390-only problems in the past (e.g. 'perf' bugs); - allows getting rid of trampolines used for module calls into kernel; - allows simplifying BPF trampoline; - minor performance improvement in branch prediction; - kernel randomization entropy is magnitude bigger, as it is derived from the amount of available virtual, not physical memory; The whole change could be described in two pictures below: before and after the change. Some aspects of the virtual memory layout setup are not clarified (number of page levels, alignment, DMA memory), since these are not a part of this change or secondary with regard to how the uncoupling itself is implemented. The focus of the pictures is to explain why __va() and __pa() macros are implemented the way they are. Memory layout in V==R mode: | Physical | Virtual | +- 0 --------------+- 0 --------------+ identity mapping start | | S390_lowcore | Low-address memory | +- 8 KB -----------+ | | | | | identity | phys == virt | | mapping | virt == phys | | | +- AMODE31_START --+- AMODE31_START --+ .amode31 rand. phys/virt start |.amode31 text/data|.amode31 text/data| +- AMODE31_END ----+- AMODE31_END ----+ .amode31 rand. phys/virt start | | | | | | +- __kaslr_offset, __kaslr_offset_phys| kernel rand. phys/virt start | | | | kernel text/data | kernel text/data | phys == kvirt | | | +------------------+------------------+ kernel phys/virt end | | | | | | | | | | | | +- ident_map_size -+- ident_map_size -+ identity mapping end | | | ... unused gap | | | +---- vmemmap -----+ 'struct page' array start | | | virtually mapped | | memory map | | | +- __abs_lowcore --+ | | | Absolute Lowcore | | | +- __memcpy_real_area | | | Real Memory Copy| | | +- VMALLOC_START --+ vmalloc area start | | | vmalloc area | | | +- MODULES_VADDR --+ modules area start | | | modules area | | | +------------------+ UltraVisor Secure Storage limit | | | ... unused gap | | | +KASAN_SHADOW_START+ KASAN shadow memory start | | | KASAN shadow | | | +------------------+ ASCE limit Memory layout in V!=R mode: | Physical | Virtual | +- 0 --------------+- 0 --------------+ | | S390_lowcore | Low-address memory | +- 8 KB -----------+ | | | | | | | | ... unused gap | | | | +- AMODE31_START --+- AMODE31_START --+ .amode31 rand. phys/virt start |.amode31 text/data|.amode31 text/data| +- AMODE31_END ----+- AMODE31_END ----+ .amode31 rand. phys/virt end (<2GB) | | | | | | +- __kaslr_offset_phys | kernel rand. phys start | | | | kernel text/data | | | | | +------------------+ | kernel phys end | | | | | | | | | | | | +- ident_map_size -+ | | | | ... unused gap | | | +- __identity_base + identity mapping start (>= 2GB) | | | identity | phys == virt - __identity_base | mapping | virt == phys + __identity_base | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | +---- vmemmap -----+ 'struct page' array start | | | virtually mapped | | memory map | | | +- __abs_lowcore --+ | | | Absolute Lowcore | | | +- __memcpy_real_area | | | Real Memory Copy| | | +- VMALLOC_START --+ vmalloc area start | | | vmalloc area | | | +- MODULES_VADDR --+ modules area start | | | modules area | | | +- __kaslr_offset -+ kernel rand. virt start | | | kernel text/data | phys == (kvirt - __kaslr_offset) + | | __kaslr_offset_phys +- kernel .bss end + kernel rand. virt end | | | ... unused gap | | | +------------------+ UltraVisor Secure Storage limit | | | ... unused gap | | | +KASAN_SHADOW_START+ KASAN shadow memory start | | | KASAN shadow | | | +------------------+ ASCE limit Unused gaps in the virtual memory layout could be present or not - depending on how partucular system is configured. No page tables are created for the unused gaps. The relative order of vmalloc, modules and kernel image in virtual memory is defined by following considerations: - start of the modules area and end of the kernel should reside within 4GB to accommodate relative 32-bit jumps. The best way to achieve that is to place kernel next to modules; - vmalloc and module areas should locate next to each other to prevent failures and extra reworks in user level tools (makedumpfile, crash, etc.) which treat vmalloc and module addresses similarily; - kernel needs to be the last area in the virtual memory layout to easily distinguish between kernel and non-kernel virtual addresses. That is needed to (again) simplify handling of addresses in user level tools and make __pa() macro faster (see below); Concluding the above, the relative order of the considered virtual areas in memory is: vmalloc - modules - kernel. Therefore, the only change to the current memory layout is moving kernel to the end of virtual address space. With that approach the implementation of __pa() macro is straightforward - all linear virtual addresses less than kernel base are considered identity mapping: phys == virt - __identity_base All addresses greater than kernel base are kernel ones: phys == (kvirt - __kaslr_offset) + __kaslr_offset_phys By contrast, __va() macro deals only with identity mapping addresses: virt == phys + __identity_base .amode31 section is mapped separately and is not covered by __pa() macro. In fact, it could have been handled easily by checking whether a virtual address is within the section or not, but there is no need for that. Thus, let __pa() code do as little machine cycles as possible. The KASAN shadow memory is located at the very end of the virtual memory layout, at addresses higher than the kernel. However, that is not a linear mapping and no code other than KASAN instrumentation or API is expected to access it. When KASLR mode is enabled the kernel base address randomized within a memory window that spans whole unused virtual address space. The size of that window depends from the amount of physical memory available to the system, the limit imposed by UltraVisor (if present) and the vmalloc area size as provided by vmalloc= kernel command line parameter. In case the virtual memory is exhausted the minimum size of the randomization window is forcefully set to 2GB, which amounts to in 15 bits of entropy if KASAN is enabled or 17 bits of entropy in default configuration. The default kernel offset 0x100000 is used as a magic value both in the decompressor code and vmlinux linker script, but it will be removed with a follow-up change. Acked-by: Heiko Carstens <hca@linux.ibm.com> Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com>
2024-03-01 07:15:22 +01:00
s390/boot: rework decompressor reserved tracking Currently several approaches for finding unused memory in decompressor are utilized. While "safe_addr" grows towards higher addresses, vmem code allocates paging structures top down. The former requires careful ordering. In addition to that ipl report handling code verifies potential intersections with secure boot certificates on its own. Neither of two approaches are memory holes aware and consistent with each other in low memory conditions. To solve that, existing approaches are generalized and combined together, as well as online memory ranges are now taken into consideration. physmem_info has been extended to contain reserved memory ranges. New set of functions allow to handle reserves and find unused memory. All reserves and memory allocations are "typed". In case of out of memory condition decompressor fails with detailed info on current reserved ranges and usable online memory. Linux version 6.2.0 ... Kernel command line: ... mem=100M Our of memory allocating 100000 bytes 100000 aligned in range 0:5800000 Reserved memory ranges: 0000000000000000 0000000003e33000 DECOMPRESSOR 0000000003f00000 00000000057648a3 INITRD 00000000063e0000 00000000063e8000 VMEM 00000000063eb000 00000000063f4000 VMEM 00000000063f7800 0000000006400000 VMEM 0000000005800000 0000000006300000 KASAN Usable online memory ranges (info source: sclp read info [3]): 0000000000000000 0000000006400000 Usable online memory total: 6400000 Reserved: 61b10a3 Free: 24ef5d Call Trace: (sp:000000000002bd58 [<0000000000012a70>] physmem_alloc_top_down+0x60/0x14c) sp:000000000002bdc8 [<0000000000013756>] _pa+0x56/0x6a sp:000000000002bdf0 [<0000000000013bcc>] pgtable_populate+0x45c/0x65e sp:000000000002be90 [<00000000000140aa>] setup_vmem+0x2da/0x424 sp:000000000002bec8 [<0000000000011c20>] startup_kernel+0x428/0x8b4 sp:000000000002bf60 [<00000000000100f4>] startup_normal+0xd4/0xd4 physmem_alloc_range allows to find free memory in specified range. It should be used for one time allocations only like finding position for amode31 and vmlinux. physmem_alloc_top_down can be used just like physmem_alloc_range, but it also allows multiple allocations per type and tries to merge sequential allocations together. Which is useful for paging structures allocations. If sequential allocations cannot be merged together they are "chained", allowing easy per type reserved ranges enumeration and migration to memblock later. Extra "struct reserved_range" allocated for chaining are not tracked or reserved but rely on the fact that both physmem_alloc_range and physmem_alloc_top_down search for free memory only below current top down allocator position. All reserved ranges should be transferred to memblock before memblock allocations are enabled. The startup code has been reordered to delay any memory allocations until online memory ranges are detected and occupied memory ranges are marked as reserved to be excluded from follow-up allocations. Ipl report certificates are a special case, ipl report certificates list is checked together with other memory reserves until certificates are saved elsewhere. KASAN required memory for shadow memory allocation and mapping is reserved as 1 large chunk which is later passed to KASAN early initialization code. Acked-by: Heiko Carstens <hca@linux.ibm.com> Reviewed-by: Alexander Gordeev <agordeev@linux.ibm.com> Signed-off-by: Vasily Gorbik <gor@linux.ibm.com> Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2023-02-02 13:59:36 +01:00
s390/boot: rework decompressor reserved tracking Currently several approaches for finding unused memory in decompressor are utilized. While "safe_addr" grows towards higher addresses, vmem code allocates paging structures top down. The former requires careful ordering. In addition to that ipl report handling code verifies potential intersections with secure boot certificates on its own. Neither of two approaches are memory holes aware and consistent with each other in low memory conditions. To solve that, existing approaches are generalized and combined together, as well as online memory ranges are now taken into consideration. physmem_info has been extended to contain reserved memory ranges. New set of functions allow to handle reserves and find unused memory. All reserves and memory allocations are "typed". In case of out of memory condition decompressor fails with detailed info on current reserved ranges and usable online memory. Linux version 6.2.0 ... Kernel command line: ... mem=100M Our of memory allocating 100000 bytes 100000 aligned in range 0:5800000 Reserved memory ranges: 0000000000000000 0000000003e33000 DECOMPRESSOR 0000000003f00000 00000000057648a3 INITRD 00000000063e0000 00000000063e8000 VMEM 00000000063eb000 00000000063f4000 VMEM 00000000063f7800 0000000006400000 VMEM 0000000005800000 0000000006300000 KASAN Usable online memory ranges (info source: sclp read info [3]): 0000000000000000 0000000006400000 Usable online memory total: 6400000 Reserved: 61b10a3 Free: 24ef5d Call Trace: (sp:000000000002bd58 [<0000000000012a70>] physmem_alloc_top_down+0x60/0x14c) sp:000000000002bdc8 [<0000000000013756>] _pa+0x56/0x6a sp:000000000002bdf0 [<0000000000013bcc>] pgtable_populate+0x45c/0x65e sp:000000000002be90 [<00000000000140aa>] setup_vmem+0x2da/0x424 sp:000000000002bec8 [<0000000000011c20>] startup_kernel+0x428/0x8b4 sp:000000000002bf60 [<00000000000100f4>] startup_normal+0xd4/0xd4 physmem_alloc_range allows to find free memory in specified range. It should be used for one time allocations only like finding position for amode31 and vmlinux. physmem_alloc_top_down can be used just like physmem_alloc_range, but it also allows multiple allocations per type and tries to merge sequential allocations together. Which is useful for paging structures allocations. If sequential allocations cannot be merged together they are "chained", allowing easy per type reserved ranges enumeration and migration to memblock later. Extra "struct reserved_range" allocated for chaining are not tracked or reserved but rely on the fact that both physmem_alloc_range and physmem_alloc_top_down search for free memory only below current top down allocator position. All reserved ranges should be transferred to memblock before memblock allocations are enabled. The startup code has been reordered to delay any memory allocations until online memory ranges are detected and occupied memory ranges are marked as reserved to be excluded from follow-up allocations. Ipl report certificates are a special case, ipl report certificates list is checked together with other memory reserves until certificates are saved elsewhere. KASAN required memory for shadow memory allocation and mapping is reserved as 1 large chunk which is later passed to KASAN early initialization code. Acked-by: Heiko Carstens <hca@linux.ibm.com> Reviewed-by: Alexander Gordeev <agordeev@linux.ibm.com> Signed-off-by: Vasily Gorbik <gor@linux.ibm.com> Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2023-02-02 13:59:36 +01:00
s390/boot: Rework deployment of the kernel image Rework deployment of kernel image for both compressed and uncompressed variants as defined by CONFIG_KERNEL_UNCOMPRESSED kernel configuration variable. In case CONFIG_KERNEL_UNCOMPRESSED is disabled avoid uncompressing the kernel to a temporary buffer and copying it to the target address. Instead, uncompress it directly to the target destination. In case CONFIG_KERNEL_UNCOMPRESSED is enabled avoid moving the kernel to default 0x100000 location when KASLR is disabled or failed. Instead, use the uncompressed kernel image directly. In case KASLR is disabled or failed .amode31 section location in memory is not randomized and precedes the kernel image. In case CONFIG_KERNEL_UNCOMPRESSED is disabled that location overlaps the area used by the decompression algorithm. That is fine, since that area is not used after the decompression finished and the size of .amode31 section is not expected to exceed BOOT_HEAP_SIZE ever. There is no decompression in case CONFIG_KERNEL_UNCOMPRESSED is enabled. Therefore, rename decompress_kernel() to deploy_kernel(), which better describes both uncompressed and compressed cases. Introduce AMODE31_SIZE macro to avoid immediate value of 0x3000 (the size of .amode31 section) in the decompressor linker script. Modify the vmlinux linker script to force the size of .amode31 section to AMODE31_SIZE (the value of (_eamode31 - _samode31) could otherwise differ as result of compiler options used). Introduce __START_KERNEL macro that defines the kernel ELF image entry point and set it to the currrent value of 0x100000. Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com>
2024-03-22 14:39:57 +01:00
s390/boot: rework decompressor reserved tracking Currently several approaches for finding unused memory in decompressor are utilized. While "safe_addr" grows towards higher addresses, vmem code allocates paging structures top down. The former requires careful ordering. In addition to that ipl report handling code verifies potential intersections with secure boot certificates on its own. Neither of two approaches are memory holes aware and consistent with each other in low memory conditions. To solve that, existing approaches are generalized and combined together, as well as online memory ranges are now taken into consideration. physmem_info has been extended to contain reserved memory ranges. New set of functions allow to handle reserves and find unused memory. All reserves and memory allocations are "typed". In case of out of memory condition decompressor fails with detailed info on current reserved ranges and usable online memory. Linux version 6.2.0 ... Kernel command line: ... mem=100M Our of memory allocating 100000 bytes 100000 aligned in range 0:5800000 Reserved memory ranges: 0000000000000000 0000000003e33000 DECOMPRESSOR 0000000003f00000 00000000057648a3 INITRD 00000000063e0000 00000000063e8000 VMEM 00000000063eb000 00000000063f4000 VMEM 00000000063f7800 0000000006400000 VMEM 0000000005800000 0000000006300000 KASAN Usable online memory ranges (info source: sclp read info [3]): 0000000000000000 0000000006400000 Usable online memory total: 6400000 Reserved: 61b10a3 Free: 24ef5d Call Trace: (sp:000000000002bd58 [<0000000000012a70>] physmem_alloc_top_down+0x60/0x14c) sp:000000000002bdc8 [<0000000000013756>] _pa+0x56/0x6a sp:000000000002bdf0 [<0000000000013bcc>] pgtable_populate+0x45c/0x65e sp:000000000002be90 [<00000000000140aa>] setup_vmem+0x2da/0x424 sp:000000000002bec8 [<0000000000011c20>] startup_kernel+0x428/0x8b4 sp:000000000002bf60 [<00000000000100f4>] startup_normal+0xd4/0xd4 physmem_alloc_range allows to find free memory in specified range. It should be used for one time allocations only like finding position for amode31 and vmlinux. physmem_alloc_top_down can be used just like physmem_alloc_range, but it also allows multiple allocations per type and tries to merge sequential allocations together. Which is useful for paging structures allocations. If sequential allocations cannot be merged together they are "chained", allowing easy per type reserved ranges enumeration and migration to memblock later. Extra "struct reserved_range" allocated for chaining are not tracked or reserved but rely on the fact that both physmem_alloc_range and physmem_alloc_top_down search for free memory only below current top down allocator position. All reserved ranges should be transferred to memblock before memblock allocations are enabled. The startup code has been reordered to delay any memory allocations until online memory ranges are detected and occupied memory ranges are marked as reserved to be excluded from follow-up allocations. Ipl report certificates are a special case, ipl report certificates list is checked together with other memory reserves until certificates are saved elsewhere. KASAN required memory for shadow memory allocation and mapping is reserved as 1 large chunk which is later passed to KASAN early initialization code. Acked-by: Heiko Carstens <hca@linux.ibm.com> Reviewed-by: Alexander Gordeev <agordeev@linux.ibm.com> Signed-off-by: Vasily Gorbik <gor@linux.ibm.com> Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2023-02-02 13:59:36 +01:00
s390/boot: Rework deployment of the kernel image Rework deployment of kernel image for both compressed and uncompressed variants as defined by CONFIG_KERNEL_UNCOMPRESSED kernel configuration variable. In case CONFIG_KERNEL_UNCOMPRESSED is disabled avoid uncompressing the kernel to a temporary buffer and copying it to the target address. Instead, uncompress it directly to the target destination. In case CONFIG_KERNEL_UNCOMPRESSED is enabled avoid moving the kernel to default 0x100000 location when KASLR is disabled or failed. Instead, use the uncompressed kernel image directly. In case KASLR is disabled or failed .amode31 section location in memory is not randomized and precedes the kernel image. In case CONFIG_KERNEL_UNCOMPRESSED is disabled that location overlaps the area used by the decompression algorithm. That is fine, since that area is not used after the decompression finished and the size of .amode31 section is not expected to exceed BOOT_HEAP_SIZE ever. There is no decompression in case CONFIG_KERNEL_UNCOMPRESSED is enabled. Therefore, rename decompress_kernel() to deploy_kernel(), which better describes both uncompressed and compressed cases. Introduce AMODE31_SIZE macro to avoid immediate value of 0x3000 (the size of .amode31 section) in the decompressor linker script. Modify the vmlinux linker script to force the size of .amode31 section to AMODE31_SIZE (the value of (_eamode31 - _samode31) could otherwise differ as result of compiler options used). Introduce __START_KERNEL macro that defines the kernel ELF image entry point and set it to the currrent value of 0x100000. Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com>
2024-03-22 14:39:57 +01:00
s390/mm: start kernel with DAT enabled The setup of the kernel virtual address space is spread throughout the sources, boot stages and config options like this: 1. The available physical memory regions are queried and stored as mem_detect information for later use in the decompressor. 2. Based on the physical memory availability the virtual memory layout is established in the decompressor; 3. If CONFIG_KASAN is disabled the kernel paging setup code populates kernel pgtables and turns DAT mode on. It uses the information stored at step [1]. 4. If CONFIG_KASAN is enabled the kernel early boot kasan setup populates kernel pgtables and turns DAT mode on. It uses the information stored at step [1]. The kasan setup creates early_pg_dir directory and directly overwrites swapper_pg_dir entries to make shadow memory pages available. Move the kernel virtual memory setup to the decompressor and start the kernel with DAT turned on right from the very first istruction. That completely eliminates the boot phase when the kernel runs in DAT-off mode, simplies the overall design and consolidates pgtables setup. The identity mapping is created in the decompressor, while kasan shadow mappings are still created by the early boot kernel code. Share with decompressor the existing kasan memory allocator. It decreases the size of a newly requested memory block from pgalloc_pos and ensures that kernel image is not overwritten. pgalloc_low and pgalloc_pos pointers are made preserved boot variables for that. Use the bootdata infrastructure to setup swapper_pg_dir and invalid_pg_dir directories used by the kernel later. The interim early_pg_dir directory established by the kasan initialization code gets eliminated as result. As the kernel runs in DAT-on mode only the PSW_KERNEL_BITS define gets PSW_MASK_DAT bit by default. Additionally, the setup_lowcore_dat_off() and setup_lowcore_dat_on() routines get merged, since there is no DAT-off mode stage anymore. The memory mappings are created with RW+X protection that allows the early boot code setting up all necessary data and services for the kernel being booted. Just before the paging is enabled the memory protection is changed to RO+X for text, RO+NX for read-only data and RW+NX for kernel data and the identity mapping. Reviewed-by: Heiko Carstens <hca@linux.ibm.com> Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com> Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2022-12-13 11:35:11 +01:00
s390/mm: start kernel with DAT enabled The setup of the kernel virtual address space is spread throughout the sources, boot stages and config options like this: 1. The available physical memory regions are queried and stored as mem_detect information for later use in the decompressor. 2. Based on the physical memory availability the virtual memory layout is established in the decompressor; 3. If CONFIG_KASAN is disabled the kernel paging setup code populates kernel pgtables and turns DAT mode on. It uses the information stored at step [1]. 4. If CONFIG_KASAN is enabled the kernel early boot kasan setup populates kernel pgtables and turns DAT mode on. It uses the information stored at step [1]. The kasan setup creates early_pg_dir directory and directly overwrites swapper_pg_dir entries to make shadow memory pages available. Move the kernel virtual memory setup to the decompressor and start the kernel with DAT turned on right from the very first istruction. That completely eliminates the boot phase when the kernel runs in DAT-off mode, simplies the overall design and consolidates pgtables setup. The identity mapping is created in the decompressor, while kasan shadow mappings are still created by the early boot kernel code. Share with decompressor the existing kasan memory allocator. It decreases the size of a newly requested memory block from pgalloc_pos and ensures that kernel image is not overwritten. pgalloc_low and pgalloc_pos pointers are made preserved boot variables for that. Use the bootdata infrastructure to setup swapper_pg_dir and invalid_pg_dir directories used by the kernel later. The interim early_pg_dir directory established by the kasan initialization code gets eliminated as result. As the kernel runs in DAT-on mode only the PSW_KERNEL_BITS define gets PSW_MASK_DAT bit by default. Additionally, the setup_lowcore_dat_off() and setup_lowcore_dat_on() routines get merged, since there is no DAT-off mode stage anymore. The memory mappings are created with RW+X protection that allows the early boot code setting up all necessary data and services for the kernel being booted. Just before the paging is enabled the memory protection is changed to RO+X for text, RO+NX for read-only data and RW+NX for kernel data and the identity mapping. Reviewed-by: Heiko Carstens <hca@linux.ibm.com> Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com> Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2022-12-13 11:35:11 +01:00
s390: compile relocatable kernel without -fPIE On s390, currently kernel uses the '-fPIE' compiler flag for compiling vmlinux. This has a few problems: - It uses dynamic symbols (.dynsym), for which the linker refuses to allow more than 64k sections. This can break features which use '-ffunction-sections' and '-fdata-sections', including kpatch-build [1] and Function Granular KASLR. - It unnecessarily uses GOT relocations, adding an extra layer of indirection for many memory accesses. Instead of using '-fPIE', resolve all the relocations at link time and then manually adjust any absolute relocations (R_390_64) during boot. This is done by first telling the linker to preserve all relocations during the vmlinux link. (Note this is harmless: they are later stripped in the vmlinux.bin link.) Then use the 'relocs' tool to find all absolute relocations (R_390_64) which apply to allocatable sections. The offsets of those relocations are saved in a special section which is then used to adjust the relocations during boot. (Note: For some reason, Clang occasionally creates a GOT reference, even without '-fPIE'. So Clang-compiled kernels have a GOT, which needs to be adjusted.) On my mostly-defconfig kernel, this reduces kernel text size by ~1.3%. [1] https://github.com/dynup/kpatch/issues/1284 [2] https://gcc.gnu.org/pipermail/gcc-patches/2023-June/622872.html [3] https://gcc.gnu.org/pipermail/gcc-patches/2023-August/625986.html Compiler consideration: Gcc recently implemented an optimization [2] for loading symbols without explicit alignment, aligning with the IBM Z ELF ABI. This ABI mandates symbols to reside on a 2-byte boundary, enabling the use of the larl instruction. However, kernel linker scripts may still generate unaligned symbols. To address this, a new -munaligned-symbols option has been introduced [3] in recent gcc versions. This option has to be used with future gcc versions. Older Clang lacks support for handling unaligned symbols generated by kernel linker scripts when the kernel is built without -fPIE. However, future versions of Clang will include support for the -munaligned-symbols option. When the support is unavailable, compile the kernel with -fPIE to maintain the existing behavior. In addition to it: move vmlinux.relocs to safe relocation When the kernel is built with CONFIG_KERNEL_UNCOMPRESSED, the entire uncompressed vmlinux.bin is positioned in the bzImage decompressor image at the default kernel LMA of 0x100000, enabling it to be executed in-place. However, the size of .vmlinux.relocs could be large enough to cause an overlap with the uncompressed kernel at the address 0x100000. To address this issue, .vmlinux.relocs is positioned after the .rodata.compressed in the bzImage. Nevertheless, in this configuration, vmlinux.relocs will overlap with the .bss section of vmlinux.bin. To overcome that, move vmlinux.relocs to a safe location before clearing .bss and handling relocs. Compile warning fix from Sumanth Korikkar: When kernel is built with CONFIG_LD_ORPHAN_WARN and -fno-PIE, there are several warnings: ld: warning: orphan section `.rela.iplt' from `arch/s390/kernel/head64.o' being placed in section `.rela.dyn' ld: warning: orphan section `.rela.head.text' from `arch/s390/kernel/head64.o' being placed in section `.rela.dyn' ld: warning: orphan section `.rela.init.text' from `arch/s390/kernel/head64.o' being placed in section `.rela.dyn' ld: warning: orphan section `.rela.rodata.cst8' from `arch/s390/kernel/head64.o' being placed in section `.rela.dyn' Orphan sections are sections that exist in an object file but don't have a corresponding output section in the final executable. ld raises a warning when it identifies such sections. Eliminate the warning by placing all .rela orphan sections in .rela.dyn and raise an error when size of .rela.dyn is greater than zero. i.e. Dont just neglect orphan sections. This is similar to adjustment performed in x86, where kernel is built with -fno-PIE. commit 5354e84598f2 ("x86/build: Add asserts for unwanted sections") [sumanthk@linux.ibm.com: rebased Josh Poimboeuf patches and move vmlinux.relocs to safe location] [hca@linux.ibm.com: merged compile warning fix from Sumanth] Tested-by: Sumanth Korikkar <sumanthk@linux.ibm.com> Acked-by: Vasily Gorbik <gor@linux.ibm.com> Signed-off-by: Josh Poimboeuf <jpoimboe@kernel.org> Signed-off-by: Sumanth Korikkar <sumanthk@linux.ibm.com> Link: https://lore.kernel.org/r/20240219132734.22881-4-sumanthk@linux.ibm.com Link: https://lore.kernel.org/r/20240219132734.22881-5-sumanthk@linux.ibm.com Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2024-02-19 14:27:33 +01:00
s390/mm: start kernel with DAT enabled The setup of the kernel virtual address space is spread throughout the sources, boot stages and config options like this: 1. The available physical memory regions are queried and stored as mem_detect information for later use in the decompressor. 2. Based on the physical memory availability the virtual memory layout is established in the decompressor; 3. If CONFIG_KASAN is disabled the kernel paging setup code populates kernel pgtables and turns DAT mode on. It uses the information stored at step [1]. 4. If CONFIG_KASAN is enabled the kernel early boot kasan setup populates kernel pgtables and turns DAT mode on. It uses the information stored at step [1]. The kasan setup creates early_pg_dir directory and directly overwrites swapper_pg_dir entries to make shadow memory pages available. Move the kernel virtual memory setup to the decompressor and start the kernel with DAT turned on right from the very first istruction. That completely eliminates the boot phase when the kernel runs in DAT-off mode, simplies the overall design and consolidates pgtables setup. The identity mapping is created in the decompressor, while kasan shadow mappings are still created by the early boot kernel code. Share with decompressor the existing kasan memory allocator. It decreases the size of a newly requested memory block from pgalloc_pos and ensures that kernel image is not overwritten. pgalloc_low and pgalloc_pos pointers are made preserved boot variables for that. Use the bootdata infrastructure to setup swapper_pg_dir and invalid_pg_dir directories used by the kernel later. The interim early_pg_dir directory established by the kasan initialization code gets eliminated as result. As the kernel runs in DAT-on mode only the PSW_KERNEL_BITS define gets PSW_MASK_DAT bit by default. Additionally, the setup_lowcore_dat_off() and setup_lowcore_dat_on() routines get merged, since there is no DAT-off mode stage anymore. The memory mappings are created with RW+X protection that allows the early boot code setting up all necessary data and services for the kernel being booted. Just before the paging is enabled the memory protection is changed to RO+X for text, RO+NX for read-only data and RW+NX for kernel data and the identity mapping. Reviewed-by: Heiko Carstens <hca@linux.ibm.com> Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com> Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2022-12-13 11:35:11 +01:00
s390: compile relocatable kernel without -fPIE On s390, currently kernel uses the '-fPIE' compiler flag for compiling vmlinux. This has a few problems: - It uses dynamic symbols (.dynsym), for which the linker refuses to allow more than 64k sections. This can break features which use '-ffunction-sections' and '-fdata-sections', including kpatch-build [1] and Function Granular KASLR. - It unnecessarily uses GOT relocations, adding an extra layer of indirection for many memory accesses. Instead of using '-fPIE', resolve all the relocations at link time and then manually adjust any absolute relocations (R_390_64) during boot. This is done by first telling the linker to preserve all relocations during the vmlinux link. (Note this is harmless: they are later stripped in the vmlinux.bin link.) Then use the 'relocs' tool to find all absolute relocations (R_390_64) which apply to allocatable sections. The offsets of those relocations are saved in a special section which is then used to adjust the relocations during boot. (Note: For some reason, Clang occasionally creates a GOT reference, even without '-fPIE'. So Clang-compiled kernels have a GOT, which needs to be adjusted.) On my mostly-defconfig kernel, this reduces kernel text size by ~1.3%. [1] https://github.com/dynup/kpatch/issues/1284 [2] https://gcc.gnu.org/pipermail/gcc-patches/2023-June/622872.html [3] https://gcc.gnu.org/pipermail/gcc-patches/2023-August/625986.html Compiler consideration: Gcc recently implemented an optimization [2] for loading symbols without explicit alignment, aligning with the IBM Z ELF ABI. This ABI mandates symbols to reside on a 2-byte boundary, enabling the use of the larl instruction. However, kernel linker scripts may still generate unaligned symbols. To address this, a new -munaligned-symbols option has been introduced [3] in recent gcc versions. This option has to be used with future gcc versions. Older Clang lacks support for handling unaligned symbols generated by kernel linker scripts when the kernel is built without -fPIE. However, future versions of Clang will include support for the -munaligned-symbols option. When the support is unavailable, compile the kernel with -fPIE to maintain the existing behavior. In addition to it: move vmlinux.relocs to safe relocation When the kernel is built with CONFIG_KERNEL_UNCOMPRESSED, the entire uncompressed vmlinux.bin is positioned in the bzImage decompressor image at the default kernel LMA of 0x100000, enabling it to be executed in-place. However, the size of .vmlinux.relocs could be large enough to cause an overlap with the uncompressed kernel at the address 0x100000. To address this issue, .vmlinux.relocs is positioned after the .rodata.compressed in the bzImage. Nevertheless, in this configuration, vmlinux.relocs will overlap with the .bss section of vmlinux.bin. To overcome that, move vmlinux.relocs to a safe location before clearing .bss and handling relocs. Compile warning fix from Sumanth Korikkar: When kernel is built with CONFIG_LD_ORPHAN_WARN and -fno-PIE, there are several warnings: ld: warning: orphan section `.rela.iplt' from `arch/s390/kernel/head64.o' being placed in section `.rela.dyn' ld: warning: orphan section `.rela.head.text' from `arch/s390/kernel/head64.o' being placed in section `.rela.dyn' ld: warning: orphan section `.rela.init.text' from `arch/s390/kernel/head64.o' being placed in section `.rela.dyn' ld: warning: orphan section `.rela.rodata.cst8' from `arch/s390/kernel/head64.o' being placed in section `.rela.dyn' Orphan sections are sections that exist in an object file but don't have a corresponding output section in the final executable. ld raises a warning when it identifies such sections. Eliminate the warning by placing all .rela orphan sections in .rela.dyn and raise an error when size of .rela.dyn is greater than zero. i.e. Dont just neglect orphan sections. This is similar to adjustment performed in x86, where kernel is built with -fno-PIE. commit 5354e84598f2 ("x86/build: Add asserts for unwanted sections") [sumanthk@linux.ibm.com: rebased Josh Poimboeuf patches and move vmlinux.relocs to safe location] [hca@linux.ibm.com: merged compile warning fix from Sumanth] Tested-by: Sumanth Korikkar <sumanthk@linux.ibm.com> Acked-by: Vasily Gorbik <gor@linux.ibm.com> Signed-off-by: Josh Poimboeuf <jpoimboe@kernel.org> Signed-off-by: Sumanth Korikkar <sumanthk@linux.ibm.com> Link: https://lore.kernel.org/r/20240219132734.22881-4-sumanthk@linux.ibm.com Link: https://lore.kernel.org/r/20240219132734.22881-5-sumanthk@linux.ibm.com Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2024-02-19 14:27:33 +01:00
s390/mm: start kernel with DAT enabled The setup of the kernel virtual address space is spread throughout the sources, boot stages and config options like this: 1. The available physical memory regions are queried and stored as mem_detect information for later use in the decompressor. 2. Based on the physical memory availability the virtual memory layout is established in the decompressor; 3. If CONFIG_KASAN is disabled the kernel paging setup code populates kernel pgtables and turns DAT mode on. It uses the information stored at step [1]. 4. If CONFIG_KASAN is enabled the kernel early boot kasan setup populates kernel pgtables and turns DAT mode on. It uses the information stored at step [1]. The kasan setup creates early_pg_dir directory and directly overwrites swapper_pg_dir entries to make shadow memory pages available. Move the kernel virtual memory setup to the decompressor and start the kernel with DAT turned on right from the very first istruction. That completely eliminates the boot phase when the kernel runs in DAT-off mode, simplies the overall design and consolidates pgtables setup. The identity mapping is created in the decompressor, while kasan shadow mappings are still created by the early boot kernel code. Share with decompressor the existing kasan memory allocator. It decreases the size of a newly requested memory block from pgalloc_pos and ensures that kernel image is not overwritten. pgalloc_low and pgalloc_pos pointers are made preserved boot variables for that. Use the bootdata infrastructure to setup swapper_pg_dir and invalid_pg_dir directories used by the kernel later. The interim early_pg_dir directory established by the kasan initialization code gets eliminated as result. As the kernel runs in DAT-on mode only the PSW_KERNEL_BITS define gets PSW_MASK_DAT bit by default. Additionally, the setup_lowcore_dat_off() and setup_lowcore_dat_on() routines get merged, since there is no DAT-off mode stage anymore. The memory mappings are created with RW+X protection that allows the early boot code setting up all necessary data and services for the kernel being booted. Just before the paging is enabled the memory protection is changed to RO+X for text, RO+NX for read-only data and RW+NX for kernel data and the identity mapping. Reviewed-by: Heiko Carstens <hca@linux.ibm.com> Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com> Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2022-12-13 11:35:11 +01:00
s390: compile relocatable kernel without -fPIE On s390, currently kernel uses the '-fPIE' compiler flag for compiling vmlinux. This has a few problems: - It uses dynamic symbols (.dynsym), for which the linker refuses to allow more than 64k sections. This can break features which use '-ffunction-sections' and '-fdata-sections', including kpatch-build [1] and Function Granular KASLR. - It unnecessarily uses GOT relocations, adding an extra layer of indirection for many memory accesses. Instead of using '-fPIE', resolve all the relocations at link time and then manually adjust any absolute relocations (R_390_64) during boot. This is done by first telling the linker to preserve all relocations during the vmlinux link. (Note this is harmless: they are later stripped in the vmlinux.bin link.) Then use the 'relocs' tool to find all absolute relocations (R_390_64) which apply to allocatable sections. The offsets of those relocations are saved in a special section which is then used to adjust the relocations during boot. (Note: For some reason, Clang occasionally creates a GOT reference, even without '-fPIE'. So Clang-compiled kernels have a GOT, which needs to be adjusted.) On my mostly-defconfig kernel, this reduces kernel text size by ~1.3%. [1] https://github.com/dynup/kpatch/issues/1284 [2] https://gcc.gnu.org/pipermail/gcc-patches/2023-June/622872.html [3] https://gcc.gnu.org/pipermail/gcc-patches/2023-August/625986.html Compiler consideration: Gcc recently implemented an optimization [2] for loading symbols without explicit alignment, aligning with the IBM Z ELF ABI. This ABI mandates symbols to reside on a 2-byte boundary, enabling the use of the larl instruction. However, kernel linker scripts may still generate unaligned symbols. To address this, a new -munaligned-symbols option has been introduced [3] in recent gcc versions. This option has to be used with future gcc versions. Older Clang lacks support for handling unaligned symbols generated by kernel linker scripts when the kernel is built without -fPIE. However, future versions of Clang will include support for the -munaligned-symbols option. When the support is unavailable, compile the kernel with -fPIE to maintain the existing behavior. In addition to it: move vmlinux.relocs to safe relocation When the kernel is built with CONFIG_KERNEL_UNCOMPRESSED, the entire uncompressed vmlinux.bin is positioned in the bzImage decompressor image at the default kernel LMA of 0x100000, enabling it to be executed in-place. However, the size of .vmlinux.relocs could be large enough to cause an overlap with the uncompressed kernel at the address 0x100000. To address this issue, .vmlinux.relocs is positioned after the .rodata.compressed in the bzImage. Nevertheless, in this configuration, vmlinux.relocs will overlap with the .bss section of vmlinux.bin. To overcome that, move vmlinux.relocs to a safe location before clearing .bss and handling relocs. Compile warning fix from Sumanth Korikkar: When kernel is built with CONFIG_LD_ORPHAN_WARN and -fno-PIE, there are several warnings: ld: warning: orphan section `.rela.iplt' from `arch/s390/kernel/head64.o' being placed in section `.rela.dyn' ld: warning: orphan section `.rela.head.text' from `arch/s390/kernel/head64.o' being placed in section `.rela.dyn' ld: warning: orphan section `.rela.init.text' from `arch/s390/kernel/head64.o' being placed in section `.rela.dyn' ld: warning: orphan section `.rela.rodata.cst8' from `arch/s390/kernel/head64.o' being placed in section `.rela.dyn' Orphan sections are sections that exist in an object file but don't have a corresponding output section in the final executable. ld raises a warning when it identifies such sections. Eliminate the warning by placing all .rela orphan sections in .rela.dyn and raise an error when size of .rela.dyn is greater than zero. i.e. Dont just neglect orphan sections. This is similar to adjustment performed in x86, where kernel is built with -fno-PIE. commit 5354e84598f2 ("x86/build: Add asserts for unwanted sections") [sumanthk@linux.ibm.com: rebased Josh Poimboeuf patches and move vmlinux.relocs to safe location] [hca@linux.ibm.com: merged compile warning fix from Sumanth] Tested-by: Sumanth Korikkar <sumanthk@linux.ibm.com> Acked-by: Vasily Gorbik <gor@linux.ibm.com> Signed-off-by: Josh Poimboeuf <jpoimboe@kernel.org> Signed-off-by: Sumanth Korikkar <sumanthk@linux.ibm.com> Link: https://lore.kernel.org/r/20240219132734.22881-4-sumanthk@linux.ibm.com Link: https://lore.kernel.org/r/20240219132734.22881-5-sumanthk@linux.ibm.com Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2024-02-19 14:27:33 +01:00
s390/mm: Uncouple physical vs virtual address spaces The uncoupling physical vs virtual address spaces brings the following benefits to s390: - virtual memory layout flexibility; - closes the address gap between kernel and modules, it caused s390-only problems in the past (e.g. 'perf' bugs); - allows getting rid of trampolines used for module calls into kernel; - allows simplifying BPF trampoline; - minor performance improvement in branch prediction; - kernel randomization entropy is magnitude bigger, as it is derived from the amount of available virtual, not physical memory; The whole change could be described in two pictures below: before and after the change. Some aspects of the virtual memory layout setup are not clarified (number of page levels, alignment, DMA memory), since these are not a part of this change or secondary with regard to how the uncoupling itself is implemented. The focus of the pictures is to explain why __va() and __pa() macros are implemented the way they are. Memory layout in V==R mode: | Physical | Virtual | +- 0 --------------+- 0 --------------+ identity mapping start | | S390_lowcore | Low-address memory | +- 8 KB -----------+ | | | | | identity | phys == virt | | mapping | virt == phys | | | +- AMODE31_START --+- AMODE31_START --+ .amode31 rand. phys/virt start |.amode31 text/data|.amode31 text/data| +- AMODE31_END ----+- AMODE31_END ----+ .amode31 rand. phys/virt start | | | | | | +- __kaslr_offset, __kaslr_offset_phys| kernel rand. phys/virt start | | | | kernel text/data | kernel text/data | phys == kvirt | | | +------------------+------------------+ kernel phys/virt end | | | | | | | | | | | | +- ident_map_size -+- ident_map_size -+ identity mapping end | | | ... unused gap | | | +---- vmemmap -----+ 'struct page' array start | | | virtually mapped | | memory map | | | +- __abs_lowcore --+ | | | Absolute Lowcore | | | +- __memcpy_real_area | | | Real Memory Copy| | | +- VMALLOC_START --+ vmalloc area start | | | vmalloc area | | | +- MODULES_VADDR --+ modules area start | | | modules area | | | +------------------+ UltraVisor Secure Storage limit | | | ... unused gap | | | +KASAN_SHADOW_START+ KASAN shadow memory start | | | KASAN shadow | | | +------------------+ ASCE limit Memory layout in V!=R mode: | Physical | Virtual | +- 0 --------------+- 0 --------------+ | | S390_lowcore | Low-address memory | +- 8 KB -----------+ | | | | | | | | ... unused gap | | | | +- AMODE31_START --+- AMODE31_START --+ .amode31 rand. phys/virt start |.amode31 text/data|.amode31 text/data| +- AMODE31_END ----+- AMODE31_END ----+ .amode31 rand. phys/virt end (<2GB) | | | | | | +- __kaslr_offset_phys | kernel rand. phys start | | | | kernel text/data | | | | | +------------------+ | kernel phys end | | | | | | | | | | | | +- ident_map_size -+ | | | | ... unused gap | | | +- __identity_base + identity mapping start (>= 2GB) | | | identity | phys == virt - __identity_base | mapping | virt == phys + __identity_base | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | +---- vmemmap -----+ 'struct page' array start | | | virtually mapped | | memory map | | | +- __abs_lowcore --+ | | | Absolute Lowcore | | | +- __memcpy_real_area | | | Real Memory Copy| | | +- VMALLOC_START --+ vmalloc area start | | | vmalloc area | | | +- MODULES_VADDR --+ modules area start | | | modules area | | | +- __kaslr_offset -+ kernel rand. virt start | | | kernel text/data | phys == (kvirt - __kaslr_offset) + | | __kaslr_offset_phys +- kernel .bss end + kernel rand. virt end | | | ... unused gap | | | +------------------+ UltraVisor Secure Storage limit | | | ... unused gap | | | +KASAN_SHADOW_START+ KASAN shadow memory start | | | KASAN shadow | | | +------------------+ ASCE limit Unused gaps in the virtual memory layout could be present or not - depending on how partucular system is configured. No page tables are created for the unused gaps. The relative order of vmalloc, modules and kernel image in virtual memory is defined by following considerations: - start of the modules area and end of the kernel should reside within 4GB to accommodate relative 32-bit jumps. The best way to achieve that is to place kernel next to modules; - vmalloc and module areas should locate next to each other to prevent failures and extra reworks in user level tools (makedumpfile, crash, etc.) which treat vmalloc and module addresses similarily; - kernel needs to be the last area in the virtual memory layout to easily distinguish between kernel and non-kernel virtual addresses. That is needed to (again) simplify handling of addresses in user level tools and make __pa() macro faster (see below); Concluding the above, the relative order of the considered virtual areas in memory is: vmalloc - modules - kernel. Therefore, the only change to the current memory layout is moving kernel to the end of virtual address space. With that approach the implementation of __pa() macro is straightforward - all linear virtual addresses less than kernel base are considered identity mapping: phys == virt - __identity_base All addresses greater than kernel base are kernel ones: phys == (kvirt - __kaslr_offset) + __kaslr_offset_phys By contrast, __va() macro deals only with identity mapping addresses: virt == phys + __identity_base .amode31 section is mapped separately and is not covered by __pa() macro. In fact, it could have been handled easily by checking whether a virtual address is within the section or not, but there is no need for that. Thus, let __pa() code do as little machine cycles as possible. The KASAN shadow memory is located at the very end of the virtual memory layout, at addresses higher than the kernel. However, that is not a linear mapping and no code other than KASAN instrumentation or API is expected to access it. When KASLR mode is enabled the kernel base address randomized within a memory window that spans whole unused virtual address space. The size of that window depends from the amount of physical memory available to the system, the limit imposed by UltraVisor (if present) and the vmalloc area size as provided by vmalloc= kernel command line parameter. In case the virtual memory is exhausted the minimum size of the randomization window is forcefully set to 2GB, which amounts to in 15 bits of entropy if KASAN is enabled or 17 bits of entropy in default configuration. The default kernel offset 0x100000 is used as a magic value both in the decompressor code and vmlinux linker script, but it will be removed with a follow-up change. Acked-by: Heiko Carstens <hca@linux.ibm.com> Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com>
2024-03-01 07:15:22 +01:00
s390/mm: start kernel with DAT enabled The setup of the kernel virtual address space is spread throughout the sources, boot stages and config options like this: 1. The available physical memory regions are queried and stored as mem_detect information for later use in the decompressor. 2. Based on the physical memory availability the virtual memory layout is established in the decompressor; 3. If CONFIG_KASAN is disabled the kernel paging setup code populates kernel pgtables and turns DAT mode on. It uses the information stored at step [1]. 4. If CONFIG_KASAN is enabled the kernel early boot kasan setup populates kernel pgtables and turns DAT mode on. It uses the information stored at step [1]. The kasan setup creates early_pg_dir directory and directly overwrites swapper_pg_dir entries to make shadow memory pages available. Move the kernel virtual memory setup to the decompressor and start the kernel with DAT turned on right from the very first istruction. That completely eliminates the boot phase when the kernel runs in DAT-off mode, simplies the overall design and consolidates pgtables setup. The identity mapping is created in the decompressor, while kasan shadow mappings are still created by the early boot kernel code. Share with decompressor the existing kasan memory allocator. It decreases the size of a newly requested memory block from pgalloc_pos and ensures that kernel image is not overwritten. pgalloc_low and pgalloc_pos pointers are made preserved boot variables for that. Use the bootdata infrastructure to setup swapper_pg_dir and invalid_pg_dir directories used by the kernel later. The interim early_pg_dir directory established by the kasan initialization code gets eliminated as result. As the kernel runs in DAT-on mode only the PSW_KERNEL_BITS define gets PSW_MASK_DAT bit by default. Additionally, the setup_lowcore_dat_off() and setup_lowcore_dat_on() routines get merged, since there is no DAT-off mode stage anymore. The memory mappings are created with RW+X protection that allows the early boot code setting up all necessary data and services for the kernel being booted. Just before the paging is enabled the memory protection is changed to RO+X for text, RO+NX for read-only data and RW+NX for kernel data and the identity mapping. Reviewed-by: Heiko Carstens <hca@linux.ibm.com> Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com> Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2022-12-13 11:35:11 +01:00
s390: Add stackprotector support Stackprotector support was previously unavailable on s390 because by default compilers generate code which is not suitable for the kernel: the canary value is accessed via thread local storage, where the address of thread local storage is within access registers 0 and 1. Using those registers also for the kernel would come with a significant performance impact and more complicated kernel entry/exit code, since access registers contents would have to be exchanged on every kernel entry and exit. With the upcoming gcc 16 release new compiler options will become available which allow to generate code suitable for the kernel. [1] Compiler option -mstack-protector-guard=global instructs gcc to generate stackprotector code that refers to a global stackprotector canary value via symbol __stack_chk_guard. Access to this value is guaranteed to occur via larl and lgrl instructions. Furthermore, compiler option -mstack-protector-guard-record generates a section containing all code addresses that reference the canary value. To allow for per task canary values the instructions which load the address of __stack_chk_guard are patched so they access a lowcore field instead: a per task canary value is available within the task_struct of each task, and is written to the per-cpu lowcore location on each context switch. Also add sanity checks and debugging option to be consistent with other kernel code patching mechanisms. Full debugging output can be enabled with the following kernel command line options: debug_stackprotector bootdebug ignore_loglevel earlyprintk dyndbg="file stackprotector.c +p" Example debug output: stackprot: 0000021e402d4eda: c010005a9ae3 -> c01f00070240 where "<insn address>: <old insn> -> <new insn>". [1] gcc commit 0cd1f03939d5 ("s390: Support global stack protector") Reviewed-by: Sven Schnelle <svens@linux.ibm.com> Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2025-11-17 15:09:53 +01:00
s390/boot: rework decompressor reserved tracking Currently several approaches for finding unused memory in decompressor are utilized. While "safe_addr" grows towards higher addresses, vmem code allocates paging structures top down. The former requires careful ordering. In addition to that ipl report handling code verifies potential intersections with secure boot certificates on its own. Neither of two approaches are memory holes aware and consistent with each other in low memory conditions. To solve that, existing approaches are generalized and combined together, as well as online memory ranges are now taken into consideration. physmem_info has been extended to contain reserved memory ranges. New set of functions allow to handle reserves and find unused memory. All reserves and memory allocations are "typed". In case of out of memory condition decompressor fails with detailed info on current reserved ranges and usable online memory. Linux version 6.2.0 ... Kernel command line: ... mem=100M Our of memory allocating 100000 bytes 100000 aligned in range 0:5800000 Reserved memory ranges: 0000000000000000 0000000003e33000 DECOMPRESSOR 0000000003f00000 00000000057648a3 INITRD 00000000063e0000 00000000063e8000 VMEM 00000000063eb000 00000000063f4000 VMEM 00000000063f7800 0000000006400000 VMEM 0000000005800000 0000000006300000 KASAN Usable online memory ranges (info source: sclp read info [3]): 0000000000000000 0000000006400000 Usable online memory total: 6400000 Reserved: 61b10a3 Free: 24ef5d Call Trace: (sp:000000000002bd58 [<0000000000012a70>] physmem_alloc_top_down+0x60/0x14c) sp:000000000002bdc8 [<0000000000013756>] _pa+0x56/0x6a sp:000000000002bdf0 [<0000000000013bcc>] pgtable_populate+0x45c/0x65e sp:000000000002be90 [<00000000000140aa>] setup_vmem+0x2da/0x424 sp:000000000002bec8 [<0000000000011c20>] startup_kernel+0x428/0x8b4 sp:000000000002bf60 [<00000000000100f4>] startup_normal+0xd4/0xd4 physmem_alloc_range allows to find free memory in specified range. It should be used for one time allocations only like finding position for amode31 and vmlinux. physmem_alloc_top_down can be used just like physmem_alloc_range, but it also allows multiple allocations per type and tries to merge sequential allocations together. Which is useful for paging structures allocations. If sequential allocations cannot be merged together they are "chained", allowing easy per type reserved ranges enumeration and migration to memblock later. Extra "struct reserved_range" allocated for chaining are not tracked or reserved but rely on the fact that both physmem_alloc_range and physmem_alloc_top_down search for free memory only below current top down allocator position. All reserved ranges should be transferred to memblock before memblock allocations are enabled. The startup code has been reordered to delay any memory allocations until online memory ranges are detected and occupied memory ranges are marked as reserved to be excluded from follow-up allocations. Ipl report certificates are a special case, ipl report certificates list is checked together with other memory reserves until certificates are saved elsewhere. KASAN required memory for shadow memory allocation and mapping is reserved as 1 large chunk which is later passed to KASAN early initialization code. Acked-by: Heiko Carstens <hca@linux.ibm.com> Reviewed-by: Alexander Gordeev <agordeev@linux.ibm.com> Signed-off-by: Vasily Gorbik <gor@linux.ibm.com> Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2023-02-02 13:59:36 +01:00
s390/mm: start kernel with DAT enabled The setup of the kernel virtual address space is spread throughout the sources, boot stages and config options like this: 1. The available physical memory regions are queried and stored as mem_detect information for later use in the decompressor. 2. Based on the physical memory availability the virtual memory layout is established in the decompressor; 3. If CONFIG_KASAN is disabled the kernel paging setup code populates kernel pgtables and turns DAT mode on. It uses the information stored at step [1]. 4. If CONFIG_KASAN is enabled the kernel early boot kasan setup populates kernel pgtables and turns DAT mode on. It uses the information stored at step [1]. The kasan setup creates early_pg_dir directory and directly overwrites swapper_pg_dir entries to make shadow memory pages available. Move the kernel virtual memory setup to the decompressor and start the kernel with DAT turned on right from the very first istruction. That completely eliminates the boot phase when the kernel runs in DAT-off mode, simplies the overall design and consolidates pgtables setup. The identity mapping is created in the decompressor, while kasan shadow mappings are still created by the early boot kernel code. Share with decompressor the existing kasan memory allocator. It decreases the size of a newly requested memory block from pgalloc_pos and ensures that kernel image is not overwritten. pgalloc_low and pgalloc_pos pointers are made preserved boot variables for that. Use the bootdata infrastructure to setup swapper_pg_dir and invalid_pg_dir directories used by the kernel later. The interim early_pg_dir directory established by the kasan initialization code gets eliminated as result. As the kernel runs in DAT-on mode only the PSW_KERNEL_BITS define gets PSW_MASK_DAT bit by default. Additionally, the setup_lowcore_dat_off() and setup_lowcore_dat_on() routines get merged, since there is no DAT-off mode stage anymore. The memory mappings are created with RW+X protection that allows the early boot code setting up all necessary data and services for the kernel being booted. Just before the paging is enabled the memory protection is changed to RO+X for text, RO+NX for read-only data and RW+NX for kernel data and the identity mapping. Reviewed-by: Heiko Carstens <hca@linux.ibm.com> Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com> Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2022-12-13 11:35:11 +01:00
s390/mm: Uncouple physical vs virtual address spaces The uncoupling physical vs virtual address spaces brings the following benefits to s390: - virtual memory layout flexibility; - closes the address gap between kernel and modules, it caused s390-only problems in the past (e.g. 'perf' bugs); - allows getting rid of trampolines used for module calls into kernel; - allows simplifying BPF trampoline; - minor performance improvement in branch prediction; - kernel randomization entropy is magnitude bigger, as it is derived from the amount of available virtual, not physical memory; The whole change could be described in two pictures below: before and after the change. Some aspects of the virtual memory layout setup are not clarified (number of page levels, alignment, DMA memory), since these are not a part of this change or secondary with regard to how the uncoupling itself is implemented. The focus of the pictures is to explain why __va() and __pa() macros are implemented the way they are. Memory layout in V==R mode: | Physical | Virtual | +- 0 --------------+- 0 --------------+ identity mapping start | | S390_lowcore | Low-address memory | +- 8 KB -----------+ | | | | | identity | phys == virt | | mapping | virt == phys | | | +- AMODE31_START --+- AMODE31_START --+ .amode31 rand. phys/virt start |.amode31 text/data|.amode31 text/data| +- AMODE31_END ----+- AMODE31_END ----+ .amode31 rand. phys/virt start | | | | | | +- __kaslr_offset, __kaslr_offset_phys| kernel rand. phys/virt start | | | | kernel text/data | kernel text/data | phys == kvirt | | | +------------------+------------------+ kernel phys/virt end | | | | | | | | | | | | +- ident_map_size -+- ident_map_size -+ identity mapping end | | | ... unused gap | | | +---- vmemmap -----+ 'struct page' array start | | | virtually mapped | | memory map | | | +- __abs_lowcore --+ | | | Absolute Lowcore | | | +- __memcpy_real_area | | | Real Memory Copy| | | +- VMALLOC_START --+ vmalloc area start | | | vmalloc area | | | +- MODULES_VADDR --+ modules area start | | | modules area | | | +------------------+ UltraVisor Secure Storage limit | | | ... unused gap | | | +KASAN_SHADOW_START+ KASAN shadow memory start | | | KASAN shadow | | | +------------------+ ASCE limit Memory layout in V!=R mode: | Physical | Virtual | +- 0 --------------+- 0 --------------+ | | S390_lowcore | Low-address memory | +- 8 KB -----------+ | | | | | | | | ... unused gap | | | | +- AMODE31_START --+- AMODE31_START --+ .amode31 rand. phys/virt start |.amode31 text/data|.amode31 text/data| +- AMODE31_END ----+- AMODE31_END ----+ .amode31 rand. phys/virt end (<2GB) | | | | | | +- __kaslr_offset_phys | kernel rand. phys start | | | | kernel text/data | | | | | +------------------+ | kernel phys end | | | | | | | | | | | | +- ident_map_size -+ | | | | ... unused gap | | | +- __identity_base + identity mapping start (>= 2GB) | | | identity | phys == virt - __identity_base | mapping | virt == phys + __identity_base | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | +---- vmemmap -----+ 'struct page' array start | | | virtually mapped | | memory map | | | +- __abs_lowcore --+ | | | Absolute Lowcore | | | +- __memcpy_real_area | | | Real Memory Copy| | | +- VMALLOC_START --+ vmalloc area start | | | vmalloc area | | | +- MODULES_VADDR --+ modules area start | | | modules area | | | +- __kaslr_offset -+ kernel rand. virt start | | | kernel text/data | phys == (kvirt - __kaslr_offset) + | | __kaslr_offset_phys +- kernel .bss end + kernel rand. virt end | | | ... unused gap | | | +------------------+ UltraVisor Secure Storage limit | | | ... unused gap | | | +KASAN_SHADOW_START+ KASAN shadow memory start | | | KASAN shadow | | | +------------------+ ASCE limit Unused gaps in the virtual memory layout could be present or not - depending on how partucular system is configured. No page tables are created for the unused gaps. The relative order of vmalloc, modules and kernel image in virtual memory is defined by following considerations: - start of the modules area and end of the kernel should reside within 4GB to accommodate relative 32-bit jumps. The best way to achieve that is to place kernel next to modules; - vmalloc and module areas should locate next to each other to prevent failures and extra reworks in user level tools (makedumpfile, crash, etc.) which treat vmalloc and module addresses similarily; - kernel needs to be the last area in the virtual memory layout to easily distinguish between kernel and non-kernel virtual addresses. That is needed to (again) simplify handling of addresses in user level tools and make __pa() macro faster (see below); Concluding the above, the relative order of the considered virtual areas in memory is: vmalloc - modules - kernel. Therefore, the only change to the current memory layout is moving kernel to the end of virtual address space. With that approach the implementation of __pa() macro is straightforward - all linear virtual addresses less than kernel base are considered identity mapping: phys == virt - __identity_base All addresses greater than kernel base are kernel ones: phys == (kvirt - __kaslr_offset) + __kaslr_offset_phys By contrast, __va() macro deals only with identity mapping addresses: virt == phys + __identity_base .amode31 section is mapped separately and is not covered by __pa() macro. In fact, it could have been handled easily by checking whether a virtual address is within the section or not, but there is no need for that. Thus, let __pa() code do as little machine cycles as possible. The KASAN shadow memory is located at the very end of the virtual memory layout, at addresses higher than the kernel. However, that is not a linear mapping and no code other than KASAN instrumentation or API is expected to access it. When KASLR mode is enabled the kernel base address randomized within a memory window that spans whole unused virtual address space. The size of that window depends from the amount of physical memory available to the system, the limit imposed by UltraVisor (if present) and the vmalloc area size as provided by vmalloc= kernel command line parameter. In case the virtual memory is exhausted the minimum size of the randomization window is forcefully set to 2GB, which amounts to in 15 bits of entropy if KASAN is enabled or 17 bits of entropy in default configuration. The default kernel offset 0x100000 is used as a magic value both in the decompressor code and vmlinux linker script, but it will be removed with a follow-up change. Acked-by: Heiko Carstens <hca@linux.ibm.com> Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com>
2024-03-01 07:15:22 +01:00
s390/mm: start kernel with DAT enabled The setup of the kernel virtual address space is spread throughout the sources, boot stages and config options like this: 1. The available physical memory regions are queried and stored as mem_detect information for later use in the decompressor. 2. Based on the physical memory availability the virtual memory layout is established in the decompressor; 3. If CONFIG_KASAN is disabled the kernel paging setup code populates kernel pgtables and turns DAT mode on. It uses the information stored at step [1]. 4. If CONFIG_KASAN is enabled the kernel early boot kasan setup populates kernel pgtables and turns DAT mode on. It uses the information stored at step [1]. The kasan setup creates early_pg_dir directory and directly overwrites swapper_pg_dir entries to make shadow memory pages available. Move the kernel virtual memory setup to the decompressor and start the kernel with DAT turned on right from the very first istruction. That completely eliminates the boot phase when the kernel runs in DAT-off mode, simplies the overall design and consolidates pgtables setup. The identity mapping is created in the decompressor, while kasan shadow mappings are still created by the early boot kernel code. Share with decompressor the existing kasan memory allocator. It decreases the size of a newly requested memory block from pgalloc_pos and ensures that kernel image is not overwritten. pgalloc_low and pgalloc_pos pointers are made preserved boot variables for that. Use the bootdata infrastructure to setup swapper_pg_dir and invalid_pg_dir directories used by the kernel later. The interim early_pg_dir directory established by the kasan initialization code gets eliminated as result. As the kernel runs in DAT-on mode only the PSW_KERNEL_BITS define gets PSW_MASK_DAT bit by default. Additionally, the setup_lowcore_dat_off() and setup_lowcore_dat_on() routines get merged, since there is no DAT-off mode stage anymore. The memory mappings are created with RW+X protection that allows the early boot code setting up all necessary data and services for the kernel being booted. Just before the paging is enabled the memory protection is changed to RO+X for text, RO+NX for read-only data and RW+NX for kernel data and the identity mapping. Reviewed-by: Heiko Carstens <hca@linux.ibm.com> Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com> Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
2022-12-13 11:35:11 +01:00