With latest llvm17, selftest fexit_bpf2bpf/func_replace_return_code
has the following verification failure:
0: R1=ctx(off=0,imm=0) R10=fp0
; int connect_v4_prog(struct bpf_sock_addr *ctx)
0: (bf) r7 = r1 ; R1=ctx(off=0,imm=0) R7_w=ctx(off=0,imm=0)
1: (b4) w6 = 0 ; R6_w=0
; memset(&tuple.ipv4.saddr, 0, sizeof(tuple.ipv4.saddr));
...
; return do_bind(ctx) ? 1 : 0;
179: (bf) r1 = r7 ; R1=ctx(off=0,imm=0) R7=ctx(off=0,imm=0)
180: (85) call pc+147
Func#3 is global and valid. Skipping.
181: R0_w=scalar()
181: (bc) w6 = w0 ; R0_w=scalar() R6_w=scalar(umax=4294967295,var_off=(0x0; 0xffffffff))
182: (05) goto pc-129
; }
54: (bc) w0 = w6 ; R0_w=scalar(umax=4294967295,var_off=(0x0; 0xffffffff)) R6_w=scalar(umax=4294967295,var_off=(0x0; 0xffffffff))
55: (95) exit
At program exit the register R0 has value (0x0; 0xffffffff) should have been in (0x0; 0x1)
processed 281 insns (limit 1000000) max_states_per_insn 1 total_states 26 peak_states 26 mark_read 13
-- END PROG LOAD LOG --
libbpf: prog 'connect_v4_prog': failed to load: -22
The corresponding source code:
__attribute__ ((noinline))
int do_bind(struct bpf_sock_addr *ctx)
{
struct sockaddr_in sa = {};
sa.sin_family = AF_INET;
sa.sin_port = bpf_htons(0);
sa.sin_addr.s_addr = bpf_htonl(SRC_REWRITE_IP4);
if (bpf_bind(ctx, (struct sockaddr *)&sa, sizeof(sa)) != 0)
return 0;
return 1;
}
...
SEC("cgroup/connect4")
int connect_v4_prog(struct bpf_sock_addr *ctx)
{
...
return do_bind(ctx) ? 1 : 0;
}
Insn 180 is a call to 'do_bind'. The call's return value is also the return value
for the program. Since do_bind() returns 0/1, so it is legitimate for compiler to
optimize 'return do_bind(ctx) ? 1 : 0' to 'return do_bind(ctx)'. However, such
optimization breaks verifier as the return value of 'do_bind()' is marked as any
scalar which violates the requirement of prog return value 0/1.
There are two ways to fix this problem, (1) changing 'return 1' in do_bind() to
e.g. 'return 10' so the compiler has to do 'do_bind(ctx) ? 1 :0', or (2)
suggested by Andrii, marking do_bind() with __weak attribute so the compiler
cannot make any assumption on do_bind() return value.
This patch adopted adding __weak approach which is simpler and more resistant
to potential compiler optimizations.
Suggested-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20230310012410.2920570-1-yhs@fb.com
chenxiang <chenxiang66@hisilicon.com> says:
To support IO_URING IOPOLL support for hisi_sas, we need to:
- Add and fill mq_poll interface to poll queue;
- Ensure internal I/Os (including internal abort I/Os) are delivered and
completed through non-iopoll queue (queue 0);
Sending internal abort commands to non-poll queue actually requires to
sending the abort command to every queue. This carries a a risk. Make iopoll
support module parameter "experimental".
I have tested performance on v3 hw with different modes as follows. 4K
READs and 4K WRITEs both see an improvement when enabling poll mode:
4K READ 4K RANDREAD 4K WRITE 4K RANDWRITE
interrupt + libaio 1770k 1316k 1197k 831k
interrupt + io_uring 1848k 1390k 1238k 857k
iopoll + io_uring 2117k 1364k 1874k 849k
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Currently we sync irq to avoid freeing task before using task in I/O
completion. After adding io_uring support, we need to do something similar
for poll queues. As the process of CQ entries on poll queue are protected
by spinlock cq->lock, we can use spin_lock() + spin_unlock() on cq->lock to
make sure that CQ entries are processed to completion and then the complete
queue is synced.
Signed-off-by: Xiang Chen <chenxiang66@hisilicon.com>
Link: https://lore.kernel.org/r/1678169355-76215-4-git-send-email-chenxiang66@hisilicon.com
Reviewed-by: John Garry <john.garry@huawei.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Add a module parameter to set how many queues are used for iopoll. Also
fill the interface mq_poll. For internal I/Os from libsas and libata we use
non-iopoll queue (queue 0) to deliver and complete them. But for internal
abort I/Os, just don't send them for poll queues.
There is still a risk associated as this sends internal abort commands to
non-iopoll queues which actually requires sending an internal abort command
to every queue. As a result, make the module parameter as "experimental"
for now.
Signed-off-by: Xiang Chen <chenxiang66@hisilicon.com>
Link: https://lore.kernel.org/r/1678169355-76215-3-git-send-email-chenxiang66@hisilicon.com
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Put the work of processing cq slots in a separate function,
complete_v3_hw(), which can then be used by cq_thread_v3_hw() and other
functions when adding poll support.
Co-developed-by: John Garry <john.garry@huawei.com>
Signed-off-by: John Garry <john.garry@huawei.com>
Signed-off-by: Xiang Chen <chenxiang66@hisilicon.com>
Link: https://lore.kernel.org/r/1678169355-76215-2-git-send-email-chenxiang66@hisilicon.com
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Dmitry Bogdanov <d.bogdanov@yadro.com> says:
SAM-5 4.6.5.2 (Relative Port Identifier attribute) defines the attribute as
unique across SCSI target ports:
The Relative Port Identifier attribute identifies a SCSI target port or a
SCSI initiator port relative to other SCSI ports in a SCSI target device
and any SCSI initiator devices contained within that SCSI target device. A
SCSI target device may assign relative port identifiers to its SCSI target
ports and any SCSI initiator ports. If relative port identifiers are
assigned, the SCSI target device shall assign each of its SCSI target
ports and any SCSI initiator ports a unique relative port identifier from
1 to 65 535. SCSI target ports and SCSI initiator ports share the same
number space.
In the current TCM implementation, auto-incremented lun_rtpi weakly follows
the model outlined by SAM-5 and SPC-4. In case of multiple SCSI target
ports (se_portal_groups), which is common to scenario with multiple HBAs or
multiple iSCSI/FC targets, it's possible to have two backstores (se_devices)
with different values of lun_rtpi on the same SCSI target port.
Similar issue happens during re-export. If a LUN of a backstore is removed
from a target port and added again to the same target port, RTPI is
incremented again and will be different from the first time.
The two issues happen because each se_device increments RTPI for its own
LUNs independently.
The behaviour means that a SCSI application client can't reliably make any
sense of RTPI values reported by a LUN as it's not really related to SCSI
target ports. A conforming target implementation must ensure that RTPI field
is unique per port. The patchset resolves the issue.
Make RTPI be part of se_tpg instead of se_lun. Make it configurable.
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
RELATIVE TARGET PORT IDENTIFIER can be read and configured via configfs:
$ echo 0x10 > $TARGET/tpgt_N/rtpi
RTPI can be changed only on disabled target ports.
Co-developed-by: Roman Bolshakov <r.bolshakov@yadro.com>
Signed-off-by: Roman Bolshakov <r.bolshakov@yadro.com>
Signed-off-by: Dmitry Bogdanov <d.bogdanov@yadro.com>
Link: https://lore.kernel.org/r/20230301084512.21956-5-d.bogdanov@yadro.com
Reviewed-by: Mike Christie <michael.christie@oracle.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
The code is not needed since target port-based RTPI allocation replaced it.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Roman Bolshakov <r.bolshakov@yadro.com>
Signed-off-by: Dmitry Bogdanov <d.bogdanov@yadro.com>
Link: https://lore.kernel.org/r/20230301084512.21956-4-d.bogdanov@yadro.com
Reviewed-by: Mike Christie <michael.christie@oracle.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Replace all references to RTPI from LUN field to se_portal_group field. It
introduces consistent reporting of RTPI for all LUNs and all target ports.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Roman Bolshakov <r.bolshakov@yadro.com>
Signed-off-by: Dmitry Bogdanov <d.bogdanov@yadro.com>
Link: https://lore.kernel.org/r/20230301084512.21956-3-d.bogdanov@yadro.com
Reviewed-by: Mike Christie <michael.christie@oracle.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
SAM-5 4.6.5.2 (Relative Port Identifier attribute) defines the attribute as
unique across SCSI target ports.
The change introduces RTPI attribute to se_portal group. The value is
unique across all enabled SCSI target ports. It also limits number of SCSI
target ports to 65535.
Signed-off-by: Dmitry Bogdanov <d.bogdanov@yadro.com>
Link: https://lore.kernel.org/r/20230301084512.21956-2-d.bogdanov@yadro.com
Reviewed-by: Mike Christie <michael.christie@oracle.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Justin Tee <justin.tee@broadcom.com> says:
Update lpfc to revision 14.2.0.11
This patch set contains bug fixes for buffer overflow, resource
management, discovery, and HBA error status handling.
The patches were cut against Martin's 6.3/scsi-queue tree.
Justin Tee (10):
lpfc: Protect against potential lpfc_debugfs_lockstat_write buffer overflow
lpfc: Reorder freeing of various dma buffers and their list removal
lpfc: Fix lockdep warning for rx_monitor lock when unloading driver
lpfc: Record LOGO state with discovery engine even if aborted
lpfc: Defer issuing new PLOGI if received RSCN before completing REG_LOGIN
lpfc: Correct used_rpi count when devloss tmo fires with no recovery
lpfc: Skip waiting for register ready bits when in unrecoverable state
lpfc: Revise lpfc_error_lost_link reason code evaluation logic
lpfc: Update lpfc version to 14.2.0.11
lpfc: Copyright updates for 14.2.0.11 patches
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Extended status reason code errors should mask off the IOERR_PARAM_MASK
before checking strict equalities for IOERR values.
Update the lpfc_error_lost_link() routine as such.
Signed-off-by: Justin Tee <justin.tee@broadcom.com>
Link: https://lore.kernel.org/r/20230301231626.9621-9-justintee8345@gmail.com
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
During tolerance tests that force an HBA to become unresponsive, rmmod
hangs resulting in the inability to remove the driver.
The lpfc_pci_remove_one_s4() routine attempts to submit a clean up mailbox
command via the lpfc_sli4_post_sync_mbox() routine, but ends up waiting
forever for a mailbox register to set its ready bit. Because the HBA is in
an unrecoverable and unresponsive state, the ready bit will never be set.
Create a new routine called lpfc_sli4_unrecoverable_port(), which checks a
port status register's error notification bits.
Use the lpfc_sli4_unrecoverable_port() routine in ready bit check routines
to early return error if port is deemed unrecoverable.
Also, when the lpfc_handle_eratt_s4() handler detects an unrecoverable
state, call the lpfc_sli4_offline_eratt() routine to kick off flushing
outstanding I/O.
Signed-off-by: Justin Tee <justin.tee@broadcom.com>
Link: https://lore.kernel.org/r/20230301231626.9621-8-justintee8345@gmail.com
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
A fabric controller can sometimes send an RDP request right before a link
down event. Because of this outstanding RDP request, the driver does not
remove the last reference count on its ndlp causing a potential leak of RPI
resources when devloss tmo fires.
In lpfc_cmpl_els_rsp(), modify the NPIV clause to always allow the
lpfc_drop_node() routine to execute when not registered with SCSI
transport.
This relaxes the contraint that an NPIV ndlp must be in a specific state in
order to call lpfc_drop node. Logic is revised such that the
lpfc_drop_node() routine is always called to ensure the last ndlp decrement
occurs.
Signed-off-by: Justin Tee <justin.tee@broadcom.com>
Link: https://lore.kernel.org/r/20230301231626.9621-7-justintee8345@gmail.com
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
When mapped to a target with multiple virtual ports, a link bounce
sometimes results in unsuccessful rediscovery of all of the target's
virtual ports. This is because a succession of repeat RSCNs for the
virtual target ports leaves ndlps in the REG_LOGIN state with the
NLP_REG_LOGIN_SEND flag set. With NLP_REG_LOGIN_SEND set, during the next
PLOGI, the driver will UNREG_RPI. When UNREG_RPI is processed, the driver
can be in the middle of PRLI_ISSUE or MAPPED state resulting in an illegal
state transition by the discovery engine and stalling.
Fix by calling the discovery state machine with DEVICE_RECOVERY event
during RSCN processing. This will set the NLP_IGNR_REG_CMPL bit and
prevent the old REG_LOGIN state from advancing. Then for the new PLOGI
issue, add the check for the NLP_IGNR_REG_CMPL bit to delay issuing the new
PLOGI until the queued REG_LOGIN and UNREG_LOGIN have been processed.
Signed-off-by: Justin Tee <justin.tee@broadcom.com>
Link: https://lore.kernel.org/r/20230301231626.9621-6-justintee8345@gmail.com
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
A target vendor array reboot in P2P topology can sometimes result in
unsuccessful rediscovery.
Rework the lpfc_cmpl_els_logo() routine such that when the LOGO completes
as a failure because of driver abort, the LOGO state is still recorded with
the discovery state machine.
This is a small rework to set LOGO completion without forcing a device
removal state change.
Signed-off-by: Justin Tee <justin.tee@broadcom.com>
Link: https://lore.kernel.org/r/20230301231626.9621-5-justintee8345@gmail.com
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Lockdep enabled kernels report a theoretical deadlock state where the
cmf_timer interrupt occurs while the rx_monitor ring is being destroyed.
During rmmod, the cmf_timer is cancelled prior to the
lpfc_rx_monitor_destroy_ring call. This actually eliminates the need to
take the rx_monitor ring lock in lpfc_rx_monitor_destroy_ring. Thus, just
remove lock/unlock of rx_monitor in lpfc_rx_monitor_destroy_ring.
Signed-off-by: Justin Tee <justin.tee@broadcom.com>
Link: https://lore.kernel.org/r/20230301231626.9621-4-justintee8345@gmail.com
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Code sections where DMA resources are freed before list removal are
reworked to ensure item removal before being freed.
Signed-off-by: Justin Tee <justin.tee@broadcom.com>
Link: https://lore.kernel.org/r/20230301231626.9621-3-justintee8345@gmail.com
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
A static code analysis tool flagged the possibility of buffer overflow when
using copy_from_user() for a debugfs entry.
Currently, it is possible that copy_from_user() copies more bytes than what
would fit in the mybuf char array. Add a min() restriction check between
sizeof(mybuf) - 1 and nbytes passed from the userspace buffer to protect
against buffer overflow.
Link: https://lore.kernel.org/r/20230301231626.9621-2-justintee8345@gmail.com
Signed-off-by: Justin Tee <justin.tee@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Like commit c4f135d643 ("workqueue: Wrap flush_workqueue() using a
macro") says, flush_scheduled_work() is dangerous and will be forbidden.
Commit 4d4109d0eb ("[SCSI] mpt fusion: Power Management fixes for MPT SAS
PCI-E controllers") added flush_scheduled_work() call into
mptscsih_suspend(). As of commit 4d4109d0eb, there were several
schedule{,_delayed}_work() calls where flush_scheduled_work() from
mptscsih_suspend() meant to wait for completion. However, which work items
this flush_scheduled_work() call was for was not explained.
Then, schedule_work("struct mptsas_hotplug_event"->work) and
schedule_work(MPT_ADAPTER->sas_persist_task) have been removed by commit
3eb0822c67 ("[SCSI] mpt fusion: Firmware event implementation using
seperate WorkQueue"), and schedule_work("struct
mptsas_send_discovery_event"->work) has been removed by commit f9c34022ea
("[SCSI] mpt fusion: SAS topology scan changes, expander events").
There still remains schedule_work("struct work_queue_wrapper"->work) call
in mptspi.c and schedule_delayed_work("struct mpt_lan_priv"->work) call in
mptlan.c , but I guess that these are not work items which commit
4d4109d0eb meant to wait for completion because these are not per
MPT_ADAPTER work items. If my guess is correct, we no longer need to call
flush_scheduled_work() from mptscsih_suspend().
Link: https://lore.kernel.org/r/0b9ebcfb-b647-1381-0653-b54528a64a86@I-love.SAKURA.ne.jp
Signed-off-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Qinglin Pan <panqinglin00@gmail.com> says:
Svnapot is a RISC-V extension for marking contiguous 4K pages as a non-4K
page. This patch set is for using Svnapot in hugetlb fs and huge vmap.
This patchset adds a Kconfig item for using Svnapot in
"Platform type"->"SVNAPOT extension support". Its default value is on,
and people can set it off if they don't allow kernel to detect Svnapot
hardware support and leverage it.
Tested on:
- qemu rv64 with "Svnapot support" off and svnapot=true.
- qemu rv64 with "Svnapot support" on and svnapot=true.
- qemu rv64 with "Svnapot support" off and svnapot=false.
- qemu rv64 with "Svnapot support" on and svnapot=false.
* b4-shazam-merge:
riscv: mm: support Svnapot in huge vmap
riscv: mm: support Svnapot in hugetlb page
riscv: mm: modify pte format for Svnapot
Link: https://lore.kernel.org/r/20230209131647.17245-1-panqinglin00@gmail.com
[Palmer: fix up the feature ordering in the merge]
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
Trusted boot support requires the platform event log passed up
from the bootloader. In U-Boot, this can now be accomplished with
a reserved memory region, so add a region for this purpose to the
Everest BMC devicetree.
Signed-off-by: Eddie James <eajames@linux.ibm.com>
Link: https://lore.kernel.org/r/20230222015056.3903734-1-eajames@linux.ibm.com
Signed-off-by: Joel Stanley <joel@jms.id.au>
Multi Circular Queue doesn't use outstanding_reqs. However, the UFS clock
scaling functions use outstanding_reqs to determine if there are requests
pending. When MCQ is enabled, this check always returns false.
Hence use active_reqs to check if there are pending requests.
Fixes: eacb139b77 ("scsi: ufs: core: mcq: Enable multi-circular queue")
Signed-off-by: Asutosh Das <quic_asutoshd@quicinc.com>
Reviewed-by: Manivannan Sadhasivam <mani@kernel.org>
Link: https://lore.kernel.org/r/a24e0d646aac70eae0fc5e05fac0c58bb7e6e680.1678317160.git.quic_asutoshd@quicinc.com
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Some storage, such as AIX VDASD (virtual storage) and IBM 2076 (front
end), fail as a result of commit c92a6b5d63 ("scsi: core: Query VPD
size before getting full page").
That commit changed getting SCSI VPD pages so that we now read just
enough of the page to get the actual page size, then read the whole
page in a second read. The problem is that the above mentioned
hardware returns zero for the page size, because of a firmware
error. In such cases, until the firmware is fixed, this new blacklist
flag says to revert to the original method of reading the VPD pages,
i.e. try to read a whole buffer's worth on the first try.
[mkp: reworked somewhat]
Fixes: c92a6b5d63 ("scsi: core: Query VPD size before getting full page")
Reported-by: Martin Wilck <mwilck@suse.com>
Suggested-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Lee Duncan <lduncan@suse.com>
Link: https://lore.kernel.org/r/20220928181350.9948-1-leeman.duncan@gmail.com
Tested-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
After commit c28cd1f343 ("clk: Mark a fwnode as initialized when using
CLK_OF_DECLARE() macro"), drivers/clk/mvebu/kirkwood.c fails to build:
drivers/clk/mvebu/kirkwood.c:358:1: error: expected identifier or '('
CLK_OF_DECLARE(98dx1135_clk, "marvell,mv98dx1135-core-clock",
^
include/linux/clk-provider.h:1367:21: note: expanded from macro 'CLK_OF_DECLARE'
static void __init name##_of_clk_init_declare(struct device_node *np) \
^
<scratch space>:124:1: note: expanded from here
98dx1135_clk_of_clk_init_declare
^
drivers/clk/mvebu/kirkwood.c:358:1: error: invalid digit 'd' in decimal constant
include/linux/clk-provider.h:1372:34: note: expanded from macro 'CLK_OF_DECLARE'
OF_DECLARE_1(clk, name, compat, name##_of_clk_init_declare)
^
<scratch space>:125:3: note: expanded from here
98dx1135_clk_of_clk_init_declare
^
drivers/clk/mvebu/kirkwood.c:358:1: error: invalid digit 'd' in decimal constant
include/linux/clk-provider.h:1372:34: note: expanded from macro 'CLK_OF_DECLARE'
OF_DECLARE_1(clk, name, compat, name##_of_clk_init_declare)
^
<scratch space>:125:3: note: expanded from here
98dx1135_clk_of_clk_init_declare
^
drivers/clk/mvebu/kirkwood.c:358:1: error: invalid digit 'd' in decimal constant
include/linux/clk-provider.h:1372:34: note: expanded from macro 'CLK_OF_DECLARE'
OF_DECLARE_1(clk, name, compat, name##_of_clk_init_declare)
^
<scratch space>:125:3: note: expanded from here
98dx1135_clk_of_clk_init_declare
^
C function names must start with either an alphabetic letter or an
underscore. To avoid generating invalid function names from clock names,
add two underscores to the beginning of the identifier.
Fixes: c28cd1f343 ("clk: Mark a fwnode as initialized when using CLK_OF_DECLARE() macro")
Suggested-by: Saravana Kannan <saravanak@google.com>
Signed-off-by: Nathan Chancellor <nathan@kernel.org>
Link: https://lore.kernel.org/r/20230308-clk_of_declare-fix-v1-1-317b741e2532@kernel.org
Reviewed-by: Saravana Kannan <saravanak@google.com>
Reported-by: Naresh Kamboju <naresh.kamboju@linaro.org>
Signed-off-by: Stephen Boyd <sboyd@kernel.org>
In bq24190_probe, &bdi->input_current_limit_work is bound
with bq24190_input_current_limit_work. When external power
changed, it will call bq24190_charger_external_power_changed
to start the work.
If we remove the module which will call bq24190_remove to make
cleanup, there may be a unfinished work. The possible
sequence is as follows:
CPU0 CPUc1
|bq24190_input_current_limit_work
bq24190_remove |
power_supply_unregister |
device_unregister |
power_supply_dev_release|
kfree(psy) |
|
| power_supply_get_property_from_supplier
| //use
Fix it by finishing the work before cleanup in the bq24190_remove
Fixes: 9777467257 ("power_supply: Initialize changed_work before calling device_add")
Signed-off-by: Zheng Wang <zyytlz.wz@163.com>
Signed-off-by: Sebastian Reichel <sebastian.reichel@collabora.com>
Sergey Matyukevich <geomatsi@gmail.com> says:
Some time ago two different patches have been posted to fix stale TLB
entries that caused applications crashes.
The patch [0] suggested 'aggregating' mm_cpumask, i.e. current cpu is not
cleared for the switched-out task in switch_mm function. For additional
explanations see the commit message by Guo Ren. The same approach is
used by arc architecture, so another good comment is for switch_mm
in arch/arc/include/asm/mmu_context.h.
The patch [1] attempted to reduce the number of TLB flushes by deferring
(and possibly avoiding) them for CPUs not running the task.
Patch [1] has been merged. However we already have two bug reports from
different vendors. So apparently something is missing in the approach
suggested in [1]. In both cases the patch [0] fixed the issue.
This patch series reverts [1] and replaces it by [0].
[0] https://lore.kernel.org/linux-riscv/20221111075902.798571-1-guoren@kernel.org/
[1] https://lore.kernel.org/linux-riscv/20220829205219.283543-1-geomatsi@gmail.com/
* b4-shazam-merge:
riscv: asid: Fixup stale TLB entry cause application crash
Revert "riscv: mm: notify remote harts about mmu cache updates"
Link: https://lore.kernel.org/r/20230226150137.1919750-1-geomatsi@gmail.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
After use_asid_allocator is enabled, the userspace application will
crash by stale TLB entries. Because only using cpumask_clear_cpu without
local_flush_tlb_all couldn't guarantee CPU's TLB entries were fresh.
Then set_mm_asid would cause the user space application to get a stale
value by stale TLB entry, but set_mm_noasid is okay.
Here is the symptom of the bug:
unhandled signal 11 code 0x1 (coredump)
0x0000003fd6d22524 <+4>: auipc s0,0x70
0x0000003fd6d22528 <+8>: ld s0,-148(s0) # 0x3fd6d92490
=> 0x0000003fd6d2252c <+12>: ld a5,0(s0)
(gdb) i r s0
s0 0x8082ed1cc3198b21 0x8082ed1cc3198b21
(gdb) x /2x 0x3fd6d92490
0x3fd6d92490: 0xd80ac8a8 0x0000003f
The core dump file shows that register s0 is wrong, but the value in
memory is correct. Because 'ld s0, -148(s0)' used a stale mapping entry
in TLB and got a wrong result from an incorrect physical address.
When the task ran on CPU0, which loaded/speculative-loaded the value of
address(0x3fd6d92490), then the first version of the mapping entry was
PTWed into CPU0's TLB.
When the task switched from CPU0 to CPU1 (No local_tlb_flush_all here by
asid), it happened to write a value on the address (0x3fd6d92490). It
caused do_page_fault -> wp_page_copy -> ptep_clear_flush ->
ptep_get_and_clear & flush_tlb_page.
The flush_tlb_page used mm_cpumask(mm) to determine which CPUs need TLB
flush, but CPU0 had cleared the CPU0's mm_cpumask in the previous
switch_mm. So we only flushed the CPU1 TLB and set the second version
mapping of the PTE. When the task switched from CPU1 to CPU0 again, CPU0
still used a stale TLB mapping entry which contained a wrong target
physical address. It raised a bug when the task happened to read that
value.
CPU0 CPU1
- switch 'task' in
- read addr (Fill stale mapping
entry into TLB)
- switch 'task' out (no tlb_flush)
- switch 'task' in (no tlb_flush)
- write addr cause pagefault
do_page_fault() (change to
new addr mapping)
wp_page_copy()
ptep_clear_flush()
ptep_get_and_clear()
& flush_tlb_page()
write new value into addr
- switch 'task' out (no tlb_flush)
- switch 'task' in (no tlb_flush)
- read addr again (Use stale
mapping entry in TLB)
get wrong value from old phyical
addr, BUG!
The solution is to keep all CPUs' footmarks of cpumask(mm) in switch_mm,
which could guarantee to invalidate all stale TLB entries during TLB
flush.
Fixes: 65d4b9c530 ("RISC-V: Implement ASID allocator")
Signed-off-by: Guo Ren <guoren@linux.alibaba.com>
Signed-off-by: Guo Ren <guoren@kernel.org>
Tested-by: Lad Prabhakar <prabhakar.mahadev-lad.rj@bp.renesas.com>
Tested-by: Zong Li <zong.li@sifive.com>
Tested-by: Sergey Matyukevich <sergey.matyukevich@syntacore.com>
Cc: Anup Patel <apatel@ventanamicro.com>
Cc: Palmer Dabbelt <palmer@rivosinc.com>
Cc: stable@vger.kernel.org
Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Link: https://lore.kernel.org/r/20230226150137.1919750-3-geomatsi@gmail.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
This reverts the remaining bits of commit 4bd1d80efb ("riscv: mm:
notify remote harts harts about mmu cache updates").
According to bug reports, suggested approach to fix stale TLB entries
is not sufficient. It needs to be replaced by a more robust solution.
Fixes: 4bd1d80efb ("riscv: mm: notify remote harts about mmu cache updates")
Reported-by: Zong Li <zong.li@sifive.com>
Reported-by: Lad Prabhakar <prabhakar.mahadev-lad.rj@bp.renesas.com>
Signed-off-by: Sergey Matyukevich <sergey.matyukevich@syntacore.com>
Cc: stable@vger.kernel.org
Reviewed-by: Guo Ren <guoren@kernel.org>
Link: https://lore.kernel.org/r/20230226150137.1919750-2-geomatsi@gmail.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
Variable 'pirq', which may receive negative value
in platform_get_irq().
Used as an index in a function regmap_irq_get_virq().
Found by Linux Verification Center (linuxtesting.org) with SVACE.
Signed-off-by: Denis Arefev <arefev@swemel.ru>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Sebastian Reichel <sebastian.reichel@collabora.com>
This doesn't need to be printed every second as an error:
...
<3>[17438.628385] cros-usbpd-charger cros-usbpd-charger.3.auto: Port 1: default case!
<3>[17439.634176] cros-usbpd-charger cros-usbpd-charger.3.auto: Port 1: default case!
<3>[17440.640298] cros-usbpd-charger cros-usbpd-charger.3.auto: Port 1: default case!
...
Reduce priority from ERROR to DEBUG.
Signed-off-by: Grant Grundler <grundler@chromium.org>
Reviewed-by: Guenter Roeck <groeck@chromium.org>
Signed-off-by: Sebastian Reichel <sebastian.reichel@collabora.com>
The tmp is defined as u32 type, which results in invalid processing of
tmp<0 in function rk817_read_or_set_full_charge_on_boot(). Therefore,
drop the comparison.
drivers/power/supply/rk817_charger.c:828 rk817_read_or_set_full_charge_on_boot() warn: unsigned 'tmp' is never less than zero.
drivers/power/supply/rk817_charger.c:788 rk817_read_or_set_full_charge_on_boot() warn: unsigned 'tmp' is never less than zero.
Link: https://bugzilla.openanolis.cn/show_bug.cgi?id=3444
Reported-by: Abaci Robot <abaci@linux.alibaba.com>
Signed-off-by: Jiapeng Chong <jiapeng.chong@linux.alibaba.com>
Tested-by: Chris Morgan <macromorgan@hotmail.com>
Signed-off-by: Sebastian Reichel <sebastian.reichel@collabora.com>
We're currently using stop_machine() to update ftrace & kprobes, which
means that the thread that takes text_mutex during may not be the same
as the thread that eventually patches the code. This isn't actually a
race because the lock is still held (preventing any other concurrent
accesses) and there is only one thread running during stop_machine(),
but it does trigger a lockdep failure.
This patch just elides the lockdep check during stop_machine.
Fixes: c15ac4fd60 ("riscv/ftrace: Add dynamic function tracer support")
Suggested-by: Steven Rostedt <rostedt@goodmis.org>
Reported-by: Changbin Du <changbin.du@gmail.com>
Signed-off-by: Palmer Dabbelt <palmerdabbelt@google.com>
Signed-off-by: Conor Dooley <conor.dooley@microchip.com>
Link: https://lore.kernel.org/r/20230303143754.4005217-1-conor.dooley@microchip.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
A recent commit added support for an optional interrupt which is only
available on some platforms.
Stop spamming the logs with bogus error messages on platforms that do
not use this new optional resource:
qcom_scm firmware:scm: error -ENXIO: IRQ index 0 not found
Fixes: 6bf3259922 ("firmware: qcom: scm: Add wait-queue handling logic")
Cc: Guru Das Srinagesh <quic_gurus@quicinc.com>
Cc: Sibi Sankar <quic_sibis@quicinc.com>
Signed-off-by: Johan Hovold <johan+linaro@kernel.org>
Tested-by: Steev Klimaszewski <steev@kali.org> # Thinkpad X13s
Acked-by: Guru Das Srinagesh <quic_gurus@quicinc.com>
Reviewed-by: Konrad Dybcio <konrad.dybcio@linaro.org>
Signed-off-by: Bjorn Andersson <andersson@kernel.org>
Link: https://lore.kernel.org/r/20230309111209.31606-1-johan+linaro@kernel.org
The UFS controller on SM8550 supports cache coherency, hence add the
"dma-coherent" property to mark it as such.
Fixes: 35cf1aaab1 ("arm64: dts: qcom: sm8550: Add UFS host controller and phy nodes")
Signed-off-by: Manivannan Sadhasivam <manivannan.sadhasivam@linaro.org>
Reviewed-by: Neil Armstrong <neil.armstrong@linaro.org>
Reviewed-by: Konrad Dybcio <konrad.dybcio@linaro.org>
Signed-off-by: Bjorn Andersson <andersson@kernel.org>
Link: https://lore.kernel.org/r/20230308054630.7202-1-manivannan.sadhasivam@linaro.org
The cdsp.mbn firmware that's referenced in sa8540p-ride.dts is actually
named cdsp0.mbn in the deliverables from Qualcomm. Let's go ahead and
correct the name to match what's in Qualcomm's deliverable.
Signed-off-by: Brian Masney <bmasney@redhat.com>
Signed-off-by: Bjorn Andersson <andersson@kernel.org>
Link: https://lore.kernel.org/r/20230307232340.2370476-1-bmasney@redhat.com
A user should be allowed to take out a lease via an idmapped mount if
the fsuid matches the mapped uid of the inode. generic_setlease() is
checking the unmapped inode uid, causing these operations to be denied.
Fix this by comparing against the mapped inode uid instead of the
unmapped uid.
Fixes: 9caccd4154 ("fs: introduce MOUNT_ATTR_IDMAP")
Cc: stable@vger.kernel.org
Signed-off-by: Seth Forshee (DigitalOcean) <sforshee@kernel.org>
Signed-off-by: Christian Brauner (Microsoft) <brauner@kernel.org>
This patch adds multi-buffer support for the i40e_driver.
i40e_clean_rx_irq() is modified to collate all the buffers of a packet
before calling the XDP program. xdp_buff is built for the first frag of
the packet and subsequent frags are added to it. 'next_to_process' is
incremented for all non-EOP frags while 'next_to_clean' stays at the
first descriptor of the packet. XDP program is called only on receiving
EOP frag.
New functions are added for adding frags to xdp_buff and for post
processing of the buffers once the xdp prog has run. For XDP_PASS this
results in a skb with multiple fragments.
i40e_build_skb() builds the skb around xdp buffer that already contains
frags data. So i40e_add_rx_frag() helper function is now removed. Since
fields before 'dataref' in skb_shared_info are cleared during
napi_skb_build(), xdp_update_skb_shared_info() is called to set those.
For i40e_construct_skb(), all the frags data needs to be copied from
xdp_buffer's shared_skb_info to newly constructed skb's shared_skb_info.
This also means 'skb' does not need to be preserved across i40e_napi_poll()
calls and hence is removed from i40e_ring structure.
Previously i40e_alloc_rx_buffers() was called for every 32 cleaned
buffers. For multi-buffers this may not be optimal as there may be more
cleaned buffers in each i40e_clean_rx_irq() call. So this is now called
when at least half of the ring size has been cleaned.
Signed-off-by: Tirthendu Sarkar <tirthendu.sarkar@intel.com>
Tested-by: Chandan Kumar Rout <chandanx.rout@intel.com> (A Contingent Worker at Intel)
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Store xdp_buff on Rx ring struct in preparation for XDP multi-buffer
support. This will allow us to combine fragmented frames across
separate NAPI cycles in the same way as currently skb fragments are
handled. This means that skb pointer on Rx ring will become redundant
and will be removed in a later patch. As a consequence i40e_trace() now
uses xdp instead of skb pointer.
Truesize only needs to be calculated for page sizes bigger than 4k as it
is always half-page for 4k pages. With xdp_buff on ring, frame size can
now be set during xdp_init_buff() and need not be repopulated in each
NAPI call for 4k pages. As a consequence i40e_rx_frame_truesize() is now
used only for bigger pages.
Signed-off-by: Tirthendu Sarkar <tirthendu.sarkar@intel.com>
Tested-by: Chandan Kumar Rout <chandanx.rout@intel.com> (A Contingent Worker at Intel)
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Add a new field called next_to_process in the i40e_ring that is
advanced for every buffer and change the semantics of next_to_clean to
point to the first buffer of a packet. Driver will use next_to_process
in the same way next_to_clean was used previously.
For the non multi-buffer case, next_to_process and next_to_clean will
always be the same since each packet consists of a single buffer.
Signed-off-by: Tirthendu Sarkar <tirthendu.sarkar@intel.com>
Tested-by: Chandan Kumar Rout <chandanx.rout@intel.com> (A Contingent Worker at Intel)
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
In skb path truesize is calculated while building skb. This is now
avoided and xdp->frame_is used instead for both i40e_build_skb() and
i40e_construct_skb().
Signed-off-by: Tirthendu Sarkar <tirthendu.sarkar@intel.com>
Tested-by: Chandan Kumar Rout <chandanx.rout@intel.com> (A Contingent Worker at Intel)
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
