This adds support for IORING_OP_MSG_RING, which allows an SQE to signal
another ring. That allows either waking up someone waiting on the ring,
or even passing a 64-bit value via the user_data field in the CQE.
sqe->fd must contain the fd of a ring that should receive the CQE.
sqe->off will be propagated to the cqe->user_data on the target ring,
and sqe->len will be propagated to cqe->res. The results CQE will have
IORING_CQE_F_MSG set in its flags, to indicate that this CQE was generated
from a messaging request rather than a SQE issued locally on that ring.
This effectively allows passing a 64-bit and a 32-bit quantify between
the two rings.
This request type has the following request specific error cases:
- -EBADFD. Set if the sqe->fd doesn't point to a file descriptor that is
of the io_uring type.
- -EOVERFLOW. Set if we were not able to deliver a request to the target
ring.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
driver keeps the same survey_info struct for its lifetime around.
This is used because while firmware does help by providing accounting
information of the current channel, it doesn't keep track on the
remaining channels.
Signed-off-by: Christian Lamparter <chunkeey@gmail.com>
Signed-off-by: Kalle Valo <quic_kvalo@quicinc.com>
Link: https://lore.kernel.org/r/9b9a059b384f49587f4384ba47f01d3fb2115b22.1646250537.git.chunkeey@gmail.com
Running yamllint is effectively required for binding schemas, so print a
warning if not found rather than silently skipping running it.
Signed-off-by: Rob Herring <robh@kernel.org>
Reviewed-by: Krzysztof Kozlowski <krzysztof.kozlowski@canonical.com>
Link: https://lore.kernel.org/r/20220303221417.2486268-1-robh@kernel.org
The device_node pointer is returned by of_parse_phandle() with refcount
incremented. We should use of_node_put() on it when done.
This function only calls of_node_put() in the regular path.
And it will cause refcount leak in error path.
Fixes: 727fec790e ("ath10k: Setup the msa resources before qmi init")
Signed-off-by: Miaoqian Lin <linmq006@gmail.com>
Reviewed-by: Jeff Johnson <quic_jjohnson@quicinc.com>
Signed-off-by: Kalle Valo <quic_kvalo@quicinc.com>
Link: https://lore.kernel.org/r/20220308070238.19295-1-linmq006@gmail.com
All work currently pending will be done first by calling destroy_workqueue,
so there is no need to flush it explicitly.
Reported-by: Zeal Robot <zealci@zte.com.cn>
Signed-off-by: Lv Ruyi (CGEL ZTE) <lv.ruyi@zte.com.cn>
Signed-off-by: Kalle Valo <quic_kvalo@quicinc.com>
Link: https://lore.kernel.org/r/20220301013246.2052570-1-lv.ruyi@zte.com.cn
The calls to devm_gpiod_get_optional() have the -gpios at the end of the
name but the GPIO core code is already adding the suffix during the lookup.
This means the PCIe driver is not finding the necessary reset or power
lines to allow initialisation of the PCIe.
Drop the redundant '-gpios' when we look up GPIOs from the DT.
This bug has not been noticed because if U-Boot has setup the GPIO lines
for the hardware when it does the PCIe initialisation (either by booting
from PCIe or user command to access PCIe) then the PCIe will work in Linux.
The U-Boot as supplied by SiFive does not by default initialise any PCIe
component.
Link: https://lore.kernel.org/r/20220221210347.1335004-1-ben.dooks@codethink.co.uk
Signed-off-by: Ben Dooks <ben.dooks@codethink.co.uk>
Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Use helper macro __ATTR_RW to define HYPERVISOR_ATTR_RW to make code more clear.
Minor readability improvement.
Remove extra whitespace [boris: added this comment]
Signed-off-by: zhanglianjie <zhanglianjie@uniontech.com>
Reviewed-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Link: https://lore.kernel.org/r/20220305133823.158961-1-zhanglianjie@uniontech.com
Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Fix the following W=1 kernel warnings:
arch/x86/xen/setup.c:725: warning: expecting prototype for
machine_specific_memory_setup(). Prototype was for xen_memory_setup()
instead.
Reported-by: Abaci Robot <abaci@linux.alibaba.com>
Signed-off-by: Jiapeng Chong <jiapeng.chong@linux.alibaba.com>
Reviewed-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Link: https://lore.kernel.org/r/20220307062554.8334-1-jiapeng.chong@linux.alibaba.com
Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
The sched_clock() can be used very early since commit 857baa87b6
("sched/clock: Enable sched clock early"). In addition, with commit
38669ba205 ("x86/xen/time: Output xen sched_clock time from 0"), kdump
kernel in Xen HVM guest may panic at very early stage when accessing
&__this_cpu_read(xen_vcpu)->time as in below:
setup_arch()
-> init_hypervisor_platform()
-> x86_init.hyper.init_platform = xen_hvm_guest_init()
-> xen_hvm_init_time_ops()
-> xen_clocksource_read()
-> src = &__this_cpu_read(xen_vcpu)->time;
This is because Xen HVM supports at most MAX_VIRT_CPUS=32 'vcpu_info'
embedded inside 'shared_info' during early stage until xen_vcpu_setup() is
used to allocate/relocate 'vcpu_info' for boot cpu at arbitrary address.
However, when Xen HVM guest panic on vcpu >= 32, since
xen_vcpu_info_reset(0) would set per_cpu(xen_vcpu, cpu) = NULL when
vcpu >= 32, xen_clocksource_read() on vcpu >= 32 would panic.
This patch calls xen_hvm_init_time_ops() again later in
xen_hvm_smp_prepare_boot_cpu() after the 'vcpu_info' for boot vcpu is
registered when the boot vcpu is >= 32.
This issue can be reproduced on purpose via below command at the guest
side when kdump/kexec is enabled:
"taskset -c 33 echo c > /proc/sysrq-trigger"
The bugfix for PVM is not implemented due to the lack of testing
environment.
[boris: xen_hvm_init_time_ops() returns on errors instead of jumping to end]
Cc: Joe Jin <joe.jin@oracle.com>
Signed-off-by: Dongli Zhang <dongli.zhang@oracle.com>
Reviewed-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Link: https://lore.kernel.org/r/20220302164032.14569-3-dongli.zhang@oracle.com
Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Use the helper function time_is_{before,after}_jiffies() to improve
code readability.
Signed-off-by: Wang Qing <wangqing@vivo.com>
Reviewed-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Link: https://lore.kernel.org/r/1646018104-61415-1-git-send-email-wangqing@vivo.com
Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
It is allocated with kvmalloc(), the corresponding release function
should not be kfree(), use kvfree() instead.
Generated by: scripts/coccinelle/api/kfree_mismatch.cocci
Fixes: b530e9e106 ("bpf: Add "live packet" mode for XDP in BPF_PROG_RUN")
Signed-off-by: Yihao Han <hanyihao@vivo.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Cc: Toke Høiland-Jørgensen <toke@redhat.com>
Link: https://lore.kernel.org/bpf/20220310092828.13405-1-hanyihao@vivo.com
The kernel test robot pointed out that the newly added
bpf_test_run_xdp_live() runner doesn't set the retval in the caller (by
design), which means that the variable can be passed unitialised to
bpf_test_finish(). Fix this by initialising the variable properly.
Fixes: b530e9e106 ("bpf: Add "live packet" mode for XDP in BPF_PROG_RUN")
Reported-by: kernel test robot <lkp@intel.com>
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20220310110228.161869-1-toke@redhat.com
Commit 1a56c18e6c ("bpftool: Stop supporting BPF offload-enabled
feature probing") removed the support to probe for BPF offload features.
This is still something that is useful for NFP NIC that can support
offloading of BPF programs.
The reason for the dropped support was that libbpf starting with v1.0
would drop support for passing the ifindex to the BPF prog/map/helper
feature probing APIs. In order to keep this useful feature for NFP
restore the functionality by moving it directly into bpftool.
The code restored is a simplified version of the code that existed in
libbpf which supposed passing the ifindex. The simplification is that it
only targets the cases where ifindex is given and call into libbpf for
the cases where it's not.
Before restoring support for probing offload features:
# bpftool feature probe dev ens4np0
Scanning system call availability...
bpf() syscall is available
Scanning eBPF program types...
Scanning eBPF map types...
Scanning eBPF helper functions...
eBPF helpers supported for program type sched_cls:
eBPF helpers supported for program type xdp:
Scanning miscellaneous eBPF features...
Large program size limit is NOT available
Bounded loop support is NOT available
ISA extension v2 is NOT available
ISA extension v3 is NOT available
With support for probing offload features restored:
# bpftool feature probe dev ens4np0
Scanning system call availability...
bpf() syscall is available
Scanning eBPF program types...
eBPF program_type sched_cls is available
eBPF program_type xdp is available
Scanning eBPF map types...
eBPF map_type hash is available
eBPF map_type array is available
Scanning eBPF helper functions...
eBPF helpers supported for program type sched_cls:
- bpf_map_lookup_elem
- bpf_get_prandom_u32
- bpf_perf_event_output
eBPF helpers supported for program type xdp:
- bpf_map_lookup_elem
- bpf_get_prandom_u32
- bpf_perf_event_output
- bpf_xdp_adjust_head
- bpf_xdp_adjust_tail
Scanning miscellaneous eBPF features...
Large program size limit is NOT available
Bounded loop support is NOT available
ISA extension v2 is NOT available
ISA extension v3 is NOT available
Signed-off-by: Niklas Söderlund <niklas.soderlund@corigine.com>
Signed-off-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Quentin Monnet <quentin@isovalent.com>
Link: https://lore.kernel.org/bpf/20220310121846.921256-1-niklas.soderlund@corigine.com
Rename the PIPECONF refresh rate select bits to be
less cryptic. Also nothing eDP specific about these as they
also select between FP0 vs. FP1 for the DPLL and thus can be
used to change the refresh rate on other output types as well.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220310004802.16310-9-ville.syrjala@linux.intel.com
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
So far, s390 registered a krobe on __kretprobe_trampoline which is
called everytime a kretprobe fires. This kprobe would then determine
the correct return address and adjust the psw accordingly, such that
the kretprobe would branch to the appropriate address after completion.
Some other archs handle kretprobes without such an additional kprobe.
This approach is adopted to s390 with this patch.
Furthermore, the __kretprobe_trampoline now uses an assembler function
to correctly gather the register and psw content to be passed to the
registered kretprobe handler as struct pt_regs. After completion, the
register content and the psw are set based on the contents of said
pt_regs struct.
Note that a change to the psw address in struct pt_regs will not have
an impact, as the probe will still return to the original return
address of the probed function.
The return address is now recovered by using the appropriate function
arch_kretprobe_fixup_return.
The no longer needed kprobe is removed.
Reviewed-by: Vasily Gorbik <gor@linux.ibm.com>
Signed-off-by: Tobias Huschle <huschle@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
With z10 as minimum supported machine generation many ".insn" encodings
could be now converted to instruction names. There are couple of exceptions
- stfle is used from the als code built for z900 and cannot be converted
- few ".insn" directives encode unsupported instruction formats
The generated code is identical before/after this change.
Acked-by: Ilya Leoshkevich <iii@linux.ibm.com>
Reviewed-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
With z10 as minimum supported machine generation the store-clock-fast
facility (25) is always present and checked in als code.
Drop alternatives and always use stckf.
Acked-by: Ilya Leoshkevich <iii@linux.ibm.com>
Reviewed-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Assembler generated expoline thunks were in a form
__s390_indirect_jump_rXuse_rX when exrl instruction has not been available.
Now with z10 as minimum supported machine generation there
is no need for 2 register thunks, always generate
__s390_indirect_jump_rX versions.
Acked-by: Heiko Carstens <hca@linux.ibm.com>
Acked-by: Sumanth Korikkar <sumanthk@linux.ibm.com>
Acked-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Machine generations up to z9 (released in May 2006) have been officially
out of service for several years now (z9 end of service - January 31, 2019).
No distributions build kernels supporting those old machine generations
anymore, except Debian, which seems to pick the oldest supported
generation. The team supporting Debian on s390 has been notified about
the change.
Raising minimum supported machine generation to z10 helps to reduce
maintenance cost and effectively remove code, which is not getting
enough testing coverage due to lack of older hardware and distributions
support. Besides that this unblocks some optimization opportunities and
allows to use wider instruction set in asm files for future features
implementation. Due to this change spectre mitigation and usercopy
implementations could be drastically simplified and many newer instructions
could be converted from ".insn" encoding to instruction names.
Acked-by: Ilya Leoshkevich <iii@linux.ibm.com>
Reviewed-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Add copy_from/to_user_key functions, which perform storage key checking.
These functions can be used by KVM for emulating instructions that need
to be key checked.
These functions differ from their non _key counterparts in
include/linux/uaccess.h only in the additional key argument and must be
kept in sync with those.
Since the existing uaccess implementation on s390 makes use of move
instructions that support having an additional access key supplied,
we can implement raw_copy_from/to_user_key by enhancing the
existing implementation.
Signed-off-by: Janis Schoetterl-Glausch <scgl@linux.ibm.com>
Acked-by: Heiko Carstens <hca@linux.ibm.com>
Reviewed-by: Christian Borntraeger <borntraeger@linux.ibm.com>
Acked-by: Janosch Frank <frankja@linux.ibm.com>
Link: https://lore.kernel.org/r/20220211182215.2730017-2-scgl@linux.ibm.com
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Kernel has full control over how extern thunks generated by
arch/s390/lib/expoline.S look like. Align them to 16 bytes like other
symbols. Also set proper symbols size which is important for tooling.
Acked-by: Ilya Leoshkevich <iii@linux.ibm.com>
Acked-by: Sumanth Korikkar <sumanthk@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Currently with -mindirect-branch=thunk and -mfunction-return=thunk compiler
options expoline thunks are put into individual COMDAT group sections. s390
is the only architecture which has group sections and it has implications
for kpatch and objtool tools support.
Using -mindirect-branch=thunk-extern and -mfunction-return=thunk-extern
is an alternative, which comes with a need to generate all required
expoline thunks manually. Unfortunately modules area is too far away from
the kernel image, and expolines from the kernel image cannon be used.
But since all new distributions (except Debian) build kernels for machine
generations newer than z10, where "exrl" instruction is available, that
leaves only 16 expolines thunks possible.
Provide an option to build the kernel with
-mindirect-branch=thunk-extern and -mfunction-return=thunk-extern for
z10 or newer. This also requires to postlink expoline thunks into all
modules explicitly. Currently modules already contain most expolines
anyhow.
Unfortunately -mindirect-branch=thunk-extern and
-mfunction-return=thunk-extern options support is broken in gcc <= 11.2.
Additional compile test is required to verify proper gcc support.
Acked-by: Ilya Leoshkevich <iii@linux.ibm.com>
Co-developed-by: Sumanth Korikkar <sumanthk@linux.ibm.com>
Signed-off-by: Sumanth Korikkar <sumanthk@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Currently assembler generated expoline thunks are always in a form
__s390_indirect_jump_rXuse_rX even when exrl instruction is available
and no additional register is utilized.
Generate __s390_indirect_jump_rX versions using a single register if the
kernel is built for z10 or newer machine, which have exrl instruction
available. Thunks generated are identical to the ones generated by the
compiler.
This helps to reduce the number of thunks for newer machines generations.
Acked-by: Sumanth Korikkar <sumanthk@linux.ibm.com>
Acked-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Commit c1e18c17bd ("s390/pci: add zpci_set_irq()/zpci_clear_irq()")
made zpci_set_irq()/zpci_clear_irq() non-static in preparation for using
them in zpci_hot_reset_device(). The version of zpci_hot_reset_device()
that was finally merged however exploits the fact that IRQs and DMA is
implicitly disabled by clp_disable_fh() so the call to zpci_clear_irq()
was never added. There are no other calls outside pci_irq.c so lets make
both functions static.
Signed-off-by: Niklas Schnelle <schnelle@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Allow using old PCIe card on Armada 37xx
-----BEGIN PGP SIGNATURE-----
iF0EABECAB0WIQQYqXDMF3cvSLY+g9cLBhiOFHI71QUCYioC+wAKCRALBhiOFHI7
1YTCAJ9B44SBSCWSGFEBeO/aE6XxT3EbcQCgiMUGAwdvTQ7rYUNiCxc2/In72L4=
=wqVm
-----END PGP SIGNATURE-----
gpgsig -----BEGIN PGP SIGNATURE-----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=YxFV
-----END PGP SIGNATURE-----
Merge tag 'mvebu-fixes-5.17-2' of git://git.kernel.org/pub/scm/linux/kernel/git/gclement/mvebu into arm/fixes
mvebu fixes for 5.17 (part 2)
Allow using old PCIe card on Armada 37xx
* tag 'mvebu-fixes-5.17-2' of git://git.kernel.org/pub/scm/linux/kernel/git/gclement/mvebu:
arm64: dts: marvell: armada-37xx: Remap IO space to bus address 0x0
Link: https://lore.kernel.org/r/87bkydj4fn.fsf@BL-laptop
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Legacy and old PCI I/O based cards do not support 32-bit I/O addressing.
Since commit 64f160e19e ("PCI: aardvark: Configure PCIe resources from
'ranges' DT property") kernel can set different PCIe address on CPU and
different on the bus for the one A37xx address mapping without any firmware
support in case the bus address does not conflict with other A37xx mapping.
So remap I/O space to the bus address 0x0 to enable support for old legacy
I/O port based cards which have hardcoded I/O ports in low address space.
Note that DDR on A37xx is mapped to bus address 0x0. And mapping of I/O
space can be set to address 0x0 too because MEM space and I/O space are
separate and so do not conflict.
Remapping IO space on Turris Mox to different address is not possible to
due bootloader bug.
Signed-off-by: Pali Rohár <pali@kernel.org>
Reported-by: Arnd Bergmann <arnd@arndb.de>
Fixes: 76f6386b25 ("arm64: dts: marvell: Add Aardvark PCIe support for Armada 3700")
Cc: stable@vger.kernel.org # 64f160e19e ("PCI: aardvark: Configure PCIe resources from 'ranges' DT property")
Cc: stable@vger.kernel.org # 514ef1e62d ("arm64: dts: marvell: armada-37xx: Extend PCIe MEM space")
Reviewed-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Gregory CLEMENT <gregory.clement@bootlin.com>
If kernel is built with hung task detection enabled and
CONFIG_DEFAULT_HUNG_TASK_TIMEOUT set to less than 60 seconds,
snd_hdac_i915_init() will trigger the hung task timeout in case i915 is
not available and taint the kernel.
Use wait_for_completion_killable_timeout() for the wait to
avoid this problem.
Co-developed-by: Ramalingam C <ramalingam.c@intel.com>
Signed-off-by: Ramalingam C <ramalingam.c@intel.com>
Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>
Acked-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://lore.kernel.org/r/20220309182439.1053856-1-kai.vehmanen@linux.intel.com
Signed-off-by: Takashi Iwai <tiwai@suse.de>
In testing high frequency workloads with provided buffers, we spend a
lot of time in allocating and freeing the buffer units themselves.
Rather than repeatedly free and alloc them, add a recycling cache
instead. There are two caches:
- ctx->io_buffers_cache. This is the one we grab from in the submission
path, and it's protected by ctx->uring_lock. For inline completions,
we can recycle straight back to this cache and not need any extra
locking.
- ctx->io_buffers_comp. If we're not under uring_lock, then we use this
list to recycle buffers. It's protected by the completion_lock.
On adding a new buffer, check io_buffers_cache. If it's empty, check if
we can splice entries from the io_buffers_comp_cache.
This reduces about 5-10% of overhead from provided buffers, bringing it
pretty close to the non-provided path.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Lots of workloads use multiple threads, in which case the file table is
shared between them. This makes getting and putting the ring file
descriptor for each io_uring_enter(2) system call more expensive, as it
involves an atomic get and put for each call.
Similarly to how we allow registering normal file descriptors to avoid
this overhead, add support for an io_uring_register(2) API that allows
to register the ring fds themselves:
1) IORING_REGISTER_RING_FDS - takes an array of io_uring_rsrc_update
structs, and registers them with the task.
2) IORING_UNREGISTER_RING_FDS - takes an array of io_uring_src_update
structs, and unregisters them.
When a ring fd is registered, it is internally represented by an offset.
This offset is returned to the application, and the application then
uses this offset and sets IORING_ENTER_REGISTERED_RING for the
io_uring_enter(2) system call. This works just like using a registered
file descriptor, rather than a real one, in an SQE, where
IOSQE_FIXED_FILE gets set to tell io_uring that we're using an internal
offset/descriptor rather than a real file descriptor.
In initial testing, this provides a nice bump in performance for
threaded applications in real world cases where the batch count (eg
number of requests submitted per io_uring_enter(2) invocation) is low.
In a microbenchmark, submitting NOP requests, we see the following
increases in performance:
Requests per syscall Baseline Registered Increase
----------------------------------------------------------------
1 ~7030K ~8080K +15%
2 ~13120K ~14800K +13%
4 ~22740K ~25300K +11%
Co-developed-by: Xiaoguang Wang <xiaoguang.wang@linux.alibaba.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
There is a slight optimisation to be had by calculating the correct pos
pointer inside io_kiocb_update_pos and then using that later.
It seems code size drops by a bit:
000000000000a1b0 0000000000000400 t io_read
000000000000a5b0 0000000000000319 t io_write
vs
000000000000a1b0 00000000000003f6 t io_read
000000000000a5b0 0000000000000310 t io_write
Signed-off-by: Dylan Yudaken <dylany@fb.com>
Reviewed-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Update kiocb->ki_pos at execution time rather than in io_prep_rw().
io_prep_rw() happens before the job is enqueued to a worker and so the
offset might be read multiple times before being executed once.
Ensures that the file position in a set of _linked_ SQEs will be only
obtained after earlier SQEs have completed, and so will include their
incremented file position.
Signed-off-by: Dylan Yudaken <dylany@fb.com>
Reviewed-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
io_kiocb_ppos is called in both branches, and it seems that the compiler
does not fuse this. Fusing removes a few bytes from loop_rw_iter.
Before:
$ nm -S fs/io_uring.o | grep loop_rw_iter
0000000000002430 0000000000000124 t loop_rw_iter
After:
$ nm -S fs/io_uring.o | grep loop_rw_iter
0000000000002430 000000000000010d t loop_rw_iter
Signed-off-by: Dylan Yudaken <dylany@fb.com>
Reviewed-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
