This new destination type supports flow transition between different
table types, e.g. from NIC_RX to RDMA_RX or from RDMA_TX to NIC_TX.
Signed-off-by: Patrisious Haddad <phaddad@nvidia.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
ksmbd allowed the actual frame length to be smaller than the rfc1002
length. If allowed, it is possible to allocates a large amount of memory
that can be limited by credit management and can eventually cause memory
exhaustion problem. This patch do not allow it except SMB2 Negotiate
request which will be validated when message handling proceeds.
Also, Allow a message that padded to 8byte boundary.
Fixes: e2f34481b2 ("cifsd: add server-side procedures for SMB3")
Cc: stable@vger.kernel.org
Signed-off-by: Namjae Jeon <linkinjeon@kernel.org>
Signed-off-by: Steve French <stfrench@microsoft.com>
When turning debug mode on, The following error message from
ksmbd_smb2_check_message() is coming.
ksmbd: cli req padded more than expected. Length 112 not 88 for cmd:10 mid:14
data area length calculation for smb2 lock request in smb2_get_data_area_len() is
incorrect.
Fixes: e2f34481b2 ("cifsd: add server-side procedures for SMB3")
Cc: stable@vger.kernel.org
Signed-off-by: Namjae Jeon <linkinjeon@kernel.org>
Signed-off-by: Steve French <stfrench@microsoft.com>
Conor Dooley <conor@kernel.org> says:
From: Conor Dooley <conor.dooley@microchip.com>
I've yoinked patch 1 from Drew's series adding support for Zicboz &
attached two more patches here that remove the need for, and then drop
the toolchain support checks for Zicbom. The goal is to remove the need
for checking the presence of toolchain Zicbom support in the work being
done to support non instruction based CMOs [1].
I've tested compliation on a number of different configurations with
the Zicbom config option enabled. The important ones to call out I
guess are:
- clang/llvm 14 w/ LLVM=1 which doesn't support Zicbom atm.
- gcc 11 w/ binutils 2.37 which doesn't support Zicbom atm either.
- clang/llvm 15 w/ LLVM=1 BUT with binutils 2.37's ld. This is the
configuration that prompted adding the LD checks as cc/as supports
Zicbom, but ld doesn't [2].
- gcc 12 w/ binutils 2.39 & clang 15 w/ LLVM=1, both of these supported
Zicbom before and still do.
I also checked building the THEAD errata etc with
CONFIG_RISCV_ISA_ZICBOM disabled, and there were no build issues there
either.
* b4-shazam-merge:
RISC-V: remove toolchain version checks for Zicbom
RISC-V: replace cbom instructions with an insn-def
RISC-V: insn-def: Add I-type insn-def
Link: https://lore.kernel.org/r/20230108163356.3063839-1-conor@kernel.org
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
Commit b8c86872d1 ("riscv: fix detection of toolchain Zicbom
support") fixed building on systems where Zicbom was supported by the
compiler/assembler but not by the linker in an easily backportable
manner.
Now that the we have insn-defs for the 3 instructions, toolchain support
is no longer required for Zicbom.
Stop emitting "_zicbom" in -march when Zicbom is enabled & drop the
version checks entirely.
Signed-off-by: Conor Dooley <conor.dooley@microchip.com>
Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Link: https://lore.kernel.org/r/20230108163356.3063839-4-conor@kernel.org
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
Thomas Weißschuh says:
====================
net: make kobj_type structures constant
Since commit ee6d3dd4ed ("driver core: make kobj_type constant.")
the driver core allows the usage of const struct kobj_type.
Take advantage of this to constify the structure definitions to prevent
modification at runtime.
====================
Link: https://lore.kernel.org/r/20230211-kobj_type-net-v2-0-013b59e59bf3@weissschuh.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Since commit ee6d3dd4ed ("driver core: make kobj_type constant.")
the driver core allows the usage of const struct kobj_type.
Take advantage of this to constify the structure definitions to prevent
modification at runtime.
Signed-off-by: Thomas Weißschuh <linux@weissschuh.net>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Since commit ee6d3dd4ed ("driver core: make kobj_type constant.")
the driver core allows the usage of const struct kobj_type.
Take advantage of this to constify the structure definition to prevent
modification at runtime.
Signed-off-by: Thomas Weißschuh <linux@weissschuh.net>
Acked-by: Nikolay Aleksandrov <razor@blackwall.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When sending a SYN message, this kernel stack trace is observed:
...
[ 13.396352] RIP: 0010:_copy_from_iter+0xb4/0x550
...
[ 13.398494] Call Trace:
[ 13.398630] <TASK>
[ 13.398630] ? __alloc_skb+0xed/0x1a0
[ 13.398630] tipc_msg_build+0x12c/0x670 [tipc]
[ 13.398630] ? shmem_add_to_page_cache.isra.71+0x151/0x290
[ 13.398630] __tipc_sendmsg+0x2d1/0x710 [tipc]
[ 13.398630] ? tipc_connect+0x1d9/0x230 [tipc]
[ 13.398630] ? __local_bh_enable_ip+0x37/0x80
[ 13.398630] tipc_connect+0x1d9/0x230 [tipc]
[ 13.398630] ? __sys_connect+0x9f/0xd0
[ 13.398630] __sys_connect+0x9f/0xd0
[ 13.398630] ? preempt_count_add+0x4d/0xa0
[ 13.398630] ? fpregs_assert_state_consistent+0x22/0x50
[ 13.398630] __x64_sys_connect+0x16/0x20
[ 13.398630] do_syscall_64+0x42/0x90
[ 13.398630] entry_SYSCALL_64_after_hwframe+0x63/0xcd
It is because commit a41dad905e ("iov_iter: saner checks for attempt
to copy to/from iterator") has introduced sanity check for copying
from/to iov iterator. Lacking of copy direction from the iterator
viewpoint would lead to kernel stack trace like above.
This commit fixes this issue by initializing the iov iterator with
the correct copy direction when sending SYN or ACK without data.
Fixes: f25dcc7687 ("tipc: tipc ->sendmsg() conversion")
Reported-by: syzbot+d43608d061e8847ec9f3@syzkaller.appspotmail.com
Acked-by: Jon Maloy <jmaloy@redhat.com>
Signed-off-by: Tung Nguyen <tung.q.nguyen@dektech.com.au>
Link: https://lore.kernel.org/r/20230214012606.5804-1-tung.q.nguyen@dektech.com.au
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Tony Nguyen says:
====================
Intel Wired LAN Driver Updates 2023-02-13 (ice)
This series contains updates to ice driver only.
Michal fixes check of scheduling node weight and priority to be done
against desired value, not current value.
Jesse adds setting of all multicast when adding promiscuous mode to
resolve traffic being lost due to filter settings.
* '100GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/net-queue:
ice: fix lost multicast packets in promisc mode
ice: Fix check for weight and priority of a scheduling node
====================
Link: https://lore.kernel.org/r/20230213185259.3959224-1-anthony.l.nguyen@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Alex Elder says:
====================
net: ipa: define GSI register fields differently
Now that we have "reg" definitions in place to define GSI register
offsets, add the definitions for the fields of GSI registers that
have them.
There aren't many differences between versions, but a few fields are
present only in some versions of IPA, so additional "gsi_reg-vX.Y.c"
files are created to capture such differences. As in the previous
series, these files are created as near-copies of existing files
just before they're needed to represent these differences. The
first patch adds files for IPA v4.0, v4.5, and v4.9; the fifth patch
adds a file for IPA v4.11.
Note that the first and fifth patch cause some checkpatch warnings
because they align some continued lines with an open parenthesis
that at the fourth column.
====================
Link: https://lore.kernel.org/r/20230213162229.604438-1-elder@linaro.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Define field IDs for the remaining GSI registers, and populate the
register definition files accordingly. Use the reg_*() functions to
access field values for those regiters, and get rid of the previous
field definition constants.
Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The next patch adds a GSI register field that is only valid starting
at IPA v4.11. Create "gsi_v4.11.c" from "gsi_v4.9.c", changing only
the name of the public regs structure it defines.
Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Define field IDs for the EV_CH_E_CNTXT_0 and EV_CH_E_CNTXT_8 GSI
registers, and populate the register definition files accordingly.
Use the reg_*() functions to access field values for those regiters,
and get rid of the previous field definition constants.
The remaining EV_CH_E_CNTXT_* registers are written with full 32-bit
values (and have no fields).
Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Beyond the CH_C_QOS register, two other registers whose offset is
related to channel number have fields within them.
Define the fields within the CH_C_CNTXT_0 GSI register, using an
enumerated type to identify the register's fields, and define an
array of field masks to use for that register's reg structure.
For the CH_C_CNTXT_1 GSI register, ch_c_cntxt_1_length_encode()
previously hid the difference in bit width in the channel ring
length field. Instead, define a new field CH_R_LENGTH and encode
the ring size with reg_encode().
Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Define the fields within the CH_C_QOS GSI register using an array of
field masks in that register's reg structure. Use the reg functions
for encoding values in those fields.
One field in the register is present for IPA v4.0-4.2 only, two
others are present starting at IPA v4.5, and one more is there
starting at IPA v4.9.
Drop the "GSI_" prefix in symbols defined in the gsi_prefetch_mode
enumerated type, and define their values using decimal rather than
hexidecimal values.
Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Create "gsi_v4.0.c", "gsi_v4.5.c", and "gsi_v4.9.c" as essentially
identical copies of "gsi_v3.5.1.c". The only difference is the name
of the exported "gsi_regs_vX_Y" structure. The next patch will
start differentiating them.
Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
v3: Fix vmw_user_bo_lookup which was also dropping the gem reference
before the kernel was done with buffer depending on userspace doing
the right thing. Same bug, different spot.
It is possible for userspace to predict the next buffer handle and
to destroy the buffer while it's still used by the kernel. Delay
dropping the internal reference on the buffers until kernel is done
with them.
Instead of immediately dropping the gem reference in vmw_user_bo_lookup
and vmw_gem_object_create_with_handle let the callers decide when they're
ready give the control back to userspace.
Also fixes the second usage of vmw_gem_object_create_with_handle in
vmwgfx_surface.c which wasn't grabbing an explicit reference
to the gem object which could have been destroyed by the userspace
on the owning surface at any point.
Signed-off-by: Zack Rusin <zackr@vmware.com>
Fixes: 8afa13a058 ("drm/vmwgfx: Implement DRIVER_GEM")
Reviewed-by: Martin Krastev <krastevm@vmware.com>
Reviewed-by: Maaz Mombasawala <mombasawalam@vmware.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230211050514.2431155-1-zack@kde.org
(cherry picked from commit 9ef8d83e8e)
Cc: <stable@vger.kernel.org> # v5.17+
Conor Dooley <conor@kernel.org> says:
From: Conor Dooley <conor.dooley@microchip.com>
Ever since RISC-V starting using generic arch topology code, the code
paths for cpu-capacity have been there but there's no binding defined to
actually convey the information. Defining the same property as used on
arm seems to be the only logical thing to do, so do it.
[Palmer: This is on top of the fix required to make it work, which
itself wasn't merged until late in the 6.2 cycle and thus pulls in
various other fixes.]
* b4-shazam-merge:
dt-bindings: riscv: add a capacity-dmips-mhz cpu property
dt-bindings: arm: move cpu-capacity to a shared loation
riscv: Move call to init_cpu_topology() to later initialization stage
riscv/kprobe: Fix instruction simulation of JALR
riscv: fix -Wundef warning for CONFIG_RISCV_BOOT_SPINWAIT
MAINTAINERS: add an IRC entry for RISC-V
RISC-V: fix compile error from deduplicated __ALTERNATIVE_CFG_2
dt-bindings: riscv: fix single letter canonical order
dt-bindings: riscv: fix underscore requirement for multi-letter extensions
riscv: uaccess: fix type of 0 variable on error in get_user()
riscv, kprobes: Stricter c.jr/c.jalr decoding
Link: https://lore.kernel.org/r/20230104180513.1379453-1-conor@kernel.org
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
ttm_bo_init_reserved on failure puts the buffer object back which
causes it to be deleted, but kfree was still being called on the same
buffer in vmw_bo_create leading to a double free.
After the double free the vmw_gem_object_create_with_handle was
setting the gem function objects before checking the return status
of vmw_bo_create leading to null pointer access.
Fix the entire path by relaying on ttm_bo_init_reserved to delete the
buffer objects on failure and making sure the return status is checked
before setting the gem function objects on the buffer object.
Signed-off-by: Zack Rusin <zackr@vmware.com>
Fixes: 8afa13a058 ("drm/vmwgfx: Implement DRIVER_GEM")
Reviewed-by: Maaz Mombasawala <mombasawalam@vmware.com>
Reviewed-by: Martin Krastev <krastevm@vmware.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230208180050.2093426-1-zack@kde.org
(cherry picked from commit 36d421e632)
Cc: <stable@vger.kernel.org> # v5.17+
Pull one more Renesas clk driver update from Geert Uytterhoeven:
- Disable R-Car H3 ES1.*, as it was only available to an internal
development group and needed a lot of quirks and workarounds.
* tag 'renesas-clk-for-v6.3-tag3' of git://git.kernel.org/pub/scm/linux/kernel/git/geert/renesas-drivers:
clk: renesas: rcar-gen3: Disable R-Car H3 ES1.*
In the dio write path, we only take shared inode lock for the case of
aligned overwriting initialized blocks inside EOF. But for overwriting
preallocated blocks, it may only need to split unwritten extents, this
procedure has been protected under i_data_sem lock, it's safe to
release the exclusive inode lock and take shared inode lock.
This could give a significant speed up for multi-threaded writes. Test
on Intel Xeon Gold 6140 and nvme SSD with below fio parameters.
direct=1
ioengine=libaio
iodepth=10
numjobs=10
runtime=60
rw=randwrite
size=100G
And the test result are:
Before:
bs=4k IOPS=11.1k, BW=43.2MiB/s
bs=16k IOPS=11.1k, BW=173MiB/s
bs=64k IOPS=11.2k, BW=697MiB/s
After:
bs=4k IOPS=41.4k, BW=162MiB/s
bs=16k IOPS=41.3k, BW=646MiB/s
bs=64k IOPS=13.5k, BW=843MiB/s
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://lore.kernel.org/r/20221226062015.3479416-1-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
If the end position of a GETFSMAP query overlaps an allocated space and
we're using the free space info to generate fsmap info, the akeys
information gets fed into the fsmap formatter with bad results.
Zero-init the space.
Reported-by: syzbot+090ae72d552e6bd93cfe@syzkaller.appspotmail.com
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
xfs: per-ag centric allocation alogrithms
This series continues the work towards making shrinking a filesystem
possible. We need to be able to stop operations from taking place
on AGs that need to be removed by a shrink, so before shrink can be
implemented we need to have the infrastructure in place to prevent
incursion into AGs that are going to be, or are in the process, of
being removed from active duty.
The focus of this is making operations that depend on access to AGs
use the perag to access and pin the AG in active use, thereby
creating a barrier we can use to delay shrink until all active uses
of an AG have been drained and new uses are prevented.
This series starts by fixing some existing issues that are exposed
by changes later in the series. They stand alone, so can be picked
up independently of the rest of this patchset.
The most complex of these fixes is cleaning up the mess that is the
AGF deadlock avoidance algorithm. This algorithm stores the first
block that is allocated in a transaction in tp->t_firstblock, then
uses this to try to limit future allocations within the transaction
to AGs at or higher than the filesystem block stored in
tp->t_firstblock. This depends on one of the initial bug fixes in
the series to move the deadlock avoidance checks to
xfs_alloc_vextent(), and then builds on it to relax the constraints
of the avoidance algorithm to only be active when a deadlock is
possible.
We also update the algorithm to record allocations from higher AGs
that are allocated from, because we when we need to lock more than
two AGs we still have to ensure lock order is correct. Therefore we
can't lock AGs in the order 1, 3, 2, even though tp->t_firstblock
indicates that we've allocated from AG 1 and so AG is valid to lock.
It's not valid, because we already hold AG 3 locked, and so
tp->t-first_block should actually point at AG 3, not AG 1 in this
situation.
It should now be obvious that the deadlock avoidance algorithm
should record AGs, not filesystem blocks. So the series then changes
the transaction to store the highest AG we've allocated in rather
than a filesystem block we allocated. This makes it obvious what
the constraints are, and trivial to update as we lock and allocate
from various AGs.
With all the bug fixes out of the way, the series then starts
converting the code to use active references. Active reference
counts are used by high level code that needs to prevent the AG from
being taken out from under it by a shrink operation. The high level
code needs to be able to handle not getting an active reference
gracefully, and the shrink code will need to wait for active
references to drain before continuing.
Active references are implemented just as reference counts right now
- an active reference is taken at perag init during mount, and all
other active references are dependent on the active reference count
being greater than zero. This gives us an initial method of stopping
new active references without needing other infrastructure; just
drop the reference taken at filesystem mount time and when the
refcount then falls to zero no new references can be taken.
In future, this will need to take into account AG control state
(e.g. offline, no alloc, etc) as well as the reference count, but
right now we can implement a basic barrier for shrink with just
reference count manipulations. As such, patches to convert the perag
state to atomic opstate fields similar to the xfs_mount and xlog
opstate fields follow the initial active perag reference counting
patches.
The first target for active reference conversion is the
for_each_perag*() iterators. This captures a lot of high level code
that should skip offline AGs, and introduces the ability to
differentiate between a lookup that didn't have an online AG and the
end of the AG iteration range.
From there, the inode allocation AG selection is converted to active
references, and the perag is driven deeper into the inode allocation
and btree code to replace the xfs_mount. Most of the inode
allocation code operates on a single AG once it is selected, hence
it should pass the perag as the primary referenced object around for
allocation, not the xfs_mount. There is a bit of churn here, but it
emphasises that inode allocation is inherently an allocation group
based operation.
Next the bmap/alloc interface undergoes a major untangling,
reworking xfs_bmap_btalloc() into separate allocation operations for
different contexts and failure handling behaviours. This then allows
us to completely remove the xfs_alloc_vextent() layer via
restructuring the xfs_alloc_vextent/xfs_alloc_ag_vextent() into a
set of realtively simple helper function that describe the
allocation that they are doing. e.g. xfs_alloc_vextent_exact_bno().
This allows the requirements for accessing AGs to be allocation
context dependent. The allocations that require operation on a
single AG generally can't tolerate failure after the allocation
method and AG has been decided on, and hence the caller needs to
manage the active references to ensure the allocation does not race
with shrink removing the selected AG for the duration of the
operation that requires access to that allocation group.
Other allocations iterate AGs and so the first AG is just a hint -
these do not need to pin a perag first as they can tolerate not
being able to access an AG by simply skipping over it. These require
new perag iteration functions that can start at arbitrary AGs and
wrap around at arbitrary AGs, hence a new set for
for_each_perag_wrap*() helpers to do this.
Next is the rework of the filestreams allocator. This doesn't change
any functionality, but gets rid of the unnecessary multi-pass
selection algorithm when the selected AG is not available. It
currently does a lookup pass which might iterate all AGs to select
an AG, then checks if the AG is acceptible and if not does a "new
AG" pass that is essentially identical to the lookup pass. Both of
these scans also do the same "longest extent in AG" check before
selecting an AG as is done after the AG is selected.
IOWs, the filestreams algorithm can be greatly simplified into a
single new AG selection pass if the there is no current association
or the currently associated AG doesn't have enough contiguous free
space for the allocation to proceed. With this simplification of
the filestreams allocator, it's then trivial to convert it to use
for_each_perag_wrap() for the AG scan algorithm.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
* tag 'xfs-alloc-perag-conversion' of git://git.kernel.org/pub/scm/linux/kernel/git/dgc/linux-xfs: (42 commits)
xfs: refactor the filestreams allocator pick functions
xfs: return a referenced perag from filestreams allocator
xfs: pass perag to filestreams tracing
xfs: use for_each_perag_wrap in xfs_filestream_pick_ag
xfs: track an active perag reference in filestreams
xfs: factor out MRU hit case in xfs_filestream_select_ag
xfs: remove xfs_filestream_select_ag() longest extent check
xfs: merge new filestream AG selection into xfs_filestream_select_ag()
xfs: merge filestream AG lookup into xfs_filestream_select_ag()
xfs: move xfs_bmap_btalloc_filestreams() to xfs_filestreams.c
xfs: use xfs_bmap_longest_free_extent() in filestreams
xfs: get rid of notinit from xfs_bmap_longest_free_extent
xfs: factor out filestreams from xfs_bmap_btalloc_nullfb
xfs: convert trim to use for_each_perag_range
xfs: convert xfs_alloc_vextent_iterate_ags() to use perag walker
xfs: move the minimum agno checks into xfs_alloc_vextent_check_args
xfs: fold xfs_alloc_ag_vextent() into callers
xfs: move allocation accounting to xfs_alloc_vextent_set_fsbno()
xfs: introduce xfs_alloc_vextent_prepare()
xfs: introduce xfs_alloc_vextent_exact_bno()
...
Currently there're two anonymous inodes (inode and anon_inode in struct
erofs_fscache) for each blob. The former was introduced as the
address_space of page cache for bootstrap.
The latter was initially introduced as both the address_space of page
cache and also a sentinel in the shared domain. Since now the management
of cookies in share domain has been decoupled with the anonymous inode,
there's no need to maintain an extra anonymous inode. Let's unify these
two anonymous inodes.
Besides, in non-share-domain mode only bootstrap will allocate anonymous
inode. To simplify the implementation, always allocate anonymous inode
for both bootstrap and data blobs. Similarly release anonymous inodes
for data blobs when .put_super() is called, or we'll get "VFS: Busy
inodes after unmount." warning.
Also remove the redundant set_nlink() when initializing the anonymous
inode, since i_nlink has already been initialized to 1 when the inode
gets allocated.
Signed-off-by: Jingbo Xu <jefflexu@linux.alibaba.com>
Reviewed-by: Jia Zhu <zhujia.zj@bytedance.com>
Link: https://lore.kernel.org/r/20230209063913.46341-5-jefflexu@linux.alibaba.com
Signed-off-by: Gao Xiang <hsiangkao@linux.alibaba.com>
We'd better not touch sb->s_inodes list and inode->i_count directly.
Let's maintain cookies of share domain in a self-contained list in erofs.
Besides, relinquish cookie with the mutex held. Otherwise if a cookie
is registered when the old cookie with the same name in the same domain
has been removed from the list but not relinquished yet, fscache may
complain "Duplicate cookie detected".
Signed-off-by: Jingbo Xu <jefflexu@linux.alibaba.com>
Reviewed-by: Jia Zhu <zhujia.zj@bytedance.com>
Link: https://lore.kernel.org/r/20230209063913.46341-3-jefflexu@linux.alibaba.com
Signed-off-by: Gao Xiang <hsiangkao@linux.alibaba.com>
Using per-cpu thread pool we can reduce the scheduling latency compared
to workqueue implementation. With this patch scheduling latency and
variation is reduced as per-cpu threads are high priority kthread_workers.
The results were evaluated on arm64 Android devices running 5.10 kernel.
The table below shows resulting improvements of total scheduling latency
for the same app launch benchmark runs with 50 iterations. Scheduling
latency is the latency between when the task (workqueue kworker vs
kthread_worker) became eligible to run to when it actually started
running.
+-------------------------+-----------+----------------+---------+
| | workqueue | kthread_worker | diff |
+-------------------------+-----------+----------------+---------+
| Average (us) | 15253 | 2914 | -80.89% |
| Median (us) | 14001 | 2912 | -79.20% |
| Minimum (us) | 3117 | 1027 | -67.05% |
| Maximum (us) | 30170 | 3805 | -87.39% |
| Standard deviation (us) | 7166 | 359 | |
+-------------------------+-----------+----------------+---------+
Background: Boot times and cold app launch benchmarks are very
important to the Android ecosystem as they directly translate to
responsiveness from user point of view. While EROFS provides
a lot of important features like space savings, we saw some
performance penalty in cold app launch benchmarks in few scenarios.
Analysis showed that the significant variance was coming from the
scheduling cost while decompression cost was more or less the same.
Having per-cpu thread pool we can see from the above table that this
variation is reduced by ~80% on average. This problem was discussed
at LPC 2022. Link to LPC 2022 slides and talk at [1]
[1] https://lpc.events/event/16/contributions/1338/
[ Gao Xiang: At least, we have to add this until WQ_UNBOUND workqueue
issue [2] on many arm64 devices is resolved. ]
[2] https://lore.kernel.org/r/CAJkfWY490-m6wNubkxiTPsW59sfsQs37Wey279LmiRxKt7aQYg@mail.gmail.com
Signed-off-by: Sandeep Dhavale <dhavale@google.com>
Signed-off-by: Gao Xiang <hsiangkao@linux.alibaba.com>
Link: https://lore.kernel.org/r/20230208093322.75816-1-hsiangkao@linux.alibaba.com
