The current implementation of the extra_bits interface is confusing:
passing extra_bits to __stack_depot_save makes it seem that the extra
bits are somehow stored in stack depot. In reality, they are only
embedded into a stack depot handle and are not used within stack depot.
Drop the extra_bits argument from __stack_depot_save and instead provide
a new stack_depot_set_extra_bits function (similar to the exsiting
stack_depot_get_extra_bits) that saves extra bits into a stack depot
handle.
Update the callers of __stack_depot_save to use the new interace.
This change also fixes a minor issue in the old code: __stack_depot_save
does not return NULL if saving stack trace fails and extra_bits is used.
Link: https://lkml.kernel.org/r/317123b5c05e2f82854fc55d8b285e0869d3cb77.1676063693.git.andreyknvl@google.com
Signed-off-by: Andrey Konovalov <andreyknvl@google.com>
Reviewed-by: Alexander Potapenko <glider@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
depot_init_pool has two call sites:
1. In depot_alloc_stack with a potentially NULL prealloc.
2. In __stack_depot_save with a non-NULL prealloc.
At the same time depot_init_pool can only return false when prealloc is
NULL.
As the second call site makes sure that prealloc is not NULL, the WARN_ON
there can never trigger. Thus, drop the WARN_ON and also move the prealloc
check from depot_init_pool to its first call site.
Also change the return type of depot_init_pool to void as it now always
returns true.
Link: https://lkml.kernel.org/r/ce149f9bdcbc80a92549b54da67eafb27f846b7b.1676063693.git.andreyknvl@google.com
Signed-off-by: Andrey Konovalov <andreyknvl@google.com>
Reviewed-by: Alexander Potapenko <glider@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Use "pool" instead of "slab" for naming memory regions stack depot
uses to store stack traces. Using "slab" is confusing, as stack depot
pools have nothing to do with the slab allocator.
Also give better names to pool-related global variables: change
"depot_" prefix to "pool_" to point out that these variables are
related to stack depot pools.
Also rename the slabindex (poolindex) field in handle_parts to pool_index
to align its name with the pool_index global variable.
No functional changes.
Link: https://lkml.kernel.org/r/923c507edb350c3b6ef85860f36be489dfc0ad21.1676063693.git.andreyknvl@google.com
Signed-off-by: Andrey Konovalov <andreyknvl@google.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Alexander Potapenko <glider@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Give more meaningful names to hash table-related constants and variables:
1. Rename STACK_HASH_SCALE to STACK_HASH_TABLE_SCALE to point out that it
is related to scaling the hash table.
2. Rename STACK_HASH_ORDER_MIN/MAX to STACK_BUCKET_NUMBER_ORDER_MIN/MAX
to point out that it is related to the number of hash table buckets.
3. Rename stack_hash_order to stack_bucket_number_order for the same
reason as #2.
No functional changes.
Link: https://lkml.kernel.org/r/f166dd6f3cb2378aea78600714393dd568c33ee9.1676063693.git.andreyknvl@google.com
Signed-off-by: Andrey Konovalov <andreyknvl@google.com>
Reviewed-by: Alexander Potapenko <glider@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
__GFP_ATOMIC serves little purpose. Its main effect is to set
ALLOC_HARDER which adds a few little boosts to increase the chance of an
allocation succeeding, one of which is to lower the water-mark at which it
will succeed.
It is *always* paired with __GFP_HIGH which sets ALLOC_HIGH which also
adjusts this watermark. It is probable that other users of __GFP_HIGH
should benefit from the other little bonuses that __GFP_ATOMIC gets.
__GFP_ATOMIC also gives a warning if used with __GFP_DIRECT_RECLAIM.
There is little point to this. We already get a might_sleep() warning if
__GFP_DIRECT_RECLAIM is set.
__GFP_ATOMIC allows the "watermark_boost" to be side-stepped. It is
probable that testing ALLOC_HARDER is a better fit here.
__GFP_ATOMIC is used by tegra-smmu.c to check if the allocation might
sleep. This should test __GFP_DIRECT_RECLAIM instead.
This patch:
- removes __GFP_ATOMIC
- allows __GFP_HIGH allocations to ignore watermark boosting as well
as GFP_ATOMIC requests.
- makes other adjustments as suggested by the above.
The net result is not change to GFP_ATOMIC allocations. Other
allocations that use __GFP_HIGH will benefit from a few different extra
privileges. This affects:
xen, dm, md, ntfs3
the vermillion frame buffer
hibernation
ksm
swap
all of which likely produce more benefit than cost if these selected
allocation are more likely to succeed quickly.
[mgorman: Minor adjustments to rework on top of a series]
Link: https://lkml.kernel.org/r/163712397076.13692.4727608274002939094@noble.neil.brown.name
Link: https://lkml.kernel.org/r/20230113111217.14134-7-mgorman@techsingularity.net
Signed-off-by: NeilBrown <neilb@suse.de>
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Thierry Reding <thierry.reding@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Mirsad report the below error which is caused by stack_depot_init()
failure in kvcalloc. Solve this by having stackdepot use
stack_depot_early_init().
On 1/4/23 17:08, Mirsad Goran Todorovac wrote:
I hate to bring bad news again, but there seems to be a problem with the output of /sys/kernel/debug/kmemleak:
[root@pc-mtodorov ~]# cat /sys/kernel/debug/kmemleak
unreferenced object 0xffff951c118568b0 (size 16):
comm "kworker/u12:2", pid 56, jiffies 4294893952 (age 4356.548s)
hex dump (first 16 bytes):
6d 65 6d 73 74 69 63 6b 30 00 00 00 00 00 00 00 memstick0.......
backtrace:
[root@pc-mtodorov ~]#
Apparently, backtrace of called functions on the stack is no longer
printed with the list of memory leaks. This appeared on Lenovo desktop
10TX000VCR, with AlmaLinux 8.7 and BIOS version M22KT49A (11/10/2022) and
6.2-rc1 and 6.2-rc2 builds. This worked on 6.1 with the same
CONFIG_KMEMLEAK=y and MGLRU enabled on a vanilla mainstream kernel from
Mr. Torvalds' tree. I don't know if this is deliberate feature for some
reason or a bug. Please find attached the config, lshw and kmemleak
output.
[vbabka@suse.cz: remove stack_depot_init() call]
Link: https://lore.kernel.org/all/5272a819-ef74-65ff-be61-4d2d567337de@alu.unizg.hr/
Link: https://lkml.kernel.org/r/1674091345-14799-2-git-send-email-zhaoyang.huang@unisoc.com
Fixes: 56a61617dd ("mm: use stack_depot for recording kmemleak's backtrace")
Reported-by: Mirsad Todorovac <mirsad.todorovac@alu.unizg.hr>
Suggested-by: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Zhaoyang Huang <zhaoyang.huang@unisoc.com>
Acked-by: Mike Rapoport (IBM) <rppt@kernel.org>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Tested-by: Borislav Petkov (AMD) <bp@alien8.de>
Cc: ke.wang <ke.wang@unisoc.com>
Cc: Nathan Chancellor <nathan@kernel.org>
Cc: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Preallocations are common in the VMA code to avoid allocating under
certain locking conditions. The preallocations must also cover the
worst-case scenario. Removing the GFP_ZERO flag from the
kmem_cache_alloc() (and bulk variant) calls will reduce the amount of time
spent zeroing memory that may not be used. Only zero out the necessary
area to keep track of the allocations in the maple state. Zero the entire
node prior to using it in the tree.
This required internal changes to node counting on allocation, so the test
code is also updated.
This restores some micro-benchmark performance: up to +9% in mmtests mmap1
by my testing +10% to +20% in mmap, mmapaddr, mmapmany tests reported by
Red Hat
Link: https://bugzilla.redhat.com/show_bug.cgi?id=2149636
Link: https://lkml.kernel.org/r/20230105160427.2988454-1-Liam.Howlett@oracle.com
Signed-off-by: Liam Howlett <Liam.Howlett@oracle.com>
Reported-by: Jirka Hladky <jhladky@redhat.com>
Suggested-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
On the x86-64 architecture even a failing cmpxchg grants exclusive
access to the cacheline, making it preferable to retry the failed op
immediately instead of stalling with the pause instruction.
To illustrate the impact, below are benchmark results obtained by
running various will-it-scale tests on top of the 6.2-rc3 kernel and
Cascade Lake (2 sockets * 24 cores * 2 threads) CPU.
All results in ops/s. Note there is some variance in re-runs, but the
code is consistently faster when contention is present.
open3 ("Same file open/close"):
proc stock no-pause
1 805603 814942 (+%1)
2 1054980 1054781 (-0%)
8 1544802 1822858 (+18%)
24 1191064 2199665 (+84%)
48 851582 1469860 (+72%)
96 609481 1427170 (+134%)
fstat2 ("Same file fstat"):
proc stock no-pause
1 3013872 3047636 (+1%)
2 4284687 4400421 (+2%)
8 3257721 5530156 (+69%)
24 2239819 5466127 (+144%)
48 1701072 5256609 (+209%)
96 1269157 6649326 (+423%)
Additionally, a kernel with a private patch to help access() scalability:
access2 ("Same file access"):
proc stock patched patched
+nopause
24 2378041 2005501 5370335 (-15% / +125%)
That is, fixing the problems in access itself *reduces* scalability
after the cacheline ping-pong only happens in lockref with the pause
instruction.
Note that fstat and access benchmarks are not currently integrated into
will-it-scale, but interested parties can find them in pull requests to
said project.
Code at hand has a rather tortured history. First modification showed
up in commit d472d9d98b ("lockref: Relax in cmpxchg loop"), written
with Itanium in mind. Later it got patched up to use an arch-dependent
macro to stop doing it on s390 where it caused a significant regression.
Said macro had undergone revisions and was ultimately eliminated later,
going back to cpu_relax.
While I intended to only remove cpu_relax for x86-64, I got the
following comment from Linus:
I would actually prefer just removing it entirely and see if
somebody else hollers. You have the numbers to prove it hurts on
real hardware, and I don't think we have any numbers to the
contrary.
So I think it's better to trust the numbers and remove it as a
failure, than say "let's just remove it on x86-64 and leave
everybody else with the potentially broken code"
Additionally, Will Deacon (maintainer of the arm64 port, one of the
architectures previously benchmarked):
So, from the arm64 side of the fence, I'm perfectly happy just
removing the cpu_relax() calls from lockref.
As such, come back full circle in history and whack it altogether.
Signed-off-by: Mateusz Guzik <mjguzik@gmail.com>
Link: https://lore.kernel.org/all/CAGudoHHx0Nqg6DE70zAVA75eV-HXfWyhVMWZ-aSeOofkA_=WdA@mail.gmail.com/
Acked-by: Tony Luck <tony.luck@intel.com> # ia64
Acked-by: Nicholas Piggin <npiggin@gmail.com> # powerpc
Acked-by: Will Deacon <will@kernel.org> # arm64
Acked-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Pull rdma fixes from Jason Gunthorpe:
"Most noticeable is that Yishai found a big data corruption regression
due to a change in the scatterlist:
- Do not wrongly combine non-contiguous pages in scatterlist
- Fix compilation warnings on gcc 13
- Oops when using some mlx5 stats
- Bad enforcement of atomic responder resources in mlx5"
* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma:
lib/scatterlist: Fix to merge contiguous pages into the last SG properly
RDMA/mlx5: Fix validation of max_rd_atomic caps for DC
RDMA/mlx5: Fix mlx5_ib_get_hw_stats when used for device
RDMA/srp: Move large values to a new enum for gcc13
When sg_alloc_append_table_from_pages() calls to pages_are_mergeable() in
its 'sgt_append->prv' flow to check whether it can merge contiguous pages
into the last SG, it passes the page arguments in the wrong order.
The first parameter should be the next candidate page to be merged to
the last page and not the opposite.
The current code leads to a corrupted SG which resulted in OOPs and
unexpected errors when non-contiguous pages are merged wrongly.
Fix to pass the page parameters in the right order.
Fixes: 1567b49d1a ("lib/scatterlist: add check when merging zone device pages")
Link: https://lore.kernel.org/r/20230105112339.107969-1-yishaih@nvidia.com
Signed-off-by: Yishai Hadas <yishaih@nvidia.com>
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
Reviewed-by: Logan Gunthorpe <logang@deltatee.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
