IF YOU WOULD LIKE TO GET AN ACCOUNT, please write an
email to Administrator. User accounts are meant only to access repo
and report issues and/or generate pull requests.
This is a purpose-specific Git hosting for
BaseALT
projects. Thank you for your understanding!
Только зарегистрированные пользователи имеют доступ к сервису!
Для получения аккаунта, обратитесь к администратору.
The driver core now can handle a const struct bus_type pointer, and the
dma_debug_add_bus() call just passes on the pointer give to it to the
driver core, so make this pointer const as well to allow everyone to use
read-only struct bus_type pointers going forward.
Cc: Christoph Hellwig <hch@lst.de>
Cc: Marek Szyprowski <m.szyprowski@samsung.com>
Cc: Robin Murphy <robin.murphy@arm.com>
Cc: <iommu@lists.linux.dev>
Reviewed-by: Robin Murphy <robin.murphy@arm.com>
Link: https://lore.kernel.org/r/2023121941-dejected-nugget-681e@gregkh
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
For percpu data structure allocation with bpf_global_percpu_ma,
the maximum data size is 4K. But for a system with large
number of cpus, bigger data size (e.g., 2K, 4K) might consume
a lot of memory. For example, the percpu memory consumption
with unit size 2K and 1024 cpus will be 2K * 1K * 1k = 2GB
memory.
We should discourage such usage. Let us limit the maximum data
size to be 512 for bpf_global_percpu_ma allocation.
Acked-by: Hou Tao <houtao1@huawei.com>
Signed-off-by: Yonghong Song <yonghong.song@linux.dev>
Link: https://lore.kernel.org/r/20231222031801.1290841-1-yonghong.song@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Currently, refill low/high marks are set with the assumption
of normal non-percpu memory allocation. For example, for
an allocation size 256, for non-percpu memory allocation,
low mark is 32 and high mark is 96, resulting in the
batch allocation of 48 elements and the allocated memory
will be 48 * 256 = 12KB for this particular cpu.
Assuming an 128-cpu system, the total memory consumption
across all cpus will be 12K * 128 = 1.5MB memory.
This might be okay for non-percpu allocation, but may not be
good for percpu allocation, which will consume 1.5MB * 128 = 192MB
memory in the worst case if every cpu has a chance of memory
allocation.
In practice, percpu allocation is very rare compared to
non-percpu allocation. So let us have smaller low/high marks
which can avoid unnecessary memory consumption.
Signed-off-by: Yonghong Song <yonghong.song@linux.dev>
Acked-by: Hou Tao <houtao1@huawei.com>
Link: https://lore.kernel.org/r/20231222031755.1289671-1-yonghong.song@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Typically for percpu map element or data structure, once allocated,
most operations are lookup or in-place update. Deletion are really
rare. Currently, for percpu data strcture, 4 elements will be
refilled if the size is <= 256. Let us just do with one element
for percpu data. For example, for size 256 and 128 cpus, the
potential saving will be 3 * 256 * 128 * 128 = 12MB.
Acked-by: Hou Tao <houtao1@huawei.com>
Signed-off-by: Yonghong Song <yonghong.song@linux.dev>
Link: https://lore.kernel.org/r/20231222031750.1289290-1-yonghong.song@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Commit 41a5db8d8161 ("Add support for non-fix-size percpu mem allocation")
added support for non-fix-size percpu memory allocation.
Such allocation will allocate percpu memory for all buckets on all
cpus and the memory consumption is in the order to quadratic.
For example, let us say, 4 cpus, unit size 16 bytes, so each
cpu has 16 * 4 = 64 bytes, with 4 cpus, total will be 64 * 4 = 256 bytes.
Then let us say, 8 cpus with the same unit size, each cpu
has 16 * 8 = 128 bytes, with 8 cpus, total will be 128 * 8 = 1024 bytes.
So if the number of cpus doubles, the number of memory consumption
will be 4 times. So for a system with large number of cpus, the
memory consumption goes up quickly with quadratic order.
For example, for 4KB percpu allocation, 128 cpus. The total memory
consumption will 4KB * 128 * 128 = 64MB. Things will become
worse if the number of cpus is bigger (e.g., 512, 1024, etc.)
In Commit 41a5db8d8161, the non-fix-size percpu memory allocation is
done in boot time, so for system with large number of cpus, the initial
percpu memory consumption is very visible. For example, for 128 cpu
system, the total percpu memory allocation will be at least
(16 + 32 + 64 + 96 + 128 + 196 + 256 + 512 + 1024 + 2048 + 4096)
* 128 * 128 = ~138MB.
which is pretty big. It will be even bigger for larger number of cpus.
Note that the current prefill also allocates 4 entries if the unit size
is less than 256. So on top of 138MB memory consumption, this will
add more consumption with
3 * (16 + 32 + 64 + 96 + 128 + 196 + 256) * 128 * 128 = ~38MB.
Next patch will try to reduce this memory consumption.
Later on, Commit 1fda5bb66ad8 ("bpf: Do not allocate percpu memory
at init stage") moved the non-fix-size percpu memory allocation
to bpf verificaiton stage. Once a particular bpf_percpu_obj_new()
is called by bpf program, the memory allocator will try to fill in
the cache with all sizes, causing the same amount of percpu memory
consumption as in the boot stage.
To reduce the initial percpu memory consumption for non-fix-size
percpu memory allocation, instead of filling the cache with all
supported allocation sizes, this patch intends to fill the cache
only for the requested size. As typically users will not use large
percpu data structure, this can save memory significantly.
For example, the allocation size is 64 bytes with 128 cpus.
Then total percpu memory amount will be 64 * 128 * 128 = 1MB,
much less than previous 138MB.
Signed-off-by: Yonghong Song <yonghong.song@linux.dev>
Acked-by: Hou Tao <houtao1@huawei.com>
Link: https://lore.kernel.org/r/20231222031745.1289082-1-yonghong.song@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
The objcg is a bpf_mem_alloc level property since all bpf_mem_cache's
are with the same objcg. This patch made such a property explicit.
The next patch will use this property to save and restore objcg
for percpu unit allocator.
Acked-by: Hou Tao <houtao1@huawei.com>
Signed-off-by: Yonghong Song <yonghong.song@linux.dev>
Link: https://lore.kernel.org/r/20231222031739.1288590-1-yonghong.song@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Currently, for percpu memory allocation, say if the user
requests allocation size to be 32 bytes, the actually
calculated size will be 40 bytes and it further rounds
to 64 bytes, and eventually 64 bytes are allocated,
wasting 32-byte memory.
Change bpf_mem_alloc() to calculate the cache index
based on the user-provided allocation size so unnecessary
extra memory can be avoided.
Suggested-by: Hou Tao <houtao1@huawei.com>
Acked-by: Hou Tao <houtao1@huawei.com>
Signed-off-by: Yonghong Song <yonghong.song@linux.dev>
Link: https://lore.kernel.org/r/20231222031734.1288400-1-yonghong.song@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
This patch simplifies the verification of size arguments associated to
pointer arguments to helpers and kfuncs. Many helpers take a pointer
argument followed by the size of the memory access performed to be
performed through that pointer. Before this patch, the handling of the
size argument in check_mem_size_reg() was confusing and wasteful: if the
size register's lower bound was 0, then the verification was done twice:
once considering the size of the access to be the lower-bound of the
respective argument, and once considering the upper bound (even if the
two are the same). The upper bound checking is a super-set of the
lower-bound checking(*), except: the only point of the lower-bound check
is to handle the case where zero-sized-accesses are explicitly not
allowed and the lower-bound is zero. This static condition is now
checked explicitly, replacing a much more complex, expensive and
confusing verification call to check_helper_mem_access().
Error messages change in this patch. Before, messages about illegal
zero-size accesses depended on the type of the pointer and on other
conditions, and sometimes the message was plain wrong: in some tests
that changed you'll see that the old message was something like "R1 min
value is outside of the allowed memory range", where R1 is the pointer
register; the error was wrongly claiming that the pointer was bad
instead of the size being bad. Other times the information that the size
came for a register with a possible range of values was wrong, and the
error presented the size as a fixed zero. Now the errors refer to the
right register. However, the old error messages did contain useful
information about the pointer register which is now lost; recovering
this information was deemed not important enough.
(*) Besides standing to reason that the checks for a bigger size access
are a super-set of the checks for a smaller size access, I have also
mechanically verified this by reading the code for all types of
pointers. I could convince myself that it's true for all but
PTR_TO_BTF_ID (check_ptr_to_btf_access). There, simply looking
line-by-line does not immediately prove what we want. If anyone has any
qualms, let me know.
Signed-off-by: Andrei Matei <andreimatei1@gmail.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20231221232225.568730-2-andreimatei1@gmail.com
In preparation for subsequent changes, introduce a specialized variant
of async_schedule_dev() that will not invoke the argument function
synchronously when it cannot be scheduled for asynchronous execution.
The new function, async_schedule_dev_nocall(), will be used for fixing
possible deadlocks in the system-wide power management core code.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com> for the series.
Tested-by: Youngmin Nam <youngmin.nam@samsung.com>
Reviewed-by: Ulf Hansson <ulf.hansson@linaro.org>
In preparation for subsequent changes, split async_schedule_node_domain()
in two pieces so as to allow the bottom part of it to be called from a
somewhat different code path.
No functional impact.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Tested-by: Youngmin Nam <youngmin.nam@samsung.com>
Reviewed-by: Ulf Hansson <ulf.hansson@linaro.org>
Introduce thermal_zone_device_critical_reboot() to trigger an
emergency reboot.
It is a counterpart of thermal_zone_device_critical() with the
difference that it will force a reboot instead of shutdown.
The motivation for doing this is to allow the thermal subystem
to trigger a reboot when the temperature reaches the critical
temperature.
Signed-off-by: Fabio Estevam <festevam@denx.de>
Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org>
Link: https://lore.kernel.org/r/20231129124330.519423-3-festevam@gmail.com
Add some helper functions to make it easier introducing the support
for thermal reboot.
No functional change.
Signed-off-by: Fabio Estevam <festevam@denx.de>
Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org>
Link: https://lore.kernel.org/r/20231129124330.519423-2-festevam@gmail.com
-----BEGIN PGP SIGNATURE-----
iHUEABYIAB0WIQTFp0I1jqZrAX+hPRXbK58LschIgwUCZYVEqQAKCRDbK58LschI
gzH6AP9hVXLpHFTWMT0+2GK2lx69VX8zW1C0SmN7WHaxUbPN9QEAwzGnELfKk00P
0IKRHSl5abhVMX7JOM3sSOhCILeKjQg=
=wRLJ
-----END PGP SIGNATURE-----
Merge tag 'for-netdev' of https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next
Daniel Borkmann says:
====================
bpf-next-for-netdev
The following pull-request contains BPF updates for your *net-next* tree.
We've added 22 non-merge commits during the last 3 day(s) which contain
a total of 23 files changed, 652 insertions(+), 431 deletions(-).
The main changes are:
1) Add verifier support for annotating user's global BPF subprogram arguments
with few commonly requested annotations for a better developer experience,
from Andrii Nakryiko.
These tags are:
- Ability to annotate a special PTR_TO_CTX argument
- Ability to annotate a generic PTR_TO_MEM as non-NULL
2) Support BPF verifier tracking of BPF_JNE which helps cases when the compiler
transforms (unsigned) "a > 0" into "if a == 0 goto xxx" and the like, from
Menglong Dong.
3) Fix a warning in bpf_mem_cache's check_obj_size() as reported by LKP, from Hou Tao.
4) Re-support uid/gid options when mounting bpffs which had to be reverted with
the prior token series revert to avoid conflicts, from Daniel Borkmann.
5) Fix a libbpf NULL pointer dereference in bpf_object__collect_prog_relos() found
from fuzzing the library with malformed ELF files, from Mingyi Zhang.
6) Skip DWARF sections in libbpf's linker sanity check given compiler options to
generate compressed debug sections can trigger a rejection due to misalignment,
from Alyssa Ross.
7) Fix an unnecessary use of the comma operator in BPF verifier, from Simon Horman.
8) Fix format specifier for unsigned long values in cpustat sample, from Colin Ian King.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
- Fix readers that are blocked on the ring buffer when buffer_percent is
100%. They are supposed to wake up when the buffer is full, but
because the sub-buffer that the writer is on is never considered
"dirty" in the calculation, dirty pages will never equal nr_pages.
Add +1 to the dirty count in order to count for the sub-buffer that
the writer is on.
- When a reader is blocked on the "snapshot_raw" file, it is to be
woken up when a snapshot is done and be able to read the snapshot
buffer. But because the snapshot swaps the buffers (the main one
with the snapshot one), and the snapshot reader is waiting on the
old snapshot buffer, it was not woken up (because it is now on
the main buffer after the swap). Worse yet, when it reads the buffer
after a snapshot, it's not reading the snapshot buffer, it's reading
the live active main buffer.
Fix this by forcing a wakeup of all readers on the snapshot buffer when
a new snapshot happens, and then update the buffer that the reader
is reading to be back on the snapshot buffer.
- Fix the modification of the direct_function hash. There was a race
when new functions were added to the direct_function hash as when
it moved function entries from the old hash to the new one, a direct
function trace could be hit and not see its entry.
This is fixed by allocating the new hash, copy all the old entries
onto it as well as the new entries, and then use rcu_assign_pointer()
to update the new direct_function hash with it.
This also fixes a memory leak in that code.
-----BEGIN PGP SIGNATURE-----
iIoEABYIADIWIQRRSw7ePDh/lE+zeZMp5XQQmuv6qgUCZZAzTxQccm9zdGVkdEBn
b29kbWlzLm9yZwAKCRAp5XQQmuv6qs9IAP9e6wZ74aEjMED9nsbC49EpyCNTqa72
y0uDS/p9ppv52gD7Be+l+kJQzYNh6bZU0+B19hNC2QVn38jb7sOadfO/1Q8=
=NDkf
-----END PGP SIGNATURE-----
Merge tag 'trace-v6.7-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/trace/linux-trace
Pull tracing fixes from Steven Rostedt:
- Fix readers that are blocked on the ring buffer when buffer_percent
is 100%. They are supposed to wake up when the buffer is full, but
because the sub-buffer that the writer is on is never considered
"dirty" in the calculation, dirty pages will never equal nr_pages.
Add +1 to the dirty count in order to count for the sub-buffer that
the writer is on.
- When a reader is blocked on the "snapshot_raw" file, it is to be
woken up when a snapshot is done and be able to read the snapshot
buffer. But because the snapshot swaps the buffers (the main one with
the snapshot one), and the snapshot reader is waiting on the old
snapshot buffer, it was not woken up (because it is now on the main
buffer after the swap). Worse yet, when it reads the buffer after a
snapshot, it's not reading the snapshot buffer, it's reading the live
active main buffer.
Fix this by forcing a wakeup of all readers on the snapshot buffer
when a new snapshot happens, and then update the buffer that the
reader is reading to be back on the snapshot buffer.
- Fix the modification of the direct_function hash. There was a race
when new functions were added to the direct_function hash as when it
moved function entries from the old hash to the new one, a direct
function trace could be hit and not see its entry.
This is fixed by allocating the new hash, copy all the old entries
onto it as well as the new entries, and then use rcu_assign_pointer()
to update the new direct_function hash with it.
This also fixes a memory leak in that code.
- Fix eventfs ownership
* tag 'trace-v6.7-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/trace/linux-trace:
ftrace: Fix modification of direct_function hash while in use
tracing: Fix blocked reader of snapshot buffer
ring-buffer: Fix wake ups when buffer_percent is set to 100
eventfs: Fix file and directory uid and gid ownership
Directly return NULL or 'next' instead of breaking out of the loop.
Signed-off-by: David Laight <david.laight@aculab.com>
[ Split original patch into two independent parts - Linus ]
Link: https://lore.kernel.org/lkml/7c8828aec72e42eeb841ca0ee3397e9a@AcuMS.aculab.com/
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
osq_wait_next() is passed 'prev' from osq_lock() and NULL from
osq_unlock() but only needs the 'cpu' value to write to lock->tail.
Just pass prev->cpu or OSQ_UNLOCKED_VAL instead.
Should have no effect on the generated code since gcc manages to assume
that 'prev != NULL' due to an earlier dereference.
Signed-off-by: David Laight <david.laight@aculab.com>
[ Changed 'old' to 'old_cpu' by request from Waiman Long - Linus ]
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
struct optimistic_spin_node is private to the implementation.
Move it into the C file to ensure nothing is accessing it.
Signed-off-by: David Laight <david.laight@aculab.com>
Acked-by: Waiman Long <longman@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Masami Hiramatsu reported a memory leak in register_ftrace_direct() where
if the number of new entries are added is large enough to cause two
allocations in the loop:
for (i = 0; i < size; i++) {
hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
new = ftrace_add_rec_direct(entry->ip, addr, &free_hash);
if (!new)
goto out_remove;
entry->direct = addr;
}
}
Where ftrace_add_rec_direct() has:
if (ftrace_hash_empty(direct_functions) ||
direct_functions->count > 2 * (1 << direct_functions->size_bits)) {
struct ftrace_hash *new_hash;
int size = ftrace_hash_empty(direct_functions) ? 0 :
direct_functions->count + 1;
if (size < 32)
size = 32;
new_hash = dup_hash(direct_functions, size);
if (!new_hash)
return NULL;
*free_hash = direct_functions;
direct_functions = new_hash;
}
The "*free_hash = direct_functions;" can happen twice, losing the previous
allocation of direct_functions.
But this also exposed a more serious bug.
The modification of direct_functions above is not safe. As
direct_functions can be referenced at any time to find what direct caller
it should call, the time between:
new_hash = dup_hash(direct_functions, size);
and
direct_functions = new_hash;
can have a race with another CPU (or even this one if it gets interrupted),
and the entries being moved to the new hash are not referenced.
That's because the "dup_hash()" is really misnamed and is really a
"move_hash()". It moves the entries from the old hash to the new one.
Now even if that was changed, this code is not proper as direct_functions
should not be updated until the end. That is the best way to handle
function reference changes, and is the way other parts of ftrace handles
this.
The following is done:
1. Change add_hash_entry() to return the entry it created and inserted
into the hash, and not just return success or not.
2. Replace ftrace_add_rec_direct() with add_hash_entry(), and remove
the former.
3. Allocate a "new_hash" at the start that is made for holding both the
new hash entries as well as the existing entries in direct_functions.
4. Copy (not move) the direct_function entries over to the new_hash.
5. Copy the entries of the added hash to the new_hash.
6. If everything succeeds, then use rcu_pointer_assign() to update the
direct_functions with the new_hash.
This simplifies the code and fixes both the memory leak as well as the
race condition mentioned above.
Link: https://lore.kernel.org/all/170368070504.42064.8960569647118388081.stgit@devnote2/
Link: https://lore.kernel.org/linux-trace-kernel/20231229115134.08dd5174@gandalf.local.home
Cc: stable@vger.kernel.org
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Fixes: 763e34e74bb7d ("ftrace: Add register_ftrace_direct()")
Signed-off-by: Steven Rostedt (Google) <rostedt@goodmis.org>
If, as part of handling a hardlockup or softlockup, we've already dumped
all CPUs and we're just about to panic, don't reenable dumping and give
some other CPU a chance to hop in there and add some confusing logs right
as the panic is happening.
Link: https://lkml.kernel.org/r/20231220131534.4.Id3a9c7ec2d7d83e4080da6f8662ba2226b40543f@changeid
Signed-off-by: Douglas Anderson <dianders@chromium.org>
Cc: John Ogness <john.ogness@linutronix.de>
Cc: Lecopzer Chen <lecopzer.chen@mediatek.com>
Cc: Li Zhe <lizhe.67@bytedance.com>
Cc: Petr Mladek <pmladek@suse.com>
Cc: Pingfan Liu <kernelfans@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
If two CPUs end up reporting a hardlockup at the same time then their logs
could get interleaved which is hard to read.
The interleaving problem was especially bad with the "perf" hardlockup
detector where the locked up CPU is always the same as the running CPU and
we end up in show_regs(). show_regs() has no inherent serialization so we
could mix together two crawls if two hardlockups happened at the same time
(and if we didn't have `sysctl_hardlockup_all_cpu_backtrace` set). With
this change we'll fully serialize hardlockups when using the "perf"
hardlockup detector.
The interleaving problem was less bad with the "buddy" hardlockup
detector. With "buddy" we always end up calling
`trigger_single_cpu_backtrace(cpu)` on some CPU other than the running
one. trigger_single_cpu_backtrace() always at least serializes the
individual stack crawls because it eventually uses
printk_cpu_sync_get_irqsave(). Unfortunately the fact that
trigger_single_cpu_backtrace() eventually calls
printk_cpu_sync_get_irqsave() (on a different CPU) means that we have to
drop the "lock" before calling it and we can't fully serialize all
printouts associated with a given hardlockup. However, we still do get
the advantage of serializing the output of print_modules() and
print_irqtrace_events().
Aside from serializing hardlockups from each other, this change also has
the advantage of serializing hardlockups and softlockups from each other
if they happen to happen at the same time since they are both using the
same "lock".
Even though nobody is expected to hang while holding the lock associated
with printk_cpu_sync_get_irqsave(), out of an abundance of caution, we
don't call printk_cpu_sync_get_irqsave() until after we print out about
the hardlockup. This makes extra sure that, even if
printk_cpu_sync_get_irqsave() somehow never runs we at least print that we
saw the hardlockup. This is different than the choice made for softlockup
because hardlockup is really our last resort.
Link: https://lkml.kernel.org/r/20231220131534.3.I6ff691b3b40f0379bc860f80c6e729a0485b5247@changeid
Signed-off-by: Douglas Anderson <dianders@chromium.org>
Reviewed-by: John Ogness <john.ogness@linutronix.de>
Cc: Lecopzer Chen <lecopzer.chen@mediatek.com>
Cc: Li Zhe <lizhe.67@bytedance.com>
Cc: Petr Mladek <pmladek@suse.com>
Cc: Pingfan Liu <kernelfans@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Instead of introducing a spinlock, use printk_cpu_sync_get_irqsave() and
printk_cpu_sync_put_irqrestore() to serialize softlockup reporting. Alone
this doesn't have any real advantage over the spinlock, but this will
allow us to use the same function in a future change to also serialize
hardlockup crawls.
NOTE: for the most part this serialization is important because we often
end up in the show_regs() path and that has no built-in serialization if
there are multiple callers at once. However, even in the case where we
end up in the dump_stack() path this still has some advantages because the
stack will be guaranteed to be together in the logs with the lockup
message with no interleaving.
NOTE: the fact that printk_cpu_sync_get_irqsave() is allowed to be called
multiple times on the same CPU is important here. Specifically we hold
the "lock" while calling dump_stack() which also gets the same "lock".
This is explicitly documented to be OK and means we don't need to
introduce a variant of dump_stack() that doesn't grab the lock.
Link: https://lkml.kernel.org/r/20231220131534.2.Ia5906525d440d8e8383cde31b7c61c2aadc8f907@changeid
Signed-off-by: Douglas Anderson <dianders@chromium.org>
Reviewed-by: John Ogness <john.ogness@linutronix.de>
Reviewed-by: Li Zhe <lizhe.67@bytedance.com>
Cc: Lecopzer Chen <lecopzer.chen@mediatek.com>
Cc: Petr Mladek <pmladek@suse.com>
Cc: Pingfan Liu <kernelfans@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Patch series "watchdog: Better handling of concurrent lockups".
When we get multiple lockups at roughly the same time, the output in the
kernel logs can be very confusing since the reports about the lockups end
up interleaved in the logs. There is some code in the kernel to try to
handle this but it wasn't that complete.
Li Zhe recently made this a bit better for softlockups (specifically for
the case where `kernel.softlockup_all_cpu_backtrace` is not set) in commit
9d02330abd3e ("softlockup: serialized softlockup's log"), but that only
handled softlockup reports. Hardlockup reports still had similar issues.
This series also has a small fix to avoid dumping all stacks a second time
in the case of a panic. This is a bit unrelated to the interleaving fixes
but it does also improve the clarity of lockup reports.
This patch (of 4):
The hardlockup detector and softlockup detector both have the ability to
dump the stack of all CPUs (`kernel.hardlockup_all_cpu_backtrace` and
`kernel.softlockup_all_cpu_backtrace`). Both detectors also have some
logic to attempt to avoid interleaving printouts if two CPUs were trying
to do dumps of all CPUs at the same time. However:
- The hardlockup detector's logic still allowed interleaving some
information. Specifically another CPU could print modules and dump
the stack of the locked CPU at the same time we were dumping all
CPUs.
- In the case where `kernel.hardlockup_panic` was set in addition to
`kernel.hardlockup_all_cpu_backtrace`, when two CPUs both detected
hardlockups at the same time the second CPU could call panic() while
the first was still dumping stacks. This was especially bad if the
locked up CPU wasn't responding to the request for a backtrace since
the function nmi_trigger_cpumask_backtrace() can wait up to 10
seconds.
Let's resolve this by adopting the softlockup logic in the hardlockup
handler.
NOTES:
- As part of this, one might think that we should make a helper
function that both the hard and softlockup detectors call. This
turns out not to be super trivial since it would have to be
parameterized quite a bit since there are separate global variables
controlling each lockup detector and they print log messages that
are just different enough that it would be a pain. We probably don't
want to change the messages that are printed without good reason to
avoid throwing log parsers for a loop.
- One might also think that it would be a good idea to have the
hardlockup and softlockup detector use the same global variable to
prevent interleaving. This would make sure that softlockups and
hardlockups can't interleave each other. That _almost_ works but has
a dangerous flaw if `kernel.hardlockup_panic` is not the same as
`kernel.softlockup_panic` because we might skip a call to panic() if
one type of lockup was detected at the same time as another.
Link: https://lkml.kernel.org/r/20231220211640.2023645-1-dianders@chromium.org
Link: https://lkml.kernel.org/r/20231220131534.1.I4f35a69fbb124b5f0c71f75c631e11fabbe188ff@changeid
Signed-off-by: Douglas Anderson <dianders@chromium.org>
Cc: John Ogness <john.ogness@linutronix.de>
Cc: Lecopzer Chen <lecopzer.chen@mediatek.com>
Cc: Li Zhe <lizhe.67@bytedance.com>
Cc: Petr Mladek <pmladek@suse.com>
Cc: Pingfan Liu <kernelfans@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
image->control_page represents the starting address for allocating the
next control page, while hole_end represents the address of the last valid
byte of the currently allocated control page.
This bug actually does not affect the correctness of allocating control
pages, because image->control_page is currently only used in
kimage_alloc_crash_control_pages(), and this function, when allocating
control pages, will first align image->control_page up to the nearest
`(1 << order) << PAGE_SHIFT` boundary, then use this value as the
starting address of the next control page. This ensures that the newly
allocated control page will use the correct starting address and not
overlap with previously allocated control pages.
Although it does not affect the correctness of the final result, it is
better for us to set image->control_page to the correct value, in case
it might be used elsewhere in the future, potentially causing errors.
Therefore, after successfully allocating a control page,
image->control_page should be updated to `hole_end + 1`, rather than
hole_end.
Link: https://lkml.kernel.org/r/20231221042308.11076-1-ytcoode@gmail.com
Signed-off-by: Yuntao Wang <ytcoode@gmail.com>
Cc: Baoquan He <bhe@redhat.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Change @task to @tsk to prevent kernel-doc warnings:
kernel/stacktrace.c:138: warning: Excess function parameter 'task' description in 'stack_trace_save_tsk'
kernel/stacktrace.c:138: warning: Function parameter or member 'tsk' not described in 'stack_trace_save_tsk'
Link: https://lkml.kernel.org/r/20231220054945.17663-1-rdunlap@infradead.org
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Documentation/filesystems/relay.rst says to use
return debugfs_create_file(filename, mode, parent, buf,
&relay_file_operations);
and this is the only way relay_file_operations is used.
Thus: debugfs_create_file(&relay_file_operations)
-> __debugfs_create_file(&debugfs_full_proxy_file_operations,
&relay_file_operations)
-> dentry{inode: {i_fop: &debugfs_full_proxy_file_operations},
d_fsdata: &relay_file_operations
| DEBUGFS_FSDATA_IS_REAL_FOPS_BIT}
debugfs_full_proxy_file_operations.open is full_proxy_open, which extracts
the &relay_file_operations from the dentry, and allocates via
__full_proxy_fops_init() new fops, with trivial wrappers around release,
llseek, read, write, poll, and unlocked_ioctl, then replaces the fops on
the opened file therewith.
Naturally, all thusly-created debugfs files have .splice_read = NULL.
This was introduced in commit 49d200deaa68 ("debugfs: prevent access to
removed files' private data") from 2016-03-22.
AFAICT, relay_file_operations is the only struct file_operations used for
debugfs which defines a .splice_read callback. Hooking it up with
> diff --git a/fs/debugfs/file.c b/fs/debugfs/file.c
> index 5063434be0fc..952fcf5b2afa 100644
> --- a/fs/debugfs/file.c
> +++ b/fs/debugfs/file.c
> @@ -328,6 +328,11 @@ FULL_PROXY_FUNC(write, ssize_t, filp,
> loff_t *ppos),
> ARGS(filp, buf, size, ppos));
>
> +FULL_PROXY_FUNC(splice_read, long, in,
> + PROTO(struct file *in, loff_t *ppos, struct pipe_inode_info *pipe,
> + size_t len, unsigned int flags),
> + ARGS(in, ppos, pipe, len, flags));
> +
> FULL_PROXY_FUNC(unlocked_ioctl, long, filp,
> PROTO(struct file *filp, unsigned int cmd, unsigned long arg),
> ARGS(filp, cmd, arg));
> @@ -382,6 +387,8 @@ static void __full_proxy_fops_init(struct file_operations *proxy_fops,
> proxy_fops->write = full_proxy_write;
> if (real_fops->poll)
> proxy_fops->poll = full_proxy_poll;
> + if (real_fops->splice_read)
> + proxy_fops->splice_read = full_proxy_splice_read;
> if (real_fops->unlocked_ioctl)
> proxy_fops->unlocked_ioctl = full_proxy_unlocked_ioctl;
> }
shows it just doesn't work, and splicing always instantly returns empty
(subsequent reads actually return the contents).
No-one noticed it became dead code in 2016, who knows if it worked back
then. Clearly no-one cares; just delete it.
Link: https://lkml.kernel.org/r/dtexwpw6zcdx7dkx3xj5gyjp5syxmyretdcbcdtvrnukd4vvuh@tarta.nabijaczleweli.xyz
Signed-off-by: Ahelenia Ziemiańska <nabijaczleweli@nabijaczleweli.xyz>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Li kunyu <kunyu@nfschina.com>
Cc: Mike Rapoport (IBM) <rppt@kernel.org>
Cc: Rafael J. Wysocki <rafael@kernel.org>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Zhang Zhengming <zhang.zhengming@h3c.com>
Cc: Zhao Lei <zhao_lei1@hoperun.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
temp_end represents the address of the last available byte. Therefore,
the starting address of the memory segment with temp_end as its last
available byte and a size of `kbuf->memsz`, that is, the value of
temp_start, should be `temp_end - kbuf->memsz + 1` instead of `temp_end -
kbuf->memsz`.
Additionally, use the ALIGN_DOWN macro instead of open-coding it directly
in locate_mem_hole_top_down() to improve code readability.
Link: https://lkml.kernel.org/r/20231217033528.303333-3-ytcoode@gmail.com
Signed-off-by: Yuntao Wang <ytcoode@gmail.com>
Acked-by: Baoquan He <bhe@redhat.com>
Cc: Borislav Petkov (AMD) <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
The end parameter received by kimage_is_destination_range() should be the
last valid byte address of the target memory segment plus 1. However, in
the locate_mem_hole_bottom_up() and locate_mem_hole_top_down() functions,
the corresponding value passed to kimage_is_destination_range() is the
last valid byte address of the target memory segment, which is 1 less.
There are two ways to fix this bug. We can either correct the logic of
the locate_mem_hole_bottom_up() and locate_mem_hole_top_down() functions,
or we can fix kimage_is_destination_range() by making the end parameter
represent the last valid byte address of the target memory segment. Here,
we choose the second approach.
Due to the modification to kimage_is_destination_range(), we also need to
adjust its callers, such as kimage_alloc_normal_control_pages() and
kimage_alloc_page().
Link: https://lkml.kernel.org/r/20231217033528.303333-2-ytcoode@gmail.com
Signed-off-by: Yuntao Wang <ytcoode@gmail.com>
Acked-by: Baoquan He <bhe@redhat.com>
Cc: Borislav Petkov (AMD) <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
We already have the folio in these functions, we just need to use it.
folio_add_new_anon_rmap() didn't exist at the time they were converted to
folios.
Link: https://lkml.kernel.org/r/20231211162214.2146080-5-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
If an application blocks on the snapshot or snapshot_raw files, expecting
to be woken up when a snapshot occurs, it will not happen. Or it may
happen with an unexpected result.
That result is that the application will be reading the main buffer
instead of the snapshot buffer. That is because when the snapshot occurs,
the main and snapshot buffers are swapped. But the reader has a descriptor
still pointing to the buffer that it originally connected to.
This is fine for the main buffer readers, as they may be blocked waiting
for a watermark to be hit, and when a snapshot occurs, the data that the
main readers want is now on the snapshot buffer.
But for waiters of the snapshot buffer, they are waiting for an event to
occur that will trigger the snapshot and they can then consume it quickly
to save the snapshot before the next snapshot occurs. But to do this, they
need to read the new snapshot buffer, not the old one that is now
receiving new data.
Also, it does not make sense to have a watermark "buffer_percent" on the
snapshot buffer, as the snapshot buffer is static and does not receive new
data except all at once.
Link: https://lore.kernel.org/linux-trace-kernel/20231228095149.77f5b45d@gandalf.local.home
Cc: stable@vger.kernel.org
Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Acked-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Fixes: debdd57f5145f ("tracing: Make a snapshot feature available from userspace")
Signed-off-by: Steven Rostedt (Google) <rostedt@goodmis.org>
The tracefs file "buffer_percent" is to allow user space to set a
water-mark on how much of the tracing ring buffer needs to be filled in
order to wake up a blocked reader.
0 - is to wait until any data is in the buffer
1 - is to wait for 1% of the sub buffers to be filled
50 - would be half of the sub buffers are filled with data
100 - is not to wake the waiter until the ring buffer is completely full
Unfortunately the test for being full was:
dirty = ring_buffer_nr_dirty_pages(buffer, cpu);
return (dirty * 100) > (full * nr_pages);
Where "full" is the value for "buffer_percent".
There is two issues with the above when full == 100.
1. dirty * 100 > 100 * nr_pages will never be true
That is, the above is basically saying that if the user sets
buffer_percent to 100, more pages need to be dirty than exist in the
ring buffer!
2. The page that the writer is on is never considered dirty, as dirty
pages are only those that are full. When the writer goes to a new
sub-buffer, it clears the contents of that sub-buffer.
That is, even if the check was ">=" it would still not be equal as the
most pages that can be considered "dirty" is nr_pages - 1.
To fix this, add one to dirty and use ">=" in the compare.
Link: https://lore.kernel.org/linux-trace-kernel/20231226125902.4a057f1d@gandalf.local.home
Cc: stable@vger.kernel.org
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Acked-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Fixes: 03329f9939781 ("tracing: Add tracefs file buffer_percentage")
Signed-off-by: Steven Rostedt (Google) <rostedt@goodmis.org>
When a CPU is taken offline, the contribution of its cfs_rqs to task_groups'
load may remain and will negatively impact the calculation of the share of
the online CPUs.
To fix this bug, clear the contribution of an offlining CPU to task groups'
load and skip its contribution while it is inactive.
Here's the reproducer of the anomaly, by Imran Khan:
"So far I have encountered only one rather lengthy way of reproducing this issue,
which is as follows:
1. Take a KVM guest (booted with 4 CPUs and can be scaled up to 124 CPUs) and
create 2 custom cgroups: /sys/fs/cgroup/cpu/test_group_1 and /sys/fs/cgroup/
cpu/test_group_2
2. Assign a CPU intensive workload to each of these cgroups and start the
workload.
For my tests I am using following app:
int main(int argc, char *argv[])
{
unsigned long count, i, val;
if (argc != 2) {
printf("usage: ./a.out <number of random nums to generate> \n");
return 0;
}
count = strtoul(argv[1], NULL, 10);
printf("Generating %lu random numbers \n", count);
for (i = 0; i < count; i++) {
val = rand();
val = val % 2;
//usleep(1);
}
printf("Generated %lu random numbers \n", count);
return 0;
}
Also since the system is booted with 4 CPUs, in order to completely load the
system I am also launching 4 instances of same test app under:
/sys/fs/cgroup/cpu/
3. We can see that both of the cgroups get similar CPU time:
# systemd-cgtop --depth 1
Path Tasks %CPU Memory Input/s Output/s
/ 659 - 5.5G - -
/system.slice - - 5.7G - -
/test_group_1 4 - - - -
/test_group_2 3 - - - -
/user.slice 31 - 56.5M - -
Path Tasks %CPU Memory Input/s Output/s
/ 659 394.6 5.5G - -
/test_group_2 3 65.7 - - -
/user.slice 29 55.1 48.0M - -
/test_group_1 4 47.3 - - -
/system.slice - 2.2 5.7G - -
Path Tasks %CPU Memory Input/s Output/s
/ 659 394.8 5.5G - -
/test_group_1 4 62.9 - - -
/user.slice 28 44.9 54.2M - -
/test_group_2 3 44.7 - - -
/system.slice - 0.9 5.7G - -
Path Tasks %CPU Memory Input/s Output/s
/ 659 394.4 5.5G - -
/test_group_2 3 58.8 - - -
/test_group_1 4 51.9 - - -
/user.slice 30 39.3 59.6M - -
/system.slice - 1.9 5.7G - -
Path Tasks %CPU Memory Input/s Output/s
/ 659 394.7 5.5G - -
/test_group_1 4 60.9 - - -
/test_group_2 3 57.9 - - -
/user.slice 28 43.5 36.9M - -
/system.slice - 3.0 5.7G - -
Path Tasks %CPU Memory Input/s Output/s
/ 659 395.0 5.5G - -
/test_group_1 4 66.8 - - -
/test_group_2 3 56.3 - - -
/user.slice 29 43.1 51.8M - -
/system.slice - 0.7 5.7G - -
4. Now move systemd-udevd to one of these test groups, say test_group_1, and
perform scale up to 124 CPUs followed by scale down back to 4 CPUs from the
host side.
5. Run the same workload i.e 4 instances of CPU hogger under /sys/fs/cgroup/cpu
and one instance of CPU hogger each in /sys/fs/cgroup/cpu/test_group_1 and
/sys/fs/cgroup/test_group_2.
It can be seen that test_group_1 (the one where systemd-udevd was moved) is getting
much less CPU time than the test_group_2, even though at this point of time both of
these groups have only CPU hogger running:
# systemd-cgtop --depth 1
Path Tasks %CPU Memory Input/s Output/s
/ 1219 - 5.4G - -
/system.slice - - 5.6G - -
/test_group_1 4 - - - -
/test_group_2 3 - - - -
/user.slice 26 - 91.3M - -
Path Tasks %CPU Memory Input/s Output/s
/ 1221 394.3 5.4G - -
/test_group_2 3 82.7 - - -
/test_group_1 4 14.3 - - -
/system.slice - 0.8 5.6G - -
/user.slice 26 0.4 91.2M - -
Path Tasks %CPU Memory Input/s Output/s
/ 1221 394.6 5.4G - -
/test_group_2 3 67.4 - - -
/system.slice - 24.6 5.6G - -
/test_group_1 4 12.5 - - -
/user.slice 26 0.4 91.2M - -
Path Tasks %CPU Memory Input/s Output/s
/ 1221 395.2 5.4G - -
/test_group_2 3 60.9 - - -
/system.slice - 27.9 5.6G - -
/test_group_1 4 12.2 - - -
/user.slice 26 0.4 91.2M - -
Path Tasks %CPU Memory Input/s Output/s
/ 1221 395.2 5.4G - -
/test_group_2 3 69.4 - - -
/test_group_1 4 13.9 - - -
/user.slice 28 1.6 92.0M - -
/system.slice - 1.0 5.6G - -
Path Tasks %CPU Memory Input/s Output/s
/ 1221 395.6 5.4G - -
/test_group_2 3 59.3 - - -
/test_group_1 4 14.1 - - -
/user.slice 28 1.3 92.2M - -
/system.slice - 0.7 5.6G - -
Path Tasks %CPU Memory Input/s Output/s
/ 1221 395.5 5.4G - -
/test_group_2 3 67.2 - - -
/test_group_1 4 11.5 - - -
/user.slice 28 1.3 92.5M - -
/system.slice - 0.6 5.6G - -
Path Tasks %CPU Memory Input/s Output/s
/ 1221 395.1 5.4G - -
/test_group_2 3 76.8 - - -
/test_group_1 4 12.9 - - -
/user.slice 28 1.3 92.8M - -
/system.slice - 1.2 5.6G - -
From sched_debug data it can be seen that in bad case the load.weight of per-CPU
sched entities corresponding to test_group_1 has reduced significantly and
also load_avg of test_group_1 remains much higher than that of test_group_2,
even though systemd-udevd stopped running long time back and at this point of
time both cgroups just have the CPU hogger app as running entity."
[ mingo: Added details from the original discussion, plus minor edits to the patch. ]
Reported-by: Imran Khan <imran.f.khan@oracle.com>
Tested-by: Imran Khan <imran.f.khan@oracle.com>
Tested-by: Aaron Lu <aaron.lu@intel.com>
Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Imran Khan <imran.f.khan@oracle.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Borislav Petkov <bp@alien8.de>
Link: https://lore.kernel.org/r/20231223111545.62135-1-vincent.guittot@linaro.org
are not considered backporting material.
-----BEGIN PGP SIGNATURE-----
iHUEABYIAB0WIQTTMBEPP41GrTpTJgfdBJ7gKXxAjgUCZYys4AAKCRDdBJ7gKXxA
jtmaAQC+o04Ia7IfB8MIqp1p7dNZQo64x/EnGA8YjUnQ8N6IwQD+ImU7dHl9g9Oo
ROiiAbtMRBUfeJRsExX/Yzc1DV9E9QM=
=ZGcs
-----END PGP SIGNATURE-----
Merge tag 'mm-hotfixes-stable-2023-12-27-15-00' of git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm
Pull misc fixes from Andrew Morton:
"11 hotfixes. 7 are cc:stable and the other 4 address post-6.6 issues
or are not considered backporting material"
* tag 'mm-hotfixes-stable-2023-12-27-15-00' of git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm:
mailmap: add an old address for Naoya Horiguchi
mm/memory-failure: cast index to loff_t before shifting it
mm/memory-failure: check the mapcount of the precise page
mm/memory-failure: pass the folio and the page to collect_procs()
selftests: secretmem: floor the memory size to the multiple of page_size
mm: migrate high-order folios in swap cache correctly
maple_tree: do not preallocate nodes for slot stores
mm/filemap: avoid buffered read/write race to read inconsistent data
kunit: kasan_test: disable fortify string checker on kmalloc_oob_memset
kexec: select CRYPTO from KEXEC_FILE instead of depending on it
kexec: fix KEXEC_FILE dependencies
by moving cond_resched_rcu() to rcupdate_wait.h, we can kill another big
sched.h dependency.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
We're trying to get sched.h down to more or less just types only, not
code - rseq can live in its own header.
This helps us kill the dependency on preempt.h in sched.h.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This variable became unused in:
5e963f2bd465 ("sched/fair: Commit to EEVDF")
Signed-off-by: Wang Jinchao <wangjinchao@xfusion.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Vincent Guittot <vincent.guittot@linaro.org>
Link: https://lore.kernel.org/r/202312141319+0800-wangjinchao@xfusion.com
Running N CPU-bound tasks on an N CPUs platform:
- with asymmetric CPU capacity
- not being a DynamIq system (i.e. having a PKG level sched domain
without the SD_SHARE_PKG_RESOURCES flag set)
.. might result in a task placement where two tasks run on a big CPU
and none on a little CPU. This placement could be more optimal by
using all CPUs.
Testing platform:
Juno-r2:
- 2 big CPUs (1-2), maximum capacity of 1024
- 4 little CPUs (0,3-5), maximum capacity of 383
Testing workload ([1]):
Spawn 6 CPU-bound tasks. During the first 100ms (step 1), each tasks
is affine to a CPU, except for:
- one little CPU which is left idle.
- one big CPU which has 2 tasks affine.
After the 100ms (step 2), remove the cpumask affinity.
Behavior before the patch:
During step 2, the load balancer running from the idle CPU tags sched
domains as:
- little CPUs: 'group_has_spare'. Cf. group_has_capacity() and
group_is_overloaded(), 3 CPU-bound tasks run on a 4 CPUs
sched-domain, and the idle CPU provides enough spare capacity
regarding the imbalance_pct
- big CPUs: 'group_overloaded'. Indeed, 3 tasks run on a 2 CPUs
sched-domain, so the following path is used:
group_is_overloaded()
\-if (sgs->sum_nr_running <= sgs->group_weight) return true;
The following path which would change the migration type to
'migrate_task' is not taken:
calculate_imbalance()
\-if (env->idle != CPU_NOT_IDLE && env->imbalance == 0)
as the local group has some spare capacity, so the imbalance
is not 0.
The migration type requested is 'migrate_util' and the busiest
runqueue is the big CPU's runqueue having 2 tasks (each having a
utilization of 512). The idle little CPU cannot pull one of these
task as its capacity is too small for the task. The following path
is used:
detach_tasks()
\-case migrate_util:
\-if (util > env->imbalance) goto next;
After the patch:
As the number of failed balancing attempts grows (with
'nr_balance_failed'), progressively make it easier to migrate
a big task to the idling little CPU. A similar mechanism is
used for the 'migrate_load' migration type.
Improvement:
Running the testing workload [1] with the step 2 representing
a ~10s load for a big CPU:
Before patch: ~19.3s
After patch: ~18s (-6.7%)
Similar issue reported at:
https://lore.kernel.org/lkml/20230716014125.139577-1-qyousef@layalina.io/
Suggested-by: Vincent Guittot <vincent.guittot@linaro.org>
Signed-off-by: Pierre Gondois <pierre.gondois@arm.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Vincent Guittot <vincent.guittot@linaro.org>
Reviewed-by: Dietmar Eggemann <dietmar.eggemann@arm.com>
Acked-by: Qais Yousef <qyousef@layalina.io>
Link: https://lore.kernel.org/r/20231206090043.634697-1-pierre.gondois@arm.com
With UTIL_EST_FASTUP now being permanent, we can take advantage of the
fact that the ewma jumps directly to a higher utilization at dequeue to
simplify util_est and remove the enqueued field.
Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Tested-by: Lukasz Luba <lukasz.luba@arm.com>
Reviewed-by: Lukasz Luba <lukasz.luba@arm.com>
Reviewed-by: Dietmar Eggemann <dietmar.eggemann@arm.com>
Reviewed-by: Hongyan Xia <hongyan.xia2@arm.com>
Reviewed-by: Alex Shi <alexs@kernel.org>
Link: https://lore.kernel.org/r/20231201161652.1241695-3-vincent.guittot@linaro.org
sched_feat(UTIL_EST_FASTUP) has been added to easily disable the feature
in order to check for possibly related regressions. After 3 years, it has
never been used and no regression has been reported. Let's remove it
and make fast increase a permanent behavior.
Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Tested-by: Lukasz Luba <lukasz.luba@arm.com>
Reviewed-by: Lukasz Luba <lukasz.luba@arm.com>
Reviewed-by: Dietmar Eggemann <dietmar.eggemann@arm.com>
Reviewed-by: Hongyan Xia <hongyan.xia2@arm.com>
Reviewed-by: Tang Yizhou <yizhou.tang@shopee.com>
Reviewed-by: Yanteng Si <siyanteng@loongson.cn> [for the Chinese translation]
Reviewed-by: Alex Shi <alexs@kernel.org>
Link: https://lore.kernel.org/r/20231201161652.1241695-2-vincent.guittot@linaro.org
cpuinfo.max_freq can change at runtime because of boost as an example. This
implies that the value could be different than the one that has been
used when computing the capacity of a CPU.
The new arch_scale_freq_ref() returns a fixed and coherent reference
frequency that can be used when computing a frequency based on utilization.
Use this arch_scale_freq_ref() when available and fallback to
policy otherwise.
Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Tested-by: Lukasz Luba <lukasz.luba@arm.com>
Reviewed-by: Lukasz Luba <lukasz.luba@arm.com>
Reviewed-by: Dietmar Eggemann <dietmar.eggemann@arm.com>
Acked-by: Rafael J. Wysocki <rafael@kernel.org>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Link: https://lore.kernel.org/r/20231211104855.558096-4-vincent.guittot@linaro.org
Although it does not seem to have any untoward side-effects, the use
of ';' to separate to assignments seems more appropriate than ','.
Flagged by clang-17 -Wcomma
No functional change intended. Compile tested only.
Signed-off-by: Simon Horman <horms@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Dave Marchevsky <davemarchevsky@fb.com>
Link: https://lore.kernel.org/bpf/20231221-bpf-verifier-comma-v1-1-cde2530912e9@kernel.org
