Re: [PATCH v1] Documentation: Add document for false sharing

[Bagas Sanjaya <bagasdotme@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 3516 bytes --]

On Fri, Mar 24, 2023 at 03:13:16PM +0800, Feng Tang wrote:
> +There are many real-world cases of performance regressions caused by
> +false sharing, and one is a rw_semaphore 'mmap_lock' inside struct
"... . One of these is rw_semaphore 'mmap_lock' ..."

But I think in English we commonly name things as "foobar struct"
instead of "struct foobar" (that is, common noun follow the proper noun
that names something).

> +* A global datum accessed (shared) by many CPUs
Global data?

> +Following 'mitigation' section provides real-world examples.
"The real-world examples are given in 'Possible mitigations' sections."
> +  #perf c2c record -ag sleep 3
> +  #perf c2c report --call-graph none -k vmlinux

Are these commands really run as root?

> +
> +Run it when testing will-it-scale's tlb_flush1 case, and the report
> +has pieces like::

"When running above during testing ..., perf reports something like::"

> +False sharing hurting performance cases are seen more frequently with
> +core count increasing, and there have been many patches merged to
> +solve it, like in networking and memory management subsystems.  Some
> +common mitigations (with examples) are:

"... Because of these detrimental effects, many patches have been
proposed across variety of subsystems (like networking and memory
management) and merged."

> +
> +* Separate hot global data in its own dedicated cache line, even if it
> +  is just a 'short' type. The downside is more consumption of memory,
> +  cache line and TLB entries.
> +
> +  Commit 91b6d3256356 ("net: cache align tcp_memory_allocated, tcp_sockets_allocated")
> +
> +* Reorganize the data structure, separate the interfering members to
> +  different cache lines.  One downside is it may introduce new false
> +  sharing of other members.
> +
> +  Commit 802f1d522d5f ("mm: page_counter: re-layout structure to reduce false sharing")
> +
> +* Replace 'write' with 'read' when possible, especially in loops.
> +  Like for some global variable, use compare(read)-then-write instead
> +  of unconditional write. For example, use:
"... For example, write::"
> +
> +	if (!test_bit(XXX))
> +		set_bit(XXX);
> +
> +  instead of directly "set_bit(XXX);", similarly for atomic_t data.
> +
> +  Commit 7b1002f7cfe5 ("bcache: fixup bcache_dev_sectors_dirty_add() multithreaded CPU false sharing")
> +  Commit 292648ac5cf1 ("mm: gup: allow FOLL_PIN to scale in SMP")
> +
> +* Turn hot global data to 'per-cpu data + global data' when possible,
> +  or reasonably increase the threshold for syncing per-cpu data to
> +  global data, to reduce or postpone the 'write' to that global data.
> +
> +  Commit 520f897a3554 ("ext4: use percpu_counters for extent_status cache hits/misses")
> +  Commit 56f3547bfa4d ("mm: adjust vm_committed_as_batch according to vm overcommit policy")

IMO it's odd to jump to specifying example commits without some sort of
conjuction (e.g. "for example, see commit <commit>").

> +
> +Surely, all mitigations should be carefully verified to not cause side
> +effects.  And to avoid false sharing in advance during coding, it's
> +better to:
> +
> +* Be aware of cache line boundaries
> +* Group mostly read-only fields together
> +* Group things that are written at the same time together
> +* Separate known read-mostly and written-mostly fields

Proactively prevent false sharing with above tips?

Thanks.

-- 
An old man doll... just what I always wanted! - Clara

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 228 bytes --]