linux-kernel - [PATCH v2 00/13] drm+msm: Shrinker and LRU rework

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Message-Id: <20220719171900.289265-1-robdclark@gmail.com>
Date:   Tue, 19 Jul 2022 10:18:44 -0700
From:   Rob Clark <robdclark@...il.com>
To:     dri-devel@...ts.freedesktop.org
Cc:     freedreno@...ts.freedesktop.org, linux-arm-msm@...r.kernel.org,
        Rob Clark <robdclark@...omium.org>,
        linux-kernel@...r.kernel.org (open list),
        Sean Paul <sean@...rly.run>
Subject: [PATCH v2 00/13] drm+msm: Shrinker and LRU rework

From: Rob Clark <robdclark@...omium.org>

Mostly a resend with a small fix in the last patch (and a couple early
patches drop out when rebasing).

original description below:

This is mostly motivated by getting drm/msm to pass an i-g-t shrinker
test that I've been working on.  In particular the test creates and
cycles between more GEM buffers than what will fit in RAM to force
eviction and re-pin.  (There are sub-tests that cover this case both
single threaded and with many child processes in parallel.)

Getting this test to pass necessitated a few improvements:

1. Re-ordering submit path to get rid of __GFP_NORETRY (in the common
   case, doing this for syncobjs is still TODO)
2. Decoupling locks needed in the retire path from locks that could
   be held while hitting reclaim in the submit path
3. If necessary, allow stalling on active BOs for reclaim.

The latter point is because we pin objects in the synchronous part of
the submit path (before queuing on the drm gpu-scheduler), which means
in the parallel variant of the i-g-t test, we need to be able to block
in the reclaim path until some queued work has completed/retired.

In the process of re-working how drm/msm tracks buffer state in it's
various LRU lists, I refactored out a drm_gem_lru helper which, in
theory, should be usable by other drivers and drm shmem helpers for
implementing LRU tracking and shrinker.

Rob Clark (13):
  drm/msm: Reorder lock vs submit alloc
  drm/msm: Small submit cleanup
  drm/msm: Split out idr_lock
  drm/msm/gem: Check for active in shrinker path
  drm/msm/gem: Rename update_inactive
  drm/msm/gem: Rename to pin/unpin_pages
  drm/msm/gem: Consolidate pin/unpin paths
  drm/msm/gem: Remove active refcnt
  drm/gem: Add LRU/shrinker helper
  drm/msm/gem: Convert to using drm_gem_lru
  drm/msm/gem: Unpin buffers earlier
  drm/msm/gem: Consolidate shrinker trace
  drm/msm/gem: Evict active GEM objects when necessary

 drivers/gpu/drm/drm_gem.c              | 183 +++++++++++++++++++++++++
 drivers/gpu/drm/msm/msm_drv.c          |  18 ++-
 drivers/gpu/drm/msm/msm_drv.h          |  70 +++++++---
 drivers/gpu/drm/msm/msm_gem.c          | 149 +++++++-------------
 drivers/gpu/drm/msm/msm_gem.h          | 112 +--------------
 drivers/gpu/drm/msm/msm_gem_prime.c    |   4 +-
 drivers/gpu/drm/msm/msm_gem_shrinker.c | 164 ++++++++++------------
 drivers/gpu/drm/msm/msm_gem_submit.c   |  78 ++++-------
 drivers/gpu/drm/msm/msm_gpu.c          |   3 -
 drivers/gpu/drm/msm/msm_gpu.h          |  10 +-
 drivers/gpu/drm/msm/msm_gpu_trace.h    |  36 +++--
 drivers/gpu/drm/msm/msm_submitqueue.c  |   1 +
 include/drm/drm_gem.h                  |  56 ++++++++
 13 files changed, 483 insertions(+), 401 deletions(-)

-- 
2.36.1