drm/i915: Enforce read order into the hardware status page csb
Subsystems (2)
PATCH 1/1 drm/i915: Enforce read order into the hardware status page csb
Read ordering on cpu can be out of order and speculative. In addition, the access to the hardware status page is snooped. This raises a concern that we might use the HWSP in an unexpected way as we may load the head of a cbs entry before we access the tail. Concerns like that the coherency protocol is tied somehow to the tail read snoop.
To enforce that we really do read the tail before we fetch the csb entry pointed by head, insert a read memory barrier after we have read the tail.
This fixes, or masks due to added latency, context status buffer incoherence on cnl, where we see an old context status entry still on that a csb slot.
References: https://bugs.freedesktop.org/show_bug.cgi?id=105888
--- drivers/gpu/drm/i915/intel_lrc.c | 2 ++ 1 file changed, 2 insertions(+)
drivers/gpu/drm/i915/intel_lrc.c +2 -0
@@ -969,6 +969,8 @@ static void execlists_submission_tasklet(unsigned long data) head = execlists->csb_head; tail =READ_ONCE (buf[write_idx]); +/* Enforce ordering of reads into the HWSP */ +rmb (); } GEM_TRACE("%s cs-irq head=%d [%d%s], tail=%d [%d%s]\n", engine->name,
Signature · 6 lines
2.14.1 _______________________________________________ Intel-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Quoting Mika Kuoppala (2018-04-12 14:52:37)
Quoted 1 line
Read ordering on cpu can be out of order and speculative.
If it didn't include a barrier to prevent compiler mispeculation. rmb() only applies to other CPU cores and not the GPU so this is entirely hocus.
-Chris
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Quoting Chris Wilson (2018-04-12 14:57:42)
Quoted 6 lines
Quoting Mika Kuoppala (2018-04-12 14:52:37)
Quoted 1 line
Read ordering on cpu can be out of order and speculative.
If it didn't include a barrier to prevent compiler mispeculation.
rmb()only applies to other CPU cores and not the GPU so this is entirely hocus.
To put it another way this is just imposing the cost of an uncached mmio which the entire point of using HWSP in the first place is to avoid.
-Chris
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Quoting Chris Wilson (2018-04-12 14:57:42)
Quoted 6 lines
Quoting Mika Kuoppala (2018-04-12 14:52:37)
Quoted 1 line
Read ordering on cpu can be out of order and speculative.
If it didn't include a barrier to prevent compiler mispeculation.
rmb()only applies to other CPU cores and not the GPU so this is entirely hocus.
Or think of it in terms checkpatch will ask you, where is the corresponding wmb()?
-Chris
_______________________________________________
Intel-gfx mailing list [email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Quoted 6 lines · Chris Wilson
Quoting Mika Kuoppala (2018-04-12 14:52:37)
Quoted 1 line
Read ordering on cpu can be out of order and speculative.
If it didn't include a barrier to prevent compiler mispeculation.
rmb()only applies to other CPU cores and not the GPU so this is entirely hocus.
How I understand the Documentation/memory-barrier.txt/THE THINGS CPUS GET UP TO, the compiler memory barrier is not enough.
Quoted 1 line
-Chris
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Patchwork
CI✓ Fi.CI.BAT: success for drm/i915: Enforce read order into the hardware status page csb
In reply to Mika Kuoppala== Series Details ==
== Series Details ==
Series: drm/i915: Enforce read order into the hardware status page csb
URL : https://patchwork.freedesktop.org/series/41622/
State : success
== Summary ==
= CI Bug Log - changes from CI_DRM_4049 -> Patchwork_8678 =
== Summary - SUCCESS ==
No regressions found.
External URL: https://patchwork.freedesktop.org/api/1.0/series/41622/revisions/1/mbox/
== Known issues ==
Here are the changes found in Patchwork_8678 that come from known issues:
=== IGT changes ===
==== Issues hit ====
igt@gem_mmap_gtt@basic-small-bo-tiledx:
fi-gdg-551: PASS -> FAIL (fdo#102575)
igt@kms_pipe_crc_basic@suspend-read-crc-pipe-b:
fi-elk-e7500: PASS -> DMESG-WARN (fdo#105225) +1
igt@kms_pipe_crc_basic@suspend-read-crc-pipe-c:
fi-bxt-dsi: PASS -> INCOMPLETE (fdo#103927) fdo#102575 https://bugs.freedesktop.org/show_bug.cgi?id=102575
fdo#103927 https://bugs.freedesktop.org/show_bug.cgi?id=103927
fdo#105225 https://bugs.freedesktop.org/show_bug.cgi?id=105225
== Participating hosts (35 -> 33) ==
Additional (1): fi-cnl-y3
Missing (3): fi-ctg-p8600 fi-ilk-m540 fi-skl-6700hq
== Build changes ==
* Linux: CI_DRM_4049 -> Patchwork_8678
CI_DRM_4049: 3dbfa04d62f1f1214a03c3b2d30f987dccf50ab4 @ git://anongit.freedesktop.org/gfx-ci/linux
IGT_4426: d502f055ac4500cada758876a512ac4f14b34851 @ git://anongit.freedesktop.org/xorg/app/intel-gpu-tools
Patchwork_8678: 02b77742d0823525ce798b2bcfdc4a3bec295b32 @ git://anongit.freedesktop.org/gfx-ci/linux
piglit_4426: 93b35926a150e318439d2505901288594b3548f5 @ git://anongit.freedesktop.org/piglit
== Linux commits ==
02b77742d082 drm/i915: Enforce read order into the hardware status page csb
== Logs ==
For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_8678/issues.html
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Patchwork
CI✓ Fi.CI.IGT: success for drm/i915: Enforce read order into the hardware status page csb
In reply to Mika Kuoppala== Series Details ==
== Series Details ==
Series: drm/i915: Enforce read order into the hardware status page csb
URL : https://patchwork.freedesktop.org/series/41622/
State : success
== Summary ==
= CI Bug Log - changes from CI_DRM_4049_full -> Patchwork_8678_full =
== Summary - WARNING ==
Minor unknown changes coming with Patchwork_8678_full need to be verified manually.
If you think the reported changes have nothing to do with the changes introduced in Patchwork_8678_full, please notify your bug team to allow them to document this new failure mode, which will reduce false positives in CI.
External URL: https://patchwork.freedesktop.org/api/1.0/series/41622/revisions/1/mbox/
== Possible new issues ==
Here are the unknown changes that may have been introduced in Patchwork_8678_full:
=== IGT changes ===
==== Warnings ====
igt@gem_mocs_settings@mocs-rc6-render:
shard-kbl: PASS -> SKIP +1
igt@gem_mocs_settings@mocs-rc6-vebox:
shard-kbl: SKIP -> PASS== Known issues ==
Here are the changes found in Patchwork_8678_full that come from known issues:
=== IGT changes ===
==== Issues hit ====
igt@kms_flip@plain-flip-ts-check:
shard-hsw: PASS -> FAIL (fdo#100368)
igt@pm_rpm@debugfs-read:
shard-kbl: PASS -> DMESG-WARN (fdo#103313, fdo#103558) +2
==== Possible fixes ====
igt@kms_flip@2x-dpms-vs-vblank-race-interruptible:
shard-hsw: FAIL (fdo#103060) -> PASS +1
igt@kms_flip@plain-flip-ts-check-interruptible:
shard-hsw: FAIL (fdo#100368) -> PASS +1 fdo#100368 https://bugs.freedesktop.org/show_bug.cgi?id=100368
fdo#103060 https://bugs.freedesktop.org/show_bug.cgi?id=103060
fdo#103313 https://bugs.freedesktop.org/show_bug.cgi?id=103313
fdo#103558 https://bugs.freedesktop.org/show_bug.cgi?id=103558
== Participating hosts (6 -> 4) ==
Missing (2): shard-glk shard-glkb
== Build changes ==
* Linux: CI_DRM_4049 -> Patchwork_8678
CI_DRM_4049: 3dbfa04d62f1f1214a03c3b2d30f987dccf50ab4 @ git://anongit.freedesktop.org/gfx-ci/linux
IGT_4426: d502f055ac4500cada758876a512ac4f14b34851 @ git://anongit.freedesktop.org/xorg/app/intel-gpu-tools
Patchwork_8678: 02b77742d0823525ce798b2bcfdc4a3bec295b32 @ git://anongit.freedesktop.org/gfx-ci/linux
piglit_4426: 93b35926a150e318439d2505901288594b3548f5 @ git://anongit.freedesktop.org/piglit
== Logs ==
For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_8678/shards.html
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx