Re: [Intel-gfx] [PATCH RFC] drm/i915/gt: Retry RING_HEAD reset until it sticks
On Fri, Jul 15, 2022 at 09:26:16AM +0100, Mauro Carvalho Chehab wrote:
> From: Chris Wilson
>
> On Haswell, in particular, we see an issue where resets fails because
Can we then make this platform specific?
Only because some older hw doesn't behave like expected we shouldn't
make this a default & global workaround.
> the engine resumes from an incorrect RING_HEAD. Since the RING_HEAD
> doesn't point to the remaining requests to re-run, but may instead point
> into the uninitialised portion of the ring, the GPU may be then fed
> invalid instructions from a privileged context, oft pushing the GPU into
> an unrecoverable hang.
>
> If at first the write doesn't succeed, try, try again.
>
> References: https://gitlab.freedesktop.org/drm/intel/-/issues/5432
> Testcase: igt/i915_selftest/hangcheck
> Signed-off-by: Chris Wilson
> Cc: Andrzej Hajda
> Cc: Mika Kuoppala
> Signed-off-by: Mauro Carvalho Chehab
> ---
> .../gpu/drm/i915/gt/intel_ring_submission.c | 44 +--
> drivers/gpu/drm/i915/i915_utils.h | 10 +
> 2 files changed, 40 insertions(+), 14 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/gt/intel_ring_submission.c
> b/drivers/gpu/drm/i915/gt/intel_ring_submission.c
> index d5d6f1fadcae..cc53feb1f8ed 100644
> --- a/drivers/gpu/drm/i915/gt/intel_ring_submission.c
> +++ b/drivers/gpu/drm/i915/gt/intel_ring_submission.c
> @@ -190,6 +190,7 @@ static bool stop_ring(struct intel_engine_cs *engine)
> static int xcs_resume(struct intel_engine_cs *engine)
> {
> struct intel_ring *ring = engine->legacy.ring;
> + ktime_t kt;
>
> ENGINE_TRACE(engine, "ring:{HEAD:%04x, TAIL:%04x}\n",
>ring->head, ring->tail);
> @@ -228,9 +229,20 @@ static int xcs_resume(struct intel_engine_cs *engine)
> set_pp_dir(engine);
>
> /* First wake the ring up to an empty/idle ring */
> - ENGINE_WRITE_FW(engine, RING_HEAD, ring->head);
> + until_timeout_ns(kt, 2 * NSEC_PER_MSEC) {
> + ENGINE_WRITE_FW(engine, RING_HEAD, ring->head);
> + if (ENGINE_READ_FW(engine, RING_HEAD) == ring->head)
> + break;
> + }
> +
> ENGINE_WRITE_FW(engine, RING_TAIL, ring->head);
> - ENGINE_POSTING_READ(engine, RING_TAIL);
> + if (ENGINE_READ_FW(engine, RING_HEAD) != ENGINE_READ_FW(engine,
> RING_TAIL)) {
> + ENGINE_TRACE(engine, "failed to reset empty ring: [%x, %x]:
> %x\n",
> + ENGINE_READ_FW(engine, RING_HEAD),
> + ENGINE_READ_FW(engine, RING_TAIL),
> + ring->head);
> + goto err;
> + }
commit message mentions until this point I'm afraid... everything below
(except the new until_timeout_ns) looks like a different patch to me,
or deserves some mention in the commit msg.
>
> ENGINE_WRITE_FW(engine, RING_CTL,
> RING_CTL_SIZE(ring->size) | RING_VALID);
> @@ -239,12 +251,16 @@ static int xcs_resume(struct intel_engine_cs *engine)
> if (__intel_wait_for_register_fw(engine->uncore,
>RING_CTL(engine->mmio_base),
>RING_VALID, RING_VALID,
> - 5000, 0, NULL))
> + 5000, 0, NULL)) {
> + ENGINE_TRACE(engine, "failed to restart\n");
> goto err;
> + }
>
> - if (GRAPHICS_VER(engine->i915) > 2)
> + if (GRAPHICS_VER(engine->i915) > 2) {
> ENGINE_WRITE_FW(engine,
> RING_MI_MODE, _MASKED_BIT_DISABLE(STOP_RING));
> + ENGINE_POSTING_READ(engine, RING_MI_MODE);
> + }
>
> /* Now awake, let it get started */
> if (ring->tail != ring->head) {
> @@ -257,16 +273,16 @@ static int xcs_resume(struct intel_engine_cs *engine)
> return 0;
>
> err:
> - drm_err(&engine->i915->drm,
> - "%s initialization failed; "
> - "ctl %08x (valid? %d) head %08x [%08x] tail %08x [%08x] start
> %08x [expected %08x]\n",
> - engine->name,
> - ENGINE_READ(engine, RING_CTL),
> - ENGINE_READ(engine, RING_CTL) & RING_VALID,
> - ENGINE_READ(engine, RING_HEAD), ring->head,
> - ENGINE_READ(engine, RING_TAIL), ring->tail,
> - ENGINE_READ(engine, RING_START),
> - i915_ggtt_offset(ring->vma));
> + ENGINE_TRACE(engine,
> + "initialization failed; "
> + "ctl %08x (valid? %d) head %08x [%08x] tail %08x [%08x]
> start %08x [expected %08x]\n",
> + ENGINE_READ(engine, RING_CTL),
> + ENGINE_READ(engine, RING_CTL) & RING_VALID,
> + ENGINE_READ(engine, RING_HEAD), ring->head,
> + ENGINE_READ(engine, RING_TAIL), ring->tail,
> + ENGINE_READ(engine, RING_START),
> + i915_ggtt_offset(ring->vma));
> + GEM
Re: [Intel-gfx] [PATCH RFC] drm/i915/gt: Retry RING_HEAD reset until it sticks
On 15.07.2022 10:26, Mauro Carvalho Chehab wrote:
From: Chris Wilson
On Haswell, in particular, we see an issue where resets fails because
the engine resumes from an incorrect RING_HEAD. Since the RING_HEAD
doesn't point to the remaining requests to re-run, but may instead point
into the uninitialised portion of the ring, the GPU may be then fed
invalid instructions from a privileged context, oft pushing the GPU into
an unrecoverable hang.
If at first the write doesn't succeed, try, try again.
References: https://gitlab.freedesktop.org/drm/intel/-/issues/5432
Testcase: igt/i915_selftest/hangcheck
Signed-off-by: Chris Wilson
Cc: Andrzej Hajda
Cc: Mika Kuoppala
Signed-off-by: Mauro Carvalho Chehab
That is pity hw does not provide reliable way of reset.
Reviewed-by: Andrzej Hajda
Regards
Andrzej
---
.../gpu/drm/i915/gt/intel_ring_submission.c | 44 +--
drivers/gpu/drm/i915/i915_utils.h | 10 +
2 files changed, 40 insertions(+), 14 deletions(-)
diff --git a/drivers/gpu/drm/i915/gt/intel_ring_submission.c
b/drivers/gpu/drm/i915/gt/intel_ring_submission.c
index d5d6f1fadcae..cc53feb1f8ed 100644
--- a/drivers/gpu/drm/i915/gt/intel_ring_submission.c
+++ b/drivers/gpu/drm/i915/gt/intel_ring_submission.c
@@ -190,6 +190,7 @@ static bool stop_ring(struct intel_engine_cs *engine)
static int xcs_resume(struct intel_engine_cs *engine)
{
struct intel_ring *ring = engine->legacy.ring;
+ ktime_t kt;
ENGINE_TRACE(engine, "ring:{HEAD:%04x, TAIL:%04x}\n",
ring->head, ring->tail);
@@ -228,9 +229,20 @@ static int xcs_resume(struct intel_engine_cs *engine)
set_pp_dir(engine);
/* First wake the ring up to an empty/idle ring */
- ENGINE_WRITE_FW(engine, RING_HEAD, ring->head);
+ until_timeout_ns(kt, 2 * NSEC_PER_MSEC) {
+ ENGINE_WRITE_FW(engine, RING_HEAD, ring->head);
+ if (ENGINE_READ_FW(engine, RING_HEAD) == ring->head)
+ break;
+ }
+
ENGINE_WRITE_FW(engine, RING_TAIL, ring->head);
- ENGINE_POSTING_READ(engine, RING_TAIL);
+ if (ENGINE_READ_FW(engine, RING_HEAD) != ENGINE_READ_FW(engine,
RING_TAIL)) {
+ ENGINE_TRACE(engine, "failed to reset empty ring: [%x, %x]:
%x\n",
+ENGINE_READ_FW(engine, RING_HEAD),
+ENGINE_READ_FW(engine, RING_TAIL),
+ring->head);
+ goto err;
+ }
ENGINE_WRITE_FW(engine, RING_CTL,
RING_CTL_SIZE(ring->size) | RING_VALID);
@@ -239,12 +251,16 @@ static int xcs_resume(struct intel_engine_cs *engine)
if (__intel_wait_for_register_fw(engine->uncore,
RING_CTL(engine->mmio_base),
RING_VALID, RING_VALID,
-5000, 0, NULL))
+5000, 0, NULL)) {
+ ENGINE_TRACE(engine, "failed to restart\n");
goto err;
+ }
- if (GRAPHICS_VER(engine->i915) > 2)
+ if (GRAPHICS_VER(engine->i915) > 2) {
ENGINE_WRITE_FW(engine,
RING_MI_MODE, _MASKED_BIT_DISABLE(STOP_RING));
+ ENGINE_POSTING_READ(engine, RING_MI_MODE);
+ }
/* Now awake, let it get started */
if (ring->tail != ring->head) {
@@ -257,16 +273,16 @@ static int xcs_resume(struct intel_engine_cs *engine)
return 0;
err:
- drm_err(&engine->i915->drm,
- "%s initialization failed; "
- "ctl %08x (valid? %d) head %08x [%08x] tail %08x [%08x] start %08x
[expected %08x]\n",
- engine->name,
- ENGINE_READ(engine, RING_CTL),
- ENGINE_READ(engine, RING_CTL) & RING_VALID,
- ENGINE_READ(engine, RING_HEAD), ring->head,
- ENGINE_READ(engine, RING_TAIL), ring->tail,
- ENGINE_READ(engine, RING_START),
- i915_ggtt_offset(ring->vma));
+ ENGINE_TRACE(engine,
+"initialization failed; "
+"ctl %08x (valid? %d) head %08x [%08x] tail %08x [%08x] start
%08x [expected %08x]\n",
+ENGINE_READ(engine, RING_CTL),
+ENGINE_READ(engine, RING_CTL) & RING_VALID,
+ENGINE_READ(engine, RING_HEAD), ring->head,
+ENGINE_READ(engine, RING_TAIL), ring->tail,
+ENGINE_READ(engine, RING_START),
+i915_ggtt_offset(ring->vma));
+ GEM_TRACE_DUMP();
return -EIO;
}
diff --git a/drivers/gpu/drm/i915/i915_utils.h b/drivers/gpu/drm/i915/i915_utils.h
index c10d68cdc3ca..717fb6b9cc15 100644
--- a/drivers/gpu/drm/i915/i915_utils.h
+++ b/drivers/gpu/drm/i915/i915_utils.h
@@ -256,6 +256,16 @@ wait_remaining_ms_from_jiffies(unsigned long
timestamp_jiffies, int to_wait_ms)
