On Mon, Sep 13, 2021 at 02:17:42PM -0700, Matthew Brost wrote:

On Mon, Sep 13, 2021 at 02:10:16PM -0700, John.C.Harrison@Intel.com wrote:

...

From: John Harrison John.C.Harrison@Intel.com

When i915 receives a context reset notification from GuC, it triggers an error capture before resetting any outstanding requsts of that context. Unfortunately, the error capture is not a time bound operation. In certain situations it can take a long time, particularly when multiple large LMEM buffers must be read back and eoncoded. If this delay is longer than other timeouts (heartbeat, test recovery, etc.) then a full GT reset can be triggered in the middle.

That can result in the context being reset by GuC actually being destroyed before the error capture completes and the GuC submission code resumes. Thus, the GuC side can start dereferencing stale pointers and Bad Things ensue.

So add a refcount get of the context during the entire reset operation. That way, the context can't be destroyed part way through no matter what other resets or user interactions occur.

Signed-off-by: John Harrison John.C.Harrison@Intel.com

drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c | 14 ++++++++++++++ 1 file changed, 14 insertions(+)

diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c b/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c index c7a41802b448..7291fd8f68a6 100644 --- a/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c +++ b/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c @@ -2920,6 +2920,7 @@ int intel_guc_context_reset_process_msg(struct intel_guc *guc, { struct intel_context *ce; int desc_idx;

• unsigned long flags;

  if (unlikely(len != 1)) { drm_err(&guc_to_gt(guc)->i915->drm, "Invalid length %u", len);

@@ -2927,11 +2928,24 @@ int intel_guc_context_reset_process_msg(struct intel_guc *guc, }

desc_idx = msg[0];

• /*

• * The context lookup uses the xarray but lookups only require an RCU lock
  

• * not the full spinlock. So take the lock explicitly and keep it until the
  

• * context has been reference count locked to ensure it can't be destroyed
  

• * asynchronously until the reset is done.
  

• */
  

• xa_lock_irqsave(&guc->context_lookup, flags); ce = g2h_context_lookup(guc, desc_idx);

• if (ce)

• intel_context_get(ce);
  

• xa_unlock_irqrestore(&guc->context_lookup, flags);

• if (unlikely(!ce)) return -EPROTO;

  guc_handle_context_reset(guc, ce);

• intel_context_put(ce);

So this is going to directly conflict with a patch that I'm about to post as I'm going to change the error capture to async operation. In that case the intel_context_put would need to be done once that op completes. I'll likely pull this patch into that series. I'd expect it to be posted by the end of the day.

tbh this entire thing is looking very scary. Somehow we can race with other processing while we try to handle a reset. That's fragile at best.

Proper fix is to exclude these kind of problems by design, by either guaranteeing that no concurrent dequeuing of guc2host message can happen, or by holding appropriate locks, or by keeping track of anything pending in a more controlled way (something like expected g2h messages as separate structs, instead of the current spaghetti layering violation chaos we have for processing g2h message).

We maybe should use a few of these things when we're going through the locking engineering training with guc team. -Daniel

Matt

...

return 0; } -- 2.25.1