From: John Harrison <john.c.harrison@intel.com>
To: Matthew Brost <matthew.brost@intel.com>
Cc: intel-gfx@lists.freedesktop.org, dri-devel@lists.freedesktop.org
Subject: Re: [Intel-gfx] [PATCH 28/47] drm/i915: Hold reference to intel_context over life of i915_request
Date: Mon, 12 Jul 2021 14:48:51 -0700 [thread overview]
Message-ID: <f343b084-faf9-71f7-e4f0-9df42c330286@intel.com> (raw)
In-Reply-To: <20210712213600.GA15656@DUT181-TGLU.fm.intel.com>
On 7/12/2021 14:36, Matthew Brost wrote:
> On Mon, Jul 12, 2021 at 08:05:30PM +0000, Matthew Brost wrote:
>> On Mon, Jul 12, 2021 at 11:23:14AM -0700, John Harrison wrote:
>>> On 6/24/2021 00:04, Matthew Brost wrote:
>>>> Hold a reference to the intel_context over life of an i915_request.
>>>> Without this an i915_request can exist after the context has been
>>>> destroyed (e.g. request retired, context closed, but user space holds a
>>>> reference to the request from an out fence). In the case of GuC
>>>> submission + virtual engine, the engine that the request references is
>>>> also destroyed which can trigger bad pointer dref in fence ops (e.g.
>>> Maybe quickly explain a why this is different for GuC submission vs
>>> execlist? Presumably it is about only decomposing virtual engines to
>>> physical ones in execlist mode?
>>>
>> Yes, it because in execlists we always end up pointing to a physical
>> engine in the end while in GuC mode we can be pointing to dynamically
>> allocated virtual engine. I can update the comment.
>>
>>>> i915_fence_get_driver_name). We could likely change
>>>> i915_fence_get_driver_name to avoid touching the engine but let's just
>>>> be safe and hold the intel_context reference.
>>>>
>>>> Signed-off-by: Matthew Brost <matthew.brost@intel.com>
>>>> ---
>>>> drivers/gpu/drm/i915/i915_request.c | 54 ++++++++++++-----------------
>>>> 1 file changed, 22 insertions(+), 32 deletions(-)
>>>>
>>>> diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c
>>>> index de9deb95b8b1..dec5a35c9aa2 100644
>>>> --- a/drivers/gpu/drm/i915/i915_request.c
>>>> +++ b/drivers/gpu/drm/i915/i915_request.c
>>>> @@ -126,39 +126,17 @@ static void i915_fence_release(struct dma_fence *fence)
>>>> i915_sw_fence_fini(&rq->semaphore);
>>>> /*
>>>> - * Keep one request on each engine for reserved use under mempressure
>>>> - *
>>>> - * We do not hold a reference to the engine here and so have to be
>>>> - * very careful in what rq->engine we poke. The virtual engine is
>>>> - * referenced via the rq->context and we released that ref during
>>>> - * i915_request_retire(), ergo we must not dereference a virtual
>>>> - * engine here. Not that we would want to, as the only consumer of
>>>> - * the reserved engine->request_pool is the power management parking,
>>>> - * which must-not-fail, and that is only run on the physical engines.
>>>> - *
>>>> - * Since the request must have been executed to be have completed,
>>>> - * we know that it will have been processed by the HW and will
>>>> - * not be unsubmitted again, so rq->engine and rq->execution_mask
>>>> - * at this point is stable. rq->execution_mask will be a single
>>>> - * bit if the last and _only_ engine it could execution on was a
>>>> - * physical engine, if it's multiple bits then it started on and
>>>> - * could still be on a virtual engine. Thus if the mask is not a
>>>> - * power-of-two we assume that rq->engine may still be a virtual
>>>> - * engine and so a dangling invalid pointer that we cannot dereference
>>>> - *
>>>> - * For example, consider the flow of a bonded request through a virtual
>>>> - * engine. The request is created with a wide engine mask (all engines
>>>> - * that we might execute on). On processing the bond, the request mask
>>>> - * is reduced to one or more engines. If the request is subsequently
>>>> - * bound to a single engine, it will then be constrained to only
>>>> - * execute on that engine and never returned to the virtual engine
>>>> - * after timeslicing away, see __unwind_incomplete_requests(). Thus we
>>>> - * know that if the rq->execution_mask is a single bit, rq->engine
>>>> - * can be a physical engine with the exact corresponding mask.
>>>> + * Keep one request on each engine for reserved use under mempressure,
>>>> + * do not use with virtual engines as this really is only needed for
>>>> + * kernel contexts.
>>>> */
>>>> - if (is_power_of_2(rq->execution_mask) &&
>>>> - !cmpxchg(&rq->engine->request_pool, NULL, rq))
>>>> + if (!intel_engine_is_virtual(rq->engine) &&
>>>> + !cmpxchg(&rq->engine->request_pool, NULL, rq)) {
>>>> + intel_context_put(rq->context);
>>>> return;
>>>> + }
>>>> +
>>>> + intel_context_put(rq->context);
>>> The put is actually unconditional? So it could be moved before the if?
>>>
>> Yep, I think so.
>>
> Wait nope. We reference rq->engine which could a virtual engine and the
> intel_context_put could free that engine. So we need to do the put after
> we reference it.
>
> Matt
Doh! That's a pretty good reason.
Okay, with a tweaked description to explain about virtual engines being
different on GuC vs execlist...
Reviewed-by: John Harrison <John.C.Harrison@Intel.com>
>
>> Matt
>>
>>> John.
>>>
>>>> kmem_cache_free(global.slab_requests, rq);
>>>> }
>>>> @@ -977,7 +955,18 @@ __i915_request_create(struct intel_context *ce, gfp_t gfp)
>>>> }
>>>> }
>>>> - rq->context = ce;
>>>> + /*
>>>> + * Hold a reference to the intel_context over life of an i915_request.
>>>> + * Without this an i915_request can exist after the context has been
>>>> + * destroyed (e.g. request retired, context closed, but user space holds
>>>> + * a reference to the request from an out fence). In the case of GuC
>>>> + * submission + virtual engine, the engine that the request references
>>>> + * is also destroyed which can trigger bad pointer dref in fence ops
>>>> + * (e.g. i915_fence_get_driver_name). We could likely change these
>>>> + * functions to avoid touching the engine but let's just be safe and
>>>> + * hold the intel_context reference.
>>>> + */
>>>> + rq->context = intel_context_get(ce);
>>>> rq->engine = ce->engine;
>>>> rq->ring = ce->ring;
>>>> rq->execution_mask = ce->engine->mask;
>>>> @@ -1054,6 +1043,7 @@ __i915_request_create(struct intel_context *ce, gfp_t gfp)
>>>> GEM_BUG_ON(!list_empty(&rq->sched.waiters_list));
>>>> err_free:
>>>> + intel_context_put(ce);
>>>> kmem_cache_free(global.slab_requests, rq);
>>>> err_unreserve:
>>>> intel_context_unpin(ce);
>> _______________________________________________
>> Intel-gfx mailing list
>> Intel-gfx@lists.freedesktop.org
>> https://lists.freedesktop.org/mailman/listinfo/intel-gfx
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
next prev parent reply other threads:[~2021-07-12 21:48 UTC|newest]
Thread overview: 170+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-06-24 7:04 [Intel-gfx] [PATCH 00/47] GuC submission support Matthew Brost
2021-06-24 7:04 ` [Intel-gfx] [PATCH 01/47] drm/i915/guc: Relax CTB response timeout Matthew Brost
2021-06-24 17:23 ` Michal Wajdeczko
2021-06-24 7:04 ` [Intel-gfx] [PATCH 02/47] drm/i915/guc: Improve error message for unsolicited CT response Matthew Brost
2021-06-25 11:58 ` Michal Wajdeczko
2021-06-24 7:04 ` [Intel-gfx] [PATCH 03/47] drm/i915/guc: Increase size of CTB buffers Matthew Brost
2021-06-24 13:49 ` Michal Wajdeczko
2021-06-24 15:41 ` Matthew Brost
2021-06-25 12:03 ` Michal Wajdeczko
2021-06-24 7:04 ` [Intel-gfx] [PATCH 04/47] drm/i915/guc: Add non blocking CTB send function Matthew Brost
2021-06-24 14:48 ` Michal Wajdeczko
2021-06-24 15:49 ` Matthew Brost
2021-06-24 17:02 ` Michal Wajdeczko
2021-06-24 22:41 ` Matthew Brost
2021-06-25 11:50 ` Michal Wajdeczko
2021-06-25 17:53 ` Matthew Brost
2021-06-24 22:47 ` Matthew Brost
2021-06-24 7:04 ` [Intel-gfx] [PATCH 05/47] drm/i915/guc: Add stall timer to " Matthew Brost
2021-06-24 17:37 ` Michal Wajdeczko
2021-06-24 23:01 ` Matthew Brost
2021-06-24 7:04 ` [Intel-gfx] [PATCH 06/47] drm/i915/guc: Optimize CTB writes and reads Matthew Brost
2021-06-25 13:09 ` Michal Wajdeczko
2021-06-25 18:26 ` Matthew Brost
2021-06-25 20:28 ` Matthew Brost
2021-06-24 7:04 ` [Intel-gfx] [PATCH 07/47] drm/i915/guc: Module load failure test for CT buffer creation Matthew Brost
2021-06-24 7:04 ` [Intel-gfx] [PATCH 08/47] drm/i915/guc: Add new GuC interface defines and structures Matthew Brost
2021-06-29 21:11 ` John Harrison
2021-06-30 0:30 ` Matthew Brost
2021-06-24 7:04 ` [Intel-gfx] [PATCH 09/47] drm/i915/guc: Remove GuC stage descriptor, add lrc descriptor Matthew Brost
2021-06-25 19:44 ` John Harrison
2021-06-24 7:04 ` [Intel-gfx] [PATCH 10/47] drm/i915/guc: Add lrc descriptor context lookup array Matthew Brost
2021-06-25 13:17 ` Michal Wajdeczko
2021-06-25 17:26 ` Matthew Brost
2021-06-29 21:20 ` John Harrison
2021-06-24 7:04 ` [Intel-gfx] [PATCH 11/47] drm/i915/guc: Implement GuC submission tasklet Matthew Brost
2021-06-29 22:04 ` John Harrison
2021-06-30 0:41 ` Matthew Brost
2021-06-24 7:04 ` [Intel-gfx] [PATCH 12/47] drm/i915/guc: Add bypass tasklet submission path to GuC Matthew Brost
2021-06-29 22:09 ` John Harrison
2021-06-24 7:04 ` [Intel-gfx] [PATCH 13/47] drm/i915/guc: Implement GuC context operations for new inteface Matthew Brost
2021-06-25 13:25 ` Michal Wajdeczko
2021-06-25 17:46 ` Matthew Brost
2021-06-24 7:04 ` [Intel-gfx] [PATCH 14/47] drm/i915/guc: Insert fence on context when deregistering Matthew Brost
2021-07-09 22:39 ` John Harrison
2021-06-24 7:04 ` [Intel-gfx] [PATCH 15/47] drm/i915/guc: Defer context unpin until scheduling is disabled Matthew Brost
2021-07-09 22:48 ` John Harrison
2021-06-24 7:04 ` [Intel-gfx] [PATCH 16/47] drm/i915/guc: Disable engine barriers with GuC during unpin Matthew Brost
2021-07-09 22:53 ` John Harrison
2021-07-10 3:00 ` Matthew Brost
2021-07-12 17:57 ` John Harrison
2021-07-12 18:11 ` Daniel Vetter
2021-06-24 7:04 ` [Intel-gfx] [PATCH 17/47] drm/i915/guc: Extend deregistration fence to schedule disable Matthew Brost
2021-07-09 22:59 ` John Harrison
2021-07-10 3:36 ` Matthew Brost
2021-07-12 17:54 ` John Harrison
2021-06-24 7:04 ` [Intel-gfx] [PATCH 18/47] drm/i915: Disable preempt busywait when using GuC scheduling Matthew Brost
2021-07-09 23:03 ` John Harrison
2021-06-24 7:04 ` [Intel-gfx] [PATCH 19/47] drm/i915/guc: Ensure request ordering via completion fences Matthew Brost
2021-07-15 1:51 ` Daniele Ceraolo Spurio
2021-06-24 7:04 ` [Intel-gfx] [PATCH 20/47] drm/i915/guc: Disable semaphores when using GuC scheduling Matthew Brost
2021-07-09 23:53 ` John Harrison
2021-07-15 0:07 ` Matthew Brost
2021-06-24 7:04 ` [Intel-gfx] [PATCH 21/47] drm/i915/guc: Ensure G2H response has space in buffer Matthew Brost
2021-07-13 18:36 ` John Harrison
2021-07-15 0:06 ` Matthew Brost
2021-07-15 0:12 ` John Harrison
2021-06-24 7:04 ` [Intel-gfx] [PATCH 22/47] drm/i915/guc: Update intel_gt_wait_for_idle to work with GuC Matthew Brost
2021-07-10 0:16 ` John Harrison
2021-07-10 3:55 ` Matthew Brost
2021-07-17 4:09 ` Matthew Brost
2021-06-24 7:04 ` [Intel-gfx] [PATCH 23/47] drm/i915/guc: Update GuC debugfs to support new GuC Matthew Brost
2021-07-12 18:05 ` John Harrison
2021-07-12 20:59 ` Matthew Brost
2021-07-12 21:37 ` John Harrison
2021-07-13 8:51 ` Michal Wajdeczko
2021-07-14 23:56 ` Matthew Brost
2021-06-24 7:04 ` [Intel-gfx] [PATCH 24/47] drm/i915/guc: Add several request trace points Matthew Brost
2021-07-12 18:08 ` John Harrison
2021-07-13 9:06 ` Tvrtko Ursulin
2021-07-20 1:59 ` Matthew Brost
2021-07-22 13:55 ` Tvrtko Ursulin
2021-06-24 7:04 ` [Intel-gfx] [PATCH 25/47] drm/i915: Add intel_context tracing Matthew Brost
2021-07-12 18:10 ` John Harrison
2021-07-12 21:47 ` Matthew Brost
2021-07-12 21:51 ` John Harrison
2021-06-24 7:04 ` [Intel-gfx] [PATCH 26/47] drm/i915/guc: GuC virtual engines Matthew Brost
2021-07-15 1:21 ` Daniele Ceraolo Spurio
2021-06-24 7:04 ` [Intel-gfx] [PATCH 27/47] drm/i915: Track 'serial' counts for " Matthew Brost
2021-07-12 18:11 ` John Harrison
2021-07-12 20:06 ` Matthew Brost
2021-06-24 7:04 ` [Intel-gfx] [PATCH 28/47] drm/i915: Hold reference to intel_context over life of i915_request Matthew Brost
2021-07-12 18:23 ` John Harrison
2021-07-12 20:05 ` Matthew Brost
2021-07-12 21:36 ` Matthew Brost
2021-07-12 21:48 ` John Harrison [this message]
2021-06-24 7:04 ` [Intel-gfx] [PATCH 29/47] drm/i915/guc: Disable bonding extension with GuC submission Matthew Brost
2021-07-12 18:23 ` John Harrison
2021-06-24 7:04 ` [Intel-gfx] [PATCH 30/47] drm/i915/guc: Direct all breadcrumbs for a class to single breadcrumbs Matthew Brost
2021-07-12 19:19 ` John Harrison
2021-06-24 7:05 ` [Intel-gfx] [PATCH 31/47] drm/i915/guc: Reset implementation for new GuC interface Matthew Brost
2021-07-12 19:58 ` John Harrison
2021-07-15 0:53 ` Matthew Brost
2021-07-15 9:36 ` Tvrtko Ursulin
2021-07-26 22:48 ` Matthew Brost
2021-07-27 8:56 ` Tvrtko Ursulin
2021-07-27 18:30 ` Matthew Brost
2021-06-24 7:05 ` [Intel-gfx] [PATCH 32/47] drm/i915: Reset GPU immediately if submission is disabled Matthew Brost
2021-07-12 20:01 ` John Harrison
2021-06-24 7:05 ` [Intel-gfx] [PATCH 33/47] drm/i915/guc: Add disable interrupts to guc sanitize Matthew Brost
2021-07-12 20:11 ` John Harrison
2021-06-24 7:05 ` [Intel-gfx] [PATCH 34/47] drm/i915/guc: Suspend/resume implementation for new interface Matthew Brost
2021-07-12 22:56 ` John Harrison
2021-06-24 7:05 ` [Intel-gfx] [PATCH 35/47] drm/i915/guc: Handle context reset notification Matthew Brost
2021-07-12 22:58 ` John Harrison
2021-07-15 0:32 ` Matthew Brost
2021-06-24 7:05 ` [Intel-gfx] [PATCH 36/47] drm/i915/guc: Handle engine reset failure notification Matthew Brost
2021-07-12 22:59 ` John Harrison
2021-06-24 7:05 ` [Intel-gfx] [PATCH 37/47] drm/i915/guc: Enable the timer expired interrupt for GuC Matthew Brost
2021-07-12 23:00 ` John Harrison
2021-06-24 7:05 ` [Intel-gfx] [PATCH 38/47] drm/i915/guc: Provide mmio list to be saved/restored on engine reset Matthew Brost
2021-06-24 7:05 ` [Intel-gfx] [PATCH 39/47] drm/i915/guc: Don't complain about reset races Matthew Brost
2021-06-24 15:55 ` Matthew Brost
2021-06-24 7:05 ` [Intel-gfx] [PATCH 40/47] drm/i915/guc: Enable GuC engine reset Matthew Brost
2021-06-24 16:19 ` Matthew Brost
2021-06-24 7:05 ` [Intel-gfx] [PATCH 41/47] drm/i915/guc: Capture error state on context reset Matthew Brost
2021-07-12 23:05 ` John Harrison
2021-06-24 7:05 ` [Intel-gfx] [PATCH 42/47] drm/i915/guc: Fix for error capture after full GPU reset with GuC Matthew Brost
2021-07-15 0:43 ` Matthew Brost
2021-06-24 7:05 ` [Intel-gfx] [PATCH 43/47] drm/i915/guc: Hook GuC scheduling policies up Matthew Brost
2021-06-25 0:59 ` Matthew Brost
2021-06-25 19:10 ` John Harrison
2021-07-10 18:56 ` Matthew Brost
2021-06-24 7:05 ` [Intel-gfx] [PATCH 44/47] drm/i915/guc: Connect reset modparam updates to GuC policy flags Matthew Brost
2021-06-25 1:10 ` Matthew Brost
2021-06-24 7:05 ` [Intel-gfx] [PATCH 45/47] drm/i915/guc: Include scheduling policies in the debugfs state dump Matthew Brost
2021-06-24 16:34 ` Matthew Brost
2021-06-24 7:05 ` [Intel-gfx] [PATCH 46/47] drm/i915/guc: Add golden context to GuC ADS Matthew Brost
2021-06-24 7:05 ` [Intel-gfx] [PATCH 47/47] drm/i915/guc: Unblock GuC submission on Gen11+ Matthew Brost
2021-06-30 8:22 ` Martin Peres
2021-06-30 18:00 ` Matthew Brost
2021-07-01 18:24 ` Martin Peres
2021-07-02 8:13 ` Martin Peres
2021-07-02 13:06 ` Michal Wajdeczko
2021-07-02 13:12 ` Martin Peres
2021-07-02 14:08 ` Michal Wajdeczko
2021-06-30 18:58 ` John Harrison
2021-07-01 8:14 ` Pekka Paalanen
2021-07-01 18:27 ` Martin Peres
2021-07-01 19:28 ` Daniel Vetter
2021-07-02 7:29 ` Pekka Paalanen
2021-07-02 8:09 ` Martin Peres
2021-07-02 15:07 ` Michal Wajdeczko
2021-07-03 8:21 ` Martin Peres
2021-07-07 0:57 ` John Harrison
2021-07-07 7:47 ` Pekka Paalanen
2021-07-07 10:11 ` Michal Wajdeczko
2021-07-15 0:49 ` Matthew Brost
2021-06-24 7:17 ` [Intel-gfx] ✗ Fi.CI.CHECKPATCH: warning for GuC submission support Patchwork
2021-06-24 7:19 ` [Intel-gfx] ✗ Fi.CI.SPARSE: " Patchwork
2021-06-24 7:47 ` [Intel-gfx] ✗ Fi.CI.BAT: failure " Patchwork
2021-07-12 19:23 ` [Intel-gfx] ✗ Fi.CI.BUILD: failure for GuC submission support (rev2) Patchwork
2021-10-22 9:35 ` [Intel-gfx] [PATCH 00/47] GuC submission support Joonas Lahtinen
2021-10-22 16:42 ` Matthew Brost
2021-10-25 9:37 ` Joonas Lahtinen
2021-10-25 15:15 ` Matthew Brost
2021-10-26 8:59 ` Joonas Lahtinen
2021-10-26 15:43 ` Matthew Brost
2021-10-26 15:51 ` Matthew Brost
2021-10-27 9:21 ` Joonas Lahtinen
2021-10-25 17:06 ` John Harrison
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=f343b084-faf9-71f7-e4f0-9df42c330286@intel.com \
--to=john.c.harrison@intel.com \
--cc=dri-devel@lists.freedesktop.org \
--cc=intel-gfx@lists.freedesktop.org \
--cc=matthew.brost@intel.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).