[Intel-gfx] [PATCH] drm/i915/gt: Incrementally check for rewinding
Mika Kuoppala
mika.kuoppala at linux.intel.com
Wed Jun 10 12:39:34 UTC 2020
Chris Wilson <chris at chris-wilson.co.uk> writes:
> In commit 5ba32c7be81e ("drm/i915/execlists: Always force a context
> reload when rewinding RING_TAIL"), we placed the check for rewinding a
> context on actually submitting the next request in that context. This
> was so that we only had to check once, and could do so with precision
> avoiding as many forced restores as possible. For example, to ensure
> that we can resubmit the same request a couple of times, we include a
> small wa_tail such that on the next submission, the ring->tail will
> appear to move forwards when resubmitting the same request. This is very
> common as it will happen for every lite-restore to fill the second port
> after a context switch.
>
> However, intel_ring_direction() is limited in precision to movements of
> upto half the ring size. The consequence being that if we tried to
> unwind many requests, we could exceed half the ring and flip the sense
> of the direction, so missing a force restore. As no request can be
> greater than half the ring (i.e. 2048 bytes in the smallest case), we
> can check for rollback incrementally. As we check against the tail that
> would be submitted, we do not lose any sensitivity and allow lite
> restores for the simple case. We still need to double check upon
> submitting the context, to allow for multiple preemptions and
> resubmissions.
>
> Fixes: 5ba32c7be81e ("drm/i915/execlists: Always force a context reload when rewinding RING_TAIL")
> Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
> Cc: Mika Kuoppala <mika.kuoppala at linux.intel.com>
> Cc: <stable at vger.kernel.org> # v5.4+
> ---
> drivers/gpu/drm/i915/gt/intel_engine_cs.c | 4 +-
> drivers/gpu/drm/i915/gt/intel_lrc.c | 21 +++-
> drivers/gpu/drm/i915/gt/intel_ring.c | 4 +
> drivers/gpu/drm/i915/gt/selftest_mocs.c | 18 ++-
> drivers/gpu/drm/i915/gt/selftest_ring.c | 110 ++++++++++++++++++
> .../drm/i915/selftests/i915_mock_selftests.h | 1 +
> 6 files changed, 154 insertions(+), 4 deletions(-)
> create mode 100644 drivers/gpu/drm/i915/gt/selftest_ring.c
>
> diff --git a/drivers/gpu/drm/i915/gt/intel_engine_cs.c b/drivers/gpu/drm/i915/gt/intel_engine_cs.c
> index e5141a897786..0a05301e00fb 100644
> --- a/drivers/gpu/drm/i915/gt/intel_engine_cs.c
> +++ b/drivers/gpu/drm/i915/gt/intel_engine_cs.c
> @@ -646,7 +646,7 @@ static int engine_setup_common(struct intel_engine_cs *engine)
> struct measure_breadcrumb {
> struct i915_request rq;
> struct intel_ring ring;
> - u32 cs[1024];
> + u32 cs[2048];
> };
>
> static int measure_breadcrumb_dw(struct intel_context *ce)
> @@ -667,6 +667,8 @@ static int measure_breadcrumb_dw(struct intel_context *ce)
>
> frame->ring.vaddr = frame->cs;
> frame->ring.size = sizeof(frame->cs);
> + frame->ring.wrap =
> + BITS_PER_TYPE(frame->ring.size) - ilog2(frame->ring.size);
> frame->ring.effective_size = frame->ring.size;
> intel_ring_update_space(&frame->ring);
> frame->rq.ring = &frame->ring;
> diff --git a/drivers/gpu/drm/i915/gt/intel_lrc.c b/drivers/gpu/drm/i915/gt/intel_lrc.c
> index a057f7a2a521..5f33342c15e2 100644
> --- a/drivers/gpu/drm/i915/gt/intel_lrc.c
> +++ b/drivers/gpu/drm/i915/gt/intel_lrc.c
> @@ -1137,6 +1137,13 @@ __unwind_incomplete_requests(struct intel_engine_cs *engine)
> list_move(&rq->sched.link, pl);
> set_bit(I915_FENCE_FLAG_PQUEUE, &rq->fence.flags);
>
> + /* Check in case rollback so far, we wrap [size/2] */
This could be ammended a little as why it is not always the case that
on the rewind the direction is not positive.
> + if (intel_ring_direction(rq->ring,
> + intel_ring_wrap(rq->ring,
> + rq->tail),
> + rq->ring->tail) > 0)
> + rq->context->lrc.desc |= CTX_DESC_FORCE_RESTORE;
> +
> active = rq;
> } else {
> struct intel_engine_cs *owner = rq->context->engine;
> @@ -1505,8 +1512,9 @@ static u64 execlists_update_context(struct i915_request *rq)
> * HW has a tendency to ignore us rewinding the TAIL to the end of
> * an earlier request.
> */
> + GEM_BUG_ON(ce->lrc_reg_state[CTX_RING_TAIL] != rq->ring->tail);
> + prev = rq->ring->tail;
> tail = intel_ring_set_tail(rq->ring, rq->tail);
> - prev = ce->lrc_reg_state[CTX_RING_TAIL];
> if (unlikely(intel_ring_direction(rq->ring, tail, prev) <= 0))
> desc |= CTX_DESC_FORCE_RESTORE;
> ce->lrc_reg_state[CTX_RING_TAIL] = tail;
> @@ -4758,6 +4766,14 @@ static int gen12_emit_flush(struct i915_request *request, u32 mode)
> return 0;
> }
>
> +static void assert_request_valid(struct i915_request *rq)
> +{
> + struct intel_ring *ring __maybe_unused = rq->ring;
> +
> + /* Can we unwind this request without appearing to go forwards? */
> + GEM_BUG_ON(intel_ring_direction(ring, rq->wa_tail, rq->head) <= 0);
Chris explained in irc that as the wa_tail is reserved for next
resubmit, the nondirection is also possible.
> +}
> +
> /*
> * Reserve space for 2 NOOPs at the end of each request to be
> * used as a workaround for not being allowed to do lite
> @@ -4770,6 +4786,9 @@ static u32 *gen8_emit_wa_tail(struct i915_request *request, u32 *cs)
> *cs++ = MI_NOOP;
> request->wa_tail = intel_ring_offset(request, cs);
>
> + /* Check that entire request is less than half the ring */
> + assert_request_valid(request);
I was thinking about adding the check in the advance part but
that is too early. And also the tail validation is too early.
This is so tricky with the wrap handling. But it is easier to
stand behind the broad shoulders of the really appreciated
selftests.
Reviewed-by: Mika Kuoppala <mika.kuoppala at linux.intel.com>
> +
> return cs;
> }
>
> diff --git a/drivers/gpu/drm/i915/gt/intel_ring.c b/drivers/gpu/drm/i915/gt/intel_ring.c
> index 8cda1b7e17ba..bdb324167ef3 100644
> --- a/drivers/gpu/drm/i915/gt/intel_ring.c
> +++ b/drivers/gpu/drm/i915/gt/intel_ring.c
> @@ -315,3 +315,7 @@ int intel_ring_cacheline_align(struct i915_request *rq)
> GEM_BUG_ON(rq->ring->emit & (CACHELINE_BYTES - 1));
> return 0;
> }
> +
> +#if IS_ENABLED(CONFIG_DRM_I915_SELFTEST)
> +#include "selftest_ring.c"
> +#endif
> diff --git a/drivers/gpu/drm/i915/gt/selftest_mocs.c b/drivers/gpu/drm/i915/gt/selftest_mocs.c
> index 7bae64018ad9..b25eba50c88e 100644
> --- a/drivers/gpu/drm/i915/gt/selftest_mocs.c
> +++ b/drivers/gpu/drm/i915/gt/selftest_mocs.c
> @@ -18,6 +18,20 @@ struct live_mocs {
> void *vaddr;
> };
>
> +static struct intel_context *mocs_context_create(struct intel_engine_cs *engine)
> +{
> + struct intel_context *ce;
> +
> + ce = intel_context_create(engine);
> + if (IS_ERR(ce))
> + return ce;
> +
> + /* We build large requests to read the registers from the ring */
> + ce->ring = __intel_context_ring_size(SZ_16K);
> +
> + return ce;
> +}
> +
> static int request_add_sync(struct i915_request *rq, int err)
> {
> i915_request_get(rq);
> @@ -301,7 +315,7 @@ static int live_mocs_clean(void *arg)
> for_each_engine(engine, gt, id) {
> struct intel_context *ce;
>
> - ce = intel_context_create(engine);
> + ce = mocs_context_create(engine);
> if (IS_ERR(ce)) {
> err = PTR_ERR(ce);
> break;
> @@ -395,7 +409,7 @@ static int live_mocs_reset(void *arg)
> for_each_engine(engine, gt, id) {
> struct intel_context *ce;
>
> - ce = intel_context_create(engine);
> + ce = mocs_context_create(engine);
> if (IS_ERR(ce)) {
> err = PTR_ERR(ce);
> break;
> diff --git a/drivers/gpu/drm/i915/gt/selftest_ring.c b/drivers/gpu/drm/i915/gt/selftest_ring.c
> new file mode 100644
> index 000000000000..2a8c534dc125
> --- /dev/null
> +++ b/drivers/gpu/drm/i915/gt/selftest_ring.c
> @@ -0,0 +1,110 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * Copyright © 2020 Intel Corporation
> + */
> +
> +static struct intel_ring *mock_ring(unsigned long sz)
> +{
> + struct intel_ring *ring;
> +
> + ring = kzalloc(sizeof(*ring) + sz, GFP_KERNEL);
> + if (!ring)
> + return NULL;
> +
> + kref_init(&ring->ref);
> + ring->size = sz;
> + ring->wrap = BITS_PER_TYPE(ring->size) - ilog2(sz);
> + ring->effective_size = sz;
> + ring->vaddr = (void *)(ring + 1);
> + atomic_set(&ring->pin_count, 1);
> +
> + intel_ring_update_space(ring);
> +
> + return ring;
> +}
> +
> +static void mock_ring_free(struct intel_ring *ring)
> +{
> + kfree(ring);
> +}
> +
> +static int check_ring_direction(struct intel_ring *ring,
> + u32 next, u32 prev,
> + int expected)
> +{
> + int result;
> +
> + result = intel_ring_direction(ring, next, prev);
> + if (result < 0)
> + result = -1;
> + else if (result > 0)
> + result = 1;
> +
> + if (result != expected) {
> + pr_err("intel_ring_direction(%u, %u):%d != %d\n",
> + next, prev, result, expected);
> + return -EINVAL;
> + }
> +
> + return 0;
> +}
> +
> +static int check_ring_step(struct intel_ring *ring, u32 x, u32 step)
> +{
> + u32 prev = x, next = intel_ring_wrap(ring, x + step);
> + int err = 0;
> +
> + err |= check_ring_direction(ring, next, next, 0);
> + err |= check_ring_direction(ring, prev, prev, 0);
> + err |= check_ring_direction(ring, next, prev, 1);
> + err |= check_ring_direction(ring, prev, next, -1);
> +
> + return err;
> +}
> +
> +static int check_ring_offset(struct intel_ring *ring, u32 x, u32 step)
> +{
> + int err = 0;
> +
> + err |= check_ring_step(ring, x, step);
> + err |= check_ring_step(ring, intel_ring_wrap(ring, x + 1), step);
> + err |= check_ring_step(ring, intel_ring_wrap(ring, x - 1), step);
> +
> + return err;
> +}
> +
> +static int igt_ring_direction(void *dummy)
> +{
> + struct intel_ring *ring;
> + unsigned int half = 2048;
> + int step, err = 0;
> +
> + ring = mock_ring(2 * half);
> + if (!ring)
> + return -ENOMEM;
> +
> + GEM_BUG_ON(ring->size != 2 * half);
> +
> + /* Precision of wrap detection is limited to ring->size / 2 */
> + for (step = 1; step < half; step <<= 1) {
> + err |= check_ring_offset(ring, 0, step);
> + err |= check_ring_offset(ring, half, step);
> + }
> + err |= check_ring_step(ring, 0, half - 64);
> +
> + /* And check unwrapped handling for good measure */
> + err |= check_ring_offset(ring, 0, 2 * half + 64);
> + err |= check_ring_offset(ring, 3 * half, 1);
> +
> + mock_ring_free(ring);
> + return err;
> +}
> +
> +int intel_ring_mock_selftests(void)
> +{
> + static const struct i915_subtest tests[] = {
> + SUBTEST(igt_ring_direction),
> + };
> +
> + return i915_subtests(tests, NULL);
> +}
> diff --git a/drivers/gpu/drm/i915/selftests/i915_mock_selftests.h b/drivers/gpu/drm/i915/selftests/i915_mock_selftests.h
> index 1929feba4e8e..3db34d3eea58 100644
> --- a/drivers/gpu/drm/i915/selftests/i915_mock_selftests.h
> +++ b/drivers/gpu/drm/i915/selftests/i915_mock_selftests.h
> @@ -21,6 +21,7 @@ selftest(fence, i915_sw_fence_mock_selftests)
> selftest(scatterlist, scatterlist_mock_selftests)
> selftest(syncmap, i915_syncmap_mock_selftests)
> selftest(uncore, intel_uncore_mock_selftests)
> +selftest(ring, intel_ring_mock_selftests)
> selftest(engine, intel_engine_cs_mock_selftests)
> selftest(timelines, intel_timeline_mock_selftests)
> selftest(requests, i915_request_mock_selftests)
> --
> 2.20.1
More information about the Intel-gfx
mailing list