linux_dsm_epyc7002/drivers/gpu/drm/i915/selftests/i915_timeline.c

735 lines
17 KiB
C
Raw Normal View History

drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
/*
* SPDX-License-Identifier: MIT
drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
*
* Copyright © 2017-2018 Intel Corporation
drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
*/
#include <linux/prime_numbers.h>
drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
#include "../i915_selftest.h"
#include "i915_random.h"
#include "igt_flush_test.h"
drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
#include "mock_gem_device.h"
#include "mock_timeline.h"
static struct page *hwsp_page(struct i915_timeline *tl)
{
struct drm_i915_gem_object *obj = tl->hwsp_ggtt->obj;
GEM_BUG_ON(!i915_gem_object_has_pinned_pages(obj));
return sg_page(obj->mm.pages->sgl);
}
static unsigned long hwsp_cacheline(struct i915_timeline *tl)
{
unsigned long address = (unsigned long)page_address(hwsp_page(tl));
return (address + tl->hwsp_offset) / CACHELINE_BYTES;
}
#define CACHELINES_PER_PAGE (PAGE_SIZE / CACHELINE_BYTES)
struct mock_hwsp_freelist {
struct drm_i915_private *i915;
struct radix_tree_root cachelines;
struct i915_timeline **history;
unsigned long count, max;
struct rnd_state prng;
};
enum {
SHUFFLE = BIT(0),
};
static void __mock_hwsp_record(struct mock_hwsp_freelist *state,
unsigned int idx,
struct i915_timeline *tl)
{
tl = xchg(&state->history[idx], tl);
if (tl) {
radix_tree_delete(&state->cachelines, hwsp_cacheline(tl));
i915_timeline_put(tl);
}
}
static int __mock_hwsp_timeline(struct mock_hwsp_freelist *state,
unsigned int count,
unsigned int flags)
{
struct i915_timeline *tl;
unsigned int idx;
while (count--) {
unsigned long cacheline;
int err;
tl = i915_timeline_create(state->i915, "mock", NULL);
if (IS_ERR(tl))
return PTR_ERR(tl);
cacheline = hwsp_cacheline(tl);
err = radix_tree_insert(&state->cachelines, cacheline, tl);
if (err) {
if (err == -EEXIST) {
pr_err("HWSP cacheline %lu already used; duplicate allocation!\n",
cacheline);
}
i915_timeline_put(tl);
return err;
}
idx = state->count++ % state->max;
__mock_hwsp_record(state, idx, tl);
}
if (flags & SHUFFLE)
i915_prandom_shuffle(state->history,
sizeof(*state->history),
min(state->count, state->max),
&state->prng);
count = i915_prandom_u32_max_state(min(state->count, state->max),
&state->prng);
while (count--) {
idx = --state->count % state->max;
__mock_hwsp_record(state, idx, NULL);
}
return 0;
}
static int mock_hwsp_freelist(void *arg)
{
struct mock_hwsp_freelist state;
const struct {
const char *name;
unsigned int flags;
} phases[] = {
{ "linear", 0 },
{ "shuffled", SHUFFLE },
{ },
}, *p;
unsigned int na;
int err = 0;
INIT_RADIX_TREE(&state.cachelines, GFP_KERNEL);
state.prng = I915_RND_STATE_INITIALIZER(i915_selftest.random_seed);
state.i915 = mock_gem_device();
if (!state.i915)
return -ENOMEM;
/*
* Create a bunch of timelines and check that their HWSP do not overlap.
* Free some, and try again.
*/
state.max = PAGE_SIZE / sizeof(*state.history);
state.count = 0;
state.history = kcalloc(state.max, sizeof(*state.history), GFP_KERNEL);
if (!state.history) {
err = -ENOMEM;
goto err_put;
}
mutex_lock(&state.i915->drm.struct_mutex);
for (p = phases; p->name; p++) {
pr_debug("%s(%s)\n", __func__, p->name);
for_each_prime_number_from(na, 1, 2 * CACHELINES_PER_PAGE) {
err = __mock_hwsp_timeline(&state, na, p->flags);
if (err)
goto out;
}
}
out:
for (na = 0; na < state.max; na++)
__mock_hwsp_record(&state, na, NULL);
mutex_unlock(&state.i915->drm.struct_mutex);
kfree(state.history);
err_put:
drm_dev_put(&state.i915->drm);
return err;
}
drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
struct __igt_sync {
const char *name;
u32 seqno;
bool expected;
bool set;
};
static int __igt_sync(struct i915_timeline *tl,
drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
u64 ctx,
const struct __igt_sync *p,
const char *name)
{
int ret;
if (__i915_timeline_sync_is_later(tl, ctx, p->seqno) != p->expected) {
drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
pr_err("%s: %s(ctx=%llu, seqno=%u) expected passed %s but failed\n",
name, p->name, ctx, p->seqno, yesno(p->expected));
return -EINVAL;
}
if (p->set) {
ret = __i915_timeline_sync_set(tl, ctx, p->seqno);
drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
if (ret)
return ret;
}
return 0;
}
static int igt_sync(void *arg)
{
const struct __igt_sync pass[] = {
{ "unset", 0, false, false },
{ "new", 0, false, true },
{ "0a", 0, true, true },
{ "1a", 1, false, true },
{ "1b", 1, true, true },
{ "0b", 0, true, false },
{ "2a", 2, false, true },
{ "4", 4, false, true },
{ "INT_MAX", INT_MAX, false, true },
{ "INT_MAX-1", INT_MAX-1, true, false },
{ "INT_MAX+1", (u32)INT_MAX+1, false, true },
{ "INT_MAX", INT_MAX, true, false },
{ "UINT_MAX", UINT_MAX, false, true },
{ "wrap", 0, false, true },
{ "unwrap", UINT_MAX, true, false },
{},
}, *p;
struct i915_timeline tl;
drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
int order, offset;
int ret = -ENODEV;
drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
mock_timeline_init(&tl, 0);
drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
for (p = pass; p->name; p++) {
for (order = 1; order < 64; order++) {
for (offset = -1; offset <= (order > 1); offset++) {
u64 ctx = BIT_ULL(order) + offset;
ret = __igt_sync(&tl, ctx, p, "1");
drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
if (ret)
goto out;
}
}
}
mock_timeline_fini(&tl);
drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
mock_timeline_init(&tl, 0);
drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
for (order = 1; order < 64; order++) {
for (offset = -1; offset <= (order > 1); offset++) {
u64 ctx = BIT_ULL(order) + offset;
for (p = pass; p->name; p++) {
ret = __igt_sync(&tl, ctx, p, "2");
drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
if (ret)
goto out;
}
}
}
out:
mock_timeline_fini(&tl);
drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
return ret;
}
static unsigned int random_engine(struct rnd_state *rnd)
{
return i915_prandom_u32_max_state(I915_NUM_ENGINES, rnd);
drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
}
static int bench_sync(void *arg)
{
struct rnd_state prng;
struct i915_timeline tl;
drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
unsigned long end_time, count;
u64 prng32_1M;
ktime_t kt;
int order, last_order;
mock_timeline_init(&tl, 0);
drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
/* Lookups from cache are very fast and so the random number generation
* and the loop itself becomes a significant factor in the per-iteration
* timings. We try to compensate the results by measuring the overhead
* of the prng and subtract it from the reported results.
*/
prandom_seed_state(&prng, i915_selftest.random_seed);
count = 0;
kt = ktime_get();
end_time = jiffies + HZ/10;
do {
u32 x;
/* Make sure the compiler doesn't optimise away the prng call */
WRITE_ONCE(x, prandom_u32_state(&prng));
count++;
} while (!time_after(jiffies, end_time));
kt = ktime_sub(ktime_get(), kt);
pr_debug("%s: %lu random evaluations, %lluns/prng\n",
__func__, count, (long long)div64_ul(ktime_to_ns(kt), count));
prng32_1M = div64_ul(ktime_to_ns(kt) << 20, count);
drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
/* Benchmark (only) setting random context ids */
prandom_seed_state(&prng, i915_selftest.random_seed);
count = 0;
kt = ktime_get();
end_time = jiffies + HZ/10;
do {
u64 id = i915_prandom_u64_state(&prng);
__i915_timeline_sync_set(&tl, id, 0);
drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
count++;
} while (!time_after(jiffies, end_time));
kt = ktime_sub(ktime_get(), kt);
kt = ktime_sub_ns(kt, (count * prng32_1M * 2) >> 20);
drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
pr_info("%s: %lu random insertions, %lluns/insert\n",
__func__, count, (long long)div64_ul(ktime_to_ns(kt), count));
/* Benchmark looking up the exact same context ids as we just set */
prandom_seed_state(&prng, i915_selftest.random_seed);
end_time = count;
kt = ktime_get();
while (end_time--) {
u64 id = i915_prandom_u64_state(&prng);
if (!__i915_timeline_sync_is_later(&tl, id, 0)) {
mock_timeline_fini(&tl);
drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
pr_err("Lookup of %llu failed\n", id);
return -EINVAL;
}
}
kt = ktime_sub(ktime_get(), kt);
kt = ktime_sub_ns(kt, (count * prng32_1M * 2) >> 20);
drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
pr_info("%s: %lu random lookups, %lluns/lookup\n",
__func__, count, (long long)div64_ul(ktime_to_ns(kt), count));
mock_timeline_fini(&tl);
drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
cond_resched();
mock_timeline_init(&tl, 0);
drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
/* Benchmark setting the first N (in order) contexts */
count = 0;
kt = ktime_get();
end_time = jiffies + HZ/10;
do {
__i915_timeline_sync_set(&tl, count++, 0);
drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
} while (!time_after(jiffies, end_time));
kt = ktime_sub(ktime_get(), kt);
pr_info("%s: %lu in-order insertions, %lluns/insert\n",
__func__, count, (long long)div64_ul(ktime_to_ns(kt), count));
/* Benchmark looking up the exact same context ids as we just set */
end_time = count;
kt = ktime_get();
while (end_time--) {
if (!__i915_timeline_sync_is_later(&tl, end_time, 0)) {
drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
pr_err("Lookup of %lu failed\n", end_time);
mock_timeline_fini(&tl);
drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
return -EINVAL;
}
}
kt = ktime_sub(ktime_get(), kt);
pr_info("%s: %lu in-order lookups, %lluns/lookup\n",
__func__, count, (long long)div64_ul(ktime_to_ns(kt), count));
mock_timeline_fini(&tl);
drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
cond_resched();
mock_timeline_init(&tl, 0);
drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
/* Benchmark searching for a random context id and maybe changing it */
prandom_seed_state(&prng, i915_selftest.random_seed);
count = 0;
kt = ktime_get();
end_time = jiffies + HZ/10;
do {
u32 id = random_engine(&prng);
u32 seqno = prandom_u32_state(&prng);
if (!__i915_timeline_sync_is_later(&tl, id, seqno))
__i915_timeline_sync_set(&tl, id, seqno);
drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
count++;
} while (!time_after(jiffies, end_time));
kt = ktime_sub(ktime_get(), kt);
kt = ktime_sub_ns(kt, (count * prng32_1M * 2) >> 20);
drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
pr_info("%s: %lu repeated insert/lookups, %lluns/op\n",
__func__, count, (long long)div64_ul(ktime_to_ns(kt), count));
mock_timeline_fini(&tl);
drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
cond_resched();
/* Benchmark searching for a known context id and changing the seqno */
for (last_order = 1, order = 1; order < 32;
({ int tmp = last_order; last_order = order; order += tmp; })) {
unsigned int mask = BIT(order) - 1;
mock_timeline_init(&tl, 0);
drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
count = 0;
kt = ktime_get();
end_time = jiffies + HZ/10;
do {
/* Without assuming too many details of the underlying
* implementation, try to identify its phase-changes
* (if any)!
*/
u64 id = (u64)(count & mask) << order;
__i915_timeline_sync_is_later(&tl, id, 0);
__i915_timeline_sync_set(&tl, id, 0);
drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
count++;
} while (!time_after(jiffies, end_time));
kt = ktime_sub(ktime_get(), kt);
pr_info("%s: %lu cyclic/%d insert/lookups, %lluns/op\n",
__func__, count, order,
(long long)div64_ul(ktime_to_ns(kt), count));
mock_timeline_fini(&tl);
drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
cond_resched();
}
return 0;
}
int i915_timeline_mock_selftests(void)
drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
{
static const struct i915_subtest tests[] = {
SUBTEST(mock_hwsp_freelist),
drm/i915: Squash repeated awaits on the same fence Track the latest fence waited upon on each context, and only add a new asynchronous wait if the new fence is more recent than the recorded fence for that context. This requires us to filter out unordered timelines, which are noted by DMA_FENCE_NO_CONTEXT. However, in the absence of a universal identifier, we have to use our own i915->mm.unordered_timeline token. v2: Throw around the debug crutches v3: Inline the likely case of the pre-allocation cache being full. v4: Drop the pre-allocation support, we can lose the most recent fence in case of allocation failure -- it just means we may emit more awaits than strictly necessary but will not break. v5: Trim allocation size for leaf nodes, they only need an array of u32 not pointers. v6: Create mock_timeline to tidy selftest writing v7: s/intel_timeline_sync_get/intel_timeline_sync_is_later/ (Tvrtko) v8: Prune the stale sync points when we idle. v9: Include a small benchmark in the kselftests v10: Separate the idr implementation into its own compartment. (Tvrkto) v11: Refactor igt_sync kselftests to avoid deep nesting (Tvrkto) v12: __sync_leaf_idx() to assert that p->height is 0 when checking leaves v13: kselftests to investigate struct i915_syncmap itself (Tvrtko) v14: Foray into ascii art graphs v15: Take into account that the random lookup/insert does 2 prng calls, not 1, when benchmarking, and use for_each_set_bit() (Tvrtko) v16: Improved ascii art Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170503093924.5320-4-chris@chris-wilson.co.uk
2017-05-03 16:39:21 +07:00
SUBTEST(igt_sync),
SUBTEST(bench_sync),
};
return i915_subtests(tests, NULL);
}
static int emit_ggtt_store_dw(struct i915_request *rq, u32 addr, u32 value)
{
u32 *cs;
cs = intel_ring_begin(rq, 4);
if (IS_ERR(cs))
return PTR_ERR(cs);
if (INTEL_GEN(rq->i915) >= 8) {
*cs++ = MI_STORE_DWORD_IMM_GEN4 | MI_USE_GGTT;
*cs++ = addr;
*cs++ = 0;
*cs++ = value;
} else if (INTEL_GEN(rq->i915) >= 4) {
*cs++ = MI_STORE_DWORD_IMM_GEN4 | MI_USE_GGTT;
*cs++ = 0;
*cs++ = addr;
*cs++ = value;
} else {
*cs++ = MI_STORE_DWORD_IMM | MI_MEM_VIRTUAL;
*cs++ = addr;
*cs++ = value;
*cs++ = MI_NOOP;
}
intel_ring_advance(rq, cs);
return 0;
}
static u32 hwsp_address(const struct i915_timeline *tl)
{
return i915_ggtt_offset(tl->hwsp_ggtt) + tl->hwsp_offset;
}
static struct i915_request *
tl_write(struct i915_timeline *tl, struct intel_engine_cs *engine, u32 value)
{
struct i915_request *rq;
int err;
lockdep_assert_held(&tl->i915->drm.struct_mutex); /* lazy rq refs */
err = i915_timeline_pin(tl);
if (err) {
rq = ERR_PTR(err);
goto out;
}
rq = i915_request_alloc(engine, engine->i915->kernel_context);
if (IS_ERR(rq))
goto out_unpin;
err = emit_ggtt_store_dw(rq, hwsp_address(tl), value);
i915_request_add(rq);
if (err)
rq = ERR_PTR(err);
out_unpin:
i915_timeline_unpin(tl);
out:
if (IS_ERR(rq))
pr_err("Failed to write to timeline!\n");
return rq;
}
static struct i915_timeline *
checked_i915_timeline_create(struct drm_i915_private *i915)
{
struct i915_timeline *tl;
tl = i915_timeline_create(i915, "live", NULL);
if (IS_ERR(tl))
return tl;
if (*tl->hwsp_seqno != tl->seqno) {
pr_err("Timeline created with incorrect breadcrumb, found %x, expected %x\n",
*tl->hwsp_seqno, tl->seqno);
i915_timeline_put(tl);
return ERR_PTR(-EINVAL);
}
return tl;
}
static int live_hwsp_engine(void *arg)
{
#define NUM_TIMELINES 4096
struct drm_i915_private *i915 = arg;
struct i915_timeline **timelines;
struct intel_engine_cs *engine;
enum intel_engine_id id;
intel_wakeref_t wakeref;
unsigned long count, n;
int err = 0;
/*
* Create a bunch of timelines and check we can write
* independently to each of their breadcrumb slots.
*/
timelines = kvmalloc_array(NUM_TIMELINES * I915_NUM_ENGINES,
sizeof(*timelines),
GFP_KERNEL);
if (!timelines)
return -ENOMEM;
mutex_lock(&i915->drm.struct_mutex);
wakeref = intel_runtime_pm_get(i915);
count = 0;
for_each_engine(engine, i915, id) {
if (!intel_engine_can_store_dword(engine))
continue;
for (n = 0; n < NUM_TIMELINES; n++) {
struct i915_timeline *tl;
struct i915_request *rq;
tl = checked_i915_timeline_create(i915);
if (IS_ERR(tl)) {
err = PTR_ERR(tl);
goto out;
}
rq = tl_write(tl, engine, count);
if (IS_ERR(rq)) {
i915_timeline_put(tl);
err = PTR_ERR(rq);
goto out;
}
timelines[count++] = tl;
}
}
out:
if (igt_flush_test(i915, I915_WAIT_LOCKED))
err = -EIO;
for (n = 0; n < count; n++) {
struct i915_timeline *tl = timelines[n];
if (!err && *tl->hwsp_seqno != n) {
pr_err("Invalid seqno stored in timeline %lu, found 0x%x\n",
n, *tl->hwsp_seqno);
err = -EINVAL;
}
i915_timeline_put(tl);
}
intel_runtime_pm_put(i915, wakeref);
mutex_unlock(&i915->drm.struct_mutex);
kvfree(timelines);
return err;
#undef NUM_TIMELINES
}
static int live_hwsp_alternate(void *arg)
{
#define NUM_TIMELINES 4096
struct drm_i915_private *i915 = arg;
struct i915_timeline **timelines;
struct intel_engine_cs *engine;
enum intel_engine_id id;
intel_wakeref_t wakeref;
unsigned long count, n;
int err = 0;
/*
* Create a bunch of timelines and check we can write
* independently to each of their breadcrumb slots with adjacent
* engines.
*/
timelines = kvmalloc_array(NUM_TIMELINES * I915_NUM_ENGINES,
sizeof(*timelines),
GFP_KERNEL);
if (!timelines)
return -ENOMEM;
mutex_lock(&i915->drm.struct_mutex);
wakeref = intel_runtime_pm_get(i915);
count = 0;
for (n = 0; n < NUM_TIMELINES; n++) {
for_each_engine(engine, i915, id) {
struct i915_timeline *tl;
struct i915_request *rq;
if (!intel_engine_can_store_dword(engine))
continue;
tl = checked_i915_timeline_create(i915);
if (IS_ERR(tl)) {
err = PTR_ERR(tl);
goto out;
}
rq = tl_write(tl, engine, count);
if (IS_ERR(rq)) {
i915_timeline_put(tl);
err = PTR_ERR(rq);
goto out;
}
timelines[count++] = tl;
}
}
out:
if (igt_flush_test(i915, I915_WAIT_LOCKED))
err = -EIO;
for (n = 0; n < count; n++) {
struct i915_timeline *tl = timelines[n];
if (!err && *tl->hwsp_seqno != n) {
pr_err("Invalid seqno stored in timeline %lu, found 0x%x\n",
n, *tl->hwsp_seqno);
err = -EINVAL;
}
i915_timeline_put(tl);
}
intel_runtime_pm_put(i915, wakeref);
mutex_unlock(&i915->drm.struct_mutex);
kvfree(timelines);
return err;
#undef NUM_TIMELINES
}
static int live_hwsp_recycle(void *arg)
{
struct drm_i915_private *i915 = arg;
struct intel_engine_cs *engine;
enum intel_engine_id id;
intel_wakeref_t wakeref;
unsigned long count;
int err = 0;
/*
* Check seqno writes into one timeline at a time. We expect to
* recycle the breadcrumb slot between iterations and neither
* want to confuse ourselves or the GPU.
*/
mutex_lock(&i915->drm.struct_mutex);
wakeref = intel_runtime_pm_get(i915);
count = 0;
for_each_engine(engine, i915, id) {
IGT_TIMEOUT(end_time);
if (!intel_engine_can_store_dword(engine))
continue;
do {
struct i915_timeline *tl;
struct i915_request *rq;
tl = checked_i915_timeline_create(i915);
if (IS_ERR(tl)) {
err = PTR_ERR(tl);
goto out;
}
rq = tl_write(tl, engine, count);
if (IS_ERR(rq)) {
i915_timeline_put(tl);
err = PTR_ERR(rq);
goto out;
}
if (i915_request_wait(rq,
I915_WAIT_LOCKED,
HZ / 5) < 0) {
pr_err("Wait for timeline writes timed out!\n");
i915_timeline_put(tl);
err = -EIO;
goto out;
}
if (*tl->hwsp_seqno != count) {
pr_err("Invalid seqno stored in timeline %lu, found 0x%x\n",
count, *tl->hwsp_seqno);
err = -EINVAL;
}
i915_timeline_put(tl);
count++;
if (err)
goto out;
i915_timelines_park(i915); /* Encourage recycling! */
} while (!__igt_timeout(end_time, NULL));
}
out:
if (igt_flush_test(i915, I915_WAIT_LOCKED))
err = -EIO;
intel_runtime_pm_put(i915, wakeref);
mutex_unlock(&i915->drm.struct_mutex);
return err;
}
int i915_timeline_live_selftests(struct drm_i915_private *i915)
{
static const struct i915_subtest tests[] = {
SUBTEST(live_hwsp_recycle),
SUBTEST(live_hwsp_engine),
SUBTEST(live_hwsp_alternate),
};
return i915_subtests(tests, i915);
}