Skip to content

L
linux
Commit c8c197d4 authored by Chris Wilson's avatar Chris Wilson
Browse files
Options

drm/i915/selftests: Use a random engine for GEM coherency tests 

Select a random user accessible engine for checking coherency results.
While we should check all engines, we use a random selection so that
over repeated runs we cover all.
Signed-off-by: default avatarChris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: default avatarMatthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191027225808.19437-4-chris@chris-wilson.co.uk
parent 96815f3d
Show whitespace changes
Inline Side-by-side
Showing with 89 additions and 79 deletions
drivers/gpu/drm/i915/gem/selftests/i915_gem_coherency.c
View file @ c8c197d4
......@@ -13,9 +13,12 @@
#include "i915_selftest.h"
#include "selftests/i915_random.h"
static int cpu_set(struct drm_i915_gem_object *obj,
unsigned long offset,
u32 v)
struct context {
struct drm_i915_gem_object *obj;
struct intel_engine_cs *engine;
};
static int cpu_set(struct context *ctx, unsigned long offset, u32 v)
{
unsigned int needs_clflush;
struct page *page;
......@@ -23,11 +26,11 @@ static int cpu_set(struct drm_i915_gem_object *obj,
u32 *cpu;
int err;
err = i915_gem_object_prepare_write(obj, &needs_clflush);
err = i915_gem_object_prepare_write(ctx->obj, &needs_clflush);
if (err)
return err;
page = i915_gem_object_get_page(obj, offset >> PAGE_SHIFT);
page = i915_gem_object_get_page(ctx->obj, offset >> PAGE_SHIFT);
map = kmap_atomic(page);
cpu = map + offset_in_page(offset);
......@@ -40,14 +43,12 @@ static int cpu_set(struct drm_i915_gem_object *obj,
drm_clflush_virt_range(cpu, sizeof(*cpu));
kunmap_atomic(map);
i915_gem_object_finish_access(obj);
i915_gem_object_finish_access(ctx->obj);
return 0;
}
static int cpu_get(struct drm_i915_gem_object *obj,
unsigned long offset,
u32 *v)
static int cpu_get(struct context *ctx, unsigned long offset, u32 *v)
{
unsigned int needs_clflush;
struct page *page;
......@@ -55,11 +56,11 @@ static int cpu_get(struct drm_i915_gem_object *obj,
u32 *cpu;
int err;
err = i915_gem_object_prepare_read(obj, &needs_clflush);
err = i915_gem_object_prepare_read(ctx->obj, &needs_clflush);
if (err)
return err;
page = i915_gem_object_get_page(obj, offset >> PAGE_SHIFT);
page = i915_gem_object_get_page(ctx->obj, offset >> PAGE_SHIFT);
map = kmap_atomic(page);
cpu = map + offset_in_page(offset);
......@@ -69,26 +70,24 @@ static int cpu_get(struct drm_i915_gem_object *obj,
*v = *cpu;
kunmap_atomic(map);
i915_gem_object_finish_access(obj);
i915_gem_object_finish_access(ctx->obj);
return 0;
}
static int gtt_set(struct drm_i915_gem_object *obj,
unsigned long offset,
u32 v)
static int gtt_set(struct context *ctx, unsigned long offset, u32 v)
{
struct i915_vma *vma;
u32 __iomem *map;
int err = 0;
i915_gem_object_lock(obj);
err = i915_gem_object_set_to_gtt_domain(obj, true);
i915_gem_object_unlock(obj);
i915_gem_object_lock(ctx->obj);
err = i915_gem_object_set_to_gtt_domain(ctx->obj, true);
i915_gem_object_unlock(ctx->obj);
if (err)
return err;
vma = i915_gem_object_ggtt_pin(obj, NULL, 0, 0, PIN_MAPPABLE);
vma = i915_gem_object_ggtt_pin(ctx->obj, NULL, 0, 0, PIN_MAPPABLE);
if (IS_ERR(vma))
return PTR_ERR(vma);
......@@ -109,21 +108,19 @@ static int gtt_set(struct drm_i915_gem_object *obj,
return err;
}
static int gtt_get(struct drm_i915_gem_object *obj,
unsigned long offset,
u32 *v)
static int gtt_get(struct context *ctx, unsigned long offset, u32 *v)
{
struct i915_vma *vma;
u32 __iomem *map;
int err = 0;
i915_gem_object_lock(obj);
err = i915_gem_object_set_to_gtt_domain(obj, false);
i915_gem_object_unlock(obj);
i915_gem_object_lock(ctx->obj);
err = i915_gem_object_set_to_gtt_domain(ctx->obj, false);
i915_gem_object_unlock(ctx->obj);
if (err)
return err;
vma = i915_gem_object_ggtt_pin(obj, NULL, 0, 0, PIN_MAPPABLE);
vma = i915_gem_object_ggtt_pin(ctx->obj, NULL, 0, 0, PIN_MAPPABLE);
if (IS_ERR(vma))
return PTR_ERR(vma);
......@@ -144,73 +141,66 @@ static int gtt_get(struct drm_i915_gem_object *obj,
return err;
}
static int wc_set(struct drm_i915_gem_object *obj,
unsigned long offset,
u32 v)
static int wc_set(struct context *ctx, unsigned long offset, u32 v)
{
u32 *map;
int err;
i915_gem_object_lock(obj);
err = i915_gem_object_set_to_wc_domain(obj, true);
i915_gem_object_unlock(obj);
i915_gem_object_lock(ctx->obj);
err = i915_gem_object_set_to_wc_domain(ctx->obj, true);
i915_gem_object_unlock(ctx->obj);
if (err)
return err;
map = i915_gem_object_pin_map(obj, I915_MAP_WC);
map = i915_gem_object_pin_map(ctx->obj, I915_MAP_WC);
if (IS_ERR(map))
return PTR_ERR(map);
map[offset / sizeof(*map)] = v;
i915_gem_object_unpin_map(obj);
i915_gem_object_unpin_map(ctx->obj);
return 0;
}
static int wc_get(struct drm_i915_gem_object *obj,
unsigned long offset,
u32 *v)
static int wc_get(struct context *ctx, unsigned long offset, u32 *v)
{
u32 *map;
int err;
i915_gem_object_lock(obj);
err = i915_gem_object_set_to_wc_domain(obj, false);
i915_gem_object_unlock(obj);
i915_gem_object_lock(ctx->obj);
err = i915_gem_object_set_to_wc_domain(ctx->obj, false);
i915_gem_object_unlock(ctx->obj);
if (err)
return err;
map = i915_gem_object_pin_map(obj, I915_MAP_WC);
map = i915_gem_object_pin_map(ctx->obj, I915_MAP_WC);
if (IS_ERR(map))
return PTR_ERR(map);
*v = map[offset / sizeof(*map)];
i915_gem_object_unpin_map(obj);
i915_gem_object_unpin_map(ctx->obj);
return 0;
}
static int gpu_set(struct drm_i915_gem_object *obj,
unsigned long offset,
u32 v)
static int gpu_set(struct context *ctx, unsigned long offset, u32 v)
{
struct drm_i915_private *i915 = to_i915(obj->base.dev);
struct i915_request *rq;
struct i915_vma *vma;
u32 *cs;
int err;
i915_gem_object_lock(obj);
err = i915_gem_object_set_to_gtt_domain(obj, true);
i915_gem_object_unlock(obj);
i915_gem_object_lock(ctx->obj);
err = i915_gem_object_set_to_gtt_domain(ctx->obj, true);
i915_gem_object_unlock(ctx->obj);
if (err)
return err;
vma = i915_gem_object_ggtt_pin(obj, NULL, 0, 0, 0);
vma = i915_gem_object_ggtt_pin(ctx->obj, NULL, 0, 0, 0);
if (IS_ERR(vma))
return PTR_ERR(vma);
rq = i915_request_create(i915->engine[RCS0]->kernel_context);
rq = i915_request_create(ctx->engine->kernel_context);
if (IS_ERR(rq)) {
i915_vma_unpin(vma);
return PTR_ERR(rq);
......@@ -223,12 +213,12 @@ static int gpu_set(struct drm_i915_gem_object *obj,
return PTR_ERR(cs);
}
if (INTEL_GEN(i915) >= 8) {
if (INTEL_GEN(ctx->engine->i915) >= 8) {
*cs++ = MI_STORE_DWORD_IMM_GEN4 | 1 << 22;
*cs++ = lower_32_bits(i915_ggtt_offset(vma) + offset);
*cs++ = upper_32_bits(i915_ggtt_offset(vma) + offset);
*cs++ = v;
} else if (INTEL_GEN(i915) >= 4) {
} else if (INTEL_GEN(ctx->engine->i915) >= 4) {
*cs++ = MI_STORE_DWORD_IMM_GEN4 | MI_USE_GGTT;
*cs++ = 0;
*cs++ = i915_ggtt_offset(vma) + offset;
......@@ -253,32 +243,29 @@ static int gpu_set(struct drm_i915_gem_object *obj,
return err;
}
static bool always_valid(struct drm_i915_private *i915)
static bool always_valid(struct context *ctx)
{
return true;
}
static bool needs_fence_registers(struct drm_i915_private *i915)
static bool needs_fence_registers(struct context *ctx)
{
return !intel_gt_is_wedged(&i915->gt);
return !intel_gt_is_wedged(ctx->engine->gt);
}
static bool needs_mi_store_dword(struct drm_i915_private *i915)
static bool needs_mi_store_dword(struct context *ctx)
{
if (intel_gt_is_wedged(&i915->gt))
return false;
if (!HAS_ENGINE(i915, RCS0))
if (intel_gt_is_wedged(ctx->engine->gt))
return false;
return intel_engine_can_store_dword(i915->engine[RCS0]);
return intel_engine_can_store_dword(ctx->engine);
}
static const struct igt_coherency_mode {
const char *name;
int (*set)(struct drm_i915_gem_object *, unsigned long offset, u32 v);
int (*get)(struct drm_i915_gem_object *, unsigned long offset, u32 *v);
bool (*valid)(struct drm_i915_private *i915);
int (*set)(struct context *ctx, unsigned long offset, u32 v);
int (*get)(struct context *ctx, unsigned long offset, u32 *v);
bool (*valid)(struct context *ctx);
} igt_coherency_mode[] = {
{ "cpu", cpu_set, cpu_get, always_valid },
{ "gtt", gtt_set, gtt_get, needs_fence_registers },
......@@ -287,18 +274,37 @@ static const struct igt_coherency_mode {
{ },
};
static struct intel_engine_cs *
random_engine(struct drm_i915_private *i915, struct rnd_state *prng)
{
struct intel_engine_cs *engine;
unsigned int count;
count = 0;
for_each_uabi_engine(engine, i915)
count++;
count = i915_prandom_u32_max_state(count, prng);
for_each_uabi_engine(engine, i915)
if (count-- == 0)
return engine;
return NULL;
}
static int igt_gem_coherency(void *arg)
{
const unsigned int ncachelines = PAGE_SIZE/64;
I915_RND_STATE(prng);
struct drm_i915_private *i915 = arg;
const struct igt_coherency_mode *read, *write, *over;
struct drm_i915_gem_object *obj;
unsigned long count, n;
u32 *offsets, *values;
I915_RND_STATE(prng);
struct context ctx;
int err = 0;
/* We repeatedly write, overwrite and read from a sequence of
/*
* We repeatedly write, overwrite and read from a sequence of
* cachelines in order to try and detect incoherency (unflushed writes
* from either the CPU or GPU). Each setter/getter uses our cache
* domain API which should prevent incoherency.
......@@ -312,31 +318,35 @@ static int igt_gem_coherency(void *arg)
values = offsets + ncachelines;
ctx.engine = random_engine(i915, &prng);
GEM_BUG_ON(!ctx.engine);
pr_info("%s: using %s\n", __func__, ctx.engine->name);
for (over = igt_coherency_mode; over->name; over++) {
if (!over->set)
continue;
if (!over->valid(i915))
if (!over->valid(&ctx))
continue;
for (write = igt_coherency_mode; write->name; write++) {
if (!write->set)
continue;
if (!write->valid(i915))
if (!write->valid(&ctx))
continue;
for (read = igt_coherency_mode; read->name; read++) {
if (!read->get)
continue;
if (!read->valid(i915))
if (!read->valid(&ctx))
continue;
for_each_prime_number_from(count, 1, ncachelines) {
obj = i915_gem_object_create_internal(i915, PAGE_SIZE);
if (IS_ERR(obj)) {
err = PTR_ERR(obj);
ctx.obj = i915_gem_object_create_internal(i915, PAGE_SIZE);
if (IS_ERR(ctx.obj)) {
err = PTR_ERR(ctx.obj);
goto free;
}
......@@ -345,7 +355,7 @@ static int igt_gem_coherency(void *arg)
values[n] = prandom_u32_state(&prng);
for (n = 0; n < count; n++) {
err = over->set(obj, offsets[n], ~values[n]);
err = over->set(&ctx, offsets[n], ~values[n]);
if (err) {
pr_err("Failed to set stale value[%ld/%ld] in object using %s, err=%d\n",
n, count, over->name, err);
......@@ -354,7 +364,7 @@ static int igt_gem_coherency(void *arg)
}
for (n = 0; n < count; n++) {
err = write->set(obj, offsets[n], values[n]);
err = write->set(&ctx, offsets[n], values[n]);
if (err) {
pr_err("Failed to set value[%ld/%ld] in object using %s, err=%d\n",
n, count, write->name, err);
......@@ -365,7 +375,7 @@ static int igt_gem_coherency(void *arg)
for (n = 0; n < count; n++) {
u32 found;
err = read->get(obj, offsets[n], &found);
err = read->get(&ctx, offsets[n], &found);
if (err) {
pr_err("Failed to get value[%ld/%ld] in object using %s, err=%d\n",
n, count, read->name, err);
......@@ -383,7 +393,7 @@ static int igt_gem_coherency(void *arg)
}
}
i915_gem_object_put(obj);
i915_gem_object_put(ctx.obj);
}
}
}
......@@ -393,7 +403,7 @@ static int igt_gem_coherency(void *arg)
return err;
put_object:
i915_gem_object_put(obj);
i915_gem_object_put(ctx.obj);
goto free;
}
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment
GitLab Nexedi Edition | About GitLab | About Nexedi | 沪ICP备2021021310号-2 | 沪ICP备2021021310号-7