linux/drivers/gpu/drm/i915/selftests/i915_gem_coherency.c
Chris Wilson e22d8e3c69 drm/i915: Treat WC a separate cache domain
When discussing a new WC mmap, we based the interface upon the
assumption that GTT was fully coherent. How naive! Commits 3b5724d702
("drm/i915: Wait for writes through the GTT to land before reading
back") and ed4596ea99 ("drm/i915/guc: WA to address the Ringbuffer
coherency issue") demonstrate that writes through the GTT are indeed
delayed and may be overtaken by direct WC access. To be safe, if
userspace is mixing WC mmaps with other potential GTT access (pwrite,
GTT mmaps) it should use set_domain(WC).

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=96563
Testcase: igt/gem_pwrite/small-gtt*
Testcase: igt/drv_selftest/coherency
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170412110111.26626-2-chris@chris-wilson.co.uk
2017-04-12 12:35:17 +01:00

380 lines
9.2 KiB
C

/*
* Copyright © 2017 Intel Corporation
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice (including the next
* paragraph) shall be included in all copies or substantial portions of the
* Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
* IN THE SOFTWARE.
*
*/
#include <linux/prime_numbers.h>
#include "../i915_selftest.h"
#include "i915_random.h"
static int cpu_set(struct drm_i915_gem_object *obj,
unsigned long offset,
u32 v)
{
unsigned int needs_clflush;
struct page *page;
typeof(v) *map;
int err;
err = i915_gem_obj_prepare_shmem_write(obj, &needs_clflush);
if (err)
return err;
page = i915_gem_object_get_page(obj, offset >> PAGE_SHIFT);
map = kmap_atomic(page);
if (needs_clflush & CLFLUSH_BEFORE)
clflush(map+offset_in_page(offset) / sizeof(*map));
map[offset_in_page(offset) / sizeof(*map)] = v;
if (needs_clflush & CLFLUSH_AFTER)
clflush(map+offset_in_page(offset) / sizeof(*map));
kunmap_atomic(map);
i915_gem_obj_finish_shmem_access(obj);
return 0;
}
static int cpu_get(struct drm_i915_gem_object *obj,
unsigned long offset,
u32 *v)
{
unsigned int needs_clflush;
struct page *page;
typeof(v) map;
int err;
err = i915_gem_obj_prepare_shmem_read(obj, &needs_clflush);
if (err)
return err;
page = i915_gem_object_get_page(obj, offset >> PAGE_SHIFT);
map = kmap_atomic(page);
if (needs_clflush & CLFLUSH_BEFORE)
clflush(map+offset_in_page(offset) / sizeof(*map));
*v = map[offset_in_page(offset) / sizeof(*map)];
kunmap_atomic(map);
i915_gem_obj_finish_shmem_access(obj);
return 0;
}
static int gtt_set(struct drm_i915_gem_object *obj,
unsigned long offset,
u32 v)
{
struct i915_vma *vma;
typeof(v) *map;
int err;
err = i915_gem_object_set_to_gtt_domain(obj, true);
if (err)
return err;
vma = i915_gem_object_ggtt_pin(obj, NULL, 0, 0, PIN_MAPPABLE);
if (IS_ERR(vma))
return PTR_ERR(vma);
map = i915_vma_pin_iomap(vma);
i915_vma_unpin(vma);
if (IS_ERR(map))
return PTR_ERR(map);
map[offset / sizeof(*map)] = v;
i915_vma_unpin_iomap(vma);
return 0;
}
static int gtt_get(struct drm_i915_gem_object *obj,
unsigned long offset,
u32 *v)
{
struct i915_vma *vma;
typeof(v) map;
int err;
err = i915_gem_object_set_to_gtt_domain(obj, false);
if (err)
return err;
vma = i915_gem_object_ggtt_pin(obj, NULL, 0, 0, PIN_MAPPABLE);
if (IS_ERR(vma))
return PTR_ERR(vma);
map = i915_vma_pin_iomap(vma);
i915_vma_unpin(vma);
if (IS_ERR(map))
return PTR_ERR(map);
*v = map[offset / sizeof(*map)];
i915_vma_unpin_iomap(vma);
return 0;
}
static int wc_set(struct drm_i915_gem_object *obj,
unsigned long offset,
u32 v)
{
typeof(v) *map;
int err;
err = i915_gem_object_set_to_wc_domain(obj, true);
if (err)
return err;
map = i915_gem_object_pin_map(obj, I915_MAP_WC);
if (IS_ERR(map))
return PTR_ERR(map);
map[offset / sizeof(*map)] = v;
i915_gem_object_unpin_map(obj);
return 0;
}
static int wc_get(struct drm_i915_gem_object *obj,
unsigned long offset,
u32 *v)
{
typeof(v) map;
int err;
err = i915_gem_object_set_to_wc_domain(obj, false);
if (err)
return err;
map = i915_gem_object_pin_map(obj, I915_MAP_WC);
if (IS_ERR(map))
return PTR_ERR(map);
*v = map[offset / sizeof(*map)];
i915_gem_object_unpin_map(obj);
return 0;
}
static int gpu_set(struct drm_i915_gem_object *obj,
unsigned long offset,
u32 v)
{
struct drm_i915_private *i915 = to_i915(obj->base.dev);
struct drm_i915_gem_request *rq;
struct i915_vma *vma;
u32 *cs;
int err;
err = i915_gem_object_set_to_gtt_domain(obj, true);
if (err)
return err;
vma = i915_gem_object_ggtt_pin(obj, NULL, 0, 0, 0);
if (IS_ERR(vma))
return PTR_ERR(vma);
rq = i915_gem_request_alloc(i915->engine[RCS], i915->kernel_context);
if (IS_ERR(rq)) {
i915_vma_unpin(vma);
return PTR_ERR(rq);
}
cs = intel_ring_begin(rq, 4);
if (IS_ERR(cs)) {
__i915_add_request(rq, false);
i915_vma_unpin(vma);
return PTR_ERR(cs);
}
if (INTEL_GEN(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) {
*cs++ = MI_STORE_DWORD_IMM_GEN4 | 1 << 22;
*cs++ = 0;
*cs++ = i915_ggtt_offset(vma) + offset;
*cs++ = v;
} else {
*cs++ = MI_STORE_DWORD_IMM | 1 << 22;
*cs++ = i915_ggtt_offset(vma) + offset;
*cs++ = v;
*cs++ = MI_NOOP;
}
intel_ring_advance(rq, cs);
i915_vma_move_to_active(vma, rq, EXEC_OBJECT_WRITE);
i915_vma_unpin(vma);
reservation_object_lock(obj->resv, NULL);
reservation_object_add_excl_fence(obj->resv, &rq->fence);
reservation_object_unlock(obj->resv);
__i915_add_request(rq, true);
return 0;
}
static bool always_valid(struct drm_i915_private *i915)
{
return true;
}
static bool needs_mi_store_dword(struct drm_i915_private *i915)
{
return igt_can_mi_store_dword_imm(i915);
}
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);
} igt_coherency_mode[] = {
{ "cpu", cpu_set, cpu_get, always_valid },
{ "gtt", gtt_set, gtt_get, always_valid },
{ "wc", wc_set, wc_get, always_valid },
{ "gpu", gpu_set, NULL, needs_mi_store_dword },
{ },
};
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;
int err = 0;
/* 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.
*/
offsets = kmalloc_array(ncachelines, 2*sizeof(u32), GFP_KERNEL);
if (!offsets)
return -ENOMEM;
for (count = 0; count < ncachelines; count++)
offsets[count] = count * 64 + 4 * (count % 16);
values = offsets + ncachelines;
mutex_lock(&i915->drm.struct_mutex);
for (over = igt_coherency_mode; over->name; over++) {
if (!over->set)
continue;
if (!over->valid(i915))
continue;
for (write = igt_coherency_mode; write->name; write++) {
if (!write->set)
continue;
if (!write->valid(i915))
continue;
for (read = igt_coherency_mode; read->name; read++) {
if (!read->get)
continue;
if (!read->valid(i915))
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);
goto unlock;
}
i915_random_reorder(offsets, ncachelines, &prng);
for (n = 0; n < count; n++)
values[n] = prandom_u32_state(&prng);
for (n = 0; n < count; n++) {
err = over->set(obj, 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);
goto put_object;
}
}
for (n = 0; n < count; n++) {
err = write->set(obj, 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);
goto put_object;
}
}
for (n = 0; n < count; n++) {
u32 found;
err = read->get(obj, 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);
goto put_object;
}
if (found != values[n]) {
pr_err("Value[%ld/%ld] mismatch, (overwrite with %s) wrote [%s] %x read [%s] %x (inverse %x), at offset %x\n",
n, count, over->name,
write->name, values[n],
read->name, found,
~values[n], offsets[n]);
err = -EINVAL;
goto put_object;
}
}
__i915_gem_object_release_unless_active(obj);
}
}
}
}
unlock:
mutex_unlock(&i915->drm.struct_mutex);
kfree(offsets);
return err;
put_object:
__i915_gem_object_release_unless_active(obj);
goto unlock;
}
int i915_gem_coherency_live_selftests(struct drm_i915_private *i915)
{
static const struct i915_subtest tests[] = {
SUBTEST(igt_gem_coherency),
};
return i915_subtests(tests, i915);
}