@@ -41,6 +41,7 @@ static bool prime_busy(struct pollfd *pfd, bool excl)
static void busy(int fd, unsigned ring, unsigned flags)
{
const int gen = intel_gen(intel_get_drm_devid(fd));
+ const uint32_t _bbe = MI_BATCH_BUFFER_END;
struct drm_i915_gem_exec_object2 obj[2];
struct pollfd pfd[2];
#define SCRATCH 0
@@ -55,7 +56,7 @@ static void busy(int fd, unsigned ring, unsigned flags)
gem_quiescent_gpu(fd);
memset(&execbuf, 0, sizeof(execbuf));
- execbuf.buffers_ptr = (uintptr_t)obj;
+ execbuf.buffers_ptr = to_user_pointer(obj);
execbuf.buffer_count = 2;
execbuf.flags = ring;
if (gen < 6)
@@ -65,7 +66,11 @@ static void busy(int fd, unsigned ring, unsigned flags)
obj[SCRATCH].handle = gem_create(fd, 4096);
obj[BATCH].handle = gem_create(fd, size);
- obj[BATCH].relocs_ptr = (uintptr_t)store;
+ obj[BATCH].offset = 1 << 20;
+ gem_write(fd, obj[BATCH].handle, 0, &_bbe, sizeof(_bbe));
+ igt_require(__gem_execbuf(fd, &execbuf) == 0); /* prebind the buffers */
+
+ obj[BATCH].relocs_ptr = to_user_pointer(store);
obj[BATCH].relocation_count = ARRAY_SIZE(store);
memset(store, 0, sizeof(store));
@@ -81,23 +86,28 @@ static void busy(int fd, unsigned ring, unsigned flags)
i = 0;
for (count = 0; count < 1024; count++) {
+ uint64_t offset;
+
store[count].target_handle = obj[SCRATCH].handle;
- store[count].presumed_offset = -1;
+ store[count].presumed_offset = obj[SCRATCH].offset;
store[count].offset = sizeof(uint32_t) * (i + 1);
store[count].delta = sizeof(uint32_t) * count;
- store[count].read_domains = I915_GEM_DOMAIN_INSTRUCTION;
- store[count].write_domain = I915_GEM_DOMAIN_INSTRUCTION;
+ store[count].read_domains = I915_GEM_DOMAIN_RENDER;
+ store[count].write_domain = I915_GEM_DOMAIN_RENDER;
+
+ offset = store[count].presumed_offset + store[count].delta;
+
batch[i] = MI_STORE_DWORD_IMM | (gen < 6 ? 1 << 22 : 0);
if (gen >= 8) {
- batch[++i] = 0;
- batch[++i] = 0;
+ batch[++i] = offset;
+ batch[++i] = offset >> 32;
} else if (gen >= 4) {
batch[++i] = 0;
- batch[++i] = 0;
+ batch[++i] = offset;
store[count].offset += sizeof(uint32_t);
} else {
batch[i]--;
- batch[++i] = 0;
+ batch[++i] = offset;
}
batch[++i] = count;
i++;
@@ -105,7 +115,7 @@ static void busy(int fd, unsigned ring, unsigned flags)
bbe = &batch[i];
store[count].target_handle = obj[BATCH].handle; /* recurse */
- store[count].presumed_offset = 0;
+ store[count].presumed_offset = obj[BATCH].offset;
store[count].offset = sizeof(uint32_t) * (i + 1);
store[count].delta = 0;
store[count].read_domains = I915_GEM_DOMAIN_COMMAND;
@@ -113,14 +123,14 @@ static void busy(int fd, unsigned ring, unsigned flags)
batch[i] = MI_BATCH_BUFFER_START;
if (gen >= 8) {
batch[i] |= 1 << 8 | 1;
- batch[++i] = 0;
- batch[++i] = 0;
+ batch[++i] = obj[BATCH].offset;
+ batch[++i] = obj[BATCH].offset >> 32;
} else if (gen >= 6) {
batch[i] |= 1 << 8;
- batch[++i] = 0;
+ batch[++i] = obj[BATCH].offset;
} else {
batch[i] |= 2 << 6;
- batch[++i] = 0;
+ batch[++i] = obj[BATCH].offset;
if (gen < 4) {
batch[i] |= 1;
store[count].delta = 1;
The test assumes that the dmabuf for the batch buffer has no exclusive fence. This is only true for the batch as it executes on the GPU; prior to that we may be in the process of moving it into memory and the ppGTT which cause the batch to carry an exclusive fence. Keep the test simple and prebind the batch buffer so that when we poll the dmabuf fd, we only expect it to be the spinning batch. Alternatively we could wait until we start writing values into scratch (i.e. igt_spin_busywait_until_started()). This seemed simpler. Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Matthew Auld <matthew.auld@intel.com> --- tests/prime_busy.c | 38 ++++++++++++++++++++++++-------------- 1 file changed, 24 insertions(+), 14 deletions(-)