static void test_with_two_bos(void)
{
int fd1, fd2;
uint32_t handle1, handle2, handle_import;
int dma_buf_fd;
counter = 0;
fd1 = drm_open_driver(DRIVER_INTEL);
fd2 = drm_open_driver(DRIVER_INTEL);
handle1 = gem_create(fd1, BO_SIZE);
handle2 = gem_create(fd1, BO_SIZE);
dma_buf_fd = prime_handle_to_fd(fd1, handle1);
handle_import = prime_fd_to_handle(fd2, dma_buf_fd);
close(dma_buf_fd);
gem_close(fd1, handle1);
dma_buf_fd = prime_handle_to_fd(fd1, handle2);
handle_import = prime_fd_to_handle(fd2, dma_buf_fd);
check_bo(fd1, handle2, fd2, handle_import);
gem_close(fd1, handle2);
close(dma_buf_fd);
check_bo(fd2, handle_import, fd2, handle_import);
close(fd1);
close(fd2);
}
static void run_test(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 drm_i915_gem_relocation_entry reloc[1024];
struct drm_i915_gem_execbuffer2 execbuf;
struct igt_hang_ring hang;
uint32_t *batch, *b;
int i;
gem_require_ring(fd, ring);
igt_skip_on_f(gen == 6 && (ring & ~(3<<13)) == I915_EXEC_BSD,
"MI_STORE_DATA broken on gen6 bsd\n");
gem_quiescent_gpu(fd);
memset(&execbuf, 0, sizeof(execbuf));
execbuf.buffers_ptr = (uintptr_t)obj;
execbuf.buffer_count = 2;
execbuf.flags = ring | (1 << 11);
if (gen < 6)
execbuf.flags |= I915_EXEC_SECURE;
memset(obj, 0, sizeof(obj));
obj[0].handle = gem_create(fd, 4096);
obj[0].flags |= EXEC_OBJECT_WRITE;
obj[1].handle = gem_create(fd, 1024*16 + 4096);
gem_write(fd, obj[1].handle, 0, &bbe, sizeof(bbe));
igt_require(__gem_execbuf(fd, &execbuf) == 0);
obj[1].relocs_ptr = (uintptr_t)reloc;
obj[1].relocation_count = 1024;
batch = gem_mmap__cpu(fd, obj[1].handle, 0, 16*1024 + 4096,
PROT_WRITE | PROT_READ);
gem_set_domain(fd, obj[1].handle,
I915_GEM_DOMAIN_CPU, I915_GEM_DOMAIN_CPU);
memset(reloc, 0, sizeof(reloc));
b = batch;
for (i = 0; i < 1024; i++) {
uint64_t offset;
reloc[i].target_handle = obj[0].handle;
reloc[i].presumed_offset = obj[0].offset;
reloc[i].offset = (b - batch + 1) * sizeof(*batch);
reloc[i].delta = i * sizeof(uint32_t);
reloc[i].read_domains = I915_GEM_DOMAIN_INSTRUCTION;
reloc[i].write_domain = I915_GEM_DOMAIN_INSTRUCTION;
offset = obj[0].offset + reloc[i].delta;
*b++ = MI_STORE_DWORD_IMM | (gen < 6 ? 1 << 22 : 0);
if (gen >= 8) {
*b++ = offset;
*b++ = offset >> 32;
} else if (gen >= 4) {
static void test_ring(int fd, unsigned ring, uint32_t flags)
{
uint32_t bbe = MI_BATCH_BUFFER_END;
uint32_t handle[3];
uint32_t read, write;
uint32_t active;
unsigned i;
gem_require_ring(fd, ring | flags);
handle[TEST] = gem_create(fd, 4096);
handle[BATCH] = gem_create(fd, 4096);
gem_write(fd, handle[BATCH], 0, &bbe, sizeof(bbe));
/* Create a long running batch which we can use to hog the GPU */
handle[BUSY] = busy_blt(fd);
/* Queue a batch after the busy, it should block and remain "busy" */
igt_assert(exec_noop(fd, handle, ring | flags, false));
igt_assert(still_busy(fd, handle[BUSY]));
__gem_busy(fd, handle[TEST], &read, &write);
igt_assert_eq(read, 1 << ring);
igt_assert_eq(write, 0);
/* Requeue with a write */
igt_assert(exec_noop(fd, handle, ring | flags, true));
igt_assert(still_busy(fd, handle[BUSY]));
__gem_busy(fd, handle[TEST], &read, &write);
igt_assert_eq(read, 1 << ring);
igt_assert_eq(write, ring);
/* Now queue it for a read across all available rings */
active = 0;
for (i = I915_EXEC_RENDER; i <= I915_EXEC_VEBOX; i++) {
if (exec_noop(fd, handle, i | flags, false))
active |= 1 << i;
}
igt_assert(still_busy(fd, handle[BUSY]));
__gem_busy(fd, handle[TEST], &read, &write);
igt_assert_eq(read, active);
igt_assert_eq(write, ring); /* from the earlier write */
/* Check that our long batch was long enough */
igt_assert(still_busy(fd, handle[BUSY]));
/* And make sure it becomes idle again */
gem_sync(fd, handle[TEST]);
__gem_busy(fd, handle[TEST], &read, &write);
igt_assert_eq(read, 0);
igt_assert_eq(write, 0);
for (i = TEST; i <= BATCH; i++)
gem_close(fd, handle[i]);
}
static void run(data_t *data, int child)
{
const int size = 4096 * (256 + child * child);
const int tiling = child % 2;
const int write = child % 2;
uint32_t handle = gem_create(data->fd, size);
uint32_t *ptr;
uint32_t x;
igt_assert(handle);
if (tiling != I915_TILING_NONE)
gem_set_tiling(data->fd, handle, tiling, 4096);
/* load up the unfaulted bo */
busy(data, handle, size, 100);
/* Note that we ignore the API and rely on the implict
* set-to-gtt-domain within the fault handler.
*/
if (write) {
ptr = gem_mmap__gtt(data->fd, handle, size,
PROT_READ | PROT_WRITE);
ptr[rand() % (size / 4)] = canary;
} else {
ptr = gem_mmap__gtt(data->fd, handle, size, PROT_READ);
}
x = ptr[rand() % (size / 4)];
munmap(ptr, size);
igt_assert_eq_u32(x, canary);
}
static void test_llseek_size(void)
{
int fd, i;
uint32_t handle;
int dma_buf_fd;
counter = 0;
fd = drm_open_driver(DRIVER_INTEL);
for (i = 0; i < 10; i++) {
int bufsz = 4096 << i;
handle = gem_create(fd, bufsz);
dma_buf_fd = prime_handle_to_fd(fd, handle);
gem_close(fd, handle);
igt_assert(prime_get_size(dma_buf_fd) == bufsz);
close(dma_buf_fd);
}
close(fd);
}
int main(int argc, char **argv)
{
uint32_t batch[2] = {MI_BATCH_BUFFER_END};
uint32_t handle;
uint32_t devid;
int fd;
drmtest_subtest_init(argc, argv);
fd = drm_open_any();
devid = intel_get_drm_devid(fd);
handle = gem_create(fd, 4096);
gem_write(fd, handle, 0, batch, sizeof(batch));
if (drmtest_run_subtest("render"))
loop(fd, handle, I915_EXEC_RENDER, "render");
if (drmtest_run_subtest("bsd"))
if (HAS_BSD_RING(devid))
loop(fd, handle, I915_EXEC_BSD, "bsd");
if (drmtest_run_subtest("blt"))
if (HAS_BLT_RING(devid))
loop(fd, handle, I915_EXEC_BLT, "blt");
gem_close(fd, handle);
close(fd);
return skipped_all ? 77 : 0;
}
static void
test_short(int fd)
{
struct drm_i915_gem_mmap_gtt mmap_arg;
int pages, p;
mmap_arg.handle = gem_create(fd, OBJECT_SIZE);
igt_assert(mmap_arg.handle);
igt_assert(drmIoctl(fd,
DRM_IOCTL_I915_GEM_MMAP_GTT,
&mmap_arg) == 0);
for (pages = 1; pages <= OBJECT_SIZE / PAGE_SIZE; pages <<= 1) {
uint8_t *r, *w;
w = mmap64(0, pages * PAGE_SIZE, PROT_READ | PROT_WRITE,
MAP_SHARED, fd, mmap_arg.offset);
igt_assert(w != MAP_FAILED);
r = mmap64(0, pages * PAGE_SIZE, PROT_READ,
MAP_SHARED, fd, mmap_arg.offset);
igt_assert(r != MAP_FAILED);
for (p = 0; p < pages; p++) {
w[p*PAGE_SIZE] = r[p*PAGE_SIZE];
w[p*PAGE_SIZE+(PAGE_SIZE-1)] =
r[p*PAGE_SIZE+(PAGE_SIZE-1)];
}
munmap(r, pages * PAGE_SIZE);
munmap(w, pages * PAGE_SIZE);
}
gem_close(fd, mmap_arg.handle);
}
static void
dontneed_after_mmap(void)
{
int fd = drm_open_driver(DRIVER_INTEL);
uint32_t handle;
char *ptr;
handle = gem_create(fd, OBJECT_SIZE);
ptr = __gem_mmap__gtt(fd, handle, OBJECT_SIZE, PROT_READ | PROT_WRITE);
igt_assert(ptr);
gem_madvise(fd, handle, I915_MADV_DONTNEED);
close(fd);
signal(SIGBUS, sigtrap);
switch (setjmp(jmp)) {
case SIGBUS:
break;
case 0:
*ptr = 0;
default:
igt_assert(!"reached");
break;
}
munmap(ptr, OBJECT_SIZE);
signal(SIGBUS, SIG_DFL);
}
static void *thread(void *data)
{
struct thread *t = data;
uint32_t bbe = MI_BATCH_BUFFER_END;
struct drm_i915_gem_execbuffer2 execbuf;
struct drm_i915_gem_exec_object2 obj;
uint32_t *ctx;
memset(&obj, 0, sizeof(obj));
obj.handle = gem_create(t->fd, 4096);
gem_write(t->fd, obj.handle, 0, &bbe, sizeof(bbe));
memset(&execbuf, 0, sizeof(execbuf));
execbuf.buffers_ptr = (uintptr_t)&obj;
execbuf.buffer_count = 1;
ctx = malloc(t->num_ctx * sizeof(uint32_t));
igt_assert(ctx);
memcpy(ctx, t->all_ctx, t->num_ctx * sizeof(uint32_t));
igt_permute_array(ctx, t->num_ctx, xchg_int);
for (unsigned n = 0; n < t->num_ctx; n++) {
execbuf.rsvd1 = ctx[n];
gem_execbuf(t->fd, &execbuf);
}
free(ctx);
gem_close(t->fd, obj.handle);
return NULL;
}
static void
test_access(int fd)
{
uint32_t handle, flink, handle2;
struct drm_i915_gem_mmap_gtt mmap_arg;
int fd2;
handle = gem_create(fd, OBJECT_SIZE);
igt_assert(handle);
fd2 = drm_open_driver(DRIVER_INTEL);
/* Check that fd1 can mmap. */
mmap_arg.handle = handle;
do_ioctl(fd, DRM_IOCTL_I915_GEM_MMAP_GTT, &mmap_arg);
igt_assert(mmap64(0, OBJECT_SIZE, PROT_READ | PROT_WRITE,
MAP_SHARED, fd, mmap_arg.offset));
/* Check that the same offset on the other fd doesn't work. */
igt_assert(mmap64(0, OBJECT_SIZE, PROT_READ | PROT_WRITE,
MAP_SHARED, fd2, mmap_arg.offset) == MAP_FAILED);
igt_assert(errno == EACCES);
flink = gem_flink(fd, handle);
igt_assert(flink);
handle2 = gem_open(fd2, flink);
igt_assert(handle2);
/* Recheck that it works after flink. */
/* Check that the same offset on the other fd doesn't work. */
igt_assert(mmap64(0, OBJECT_SIZE, PROT_READ | PROT_WRITE,
MAP_SHARED, fd2, mmap_arg.offset));
}
static void test_llseek_bad(void)
{
int fd;
uint32_t handle;
int dma_buf_fd;
counter = 0;
fd = drm_open_driver(DRIVER_INTEL);
handle = gem_create(fd, BO_SIZE);
dma_buf_fd = prime_handle_to_fd(fd, handle);
gem_close(fd, handle);
igt_require(lseek(dma_buf_fd, 0, SEEK_END) >= 0);
igt_assert(lseek(dma_buf_fd, -1, SEEK_END) == -1 && errno == EINVAL);
igt_assert(lseek(dma_buf_fd, 1, SEEK_SET) == -1 && errno == EINVAL);
igt_assert(lseek(dma_buf_fd, BO_SIZE, SEEK_SET) == -1 && errno == EINVAL);
igt_assert(lseek(dma_buf_fd, BO_SIZE + 1, SEEK_SET) == -1 && errno == EINVAL);
igt_assert(lseek(dma_buf_fd, BO_SIZE - 1, SEEK_SET) == -1 && errno == EINVAL);
close(dma_buf_fd);
close(fd);
}
static void
test_read_write2(int fd, enum test_read_write order)
{
uint32_t handle;
void *r, *w;
volatile uint32_t val = 0;
handle = gem_create(fd, OBJECT_SIZE);
r = gem_mmap__gtt(fd, handle, OBJECT_SIZE, PROT_READ);
w = gem_mmap__gtt(fd, handle, OBJECT_SIZE, PROT_READ | PROT_WRITE);
if (order == READ_BEFORE_WRITE) {
val = *(uint32_t *)r;
*(uint32_t *)w = val;
} else {
*(uint32_t *)w = val;
val = *(uint32_t *)r;
}
gem_close(fd, handle);
munmap(r, OBJECT_SIZE);
munmap(w, OBJECT_SIZE);
}
static void *thread_fn_export_vs_close(void *p)
{
struct drm_prime_handle prime_h2f;
struct drm_gem_close close_bo;
int fd = (uintptr_t)p;
uint32_t handle;
while (!pls_die) {
/* We want to race gem close against prime export on handle one.*/
handle = gem_create(fd, 4096);
if (handle != 1)
gem_close(fd, handle);
/* raw ioctl since we expect this to fail */
/* WTF: for gem_flink_race I've unconditionally used handle == 1
* here, but with prime it seems to help a _lot_ to use
* something more random. */
prime_h2f.handle = 1;
prime_h2f.flags = DRM_CLOEXEC;
prime_h2f.fd = -1;
ioctl(fd, DRM_IOCTL_PRIME_HANDLE_TO_FD, &prime_h2f);
close_bo.handle = 1;
ioctl(fd, DRM_IOCTL_GEM_CLOSE, &close_bo);
close(prime_h2f.fd);
}
return (void *)0;
}
static void test_with_one_bo_two_files(void)
{
int fd1, fd2;
uint32_t handle_import, handle_open, handle_orig, flink_name;
int dma_buf_fd1, dma_buf_fd2;
fd1 = drm_open_driver(DRIVER_INTEL);
fd2 = drm_open_driver(DRIVER_INTEL);
handle_orig = gem_create(fd1, BO_SIZE);
dma_buf_fd1 = prime_handle_to_fd(fd1, handle_orig);
flink_name = gem_flink(fd1, handle_orig);
handle_open = gem_open(fd2, flink_name);
dma_buf_fd2 = prime_handle_to_fd(fd2, handle_open);
handle_import = prime_fd_to_handle(fd2, dma_buf_fd2);
/* dma-buf selfimporting an flink bo should give the same handle */
igt_assert_eq_u32(handle_import, handle_open);
close(fd1);
close(fd2);
close(dma_buf_fd1);
close(dma_buf_fd2);
}
static uint32_t batch(int fd)
{
const uint32_t buf[] = {MI_BATCH_BUFFER_END};
uint32_t handle = gem_create(fd, 4096);
gem_write(fd, handle, 0, buf, sizeof(buf));
return handle;
}
static void draw_rect_render(int fd, struct cmd_data *cmd_data,
struct buf_data *buf, struct rect *rect,
uint32_t color)
{
drm_intel_bo *src, *dst;
uint32_t devid = intel_get_drm_devid(fd);
igt_render_copyfunc_t rendercopy = igt_get_render_copyfunc(devid);
struct igt_buf src_buf, dst_buf;
struct intel_batchbuffer *batch;
uint32_t tiling, swizzle;
struct buf_data tmp;
int pixel_size = buf->bpp / 8;
unsigned adjusted_w, adjusted_dst_x;
igt_skip_on(!rendercopy);
/* Rendercopy works at 32bpp, so if you try to do copies on buffers with
* smaller bpps you won't succeeed if you need to copy "half" of a 32bpp
* pixel or something similar. */
igt_skip_on(rect->x % (32 / buf->bpp) != 0 ||
rect->y % (32 / buf->bpp) != 0 ||
rect->w % (32 / buf->bpp) != 0 ||
rect->h % (32 / buf->bpp) != 0);
gem_get_tiling(fd, buf->handle, &tiling, &swizzle);
/* We create a temporary buffer and copy from it using rendercopy. */
tmp.size = rect->w * rect->h * pixel_size;
tmp.handle = gem_create(fd, tmp.size);
tmp.stride = rect->w * pixel_size;
tmp.bpp = buf->bpp;
draw_rect_mmap_cpu(fd, &tmp, &(struct rect){0, 0, rect->w, rect->h},
static void performance(void)
{
int n, loop, count;
int fd, num_fences;
double linear[2], tiled[2];
fd = drm_open_any();
num_fences = gem_available_fences(fd);
igt_require(num_fences > 0);
for (count = 2; count < 4*num_fences; count *= 2) {
struct timeval start, end;
uint32_t handle[count];
void *ptr[count];
for (n = 0; n < count; n++) {
handle[n] = gem_create(fd, OBJECT_SIZE);
ptr[n] = gem_mmap(fd, handle[n], OBJECT_SIZE, PROT_READ | PROT_WRITE);
igt_assert(ptr[n]);
}
gettimeofday(&start, NULL);
for (loop = 0; loop < 1024; loop++) {
for (n = 0; n < count; n++)
memset(ptr[n], 0, OBJECT_SIZE);
}
gettimeofday(&end, NULL);
linear[count != 2] = count * loop / elapsed(&start, &end);
igt_info("Upload rate for %d linear surfaces: %7.3fMiB/s\n", count, linear[count != 2]);
for (n = 0; n < count; n++)
gem_set_tiling(fd, handle[n], I915_TILING_X, 1024);
gettimeofday(&start, NULL);
for (loop = 0; loop < 1024; loop++) {
for (n = 0; n < count; n++)
memset(ptr[n], 0, OBJECT_SIZE);
}
gettimeofday(&end, NULL);
tiled[count != 2] = count * loop / elapsed(&start, &end);
igt_info("Upload rate for %d tiled surfaces: %7.3fMiB/s\n", count, tiled[count != 2]);
for (n = 0; n < count; n++) {
munmap(ptr[n], OBJECT_SIZE);
gem_close(fd, handle[n]);
}
}
errno = 0;
igt_assert(linear[1] > 0.75 * linear[0]);
igt_assert(tiled[1] > 0.75 * tiled[0]);
}
static void
test_create_fd_close(int fd)
{
igt_info("Testing closing with an object allocated.\n");
gem_create(fd, 16*1024);
/* leak it */
close(fd);
}
static void test_reimport_close_race(void)
{
pthread_t *threads;
int r, i, num_threads;
int fds[2];
int obj_count;
void *status;
uint32_t handle;
int fake;
/* Allocate exit handler fds in here so that we dont screw
* up the counts */
fake = drm_open_driver(DRIVER_INTEL);
gem_quiescent_gpu(fake);
obj_count = get_object_count();
num_threads = sysconf(_SC_NPROCESSORS_ONLN);
threads = calloc(num_threads, sizeof(pthread_t));
fds[0] = drm_open_driver(DRIVER_INTEL);
handle = gem_create(fds[0], BO_SIZE);
fds[1] = prime_handle_to_fd(fds[0], handle);
for (i = 0; i < num_threads; i++) {
r = pthread_create(&threads[i], NULL,
thread_fn_reimport_vs_close,
(void *)(uintptr_t)fds);
igt_assert_eq(r, 0);
}
sleep(5);
pls_die = 1;
for (i = 0; i < num_threads; i++) {
pthread_join(threads[i], &status);
igt_assert(status == 0);
}
close(fds[0]);
close(fds[1]);
gem_quiescent_gpu(fake);
obj_count = get_object_count() - obj_count;
igt_info("leaked %i objects\n", obj_count);
close(fake);
igt_assert_eq(obj_count, 0);
}
static void wc_contention(void)
{
const int loops = 4096;
int n, count;
int fd, num_fences;
double linear[2], tiled[2];
fd = drm_open_any();
gem_require_mmap_wc(fd);
num_fences = gem_available_fences(fd);
igt_require(num_fences > 0);
for (count = 1; count < 4*num_fences; count *= 2) {
struct timeval start, end;
struct thread_contention threads[count];
for (n = 0; n < count; n++) {
threads[n].handle = gem_create(fd, OBJECT_SIZE);
threads[n].loops = loops;
threads[n].fd = fd;
}
gettimeofday(&start, NULL);
for (n = 0; n < count; n++)
pthread_create(&threads[n].thread, NULL, wc_mmap, &threads[n]);
for (n = 0; n < count; n++)
pthread_join(threads[n].thread, NULL);
gettimeofday(&end, NULL);
linear[count != 2] = count * loops / elapsed(&start, &end) / (OBJECT_SIZE / 4096);
igt_info("Contended upload rate for %d linear threads/wc: %7.3fMiB/s\n", count, linear[count != 2]);
for (n = 0; n < count; n++)
gem_set_tiling(fd, threads[n].handle, I915_TILING_X, 1024);
gettimeofday(&start, NULL);
for (n = 0; n < count; n++)
pthread_create(&threads[n].thread, NULL, wc_mmap, &threads[n]);
for (n = 0; n < count; n++)
pthread_join(threads[n].thread, NULL);
gettimeofday(&end, NULL);
tiled[count != 2] = count * loops / elapsed(&start, &end) / (OBJECT_SIZE / 4096);
igt_info("Contended upload rate for %d tiled threads/wc: %7.3fMiB/s\n", count, tiled[count != 2]);
for (n = 0; n < count; n++) {
gem_close(fd, threads[n].handle);
}
}
errno = 0;
igt_assert(linear[1] > 0.75 * linear[0]);
igt_assert(tiled[1] > 0.75 * tiled[0]);
}
static void
test_create_close(int fd)
{
uint32_t handle;
igt_info("Testing creating and closing an object.\n");
handle = gem_create(fd, 16*1024);
gem_close(fd, handle);
}
static uint32_t
tiled_bo_create (int fd)
{
uint32_t handle;
handle = gem_create(fd, OBJECT_SIZE);
gem_set_tiling(fd, handle, I915_TILING_X, WIDTH*4);
return handle;
}
/*
* Creating an object with non-aligned size and trying to access it with an
* offset, which is greater than the requested size but smaller than the
* object's last page boundary. pwrite here must be successful.
*/
static void valid_nonaligned_size(int fd)
{
int handle;
char buf[PAGE_SIZE];
handle = gem_create(fd, PAGE_SIZE / 2);
gem_write(fd, handle, PAGE_SIZE / 2, buf, PAGE_SIZE / 2);
gem_close(fd, handle);
}
static void
test_huge_bo(int huge, int tiling)
{
uint64_t huge_object_size, last_offset;
char *ptr_cpu;
char *cpu_pattern;
uint32_t bo;
int i;
switch (huge) {
case -1:
huge_object_size = gem_mappable_aperture_size() / 2;
break;
case 0:
huge_object_size = gem_mappable_aperture_size() + PAGE_SIZE;
break;
default:
huge_object_size = gem_aperture_size(fd) + PAGE_SIZE;
break;
}
intel_require_memory(1, huge_object_size, CHECK_RAM);
last_offset = huge_object_size - PAGE_SIZE;
cpu_pattern = malloc(PAGE_SIZE);
igt_assert(cpu_pattern);
for (i = 0; i < PAGE_SIZE; i++)
cpu_pattern[i] = i;
bo = gem_create(fd, huge_object_size);
/* Obtain CPU mapping for the object. */
ptr_cpu = gem_mmap__cpu(fd, bo, 0, huge_object_size,
PROT_READ | PROT_WRITE);
igt_assert(ptr_cpu);
gem_set_domain(fd, bo, I915_GEM_DOMAIN_CPU, I915_GEM_DOMAIN_CPU);
gem_close(fd, bo);
/* Write first page through the mapping and assert reading it back
* works. */
memcpy(ptr_cpu, cpu_pattern, PAGE_SIZE);
igt_assert(memcmp(ptr_cpu, cpu_pattern, PAGE_SIZE) == 0);
/* Write last page through the mapping and assert reading it back
* works. */
memcpy(ptr_cpu + last_offset, cpu_pattern, PAGE_SIZE);
igt_assert(memcmp(ptr_cpu + last_offset, cpu_pattern, PAGE_SIZE) == 0);
/* Cross check that accessing two simultaneous pages works. */
igt_assert(memcmp(ptr_cpu, ptr_cpu + last_offset, PAGE_SIZE) == 0);
munmap(ptr_cpu, huge_object_size);
free(cpu_pattern);
}
static int test_can_pin(int fd)
{
struct drm_i915_gem_pin pin;
int ret;
pin.handle = gem_create(fd, 4096);;
pin.alignment = 0;
ret = drmIoctl(fd, DRM_IOCTL_I915_GEM_PIN, &pin);
gem_close(fd, pin.handle);
return ret == 0;;
}
static uint32_t
batch_create (int fd)
{
uint32_t buf[] = { MI_BATCH_BUFFER_END, 0 };
uint32_t batch_handle;
batch_handle = gem_create(fd, BATCH_SIZE);
gem_write(fd, batch_handle, 0, buf, sizeof(buf));
return batch_handle;
}
static void run_on_ring(int fd, unsigned ring_id, const char *ring_name)
{
uint32_t handle, handle_new;
uint64_t gtt_offset, gtt_offset_new;
uint32_t *batch_ptr, *batch_ptr_old;
unsigned split;
char buf[100];
int i;
gem_require_ring(fd, ring_id);
sprintf(buf, "testing %s cs tlb coherency: ", ring_name);
/* Shut up gcc, too stupid. */
batch_ptr_old = NULL;
handle = 0;
gtt_offset = 0;
for (split = 0; split < BATCH_SIZE/8 - 1; split += 2) {
igt_progress(buf, split, BATCH_SIZE/8 - 1);
handle_new = gem_create(fd, BATCH_SIZE);
batch_ptr = gem_mmap__cpu(fd, handle_new, 0, BATCH_SIZE,
PROT_READ | PROT_WRITE);
batch_ptr[split*2] = MI_BATCH_BUFFER_END;
for (i = split*2 + 2; i < BATCH_SIZE/8; i++)
batch_ptr[i] = 0xffffffff;
if (split > 0) {
gem_sync(fd, handle);
gem_close(fd, handle);
}
igt_assert_eq(exec(fd, handle_new, split, >t_offset_new, 0),
0);
if (split > 0) {
/* Check that we've managed to collide in the tlb. */
igt_assert(gtt_offset == gtt_offset_new);
/* We hang onto the storage of the old batch by keeping
* the cpu mmap around. */
munmap(batch_ptr_old, BATCH_SIZE);
}
handle = handle_new;
gtt_offset = gtt_offset_new;
batch_ptr_old = batch_ptr;
}
}
static int prepare_primary_surface(int fd, int prim_width, int prim_height,
uint32_t *prim_handle, uint32_t *prim_stride,
uint32_t *prim_size, int tiled)
{
uint32_t bytes_per_pixel = sizeof(uint32_t);
uint32_t *prim_fb_ptr;
if (bytes_per_pixel != sizeof(uint32_t)) {
printf("Bad bytes_per_pixel for primary surface: %d\n",
bytes_per_pixel);
return -EINVAL;
}
if (tiled) {
int v;
/* Round the tiling up to the next power-of-two and the
* region up to the next pot fence size so that this works
* on all generations.
*
* This can still fail if the framebuffer is too large to
* be tiled. But then that failure is expected.
*/
v = prim_width * bytes_per_pixel;
for (*prim_stride = 512; *prim_stride < v; *prim_stride *= 2)
;
v = *prim_stride * prim_height;
for (*prim_size = 1024*1024; *prim_size < v; *prim_size *= 2)
;
} else {
/* Scan-out has a 64 byte alignment restriction */
*prim_stride = (prim_width * bytes_per_pixel + 63) & ~63;
*prim_size = *prim_stride * prim_height;
}
*prim_handle = gem_create(fd, *prim_size);
if (tiled)
gem_set_tiling(fd, *prim_handle, I915_TILING_X, *prim_stride);
prim_fb_ptr = gem_mmap(fd, *prim_handle, *prim_size, PROT_READ | PROT_WRITE);
if (prim_fb_ptr != NULL) {
// Write primary surface with gray background
memset(prim_fb_ptr, 0x3f, *prim_size);
munmap(prim_fb_ptr, *prim_size);
}
return 0;
}
static void *
create_pointer(int fd)
{
uint32_t handle;
void *ptr;
handle = gem_create(fd, OBJECT_SIZE);
ptr = mmap_bo(fd, handle);
gem_close(fd, handle);
return ptr;
}
static void
dontneed_before_mmap(void)
{
int fd = drm_open_driver(DRIVER_INTEL);
uint32_t handle;
char *ptr;
handle = gem_create(fd, OBJECT_SIZE);
gem_madvise(fd, handle, I915_MADV_DONTNEED);
ptr = __gem_mmap__gtt(fd, handle, OBJECT_SIZE, PROT_READ | PROT_WRITE);
igt_assert(ptr == NULL);
igt_assert(errno == EFAULT);
close(fd);
}