[igt-dev] [PATCH i-g-t v3 3/4] tests/i915/gem_mmap_offset: Add new API test for gem_mmap_offset
Zbigniew KempczyĆski
zbigniew.kempczynski at intel.com
Thu Nov 21 17:07:52 UTC 2019
On Wed, Nov 20, 2019 at 07:50:55PM +0000, Chris Wilson wrote:
> > +igt_main
> > +{
> > + uint8_t *addr;
> > + uint32_t obj_size, batch_size;
> > + uint32_t mem_type, mem_instance;
> > + uint32_t region;
> > + int fd;
> > + const struct intel_memory_region *mr;
> > + struct local_i915_query_memory_region_info *query_info;
> > +
> > + igt_fixture {
> > + fd = drm_open_driver(DRIVER_INTEL);
> > + gem_require_mmap_offset(fd);
> > + igt_require(gem_mmap__has_wc(fd));
> > +
> > + query_info = gem_query_memory_regions(fd);
> > + igt_assert(query_info);
>
> This ioctl will be available before memory regions uAPI, and we will be
> expecting coverage...
You mean i915_query.c test should be extended to do similar?
>
> gem_mmap_offset_regions
> or
> gem_memory_regions_mmap
>
> or just igt_subtest_group {}
>
> And do we have a plan for switching to something like i915_gem_ instead?
I tried to understand and I'm sorry I don't get what you mean.
>
> > + }
> > +
> > + for (mr = intel_memory_regions; mr->region_name; mr++) {
>
> See igt_subtest_with_dynamic
I wondered to provide macro similar to existing - for_each_memory_region.
I'm taking a look how to use dynamic test here.
> > + mem_type = mr->memory_type;
> > + mem_instance = mr->memory_instance;
> > + region = INTEL_MEMORY_REGION_ID(mem_type, mem_instance);
> > + batch_size = gem_get_batch_size(fd, region);
> > + obj_size = 4 * batch_size;
> > +
> > + igt_subtest_f("bad-object-%s", mr->region_name) {
> > + uint32_t real_handle;
> > + uint32_t handles[20];
> > + int i = 0;
> > +
> > + gem_query_require_region(query_info, region);
> > +
> > + real_handle = gem_create_in_memory_regions(fd, obj_size,
> > + region);
> > +
> > + handles[i++] = 0xdeadbeef;
> > + for (int bit = 0; bit < 16; bit++)
> > + handles[i++] = real_handle | (1 << (bit + 16));
> > + handles[i] = real_handle + 1;
> > +
> > + for (; i >= 0; i--) {
> > + struct local_i915_gem_mmap_offset arg = {
> > + .handle = handles[i],
> > + .flags = LOCAL_I915_MMAP_OFFSET_WC,
> > + };
> > +
> > + igt_debug("Trying MMAP IOCTL with handle %x\n",
> > + handles[i]);
> > + igt_assert_eq(mmap_offset_ioctl(fd, &arg),
> > + -ENOENT);
> > + }
> > +
> > + gem_close(fd, real_handle);
> > + }
> > +
> > + igt_subtest_f("basic-%s", mr->region_name) {
> > + uint32_t handle;
> > + uint64_t flags;
> > + uint8_t *expected;
> > + uint8_t *buf;
> > +
> > + gem_query_require_region(query_info, region);
> > +
> > + handle = gem_create_in_memory_regions(fd, obj_size,
> > + region);
> > + flags = LOCAL_I915_MMAP_OFFSET_WC;
> > + addr = __gem_mmap_offset(fd, handle, 0, obj_size,
> > + PROT_READ | PROT_WRITE,
> > + flags);
> > + igt_assert(addr);
> > +
> > + igt_debug("Testing contents of newly created object.\n");
> > + expected = calloc(obj_size, sizeof(*expected));
> > + igt_assert_eq(memcmp(addr, expected, obj_size), 0);
> > + free(expected);
> > +
> > + igt_debug("Testing coherency of writes and mmap reads.\n");
> > + buf = calloc(obj_size, sizeof(*buf));
> > + memset(buf + 1024, 0x01, 1024);
> > + gem_write(fd, handle, 0, buf, obj_size);
> > + igt_assert_eq(memcmp(buf, addr, obj_size), 0);
> > +
> > + igt_debug("Testing that mapping stays after close\n");
> > + gem_close(fd, handle);
> > + igt_assert_eq(memcmp(buf, addr, obj_size), 0);
> > + free(buf);
> > +
> > + igt_debug("Testing unmapping\n");
> > + munmap(addr, obj_size);
> > + }
> > +
> > + igt_subtest_f("short-mmap-%s", mr->region_name) {
> > + uint32_t handle;
>
> Not supported in vanilla ioctl, coming to an extension near you :)
> -Chris
Hmm, still not get it :/
Zbigniew
More information about the igt-dev
mailing list