@@ -53,6 +53,7 @@
* SUBTEST: basic-uaf
* SUBTEST: blt-coherency
* SUBTEST: clear
+ * SUBTEST: clear-via-pagefault
* SUBTEST: close-race
* SUBTEST: isolation
* SUBTEST: mmap-boundaries
@@ -720,6 +721,8 @@ struct thread_clear {
struct drm_i915_gem_memory_class_instance region;
int timeout;
int i915;
+ unsigned flags;
+#define CLEAR_IN_EXECBUF 0x1 << 0
};
static void *thread_clear(void *data)
@@ -743,8 +746,8 @@ static void *thread_clear(void *data)
size = npages << 12;
igt_assert_eq(__gem_create_in_memory_region_list(i915, &handle, &size, 0, &arg->region, 1), 0);
- /* Zero-init bo in execbuf or pagefault handler path randomly */
- if (random() & 1)
+ /* Zero-init bo in execbuf or pagefault handler path as requested */
+ if (arg->flags & CLEAR_IN_EXECBUF)
make_resident(i915, batch, handle);
ptr = __mmap_offset(i915, handle, 0, size,
@@ -784,13 +787,14 @@ static void *thread_clear(void *data)
return (void *)(uintptr_t)checked;
}
-static void always_clear(int i915, const struct gem_memory_region *r, int timeout)
+static void always_clear(int i915, const struct gem_memory_region *r, int timeout, unsigned flags)
{
struct thread_clear arg = {
.i915 = i915,
.region = r->ci,
.max = r->cpu_size / 2 >> 12, /* in pages */
.timeout = timeout,
+ .flags = flags,
};
const int ncpus = sysconf(_SC_NPROCESSORS_ONLN);
unsigned long checked;
@@ -1200,7 +1204,14 @@ igt_main
igt_subtest_with_dynamic("clear") {
for_each_memory_region(r, i915) {
igt_dynamic_f("%s", r->name)
- always_clear(i915, r, 20);
+ always_clear(i915, r, 20, CLEAR_IN_EXECBUF);
+ }
+ }
+
+ igt_subtest_with_dynamic("clear-via-pagefault") {
+ for_each_memory_region(r, i915) {
+ igt_dynamic_f("%s", r->name)
+ always_clear(i915, r, 20, 0);
}
}
Commit e25913a1a79d ("i915/gem_mmap_offset: Ignore ENOSPC error for making residency execbuf") not only resolved the issue of unnecessary failures on ENOSPC errors, but also introduced an alternative method of clearing memory of an object, with random selection of one of those methods on each iteration. The new method expects the memory to be cleared in pagefault handler path. Since those two methods may give different results, mixing them in one exercise may make the picture unclear, especially if something goes wrong. Add a new subtest that exercises the pagefault method exclusively. Link: https://gitlab.freedesktop.org/drm/i915/kernel/-/issues/11738 Signed-off-by: Janusz Krzysztofik <janusz.krzysztofik@linux.intel.com> --- tests/intel/gem_mmap_offset.c | 19 +++++++++++++++---- 1 file changed, 15 insertions(+), 4 deletions(-)