[igt-dev] [PATCH] test/perf: Add support for TGL in perf tests
Umesh Nerlige Ramappa
umesh.nerlige.ramappa at intel.com
Wed Oct 30 23:40:30 UTC 2019
On Fri, Oct 25, 2019 at 09:14:35AM +0300, Lionel Landwerlin wrote:
>On 24/10/2019 22:22, Umesh Nerlige Ramappa wrote:
>>On Thu, Oct 24, 2019 at 02:22:51PM +0300, Lionel Landwerlin wrote:
>>>On 14/10/2019 21:57, Umesh Nerlige Ramappa wrote:
>>>>Add following changes to enable perf tests on TGL
>>>>- Support only a single OA format
>>>>- Add TGL metrics
>>>>- Update whitelist test case
>>>>- Cleanup mi-rpc test if it fails
>>>>- Skip unsupported test - gen8-unprivileged-single-ctx-counters
>>>>
>>>>Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa at intel.com>
>>>>---
>>>> tests/perf.c | 85 ++++++++++++++++++++++++++++++++++++++--------------
>>>> 1 file changed, 62 insertions(+), 23 deletions(-)
>>>>
>>>>diff --git a/tests/perf.c b/tests/perf.c
>>>>index 5ad8b2db..c5fc6878 100644
>>>>--- a/tests/perf.c
>>>>+++ b/tests/perf.c
>>>>@@ -159,6 +159,15 @@ static struct oa_format
>>>>gen8_oa_formats[I915_OA_FORMAT_MAX] = {
>>>> .b_off = 32, .n_b = 8, },
>>>> };
>>>>+static struct oa_format gen12_oa_formats[I915_OA_FORMAT_MAX] = {
>>>>+ [I915_OA_FORMAT_A32u40_A4u32_B8_C8] = {
>>>>+ "A32u40_A4u32_B8_C8", .size = 256,
>>>>+ .a40_high_off = 160, .a40_low_off = 16, .n_a40 = 32,
>>>>+ .a_off = 144, .n_a = 4, .first_a = 32,
>>>>+ .b_off = 192, .n_b = 8,
>>>>+ .c_off = 224, .n_c = 8, },
>>>>+};
>>>>+
>>>> static bool hsw_undefined_a_counters[45] = {
>>>> [4] = true,
>>>> [6] = true,
>>>>@@ -206,7 +215,10 @@ get_oa_format(enum drm_i915_oa_format format)
>>>> {
>>>> if (IS_HASWELL(devid))
>>>> return hsw_oa_formats[format];
>>>>- return gen8_oa_formats[format];
>>>>+ else if (IS_GEN12(devid))
>>>>+ return gen12_oa_formats[format];
>>>>+ else
>>>>+ return gen8_oa_formats[format];
>>>> }
>>>> static void
>>>>@@ -945,6 +957,8 @@ init_sys_info(void)
>>>> test_set_uuid = "db41edd4-d8e7-4730-ad11-b9a2d6833503";
>>>> } else if (IS_ICELAKE(devid)) {
>>>> test_set_uuid = "a291665e-244b-4b76-9b9a-01de9d3c8068";
>>>>+ } else if (IS_TIGERLAKE(devid)) {
>>>>+ test_set_uuid = "80a833f0-2504-4321-8894-e9277844ce7b";
>>>> } else {
>>>> igt_debug("unsupported GT\n");
>>>> return false;
>>>>@@ -2813,6 +2827,7 @@ test_mi_rpc(void)
>>>> drm_intel_bo *bo;
>>>> uint32_t *report32;
>>>> int ret;
>>>>+ uint32_t word0, word1, word63, word64;
>>>> stream_fd = __perf_open(drm_fd, ¶m, false);
>>>>@@ -2842,18 +2857,26 @@ test_mi_rpc(void)
>>>> igt_assert_eq(ret, 0);
>>>> report32 = bo->virtual;
>>>>- igt_assert_eq(report32[0], 0xdeadbeef); /* report ID */
>>>>- igt_assert_neq(report32[1], 0); /* timestamp */
>>>>-
>>>>- igt_assert_neq(report32[63], 0x80808080); /* end of report */
>>>>- igt_assert_eq(report32[64], 0x80808080); /* after 256 byte
>>>>report */
>>>>+ word0 = report32[0];
>>>>+ word1 = report32[1];
>>>>+ word63 = report32[63];
>>>>+ word64 = report32[64];
>>>>+ /* cleanup before you fail for any sanity checks so that
>>>>subsequent
>>>>+ * tests do not fail because of bad perf state.
>>>>+ */
>>>> drm_intel_bo_unmap(bo);
>>>> drm_intel_bo_unreference(bo);
>>>> intel_batchbuffer_free(batch);
>>>> drm_intel_gem_context_destroy(context);
>>>> drm_intel_bufmgr_destroy(bufmgr);
>>>> __perf_close(stream_fd);
>>>>+
>>>>+ igt_assert_eq(word0, 0xdeadbeef); /* report ID */
>>>>+ igt_assert_neq(word1, 0); /* timestamp */
>>>>+
>>>>+ igt_assert_neq(word63, 0x80808080); /* end of report */
>>>>+ igt_assert_eq(word64, 0x80808080); /* after 256 byte report */
>>>
>>>
>>>Any reason for this change? I can't see any functional change.
>>
>>When running the entire perf test suite, if this test fails/asserts,
>>all subsequent tests fail too (because perf_close is not called on
>>failure). To avoid that, I moved the checks to a later point
>>(temporarily, the code might need additional cleanup).
>
>
>That's strange, because __perf_open & __perf_close are supposed to
>deal with this.
>
True. Digging deeper, not all tests fail following mi-rpc. This is the
current sequence of tests in non-hsw platforms that fails.
mi-rpc
gen8-unprivileged-single-ctx-counters
the latter test forks out a child and calls perf_open in the child
process, so it tries to close the stream_fd from this child which will
not work. Hence the perf_open fails.
Also note that gen8-unprivileged-single-ctx-counters itself forks out
multiple childs if error is EAGAIN (say for delta_delta > 500). We may
need to cleanup the stream_fds appropriately if such an error occurs (or
maybe occured - https://bugs.freedesktop.org/show_bug.cgi?id=111821)
Anyways, I think I will back out the above change and post a new patch
to fix a potential failure in gen8-unprivileged-single-ctx-counters.
Thanks,
Umesh
>
>>
>>>
>>>
>>>> }
>>>> static void
>>>>@@ -3846,6 +3869,8 @@ test_whitelisted_registers_userspace_config(void)
>>>> uint32_t b_counters_regs[200];
>>>> uint32_t flex_regs[200];
>>>> uint32_t i;
>>>>+ uint32_t oa_start_trig1, oa_start_trig8;
>>>>+ uint32_t oa_report_trig1, oa_report_trig8;
>>>> uint64_t config_id;
>>>> char path[512];
>>>> int ret;
>>>>@@ -3869,14 +3894,26 @@
>>>>test_whitelisted_registers_userspace_config(void)
>>>> memset(&config, 0, sizeof(config));
>>>> memcpy(config.uuid, uuid, sizeof(config.uuid));
>>>>+ if (intel_gen(devid) >= 12) {
>>>>+ oa_start_trig1 = 0xd900;
>>>>+ oa_start_trig8 = 0xd91c;
>>>>+ oa_report_trig1 = 0xd920;
>>>>+ oa_report_trig8 = 0xd93c;
>>>>+ } else {
>>>>+ oa_start_trig1 = 0x2710;
>>>>+ oa_start_trig8 = 0x272c;
>>>>+ oa_report_trig1 = 0x2740;
>>>>+ oa_report_trig8 = 0x275c;
>>>>+ }
>>>>+
>>>> /* OASTARTTRIG[1-8] */
>>>>- for (i = 0x2710; i <= 0x272c; i += 4) {
>>>>+ for (i = oa_start_trig1; i <= oa_start_trig8; i += 4) {
>>>> b_counters_regs[config.n_boolean_regs * 2] = i;
>>>> b_counters_regs[config.n_boolean_regs * 2 + 1] = 0;
>>>> config.n_boolean_regs++;
>>>> }
>>>> /* OAREPORTTRIG[1-8] */
>>>>- for (i = 0x2740; i <= 0x275c; i += 4) {
>>>>+ for (i = oa_report_trig1; i <= oa_report_trig8; i += 4) {
>>>> b_counters_regs[config.n_boolean_regs * 2] = i;
>>>> b_counters_regs[config.n_boolean_regs * 2 + 1] = 0;
>>>> config.n_boolean_regs++;
>>>>@@ -3896,10 +3933,6 @@
>>>>test_whitelisted_registers_userspace_config(void)
>>>> /* Mux registers (too many of them, just checking bounds) */
>>>> i = 0;
>>>>- /* NOA_WRITE */
>>>>- mux_regs[i++] = 0x9800;
>>>>- mux_regs[i++] = 0;
>>>>-
>>>> if (IS_HASWELL(devid)) {
>>>> /* Haswell specific. undocumented... */
>>>> mux_regs[i++] = 0x9ec0;
>>>>@@ -3922,10 +3955,6 @@
>>>>test_whitelisted_registers_userspace_config(void)
>>>> mux_regs[i++] = 0;
>>>> }
>>>>- /* HALF_SLICE_CHICKEN2 (shared with kernel workaround) */
>>>>- mux_regs[i++] = 0xE180;
>>>>- mux_regs[i++] = 0;
>>>>-
>>>> if (IS_CHERRYVIEW(devid)) {
>>>> /* Cherryview specific. undocumented... */
>>>> mux_regs[i++] = 0x182300;
>>>>@@ -3934,12 +3963,20 @@
>>>>test_whitelisted_registers_userspace_config(void)
>>>> mux_regs[i++] = 0;
>>>> }
>>>>- /* PERFCNT[12] */
>>>>- mux_regs[i++] = 0x91B8;
>>>>- mux_regs[i++] = 0;
>>>>- /* PERFMATRIX */
>>>>- mux_regs[i++] = 0x91C8;
>>>>- mux_regs[i++] = 0;
>>>>+ if (intel_gen(devid) <= 11) {
>>>>+ /* NOA_WRITE */
>>>>+ mux_regs[i++] = 0x9800;
>>>>+ mux_regs[i++] = 0;
>>>>+ /* HALF_SLICE_CHICKEN2 (shared with kernel workaround) */
>>>>+ mux_regs[i++] = 0xE180;
>>>>+ mux_regs[i++] = 0;
>>>>+ /* PERFCNT[12] */
>>>>+ mux_regs[i++] = 0x91B8;
>>>>+ mux_regs[i++] = 0;
>>>>+ /* PERFMATRIX */
>>>>+ mux_regs[i++] = 0x91C8;
>>>>+ mux_regs[i++] = 0;
>>>>+ }
>>>> config.mux_regs_ptr = (uintptr_t) mux_regs;
>>>> config.n_mux_regs = i / 2;
>>>>@@ -4170,8 +4207,10 @@ igt_main
>>>> * functionality to HW filter timer reports for a specific
>>>> * context (SKL+) can't stop multiple applications viewing
>>>> * system-wide data via MI_REPORT_PERF_COUNT commands.
>>>>+ *
>>>>+ * For gen12 implement a separate test that uses only OAR
>>>
>>>
>>>Where is that test? :)
>>
>>:) wip. I gave it a shot - removed the code that collects samples
>>from oa buffer in this test, but it asserts later trying to compare
>>the counters accumulated. I need to debug that. I will post that in
>>another series/patch.
>>
>>Thanks,
>>Umesh
>
>
>Cool, I was looking forward to see a test that at least verifies the
>save/restore of the counters.
>
>Pretty much all of the 3d pipeline counters (A counters) should not
>progress.
>
>
>-Lionel
>
>
>>
>>>
>>>
>>>-Lionel
>>>
>>>
>>>> */
>>>>- igt_require(intel_gen(devid) >= 8);
>>>>+ igt_require(intel_gen(devid) >= 8 && intel_gen(devid) < 12);
>>>>gen8_test_single_ctx_render_target_writes_a_counter();
>>>> }
>>>
>>>
>>
>
More information about the igt-dev
mailing list