Return-Path: From: John.C.Harrison@Intel.com Date: Fri, 14 Jan 2022 10:17:03 -0800 Message-Id: <20220114181713.1798514-6-John.C.Harrison@Intel.com> In-Reply-To: <20220114181713.1798514-1-John.C.Harrison@Intel.com> References: <20220114181713.1798514-1-John.C.Harrison@Intel.com> MIME-Version: 1.0 Content-Transfer-Encoding: 8bit Subject: [igt-dev] [PATCH v5 i-g-t 05/15] tests/i915/i915_hangman: Add uevent test & fix detector List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: igt-dev-bounces@lists.freedesktop.org Sender: "igt-dev" To: IGT-Dev@Lists.FreeDesktop.Org Cc: Intel-GFX@Lists.FreeDesktop.Org List-ID: From: John Harrison Some of the IGT framework relies on receving a uevent when a hang occurs. So add a test that this actually works. While testing this, noticed that hangs could sometimes be missed because the uevent was (presumably) still in flight by the time the handler was de-registered. So add an extra delay during cleanup to give the uevent chance to arrive. Signed-off-by: John Harrison Reviewed-by: Matthew Brost --- lib/igt_aux.c | 7 +++++++ tests/i915/i915_hangman.c | 43 +++++++++++++++++++++++++++++++++++++++ 2 files changed, 50 insertions(+) diff --git a/lib/igt_aux.c b/lib/igt_aux.c index c247a1aa4..03cc38c93 100644 --- a/lib/igt_aux.c +++ b/lib/igt_aux.c @@ -523,6 +523,13 @@ void igt_fork_hang_detector(int fd) void igt_stop_hang_detector(void) { + /* + * Give the uevent time to arrive. No sleep at all misses about 20% of + * hangs (at least, in the i915_hangman/detector test). A sleep of 1ms + * seems to miss about 2%, 10ms loses <1%, so 100ms should be safe. + */ + usleep(100 * 1000); + igt_stop_helper(&hang_detector); } diff --git a/tests/i915/i915_hangman.c b/tests/i915/i915_hangman.c index 7b8390a6c..354769f39 100644 --- a/tests/i915/i915_hangman.c +++ b/tests/i915/i915_hangman.c @@ -31,6 +31,7 @@ #include #include #include +#include #include "i915/gem.h" #include "i915/gem_create.h" @@ -289,6 +290,38 @@ test_engine_hang(const intel_ctx_t *ctx, put_ahnd(ahnd); } +static int hang_count; + +static void sig_io(int sig) +{ + hang_count++; +} + +static void test_hang_detector(const intel_ctx_t *ctx, + const struct intel_execution_engine2 *e) +{ + igt_hang_t hang; + uint64_t ahnd = get_reloc_ahnd(device, ctx->id); + + hang_count = 0; + + igt_fork_hang_detector(device); + + /* Steal the signal handler */ + signal(SIGIO, sig_io); + + /* Make a hang... */ + hang = igt_hang_ctx_with_ahnd(device, ahnd, ctx->id, e->flags, 0); + + igt_post_hang_ring(device, hang); + put_ahnd(ahnd); + + igt_stop_hang_detector(); + + /* Did it work? */ + igt_assert(hang_count == 1); +} + /* This test covers the case where we end up in an uninitialised area of the * ppgtt and keep executing through it. This is particularly relevant if 48b * ppgtt is enabled because the ppgtt is massively bigger compared to the 32b @@ -408,6 +441,16 @@ igt_main igt_subtest("hangcheck-unterminated") hangcheck_unterminated(ctx); + igt_describe("Check that hang detector works"); + igt_subtest_with_dynamic("detector") { + const struct intel_execution_engine2 *e; + + for_each_ctx_engine(device, ctx, e) { + igt_dynamic_f("%s", e->name) + test_hang_detector(ctx, e); + } + } + do_tests("GT", "gt", ctx); igt_fixture { -- 2.25.1