Message ID | 20190911101501.7182-1-chris@chris-wilson.co.uk (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
Series | [i-g-t,1/3] i915/gem_eio: Race kms on/off vs reset | expand |
Hi Chris, On Wed, Sep 11, 2019 at 11:14:59AM +0100, Chris Wilson wrote: > On older platforms, performing a GPU reset clobbers the display. > Exercise the interactions between reset/wedge and the display and > hopefully prevent any races creeping in. > > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> > Cc: Ville Syrjälä <ville.syrjala@linux.intel.com> Is Ville Cc'ed for real or just nominally? looks straight forward, Reviewed-by: Andi Shyti <andi.shyti@intel.com> Andi
Quoting Andi Shyti (2019-09-19 15:08:57) > Hi Chris, > > On Wed, Sep 11, 2019 at 11:14:59AM +0100, Chris Wilson wrote: > > On older platforms, performing a GPU reset clobbers the display. > > Exercise the interactions between reset/wedge and the display and > > hopefully prevent any races creeping in. > > > > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> > > Cc: Ville Syrjälä <ville.syrjala@linux.intel.com> > > Is Ville Cc'ed for real or just nominally? He's in the original cc. As it is using "Syrjälä", that means I copied his email address from one of his commits, so it should be the right address. mailman naturally drops cc of list members... -Chris
On Wed, Sep 11, 2019 at 11:14:59AM +0100, Chris Wilson wrote: > On older platforms, performing a GPU reset clobbers the display. > Exercise the interactions between reset/wedge and the display and > hopefully prevent any races creeping in. > > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> > Cc: Ville Syrjälä <ville.syrjala@linux.intel.com> > --- > tests/i915/gem_eio.c | 79 ++++++++++++++++++++++++++++++++++++++++++++ > 1 file changed, 79 insertions(+) > > diff --git a/tests/i915/gem_eio.c b/tests/i915/gem_eio.c > index 9b086a039..e0213c76c 100644 > --- a/tests/i915/gem_eio.c > +++ b/tests/i915/gem_eio.c > @@ -42,6 +42,8 @@ > > #include "igt.h" > #include "igt_device.h" > +#include "igt_fb.h" > +#include "igt_kms.h" > #include "igt_stats.h" > #include "igt_sysfs.h" > #include "sw_sync.h" > @@ -813,6 +815,67 @@ static void test_reset_stress(int fd, unsigned int flags) > gem_context_destroy(fd, ctx0); > } > > +/* > + * Modesetting vs wedged > + */ > + > +static void display_helper(igt_display_t *dpy, int *done) > +{ > + const int commit = dpy->is_atomic ? COMMIT_ATOMIC : COMMIT_LEGACY; > + struct igt_fb fb = {}; > + > + while (!READ_ONCE(*done)) { > + drmModeModeInfoPtr mode; > + igt_plane_t *primary; > + igt_output_t *output; > + int pipe; > + > + pipe = rand() % dpy->n_pipes; I guess this ends up doing nothing if we get the same pipe twice in a row? Hmm, I guess it does actually do something but since nothing significant would change the kernel likely just downgrades it to a fastset. But maybe that's fine. There are some reset vs. modeset type of tests in kms_flip, but those are rather more carefully sequenced so they are probably not so great at finding new races. Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com> > + output = igt_get_single_output_for_pipe(dpy, pipe); > + if (!output) > + continue; > + > + igt_output_set_pipe(output, pipe); > + mode = igt_output_get_mode(output); > + > + if (fb.width != mode->hdisplay || fb.height != mode->vdisplay) { > + igt_remove_fb(dpy->drm_fd, &fb); > + igt_create_pattern_fb(dpy->drm_fd, > + mode->hdisplay, mode->vdisplay, > + DRM_FORMAT_XRGB8888, > + LOCAL_I915_FORMAT_MOD_X_TILED, > + &fb); > + } > + > + primary = igt_output_get_plane_type(output, > + DRM_PLANE_TYPE_PRIMARY); > + igt_plane_set_fb(primary, &fb); > + > + igt_display_commit2(dpy, commit); > + igt_display_reset(dpy); > + } > + > + igt_remove_fb(dpy->drm_fd, &fb); > +} > + > +static void test_kms(int i915, igt_display_t *dpy) > +{ > + int *shared; > + > + shared = mmap(NULL, 4096, PROT_WRITE, MAP_SHARED | MAP_ANON, -1, 0); > + igt_assert(shared != MAP_FAILED); > + > + igt_fork(child, 1) > + display_helper(dpy, shared); > + > + test_reset_stress(i915, 0); > + test_reset_stress(i915, TEST_WEDGE); > + > + *shared = 1; > + igt_waitchildren(); > + munmap(shared, 4096); > +} > + > static int fd = -1; > > static void > @@ -906,4 +969,20 @@ igt_main > } > } > } > + > + igt_subtest_group { > + igt_display_t display = { > + .drm_fd = -1, .n_pipes = IGT_MAX_PIPES > + }; > + > + igt_fixture { > + igt_device_set_master(fd); > + > + igt_display_require(&display, fd); > + igt_display_require_output(&display); > + } > + > + igt_subtest("kms") > + test_kms(fd, &display); > + } > } > -- > 2.23.0
diff --git a/tests/i915/gem_eio.c b/tests/i915/gem_eio.c index 9b086a039..e0213c76c 100644 --- a/tests/i915/gem_eio.c +++ b/tests/i915/gem_eio.c @@ -42,6 +42,8 @@ #include "igt.h" #include "igt_device.h" +#include "igt_fb.h" +#include "igt_kms.h" #include "igt_stats.h" #include "igt_sysfs.h" #include "sw_sync.h" @@ -813,6 +815,67 @@ static void test_reset_stress(int fd, unsigned int flags) gem_context_destroy(fd, ctx0); } +/* + * Modesetting vs wedged + */ + +static void display_helper(igt_display_t *dpy, int *done) +{ + const int commit = dpy->is_atomic ? COMMIT_ATOMIC : COMMIT_LEGACY; + struct igt_fb fb = {}; + + while (!READ_ONCE(*done)) { + drmModeModeInfoPtr mode; + igt_plane_t *primary; + igt_output_t *output; + int pipe; + + pipe = rand() % dpy->n_pipes; + output = igt_get_single_output_for_pipe(dpy, pipe); + if (!output) + continue; + + igt_output_set_pipe(output, pipe); + mode = igt_output_get_mode(output); + + if (fb.width != mode->hdisplay || fb.height != mode->vdisplay) { + igt_remove_fb(dpy->drm_fd, &fb); + igt_create_pattern_fb(dpy->drm_fd, + mode->hdisplay, mode->vdisplay, + DRM_FORMAT_XRGB8888, + LOCAL_I915_FORMAT_MOD_X_TILED, + &fb); + } + + primary = igt_output_get_plane_type(output, + DRM_PLANE_TYPE_PRIMARY); + igt_plane_set_fb(primary, &fb); + + igt_display_commit2(dpy, commit); + igt_display_reset(dpy); + } + + igt_remove_fb(dpy->drm_fd, &fb); +} + +static void test_kms(int i915, igt_display_t *dpy) +{ + int *shared; + + shared = mmap(NULL, 4096, PROT_WRITE, MAP_SHARED | MAP_ANON, -1, 0); + igt_assert(shared != MAP_FAILED); + + igt_fork(child, 1) + display_helper(dpy, shared); + + test_reset_stress(i915, 0); + test_reset_stress(i915, TEST_WEDGE); + + *shared = 1; + igt_waitchildren(); + munmap(shared, 4096); +} + static int fd = -1; static void @@ -906,4 +969,20 @@ igt_main } } } + + igt_subtest_group { + igt_display_t display = { + .drm_fd = -1, .n_pipes = IGT_MAX_PIPES + }; + + igt_fixture { + igt_device_set_master(fd); + + igt_display_require(&display, fd); + igt_display_require_output(&display); + } + + igt_subtest("kms") + test_kms(fd, &display); + } }
On older platforms, performing a GPU reset clobbers the display. Exercise the interactions between reset/wedge and the display and hopefully prevent any races creeping in. Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Ville Syrjälä <ville.syrjala@linux.intel.com> --- tests/i915/gem_eio.c | 79 ++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 79 insertions(+)