From 8f3f862b961c5b28ccdb42faff7b279aafcec652 Mon Sep 17 00:00:00 2001 From: Chris Wilson Date: Tue, 1 Sep 2009 10:09:55 +0100 Subject: Add gem_fence_thrash test This is a test case that overcommits fence registers between threads, which are copying from one fenced bo to another. In earlier versions of the driver this would cause excessive spinning as the first inactive (i.e. not in use by the GPU) would be used to service the next page. After all the fence registers had been allocated, in effect only the very first fence would then be used for all subsequent faults. --- .gitignore | 1 + tests/Makefile.am | 6 +- tests/gem_fence_thrash.c | 148 +++++++++++++++++++++++++++++++++++++++++++++++ 3 files changed, 154 insertions(+), 1 deletion(-) create mode 100644 tests/gem_fence_thrash.c diff --git a/.gitignore b/.gitignore index 6b3a4520..86d054d4 100644 --- a/.gitignore +++ b/.gitignore @@ -38,6 +38,7 @@ tests/getclient tests/getstats tests/getversion tests/gem_basic +tests/gem_fence_thrash tests/gem_flink tests/gem_largeobject tests/gem_mmap diff --git a/tests/Makefile.am b/tests/Makefile.am index c6769dd5..e9f56310 100644 --- a/tests/Makefile.am +++ b/tests/Makefile.am @@ -13,7 +13,8 @@ TESTS = getversion \ gem_bad_address \ gem_bad_blit \ gem_bad_batch \ - gem_hang + gem_hang \ + gem_fence_thrash EXTRA_PROGRAMS = $(TESTS) CLEANFILES = $(EXTRA_PROGRAMS) @@ -22,3 +23,6 @@ AM_CFLAGS = $(DRM_CFLAGS) $(WARN_CFLAGS) \ -I$(srcdir)/.. \ -I$(srcdir)/../lib LDADD = ../lib/libintel_tools.la $(PCIACCESS_LIBS) $(DRM_LIBS) + +gem_fence_thrash_CFLAGS = $(AM_CFLAGS) -pthread +gem_fence_thrash_LDADD = $(LDADD) -lpthread diff --git a/tests/gem_fence_thrash.c b/tests/gem_fence_thrash.c new file mode 100644 index 00000000..8f2203c5 --- /dev/null +++ b/tests/gem_fence_thrash.c @@ -0,0 +1,148 @@ +/* + * Copyright © 2008-9 Intel Corporation + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS + * IN THE SOFTWARE. + * + * Authors: + * Eric Anholt + * Chris Wilson + * + */ + +#define _GNU_SOURCE + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include "drm.h" +#include "i915_drm.h" +#include "drmtest.h" + +#define OBJECT_SIZE (1024*1024) /* restricted to 1MiB alignment on i915 fences */ + +/* Before introduction of the LRU list for fences, allocation of a fence for a page + * fault would use the first inactive fence (i.e. in preference one with no outstanding + * GPU activity, or it would wait on the first to finish). Given the choice, it would simply + * reuse the fence that had just been allocated for the previous page-fault - the worst choice + * when copying between two buffers and thus constantly swapping fences. + */ + +static void * +bo_create (int fd) +{ + struct drm_i915_gem_create create; + struct drm_i915_gem_set_tiling tiling; + struct drm_i915_gem_set_domain domain; + struct drm_i915_gem_mmap_gtt mmap_arg; + void *ptr; + int handle; + int ret; + + memset(&create, 0, sizeof(create)); + create.size = OBJECT_SIZE; + ret = ioctl(fd, DRM_IOCTL_I915_GEM_CREATE, &create); + assert(ret == 0); + handle = create.handle; + + memset(&tiling, 0, sizeof(tiling)); + tiling.handle = handle; + tiling.tiling_mode = I915_TILING_X; + tiling.stride = 1024; + ret = ioctl(fd, DRM_IOCTL_I915_GEM_SET_TILING, &tiling); + assert(ret == 0); + + memset(&mmap_arg, 0, sizeof(mmap_arg)); + mmap_arg.handle = handle; + + /* Get the fake offset back... */ + ret = ioctl(fd, DRM_IOCTL_I915_GEM_MMAP_GTT, &mmap_arg); + assert (ret == 0); + ptr = mmap64(0, OBJECT_SIZE, PROT_READ | PROT_WRITE, + MAP_SHARED, fd, mmap_arg.offset); + assert (ptr != MAP_FAILED); + + /* XXX: mmap_gtt pulls the bo into the GTT read domain. */ + memset(&domain, 0, sizeof(domain)); + domain.handle = handle; + domain.read_domains = I915_GEM_DOMAIN_GTT; + domain.write_domain = I915_GEM_DOMAIN_GTT; + ret = ioctl(fd, DRM_IOCTL_I915_GEM_SET_DOMAIN, &domain); + assert (ret == 0); + + return ptr; +} + +static void * +bo_copy (void *_arg) +{ + int fd = *(int *)_arg; + int offset = 0, n; + char *a, *b; + + a = bo_create (fd); + b = bo_create (fd); + + for (n = 0; n < OBJECT_SIZE; n++) { + memcpy (a + offset, b + offset, 1); + pthread_yield (); + offset += 4097; + offset %= OBJECT_SIZE; + } + + return NULL; +} + +int +main(int argc, char **argv) +{ + drm_i915_getparam_t gp; + pthread_t threads[32]; + int n, num_fences; + int fd, ret; + + fd = drm_open_any(); + + gp.param = I915_PARAM_NUM_FENCES_AVAIL; + gp.value = &num_fences; + ret = ioctl(fd, DRM_IOCTL_I915_GETPARAM, &gp); + assert (ret == 0); + + printf ("creating %d threads\n", num_fences); + assert (num_fences < sizeof (threads) / sizeof (threads[0])); + + for (n = 0; n < num_fences; n++) + pthread_create (&threads[n], NULL, bo_copy, &fd); + + for (n = 0; n < num_fences; n++) + pthread_join (threads[n], NULL); + + close(fd); + + return 0; +} -- cgit v1.2.3