IGT updates for upstreaming prep
cover IGT updates for upstreaming prep
Main changes:
- Add tests for NULL bindings (VK Sparse)
- Test new GPUVA features (MMAP style binds, avoiding rebinds, doing rebinds)
- Update to VM bind uAPI + error handling flow
KMD series associated with these changes:
https://patchwork.freedesktop.org/series/117156/
Matthew Brost (5):
xe: Update to latest uAPI
xe_exec_basic: Add NULL VM bind section
xe_vm: MMAP style VM binds section
xe_vm: Add mmap / munmap sections that split large pages
xe: Update to new VM bind uAPI
include/drm-uapi/xe_drm.h | 106 ++---- lib/intel_batchbuffer.c | 2 +- lib/xe/xe_compute.c | 2 +- lib/xe/xe_ioctl.c | 15 +- lib/xe/xe_ioctl.h | 2 +- lib/xe/xe_query.c | 2 +- tests/xe/xe_create.c | 2 +- tests/xe/xe_evict.c | 23 +- tests/xe/xe_exec_balancer.c | 6 +- tests/xe/xe_exec_basic.c | 38 ++- tests/xe/xe_exec_compute_mode.c | 6 +- tests/xe/xe_exec_fault_mode.c | 6 +- tests/xe/xe_exec_reset.c | 8 +- tests/xe/xe_exec_threads.c | 112 ++---- tests/xe/xe_guc_pc.c | 2 +- tests/xe/xe_huc_copy.c | 2 +- tests/xe/xe_intel_bb.c | 2 +- tests/xe/xe_pm.c | 2 +- tests/xe/xe_vm.c | 583 ++++++++++++++++++++++---------- tests/xe/xe_waitfence.c | 18 +- 20 files changed, 532 insertions(+), 407 deletions(-)
Signature · 1 line
2.34.1
PATCH 1/5 xe: Update to latest uAPI
Includes NULL bindings support.
--- include/drm-uapi/xe_drm.h | 27 ++++++++++++++++++--------- 1 file changed, 18 insertions(+), 9 deletions(-)
include/drm-uapi/xe_drm.h +18 -9
@@ -37,7 +37,7 @@ extern "C" {*/ /** -* struct i915_user_extension - Base class for defining a chain of extensions +* struct xe_user_extension - Base class for defining a chain of extensions * * Many interfaces need to grow over time. In most cases we can simply * extend the struct and have userspace pass in more data. Another option,
@@ -55,20 +55,20 @@ extern "C" {* * .. code-block:: C * -* struct i915_user_extension ext3 { +* struct xe_user_extension ext3 { * .next_extension = 0, // end * .name = ..., * }; -* struct i915_user_extension ext2 { +* struct xe_user_extension ext2 { * .next_extension = (uintptr_t)&ext3, * .name = ..., * }; -* struct i915_user_extension ext1 { +* struct xe_user_extension ext1 { * .next_extension = (uintptr_t)&ext2, * .name = ..., * }; * -* Typically the struct i915_user_extension would be embedded in some uAPI +* Typically the struct xe_user_extension would be embedded in some uAPI * struct, and in this case we would feed it the head of the chain(i.e ext1), * which would then apply all of the above extensions. *
@@ -77,7 +77,7 @@ struct xe_user_extension {/** * @next_extension: * -* Pointer to the next struct i915_user_extension, or zero if the end. +* Pointer to the next struct xe_user_extension, or zero if the end. */ __u64 next_extension;/**
@@ -87,7 +87,7 @@ struct xe_user_extension {* * Also note that the name space for this is not global for the whole * driver, but rather its scope/meaning is limited to the specific piece -* of uAPI which has embedded the struct i915_user_extension. +* of uAPI which has embedded the struct xe_user_extension. */ __u32 name;/**
@@ -99,7 +99,7 @@ struct xe_user_extension { };/* -* i915 specific ioctls. +* xe specific ioctls. * * The device specific ioctl range is [DRM_COMMAND_BASE, DRM_COMMAND_END) ie * [0x40, 0xa0) (a0 is excluded). The numbers below are defined as offset
@@ -184,7 +184,8 @@ struct drm_xe_query_config {#define XE_QUERY_CONFIG_VA_BITS 3#define XE_QUERY_CONFIG_GT_COUNT 4#define XE_QUERY_CONFIG_MEM_REGION_COUNT 5 -#define XE_QUERY_CONFIG_NUM_PARAM XE_QUERY_CONFIG_MEM_REGION_COUNT + 1 +#define XE_QUERY_CONFIG_MAX_ENGINE_PRIORITY 6 +#define XE_QUERY_CONFIG_NUM_PARAM XE_QUERY_CONFIG_MAX_ENGINE_PRIORITY + 1__u64 info[]; };
@@ -446,6 +447,14 @@ struct drm_xe_vm_bind_op {* than differing the MAP to the page fault handler. */ #define XE_VM_BIND_FLAG_IMMEDIATE (0x1 << 18) +/* +* When the NULL flag is set, the page tables are setup with a special +* bit which indicates writes are dropped and all reads return zero. The +* NULL flags is only valid for XE_VM_BIND_OP_MAP operations, the BO +* handle MBZ, and the BO offset MBZ. This flag is intended to implement +* VK sparse bindings. +*/ +#define XE_VM_BIND_FLAG_NULL (0x1 << 19)/** @reserved: Reserved */ __u64 reserved[2 ];
Signature · 1 line
2.34.1
PATCH 2/5 xe_exec_basic: Add NULL VM bind section
In reply to Matthew BrostA NULL VM bind results in writes being dropped and reads returning zero, verify uAPI for NULL VM binds is working as designed.
--- tests/xe/xe_exec_basic.c | 33 ++++++++++++++++++++++++++++----- 1 file changed, 28 insertions(+), 5 deletions(-)
tests/xe/xe_exec_basic.c +28 -5
@@ -27,6 +27,7 @@#define BIND_ENGINE (0x1 << 4)#define DEFER_ALLOC (0x1 << 5)#define DEFER_BIND (0x1 << 6) +#define SPARSE (0x1 << 7)/** * SUBTEST: once-%s
@@ -67,6 +68,10 @@* @userptr-rebind: userptr rebind * @userptr-invalidate: userptr invalidate * @userptr-invalidate-race: userptr invalidate racy +* @null: null +* @null-defer-mmap: null defer mmap +* @null-defer-bind: null defer bind +* @null-rebind: null rebind * @bindengine: bind engine * @bindengine-userptr: bind engine userptr description * @bindengine-rebind: bind engine rebind description
@@ -89,6 +94,7 @@ test_exec(int fd, struct drm_xe_engine_class_instance *eci, .syncs = to_user_pointer(sync), };uint64_t addr[MAX_N_ENGINES ]; +uint64_t sparse_addr[MAX_N_ENGINES ];uint32_t vm[MAX_N_ENGINES ];uint32_t engines[MAX_N_ENGINES ];uint32_t bind_engines[MAX_N_ENGINES ];
@@ -113,8 +119,11 @@ test_exec(int fd, struct drm_xe_engine_class_instance *eci, xe_get_default_alignment(fd)); addr[0 ] =0x1a0000 ; -for (i =1 ; i <MAX_N_ENGINES ; ++i) + sparse_addr[0 ] =0x301a0000 ; +for (i =1 ; i <MAX_N_ENGINES ; ++i) { addr[i] = addr[i -1 ] + (0x1ull <<32 ); + sparse_addr[i] = sparse_addr[i -1 ] + (0x1ull <<32 ); + }if (flags &USERPTR ) {#define MAP_ADDRESS 0x00007fadeadbe000
@@ -163,6 +172,13 @@ test_exec(int fd, struct drm_xe_engine_class_instance *eci,xe_vm_bind_userptr_async (fd, vm[i], bind_engines[i],to_user_pointer (data), addr[i], bo_size, sync,1 ); +if (flags &SPARSE ) +__xe_vm_bind_assert (fd, vm[i], bind_engines[i], +0 ,0 , sparse_addr[i], bo_size, +XE_VM_BIND_OP_MAP | +XE_VM_BIND_FLAG_ASYNC | +XE_VM_BIND_FLAG_NULL , sync, +1 ,0 ,0 ); }if (flags &DEFER_BIND )
@@ -174,7 +190,8 @@ test_exec(int fd, struct drm_xe_engine_class_instance *eci,uint64_t batch_offset = (char *)&data[i].batch - (char *)data;uint64_t batch_addr = __addr + batch_offset;uint64_t sdi_offset = (char *)&data[i].data - (char *)data; -uint64_t sdi_addr = __addr + sdi_offset; +uint64_t sdi_addr = (flags &SPARSE ? sparse_addr[i % n_vm] : + __addr)+ sdi_offset;int e = i % n_engines; b =0 ;
@@ -261,9 +278,11 @@ test_exec(int fd, struct drm_xe_engine_class_instance *eci, INT64_MAX, 0, NULL)); } -for (i = (flags &INVALIDATE && n_execs) ? n_execs -1 :0 ; - i < n_execs; i++) -igt_assert_eq (data[i].data,0xc0ffee ); +if (!(flags &SPARSE )) { +for (i = (flags &INVALIDATE && n_execs) ? n_execs -1 :0 ; + i < n_execs; i++) +igt_assert_eq (data[i].data,0xc0ffee ); + }for (i =0 ; i < n_engines; i++) {syncobj_destroy (fd, syncobjs[i]);
@@ -296,6 +315,10 @@ igt_main { "basic-defer-bind", DEFER_ALLOC | DEFER_BIND }, { "userptr", USERPTR }, { "rebind", REBIND }, + { "null", SPARSE }, + { "null-defer-mmap", SPARSE | DEFER_ALLOC }, + { "null-defer-bind", SPARSE | DEFER_ALLOC | DEFER_BIND }, + { "null-rebind", SPARSE | REBIND }, { "userptr-rebind", USERPTR | REBIND }, { "userptr-invalidate", USERPTR | INVALIDATE }, { "userptr-invalidate-race", USERPTR | INVALIDATE | RACE },
Signature · 1 line
2.34.1
PATCH 3/5 xe_vm: MMAP style VM binds section
In reply to Matthew BrostGPUVA added support for this let's test it.
--- tests/xe/xe_vm.c | 350 ++++++++++++++++++++++++++++++++++++++++++----- 1 file changed, 314 insertions(+), 36 deletions(-)
tests/xe/xe_vm.c +314 -36
@@ -1233,9 +1233,9 @@ static void *hammer_thread(void *tdata)return NULL ; } -#define MUNMAP_FLAG_USERPTR (0x1 << 0) -#define MUNMAP_FLAG_INVALIDATE (0x1 << 1) -#define MUNMAP_FLAG_HAMMER_FIRST_PAGE (0x1 << 2) +#define MAP_FLAG_USERPTR (0x1 << 0) +#define MAP_FLAG_INVALIDATE (0x1 << 1) +#define MAP_FLAG_HAMMER_FIRST_PAGE (0x1 << 2)/**
@@ -1327,7 +1327,7 @@ test_munmap_style_unbind(int fd, struct drm_xe_engine_class_instance *eci, vm =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_BIND_OPS ,0 ); bo_size = page_size * bo_n_pages; -if (flags &MUNMAP_FLAG_USERPTR ) { +if (flags &MAP_FLAG_USERPTR ) { map =mmap (from_user_pointer (addr), bo_size,PROT_READ |PROT_WRITE ,MAP_SHARED |MAP_FIXED |MAP_ANONYMOUS ,-1 ,0 );
@@ -1346,7 +1346,7 @@ test_munmap_style_unbind(int fd, struct drm_xe_engine_class_instance *eci,/* Do initial binds */ bind_size = (page_size * bo_n_pages) / n_binds;for (i =0 ; i < n_binds; ++i) { -if (flags &MUNMAP_FLAG_USERPTR ) +if (flags &MAP_FLAG_USERPTR )xe_vm_bind_userptr_async (fd, vm,0 , addr, addr, bind_size, sync,1 );else
@@ -1361,7 +1361,7 @@ test_munmap_style_unbind(int fd, struct drm_xe_engine_class_instance *eci,* cause a fault if a rebind occurs during munmap style VM unbind * (partial VMAs unbound). */ -if (flags &MUNMAP_FLAG_HAMMER_FIRST_PAGE ) { +if (flags &MAP_FLAG_HAMMER_FIRST_PAGE ) { t.fd = fd; t.vm = vm;#define PAGE_SIZE 4096
@@ -1420,7 +1420,7 @@ test_munmap_style_unbind(int fd, struct drm_xe_engine_class_instance *eci, data = map + i * page_size;igt_assert_eq (data->data,0xc0ffee ); } -if (flags &MUNMAP_FLAG_HAMMER_FIRST_PAGE ) { +if (flags &MAP_FLAG_HAMMER_FIRST_PAGE ) {memset (map,0 ,PAGE_SIZE /2 );memset (map +PAGE_SIZE ,0 , bo_size -PAGE_SIZE ); }else {
@@ -1470,7 +1470,7 @@ try_again_after_invalidate:igt_assert_eq (data->data,0xc0ffee ); } } -if (flags &MUNMAP_FLAG_HAMMER_FIRST_PAGE ) { +if (flags &MAP_FLAG_HAMMER_FIRST_PAGE ) {memset (map,0 ,PAGE_SIZE /2 );memset (map +PAGE_SIZE ,0 , bo_size -PAGE_SIZE ); }else {
@@ -1481,7 +1481,7 @@ try_again_after_invalidate:* The munmap style VM unbind can create new VMAs, make sure those are * in the bookkeeping for another rebind after a userptr invalidate. */ -if (flags &MUNMAP_FLAG_INVALIDATE && !invalidate++) { +if (flags &MAP_FLAG_INVALIDATE && !invalidate++) { map =mmap (from_user_pointer (addr), bo_size,PROT_READ |PROT_WRITE ,MAP_SHARED |MAP_FIXED |MAP_ANONYMOUS ,-1 ,0 );
@@ -1492,7 +1492,7 @@ try_again_after_invalidate:/* Confirm unbound region can be rebound */ syncobj_reset (fd, &sync[0 ].handle,1 ); sync[0 ].flags |=DRM_XE_SYNC_SIGNAL ; -if (flags &MUNMAP_FLAG_USERPTR ) +if (flags &MAP_FLAG_USERPTR ) xe_vm_bind_userptr_async(fd, vm, 0, addr + unbind_n_page_offfset * page_size, addr + unbind_n_page_offfset * page_size,
@@ -1540,7 +1540,7 @@ try_again_after_invalidate:igt_assert_eq (data->data,0xc0ffee ); } -if (flags &MUNMAP_FLAG_HAMMER_FIRST_PAGE ) { +if (flags &MAP_FLAG_HAMMER_FIRST_PAGE ) { exit =1 ;pthread_join (t.thread,NULL );pthread_barrier_destroy (&barrier);
@@ -1555,6 +1555,251 @@ try_again_after_invalidate:xe_vm_destroy (fd, vm); } +/** +* SUBTEST: mmap-style-bind-%s +* Description: Test mmap style unbind with %arg[1] +* Run type: FULL +* TODO: change ``'Run type' == FULL`` to a better category +* +* arg[1]: +* +* @all: all +* @one-partial: one partial +* @either-side-partial: either side partial +* @either-side-full: either side full +* @either-side-partial-hammer: either side partial hammer +* @end: end +* @front: front +* @many-all: many all +* @many-either-side-partial: many either side partial +* @many-either-side-partial-hammer: many either side partial hammer +* @userptr-all: userptr all +* @userptr-one-partial: userptr one partial +* @userptr-either-side-partial: userptr either side partial +* @userptr-either-side-full: userptr either side full +*/ + +static void +test_mmap_style_bind (int fd,struct drm_xe_engine_class_instance *eci, +int bo_n_pages,int n_binds,int unbind_n_page_offfset, +int unbind_n_pages,unsigned int flags) +{ +struct drm_xe_sync sync[2 ] = { + { .flags =DRM_XE_SYNC_SYNCOBJ |DRM_XE_SYNC_SIGNAL , }, + { .flags =DRM_XE_SYNC_SYNCOBJ |DRM_XE_SYNC_SIGNAL , }, + }; +struct drm_xe_exec exec = { + .num_batch_buffer =1 , + .num_syncs =2 , + .syncs =to_user_pointer (sync), + }; +uint64_t addr =0x1a0000 , base_addr =0x1a0000 ; +uint32_t vm; +uint32_t engine; +size_t bo_size; +uint32_t bo0 =0 , bo1 =0 ; +uint64_t bind_size; +uint64_t page_size =xe_get_default_alignment (fd); +struct { +uint32_t batch[16 ]; +uint64_t pad; +uint32_t data; + } *data; +void *map0, *map1; +int i, b; +struct thread_data t; +pthread_barrier_t barrier; +int exit =0 ; + + vm =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_BIND_OPS ,0 ); + bo_size = page_size * bo_n_pages; + +if (flags &MAP_FLAG_USERPTR ) { + map0 =mmap (from_user_pointer (addr), bo_size,PROT_READ | +PROT_WRITE ,MAP_SHARED |MAP_FIXED | +MAP_ANONYMOUS ,-1 ,0 ); + map1 =mmap (from_user_pointer (addr + bo_size), + bo_size,PROT_READ |PROT_WRITE ,MAP_SHARED | +MAP_FIXED |MAP_ANONYMOUS ,-1 ,0 ); +igt_assert (map0 !=MAP_FAILED ); +igt_assert (map1 !=MAP_FAILED ); + }else { + bo0 =xe_bo_create (fd, eci->gt_id, vm, bo_size); + map0 =xe_bo_map (fd, bo0, bo_size); + bo1 =xe_bo_create (fd, eci->gt_id, vm, bo_size); + map1 =xe_bo_map (fd, bo1, bo_size); + } +memset (map0,0 , bo_size); +memset (map1,0 , bo_size); + + engine =xe_engine_create (fd, vm, eci,0 ); + + sync[0 ].handle =syncobj_create (fd,0 ); + sync[1 ].handle =syncobj_create (fd,0 ); + +/* Do initial binds */ + bind_size = (page_size * bo_n_pages) / n_binds; +for (i =0 ; i < n_binds; ++i) { +if (flags &MAP_FLAG_USERPTR ) +xe_vm_bind_userptr_async (fd, vm,0 , addr, addr, + bind_size, sync,1 ); +else +xe_vm_bind_async (fd, vm,0 , bo0, i * bind_size, + addr, bind_size, sync,1 ); + addr += bind_size; + } + addr = base_addr; + +/* +* Kick a thread to write the first page continously to ensure we can't +* cause a fault if a rebind occurs during munmap style VM unbind +* (partial VMAs unbound). +*/ +if (flags &MAP_FLAG_HAMMER_FIRST_PAGE ) { + t.fd = fd; + t.vm = vm; +#define PAGE_SIZE 4096 + t.addr = addr +PAGE_SIZE /2 ; + t.eci = eci; + t.exit = &exit; + t.map = map0 +PAGE_SIZE /2 ; + t.barrier = &barrier; +pthread_barrier_init (&barrier,NULL ,2 ); +pthread_create (&t.thread,0 , hammer_thread, &t); +pthread_barrier_wait (&barrier); + } + +/* Verify we can use every page */ +for (i =0 ; i < n_binds; ++i) { +uint64_t batch_offset = (char *)&data->batch - (char *)data; +uint64_t batch_addr = addr + batch_offset; +uint64_t sdi_offset = (char *)&data->data - (char *)data; +uint64_t sdi_addr = addr + sdi_offset; + data = map0 + i * page_size; + + b =0 ; + data->batch[b++] =MI_STORE_DWORD_IMM_GEN4 ; + data->batch[b++] = sdi_addr; + data->batch[b++] = sdi_addr >>32 ; + data->batch[b++] =0xc0ffee ; + data->batch[b++] =MI_BATCH_BUFFER_END ; +igt_assert (b <=ARRAY_SIZE (data[i].batch)); + + sync[0 ].flags &= ~DRM_XE_SYNC_SIGNAL ; +if (i) +syncobj_reset (fd, &sync[1 ].handle,1 ); + sync[1 ].flags |=DRM_XE_SYNC_SIGNAL ; + + exec.engine_id = engine; + exec.address = batch_addr; +xe_exec (fd, &exec); + + addr += page_size; + } + addr = base_addr; + +/* Bind some of the pages to different BO / userptr */ +syncobj_reset (fd, &sync[0 ].handle,1 ); + sync[0 ].flags |=DRM_XE_SYNC_SIGNAL ; + sync[1 ].flags &= ~DRM_XE_SYNC_SIGNAL ; +if (flags &MAP_FLAG_USERPTR ) +xe_vm_bind_userptr_async (fd, vm,0 , addr + bo_size + + unbind_n_page_offfset * page_size, + addr + unbind_n_page_offfset * page_size, + unbind_n_pages * page_size, sync,2 ); +else +xe_vm_bind_async (fd, vm,0 , bo1, + unbind_n_page_offfset * page_size, + addr + unbind_n_page_offfset * page_size, + unbind_n_pages * page_size, sync,2 ); +igt_assert (syncobj_wait (fd, &sync[0 ].handle,1 ,INT64_MAX ,0 ,NULL )); +igt_assert (syncobj_wait (fd, &sync[1 ].handle,1 ,INT64_MAX ,0 ,NULL )); + +/* Verify all pages written */ +for (i =0 ; i < n_binds; ++i) { + data = map0 + i * page_size; +igt_assert_eq (data->data,0xc0ffee ); + } +if (flags &MAP_FLAG_HAMMER_FIRST_PAGE ) { +memset (map0,0 ,PAGE_SIZE /2 ); +memset (map0 +PAGE_SIZE ,0 , bo_size -PAGE_SIZE ); + }else { +memset (map0,0 , bo_size); +memset (map1,0 , bo_size); + } + +/* Verify we can use every page */ +for (i =0 ; i < n_binds; ++i) { +uint64_t batch_offset = (char *)&data->batch - (char *)data; +uint64_t batch_addr = addr + batch_offset; +uint64_t sdi_offset = (char *)&data->data - (char *)data; +uint64_t sdi_addr = addr + sdi_offset; + + data = map0 + i * page_size; + b =0 ; + data->batch[b++] =MI_STORE_DWORD_IMM_GEN4 ; + data->batch[b++] = sdi_addr; + data->batch[b++] = sdi_addr >>32 ; + data->batch[b++] =0xc0ffee ; + data->batch[b++] =MI_BATCH_BUFFER_END ; +igt_assert (b <=ARRAY_SIZE (data[i].batch)); + + data = map1 + i * page_size; + b =0 ; + data->batch[b++] =MI_STORE_DWORD_IMM_GEN4 ; + data->batch[b++] = sdi_addr; + data->batch[b++] = sdi_addr >>32 ; + data->batch[b++] =0xc0ffee ; + data->batch[b++] =MI_BATCH_BUFFER_END ; +igt_assert (b <=ARRAY_SIZE (data[i].batch)); + + sync[0 ].flags &= ~DRM_XE_SYNC_SIGNAL ; +if (i) +syncobj_reset (fd, &sync[1 ].handle,1 ); + sync[1 ].flags |=DRM_XE_SYNC_SIGNAL ; + + exec.engine_id = engine; + exec.address = batch_addr; +xe_exec (fd, &exec); + + addr += page_size; + } + addr = base_addr; + +igt_assert (syncobj_wait (fd, &sync[0 ].handle,1 ,INT64_MAX ,0 ,NULL )); +igt_assert (syncobj_wait (fd, &sync[1 ].handle,1 ,INT64_MAX ,0 ,NULL )); + +/* Verify all pages written */ +for (i =0 ; i < n_binds; ++i) { +uint32_t result =0 ; + + data = map0 + i * page_size; + result |= data->data; + + data = map1 + i * page_size; + result |= data->data; + +igt_assert_eq (result,0xc0ffee ); + } + +if (flags &MAP_FLAG_HAMMER_FIRST_PAGE ) { + exit =1 ; +pthread_join (t.thread,NULL ); +pthread_barrier_destroy (&barrier); + } + +syncobj_destroy (fd, sync[0 ].handle); +syncobj_destroy (fd, sync[1 ].handle); +xe_engine_destroy (fd, engine); +munmap (map0, bo_size); +munmap (map1, bo_size); +if (bo0) +gem_close (fd, bo0); +if (bo1) +gem_close (fd, bo1); +xe_vm_destroy (fd, vm); +} + igt_main {struct drm_xe_engine_class_instance *hwe, *hwe_non_copy =NULL ;
@@ -1567,55 +1812,74 @@ igt_mainint unbind_n_page_offfset;int unbind_n_pages;unsigned int flags; - } sections[] = { + } munmap_sections[] = { { "all", 4, 2, 0, 4, 0 }, { "one-partial", 4, 1, 1, 2, 0 }, { "either-side-partial", 4, 2, 1, 2, 0 }, { "either-side-partial-hammer", 4, 2, 1, 2, - MUNMAP_FLAG_HAMMER_FIRST_PAGE }, + MAP_FLAG_HAMMER_FIRST_PAGE }, { "either-side-full", 4, 4, 1, 2, 0 }, { "end", 4, 2, 0, 3, 0 }, { "front", 4, 2, 1, 3, 0 }, { "many-all", 4 * 8, 2 * 8, 0 * 8, 4 * 8, 0 }, { "many-either-side-partial", 4 * 8, 2 * 8, 1, 4 * 8 - 2, 0 }, { "many-either-side-partial-hammer", 4 * 8, 2 * 8, 1, 4 * 8 - 2, - MUNMAP_FLAG_HAMMER_FIRST_PAGE }, + MAP_FLAG_HAMMER_FIRST_PAGE }, { "many-either-side-full", 4 * 8, 4 * 8, 1 * 8, 2 * 8, 0 }, { "many-end", 4 * 8, 4, 0 * 8, 3 * 8 + 2, 0 }, { "many-front", 4 * 8, 4, 1 * 8 - 2, 3 * 8 + 2, 0 }, - { "userptr-all", 4, 2, 0, 4, MUNMAP_FLAG_USERPTR }, - { "userptr-one-partial", 4, 1, 1, 2, MUNMAP_FLAG_USERPTR }, + { "userptr-all", 4, 2, 0, 4, MAP_FLAG_USERPTR }, + { "userptr-one-partial", 4, 1, 1, 2, MAP_FLAG_USERPTR }, { "userptr-either-side-partial", 4, 2, 1, 2, - MUNMAP_FLAG_USERPTR }, + MAP_FLAG_USERPTR }, { "userptr-either-side-full", 4, 4, 1, 2, - MUNMAP_FLAG_USERPTR }, - { "userptr-end", 4, 2, 0, 3, MUNMAP_FLAG_USERPTR }, - { "userptr-front", 4, 2, 1, 3, MUNMAP_FLAG_USERPTR }, + MAP_FLAG_USERPTR }, + { "userptr-end", 4, 2, 0, 3, MAP_FLAG_USERPTR }, + { "userptr-front", 4, 2, 1, 3, MAP_FLAG_USERPTR }, { "userptr-many-all", 4 * 8, 2 * 8, 0 * 8, 4 * 8, - MUNMAP_FLAG_USERPTR }, + MAP_FLAG_USERPTR }, { "userptr-many-either-side-full", 4 * 8, 4 * 8, 1 * 8, 2 * 8, - MUNMAP_FLAG_USERPTR }, + MAP_FLAG_USERPTR }, { "userptr-many-end", 4 * 8, 4, 0 * 8, 3 * 8 + 2, - MUNMAP_FLAG_USERPTR }, + MAP_FLAG_USERPTR }, { "userptr-many-front", 4 * 8, 4, 1 * 8 - 2, 3 * 8 + 2, - MUNMAP_FLAG_USERPTR }, + MAP_FLAG_USERPTR }, { "userptr-inval-either-side-full", 4, 4, 1, 2, - MUNMAP_FLAG_USERPTR | MUNMAP_FLAG_INVALIDATE }, - { "userptr-inval-end", 4, 2, 0, 3, MUNMAP_FLAG_USERPTR | - MUNMAP_FLAG_INVALIDATE }, - { "userptr-inval-front", 4, 2, 1, 3, MUNMAP_FLAG_USERPTR | - MUNMAP_FLAG_INVALIDATE }, + MAP_FLAG_USERPTR | MAP_FLAG_INVALIDATE }, + { "userptr-inval-end", 4, 2, 0, 3, MAP_FLAG_USERPTR | + MAP_FLAG_INVALIDATE }, + { "userptr-inval-front", 4, 2, 1, 3, MAP_FLAG_USERPTR | + MAP_FLAG_INVALIDATE }, { "userptr-inval-many-all", 4 * 8, 2 * 8, 0 * 8, 4 * 8, - MUNMAP_FLAG_USERPTR | MUNMAP_FLAG_INVALIDATE }, + MAP_FLAG_USERPTR | MAP_FLAG_INVALIDATE }, { "userptr-inval-many-either-side-partial", 4 * 8, 2 * 8, 1, - 4 * 8 - 2, MUNMAP_FLAG_USERPTR | - MUNMAP_FLAG_INVALIDATE }, + 4 * 8 - 2, MAP_FLAG_USERPTR | + MAP_FLAG_INVALIDATE }, { "userptr-inval-many-either-side-full", 4 * 8, 4 * 8, 1 * 8, - 2 * 8, MUNMAP_FLAG_USERPTR | MUNMAP_FLAG_INVALIDATE }, + 2 * 8, MAP_FLAG_USERPTR | MAP_FLAG_INVALIDATE }, { "userptr-inval-many-end", 4 * 8, 4, 0 * 8, 3 * 8 + 2, - MUNMAP_FLAG_USERPTR | MUNMAP_FLAG_INVALIDATE }, + MAP_FLAG_USERPTR | MAP_FLAG_INVALIDATE }, { "userptr-inval-many-front", 4 * 8, 4, 1 * 8 - 2, 3 * 8 + 2, - MUNMAP_FLAG_USERPTR | MUNMAP_FLAG_INVALIDATE }, + MAP_FLAG_USERPTR | MAP_FLAG_INVALIDATE }, + { NULL }, + }; + const struct section mmap_sections[] = { + { "all", 4, 2, 0, 4, 0 }, + { "one-partial", 4, 1, 1, 2, 0 }, + { "either-side-partial", 4, 2, 1, 2, 0 }, + { "either-side-full", 4, 4, 1, 2, 0 }, + { "either-side-partial-hammer", 4, 2, 1, 2, + MAP_FLAG_HAMMER_FIRST_PAGE }, + { "end", 4, 2, 0, 3, 0 }, + { "front", 4, 2, 1, 3, 0 }, + { "many-all", 4 * 8, 2 * 8, 0 * 8, 4 * 8, 0 }, + { "many-either-side-partial", 4 * 8, 2 * 8, 1, 4 * 8 - 2, 0 }, + { "many-either-side-partial-hammer", 4 * 8, 2 * 8, 1, 4 * 8 - 2, + MAP_FLAG_HAMMER_FIRST_PAGE }, + { "userptr-all", 4, 2, 0, 4, MAP_FLAG_USERPTR }, + { "userptr-one-partial", 4, 1, 1, 2, MAP_FLAG_USERPTR }, + { "userptr-either-side-partial", 4, 2, 1, 2, MAP_FLAG_USERPTR }, + { "userptr-either-side-full", 4, 4, 1, 2, MAP_FLAG_USERPTR }, { NULL }, };
@@ -1823,7 +2087,7 @@ igt_mainbreak ; } -for (const struct section *s = sections; s->name; s++) { +for (const struct section *s = munmap_sections; s->name; s++) { igt_subtest_f("munmap-style-unbind-%s", s->name) {igt_require_f (hwe_non_copy,"Requires non-copy engine to run\n" );
@@ -1837,6 +2101,20 @@ igt_main } } +for (const struct section *s = mmap_sections; s->name; s++) { + igt_subtest_f("mmap-style-bind-%s", s->name) { +igt_require_f (hwe_non_copy, +"Requires non-copy engine to run\n" ); + +test_mmap_style_bind (fd, hwe_non_copy, + s->bo_n_pages, + s->n_binds, + s->unbind_n_page_offfset, + s->unbind_n_pages, + s->flags); + } + } + igt_fixture {xe_device_put (fd);close (fd);
Signature · 1 line
2.34.1
PATCH 4/5 xe_vm: Add mmap / munmap sections that split large pages
In reply to Matthew BrostSplitting large pages involves using dma-resv slots for ordering, make sure this works.
--- tests/xe/xe_vm.c | 81 +++++++++++++++++++++++++++++++++++------------- 1 file changed, 60 insertions(+), 21 deletions(-)
tests/xe/xe_vm.c +60 -21
@@ -1236,7 +1236,8 @@ static void *hammer_thread(void *tdata)#define MAP_FLAG_USERPTR (0x1 << 0)#define MAP_FLAG_INVALIDATE (0x1 << 1)#define MAP_FLAG_HAMMER_FIRST_PAGE (0x1 << 2) - +#define MAP_FLAG_LARGE_PAGE (0x1 << 3) +#define MAP_FLAG_LARGE_PAGE_NO_SPLIT (0x1 << 4)/** * SUBTEST: munmap-style-unbind-%s
@@ -1288,12 +1289,16 @@ static void *hammer_thread(void *tdata)* userptr inval many either side full * @userptr-inval-many-end: userptr inval many end * @userptr-inval-many-front: userptr inval many front +* @either-side-partial-large-page-hammer: +* either side partial large page hammer +* @either-side-partial-split-page-hammer: +* either side partial split page hammer */ static void test_munmap_style_unbind (int fd,struct drm_xe_engine_class_instance *eci,int bo_n_pages,int n_binds, -int unbind_n_page_offfset,int unbind_n_pages, +int unbind_n_page_offset,int unbind_n_pages,unsigned int flags) {struct drm_xe_sync sync[2 ] = {
@@ -1305,7 +1310,7 @@ test_munmap_style_unbind(int fd, struct drm_xe_engine_class_instance *eci, .num_syncs = 2, .syncs = to_user_pointer(sync), }; -uint64_t addr =0x1a0000 , base_addr =0x1a0000 ; +uint64_t addr =0x1a00000 , base_addr =0x1a00000 ;uint32_t vm;uint32_t engine;size_t bo_size;
@@ -1323,6 +1328,14 @@ test_munmap_style_unbind(int fd, struct drm_xe_engine_class_instance *eci,struct thread_data t;pthread_barrier_t barrier;int exit =0 ; +int n_page_per_2mb =0x200000 /xe_get_default_alignment (fd); + +if (flags &MAP_FLAG_LARGE_PAGE ) { + bo_n_pages *= n_page_per_2mb; + unbind_n_pages *= n_page_per_2mb; +if (flags &MAP_FLAG_LARGE_PAGE_NO_SPLIT ) + unbind_n_page_offset *= n_page_per_2mb; + } vm =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_BIND_OPS ,0 ); bo_size = page_size * bo_n_pages;
@@ -1409,7 +1422,7 @@ test_munmap_style_unbind(int fd, struct drm_xe_engine_class_instance *eci, sync[0 ].flags |=DRM_XE_SYNC_SIGNAL ; sync[1 ].flags &= ~DRM_XE_SYNC_SIGNAL ;xe_vm_unbind_async (fd, vm,0 ,0 , - addr + unbind_n_page_offfset * page_size, + addr + unbind_n_page_offset * page_size, unbind_n_pages * page_size, sync,2 );igt_assert (syncobj_wait (fd, &sync[0 ].handle,1 ,INT64_MAX ,0 ,NULL ));
@@ -1438,8 +1451,8 @@ try_again_after_invalidate: data = map + i * page_size; addr += page_size; -if (i < unbind_n_page_offfset || - i +1 > unbind_n_page_offfset + unbind_n_pages) { +if (i < unbind_n_page_offset || + i +1 > unbind_n_page_offset + unbind_n_pages) { b =0 ; data->batch[b++] =MI_STORE_DWORD_IMM_GEN4 ; data->batch[b++] = sdi_addr;
@@ -1464,8 +1477,8 @@ try_again_after_invalidate:/* Verify all pages still bound written */ for (i =0 ; i < n_binds; ++i) { -if (i < unbind_n_page_offfset || - i +1 > unbind_n_page_offfset + unbind_n_pages) { +if (i < unbind_n_page_offset || + i +1 > unbind_n_page_offset + unbind_n_pages) { data = map + i * page_size;igt_assert_eq (data->data,0xc0ffee ); }
@@ -1494,13 +1507,13 @@ try_again_after_invalidate: sync[0 ].flags |=DRM_XE_SYNC_SIGNAL ;if (flags &MAP_FLAG_USERPTR )xe_vm_bind_userptr_async (fd, vm,0 , - addr + unbind_n_page_offfset * page_size, - addr + unbind_n_page_offfset * page_size, + addr + unbind_n_page_offset * page_size, + addr + unbind_n_page_offset * page_size, unbind_n_pages * page_size, sync,1 );else xe_vm_bind_async (fd, vm,0 , bo, - unbind_n_page_offfset * page_size, - addr + unbind_n_page_offfset * page_size, + unbind_n_page_offset * page_size, + addr + unbind_n_page_offset * page_size, unbind_n_pages * page_size, sync,1 );/* Verify we can use every page */
@@ -1577,11 +1590,15 @@ try_again_after_invalidate:* @userptr-one-partial: userptr one partial * @userptr-either-side-partial: userptr either side partial * @userptr-either-side-full: userptr either side full +* @either-side-partial-large-page-hammer: +* either side partial large page hammer +* @either-side-partial-split-page-hammer: +* either side partial split page hammer */ static void test_mmap_style_bind (int fd,struct drm_xe_engine_class_instance *eci, -int bo_n_pages,int n_binds,int unbind_n_page_offfset, +int bo_n_pages,int n_binds,int unbind_n_page_offset,int unbind_n_pages,unsigned int flags) {struct drm_xe_sync sync[2 ] = {
@@ -1593,7 +1610,7 @@ test_mmap_style_bind(int fd, struct drm_xe_engine_class_instance *eci, .num_syncs = 2, .syncs = to_user_pointer(sync), }; -uint64_t addr =0x1a0000 , base_addr =0x1a0000 ; +uint64_t addr =0x1a00000 , base_addr =0x1a00000 ;uint32_t vm;uint32_t engine;size_t bo_size;
@@ -1610,6 +1627,14 @@ test_mmap_style_bind(int fd, struct drm_xe_engine_class_instance *eci,struct thread_data t;pthread_barrier_t barrier;int exit =0 ; +int n_page_per_2mb =0x200000 /xe_get_default_alignment (fd); + +if (flags &MAP_FLAG_LARGE_PAGE ) { + bo_n_pages *= n_page_per_2mb; + unbind_n_pages *= n_page_per_2mb; +if (flags &MAP_FLAG_LARGE_PAGE_NO_SPLIT ) + unbind_n_page_offset *= n_page_per_2mb; + } vm =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_BIND_OPS ,0 ); bo_size = page_size * bo_n_pages;
@@ -1704,13 +1729,13 @@ test_mmap_style_bind(int fd, struct drm_xe_engine_class_instance *eci, sync[1 ].flags &= ~DRM_XE_SYNC_SIGNAL ;if (flags &MAP_FLAG_USERPTR )xe_vm_bind_userptr_async (fd, vm,0 , addr + bo_size + - unbind_n_page_offfset * page_size, - addr + unbind_n_page_offfset * page_size, + unbind_n_page_offset * page_size, + addr + unbind_n_page_offset * page_size, unbind_n_pages * page_size, sync,2 );else xe_vm_bind_async (fd, vm,0 , bo1, - unbind_n_page_offfset * page_size, - addr + unbind_n_page_offfset * page_size, + unbind_n_page_offset * page_size, + addr + unbind_n_page_offset * page_size, unbind_n_pages * page_size, sync,2 );igt_assert (syncobj_wait (fd, &sync[0 ].handle,1 ,INT64_MAX ,0 ,NULL ));igt_assert (syncobj_wait (fd, &sync[1 ].handle,1 ,INT64_MAX ,0 ,NULL ));
@@ -1809,7 +1834,7 @@ igt_mainconst char *name;int bo_n_pages;int n_binds; -int unbind_n_page_offfset; +int unbind_n_page_offset;int unbind_n_pages;unsigned int flags; } munmap_sections[] = {
@@ -1818,6 +1843,13 @@ igt_main { "either-side-partial", 4, 2, 1, 2, 0 }, { "either-side-partial-hammer", 4, 2, 1, 2, MAP_FLAG_HAMMER_FIRST_PAGE }, + { "either-side-partial-split-page-hammer", 4, 2, 1, 2, +MAP_FLAG_HAMMER_FIRST_PAGE | + MAP_FLAG_LARGE_PAGE }, + { "either-side-partial-large-page-hammer", 4, 2, 1, 2, +MAP_FLAG_HAMMER_FIRST_PAGE | +MAP_FLAG_LARGE_PAGE | + MAP_FLAG_LARGE_PAGE_NO_SPLIT }, { "either-side-full", 4, 4, 1, 2, 0 }, { "end", 4, 2, 0, 3, 0 }, { "front", 4, 2, 1, 3, 0 },
@@ -1870,6 +1902,13 @@ igt_main { "either-side-full", 4, 4, 1, 2, 0 }, { "either-side-partial-hammer", 4, 2, 1, 2, MAP_FLAG_HAMMER_FIRST_PAGE }, + { "either-side-partial-split-page-hammer", 4, 2, 1, 2, +MAP_FLAG_HAMMER_FIRST_PAGE | + MAP_FLAG_LARGE_PAGE }, + { "either-side-partial-large-page-hammer", 4, 2, 1, 2, +MAP_FLAG_HAMMER_FIRST_PAGE | +MAP_FLAG_LARGE_PAGE | + MAP_FLAG_LARGE_PAGE_NO_SPLIT }, { "end", 4, 2, 0, 3, 0 }, { "front", 4, 2, 1, 3, 0 }, { "many-all", 4 * 8, 2 * 8, 0 * 8, 4 * 8, 0 },
@@ -2095,7 +2134,7 @@ igt_maintest_munmap_style_unbind (fd, hwe_non_copy, s->bo_n_pages, s->n_binds, - s->unbind_n_page_offfset, + s->unbind_n_page_offset, s->unbind_n_pages, s->flags); }
@@ -2109,7 +2148,7 @@ igt_maintest_mmap_style_bind (fd, hwe_non_copy, s->bo_n_pages, s->n_binds, - s->unbind_n_page_offfset, + s->unbind_n_page_offset, s->unbind_n_pages, s->flags); }
Signature · 1 line
2.34.1
PATCH 5/5 xe: Update to new VM bind uAPI
In reply to Matthew BrostSync vs. async changes and new error handling.
--- include/drm-uapi/xe_drm.h | 79 ++------------- lib/intel_batchbuffer.c | 2 +- lib/xe/xe_compute.c | 2 +- lib/xe/xe_ioctl.c | 15 +-- lib/xe/xe_ioctl.h | 2 +- lib/xe/xe_query.c | 2 +- tests/xe/xe_create.c | 2 +- tests/xe/xe_evict.c | 23 +++-- tests/xe/xe_exec_balancer.c | 6 +- tests/xe/xe_exec_basic.c | 5 +- tests/xe/xe_exec_compute_mode.c | 6 +- tests/xe/xe_exec_fault_mode.c | 6 +- tests/xe/xe_exec_reset.c | 8 +- tests/xe/xe_exec_threads.c | 112 ++++++--------------- tests/xe/xe_guc_pc.c | 2 +- tests/xe/xe_huc_copy.c | 2 +- tests/xe/xe_intel_bb.c | 2 +- tests/xe/xe_pm.c | 2 +- tests/xe/xe_vm.c | 168 +++++++------------------------- tests/xe/xe_waitfence.c | 18 +--- 20 files changed, 120 insertions(+), 344 deletions(-)
include/drm-uapi/xe_drm.h +9 -70
@@ -145,10 +145,11 @@ struct drm_xe_engine_class_instance {#define DRM_XE_ENGINE_CLASS_VIDEO_ENHANCE 3#define DRM_XE_ENGINE_CLASS_COMPUTE 4/* -* Kernel only class (not actual hardware engine class). Used for +* Kernel only classes (not actual hardware engine class). Used for * creating ordered queues of VM bind operations. */ -#define DRM_XE_ENGINE_CLASS_VM_BIND 5 +#define DRM_XE_ENGINE_CLASS_VM_BIND_ASYNC 5 +#define DRM_XE_ENGINE_CLASS_VM_BIND_SYNC 6__u16 engine_instance;__u16 gt_id;
@@ -312,39 +313,8 @@ struct drm_xe_gem_mmap_offset {__u64 reserved[2 ]; }; -/** -* struct drm_xe_vm_bind_op_error_capture - format of VM bind op error capture -*/ -struct drm_xe_vm_bind_op_error_capture { -/** @error: errno that occured */ -__s32 error; -/** @op: operation that encounter an error */ -__u32 op; -/** @addr: address of bind op */ -__u64 addr; -/** @size: size of bind */ -__u64 size; -}; - -/** struct drm_xe_ext_vm_set_property - VM set property extension */ -struct drm_xe_ext_vm_set_property { -/** @base: base user extension */ -struct xe_user_extension base; - -/** @property: property to set */ -#define XE_VM_PROPERTY_BIND_OP_ERROR_CAPTURE_ADDRESS 0 -__u32 property; - -/** @value: property value */ -__u64 value; - -/** @reserved: Reserved */ -__u64 reserved[2 ]; -}; -struct drm_xe_vm_create {/** @extensions: Pointer to the first extension struct, if any */ -#define XE_VM_EXTENSION_SET_PROPERTY 0__u64 extensions;/** @flags: Flags */
@@ -352,7 +322,7 @@ struct drm_xe_vm_create {#define DRM_XE_VM_CREATE_SCRATCH_PAGE (0x1 << 0)#define DRM_XE_VM_CREATE_COMPUTE_MODE (0x1 << 1) -#define DRM_XE_VM_CREATE_ASYNC_BIND_OPS (0x1 << 2) +#define DRM_XE_VM_CREATE_ASYNC_DEFAULT (0x1 << 2)#define DRM_XE_VM_CREATE_FAULT_MODE (0x1 << 3)/** @vm_id: Returned VM ID */
@@ -417,30 +387,6 @@ struct drm_xe_vm_bind_op {#define XE_VM_BIND_OP_PREFETCH 0x5#define XE_VM_BIND_FLAG_READONLY (0x1 << 16) -/* -* A bind ops completions are always async, hence the support for out -* sync. This flag indicates the allocation of the memory for new page -* tables and the job to program the pages tables is asynchronous -* relative to the IOCTL. That part of a bind operation can fail under -* memory pressure, the job in practice can't fail unless the system is -* totally shot. -* -* If this flag is clear and the IOCTL doesn't return an error, in -* practice the bind op is good and will complete. -* -* If this flag is set and doesn't return return an error, the bind op -* can still fail and recovery is needed. If configured, the bind op that -* caused the error will be captured in drm_xe_vm_bind_op_error_capture. -* Once the user sees the error (via a ufence + -* XE_VM_PROPERTY_BIND_OP_ERROR_CAPTURE_ADDRESS), it should free memory -* via non-async unbinds, and then restart all queue'd async binds op via -* XE_VM_BIND_OP_RESTART. Or alternatively the user should destroy the -* VM. -* -* This flag is only allowed when DRM_XE_VM_CREATE_ASYNC_BIND_OPS is -* configured in the VM and must be set if the VM is configured with -* DRM_XE_VM_CREATE_ASYNC_BIND_OPS and not in an error state. -*/ #define XE_VM_BIND_FLAG_ASYNC (0x1 << 17)/* * Valid on a faulting VM only, do the MAP operation immediately rather
@@ -455,6 +401,7 @@ struct drm_xe_vm_bind_op {* VK sparse bindings. */ #define XE_VM_BIND_FLAG_NULL (0x1 << 19) +#define XE_VM_BIND_FLAG_RECLAIM (0x1 << 20)/** @reserved: Reserved */ __u64 reserved[2 ];
@@ -702,17 +649,10 @@ struct drm_xe_mmio {struct drm_xe_wait_user_fence {/** @extensions: Pointer to the first extension struct, if any */ __u64 extensions; -union { -/** -* @addr: user pointer address to wait on, must qword aligned -*/ -__u64 addr; -/** -* @vm_id: The ID of the VM which encounter an error used with -* DRM_XE_UFENCE_WAIT_VM_ERROR. Upper 32 bits must be clear. -*/ -__u64 vm_id; - }; +/** +* @addr: user pointer address to wait on, must qword aligned +*/ +__u64 addr;/** @op: wait operation (type of comparison) */ #define DRM_XE_UFENCE_WAIT_EQ 0#define DRM_XE_UFENCE_WAIT_NEQ 1
@@ -724,7 +664,6 @@ struct drm_xe_wait_user_fence {/** @flags: wait flags */ #define DRM_XE_UFENCE_WAIT_SOFT_OP (1 << 0)/* e.g. Wait on VM bind */ #define DRM_XE_UFENCE_WAIT_ABSTIME (1 << 1) -#define DRM_XE_UFENCE_WAIT_VM_ERROR (1 << 2)__u16 flags;/** @value: compare value */ __u64 value;
lib/intel_batchbuffer.c +1 -1
@@ -940,7 +940,7 @@ __intel_bb_create(int fd, uint32_t ctx, const intel_ctx_cfg_t *cfg, ibb->gtt_size =1ull <<xe_va_bits (fd);if (!ctx) - ctx =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_BIND_OPS ,0 ); + ctx =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_DEFAULT ,0 ); ibb->uses_full_ppgtt = true; ibb->allocator_handle =
lib/xe/xe_compute.c +1 -1
@@ -406,7 +406,7 @@ static void tgl_compute_exec(int fd, const unsigned char *kernel,/* Sets Kernel size */ bo_dict[0 ].size =ALIGN (size,0x1000 ); - vm =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_BIND_OPS ,0 ); + vm =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_DEFAULT ,0 ); engine =xe_engine_create_class (fd, vm,DRM_XE_ENGINE_CLASS_RENDER ); sync.flags =DRM_XE_SYNC_SYNCOBJ |DRM_XE_SYNC_SIGNAL ; sync.handle =syncobj_create (fd,0 );
lib/xe/xe_ioctl.c +4 -11
@@ -199,16 +199,8 @@ void xe_vm_unbind_async(int fd, uint32_t vm, uint32_t engine,static void __xe_vm_bind_sync (int fd,uint32_t vm,uint32_t bo,uint64_t offset,uint64_t addr,uint64_t size,uint32_t op) { -struct drm_xe_sync sync = { - .flags =DRM_XE_SYNC_SYNCOBJ |DRM_XE_SYNC_SIGNAL , - .handle =syncobj_create (fd,0 ), - }; - -__xe_vm_bind_assert (fd, vm,0 , bo, offset, addr, size, op, &sync,1 ,0 , +__xe_vm_bind_assert (fd, vm,0 , bo, offset, addr, size, op,NULL ,0 ,0 ,0 ); - -igt_assert (syncobj_wait (fd, &sync.handle,1 ,INT64_MAX ,0 ,NULL )); -syncobj_destroy (fd, sync.handle); } void xe_vm_bind_sync(int fd, uint32_t vm, uint32_t bo, uint64_t offset,
@@ -260,10 +252,11 @@ uint32_t xe_bo_create(int fd, int gt, uint32_t vm, uint64_t size)return create.handle; } -uint32_t xe_bind_engine_create (int fd,uint32_t vm,uint64_t ext) +uint32_t xe_bind_engine_create (int fd,uint32_t vm,uint64_t ext,bool async) {struct drm_xe_engine_class_instance instance = { - .engine_class =DRM_XE_ENGINE_CLASS_VM_BIND , + .engine_class = async ?DRM_XE_ENGINE_CLASS_VM_BIND_ASYNC : +DRM_XE_ENGINE_CLASS_VM_BIND_SYNC , };struct drm_xe_engine_create create = { .extensions = ext,
lib/xe/xe_ioctl.h +1 -1
@@ -69,7 +69,7 @@ uint32_t xe_bo_create(int fd, int gt, uint32_t vm, uint64_t size);uint32_t xe_engine_create (int fd,uint32_t vm,struct drm_xe_engine_class_instance *instance,uint64_t ext); -uint32_t xe_bind_engine_create (int fd,uint32_t vm,uint64_t ext); +uint32_t xe_bind_engine_create (int fd,uint32_t vm,uint64_t ext,bool async);uint32_t xe_engine_create_class (int fd,uint32_t vm,uint16_t class);void xe_engine_destroy (int fd,uint32_t engine);uint64_t xe_bo_mmap_offset (int fd,uint32_t bo);
lib/xe/xe_query.c +1 -1
@@ -300,7 +300,7 @@ bool xe_supports_faults(int fd)bool supports_faults;struct drm_xe_vm_create create = { - .flags =DRM_XE_VM_CREATE_ASYNC_BIND_OPS | + .flags =DRM_XE_VM_CREATE_ASYNC_DEFAULT |DRM_XE_VM_CREATE_FAULT_MODE , };
tests/xe/xe_create.c +1 -1
@@ -54,7 +54,7 @@ static void create_invalid_size(int fd)uint32_t handle;int ret; - vm =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_BIND_OPS ,0 ); + vm =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_DEFAULT ,0 );xe_for_each_mem_region (fd, memreg, region) { memregion =xe_mem_region (fd, region);
tests/xe/xe_evict.c +13 -10
@@ -64,15 +64,17 @@ test_evict(int fd, struct drm_xe_engine_class_instance *eci, fd =drm_open_driver (DRIVER_XE );xe_device_get (fd); - vm =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_BIND_OPS ,0 ); + vm =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_DEFAULT ,0 );if (flags &BIND_ENGINE ) - bind_engines[0 ] =xe_bind_engine_create (fd, vm,0 ); + bind_engines[0 ] =xe_bind_engine_create (fd, vm,0 , true);if (flags &MULTI_VM ) { - vm2 =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_BIND_OPS ,0 ); - vm3 =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_BIND_OPS ,0 ); + vm2 =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_DEFAULT ,0 ); + vm3 =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_DEFAULT ,0 );if (flags &BIND_ENGINE ) { - bind_engines[1 ] =xe_bind_engine_create (fd, vm2,0 ); - bind_engines[2 ] =xe_bind_engine_create (fd, vm3,0 ); + bind_engines[1 ] =xe_bind_engine_create (fd, vm2,0 , + true); + bind_engines[2 ] =xe_bind_engine_create (fd, vm3,0 , + true); } }
@@ -241,15 +243,16 @@ test_evict_cm(int fd, struct drm_xe_engine_class_instance *eci, fd =drm_open_driver (DRIVER_XE );xe_device_get (fd); - vm =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_BIND_OPS | + vm =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_DEFAULT |DRM_XE_VM_CREATE_COMPUTE_MODE ,0 );if (flags &BIND_ENGINE ) - bind_engines[0 ] =xe_bind_engine_create (fd, vm,0 ); + bind_engines[0 ] =xe_bind_engine_create (fd, vm,0 , true);if (flags &MULTI_VM ) { - vm2 =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_BIND_OPS | + vm2 =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_DEFAULT |DRM_XE_VM_CREATE_COMPUTE_MODE ,0 );if (flags &BIND_ENGINE ) - bind_engines[1 ] =xe_bind_engine_create (fd, vm2,0 ); + bind_engines[1 ] =xe_bind_engine_create (fd, vm2,0 , + true); }for (i =0 ; i < n_engines; i++) {
tests/xe/xe_exec_balancer.c +3 -3
@@ -66,7 +66,7 @@ static void test_all_active(int fd, int gt, int class)if (num_placements <2 )return ; - vm =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_BIND_OPS ,0 ); + vm =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_DEFAULT ,0 ); bo_size =sizeof (*data) * num_placements; bo_size =ALIGN (bo_size +xe_cs_prefetch_size (fd),xe_get_default_alignment (fd));
@@ -212,7 +212,7 @@ test_exec(int fd, int gt, int class, int n_engines, int n_execs,if (num_placements <2 )return ; - vm =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_BIND_OPS ,0 ); + vm =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_DEFAULT ,0 ); bo_size =sizeof (*data) * n_execs; bo_size =ALIGN (bo_size +xe_cs_prefetch_size (fd),xe_get_default_alignment (fd));
@@ -435,7 +435,7 @@ test_cm(int fd, int gt, int class, int n_engines, int n_execs,if (num_placements <2 )return ; - vm =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_BIND_OPS | + vm =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_DEFAULT |DRM_XE_VM_CREATE_COMPUTE_MODE ,0 ); bo_size =sizeof (*data) * n_execs; bo_size = ALIGN(bo_size + xe_cs_prefetch_size(fd),
tests/xe/xe_exec_basic.c +3 -2
@@ -113,7 +113,7 @@ test_exec(int fd, struct drm_xe_engine_class_instance *eci,igt_assert (n_vm <=MAX_N_ENGINES );for (i =0 ; i < n_vm; ++i) - vm[i] =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_BIND_OPS ,0 ); + vm[i] =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_DEFAULT ,0 ); bo_size =sizeof (*data) * n_execs; bo_size =ALIGN (bo_size +xe_cs_prefetch_size (fd),xe_get_default_alignment (fd));
@@ -156,7 +156,8 @@ test_exec(int fd, struct drm_xe_engine_class_instance *eci, engines[i] =xe_engine_create (fd, __vm, eci,0 );if (flags &BIND_ENGINE ) - bind_engines[i] =xe_bind_engine_create (fd, __vm,0 ); + bind_engines[i] =xe_bind_engine_create (fd, __vm,0 , + true);else bind_engines[i] =0 ; syncobjs[i] =syncobj_create (fd,0 );
tests/xe/xe_exec_compute_mode.c +3 -3
@@ -116,7 +116,7 @@ test_exec(int fd, struct drm_xe_engine_class_instance *eci,igt_assert (n_engines <=MAX_N_ENGINES ); - vm =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_BIND_OPS | + vm =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_DEFAULT |DRM_XE_VM_CREATE_COMPUTE_MODE ,0 ); bo_size =sizeof (*data) * n_execs; bo_size = ALIGN(bo_size + xe_cs_prefetch_size(fd),
@@ -134,7 +134,7 @@ test_exec(int fd, struct drm_xe_engine_class_instance *eci, to_user_pointer(&ext));if (flags &BIND_ENGINE ) bind_engines[i] = -xe_bind_engine_create (fd, vm,0 ); +xe_bind_engine_create (fd, vm,0 , true);else bind_engines[i] =0 ; };
@@ -170,7 +170,7 @@ test_exec(int fd, struct drm_xe_engine_class_instance *eci, to_user_pointer(&ext));if (flags &BIND_ENGINE ) bind_engines[i] = -xe_bind_engine_create (fd, vm,0 ); +xe_bind_engine_create (fd, vm,0 , true);else bind_engines[i] =0 ; };
tests/xe/xe_exec_fault_mode.c +3 -3
@@ -136,7 +136,7 @@ test_exec(int fd, struct drm_xe_engine_class_instance *eci,igt_assert (n_engines <=MAX_N_ENGINES ); - vm =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_BIND_OPS | + vm =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_DEFAULT |DRM_XE_VM_CREATE_FAULT_MODE ,0 ); bo_size =sizeof (*data) * n_execs; bo_size = ALIGN(bo_size + xe_cs_prefetch_size(fd),
@@ -168,7 +168,7 @@ test_exec(int fd, struct drm_xe_engine_class_instance *eci, engines[i] =xe_engine_create (fd, vm, eci,0 );if (flags &BIND_ENGINE ) bind_engines[i] = -xe_bind_engine_create (fd, vm,0 ); +xe_bind_engine_create (fd, vm,0 , true);else bind_engines[i] =0 ; };
@@ -382,7 +382,7 @@ test_atomic(int fd, struct drm_xe_engine_class_instance *eci,uint32_t *ptr;int i, b, wait_idx =0 ; - vm =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_BIND_OPS | + vm =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_DEFAULT |DRM_XE_VM_CREATE_FAULT_MODE ,0 ); bo_size =sizeof (*data) * n_atomic; bo_size = ALIGN(bo_size + xe_cs_prefetch_size(fd),
tests/xe/xe_exec_reset.c +4 -4
@@ -46,7 +46,7 @@ static void test_spin(int fd, struct drm_xe_engine_class_instance *eci)uint32_t bo =0 ;struct xe_spin *spin; - vm =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_BIND_OPS ,0 ); + vm =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_DEFAULT ,0 ); bo_size =sizeof (*spin); bo_size =ALIGN (bo_size +xe_cs_prefetch_size (fd),xe_get_default_alignment (fd));
@@ -192,7 +192,7 @@ test_balancer(int fd, int gt, int class, int n_engines, int n_execs,if (num_placements <2 )return ; - vm =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_BIND_OPS ,0 ); + vm =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_DEFAULT ,0 ); bo_size =sizeof (*data) * n_execs; bo_size =ALIGN (bo_size +xe_cs_prefetch_size (fd),xe_get_default_alignment (fd));
@@ -393,7 +393,7 @@ test_legacy_mode(int fd, struct drm_xe_engine_class_instance *eci,xe_device_get (fd); } - vm =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_BIND_OPS ,0 ); + vm =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_DEFAULT ,0 ); bo_size =sizeof (*data) * n_execs; bo_size =ALIGN (bo_size +xe_cs_prefetch_size (fd),xe_get_default_alignment (fd));
@@ -571,7 +571,7 @@ test_compute_mode(int fd, struct drm_xe_engine_class_instance *eci,xe_device_get (fd); } - vm =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_BIND_OPS | + vm =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_DEFAULT |DRM_XE_VM_CREATE_COMPUTE_MODE ,0 ); bo_size =sizeof (*data) * n_execs; bo_size = ALIGN(bo_size + xe_cs_prefetch_size(fd),
tests/xe/xe_exec_threads.c +30 -82
203 lines not on this page · raw message
tests/xe/xe_guc_pc.c +1 -1
@@ -59,7 +59,7 @@ static void exec_basic(int fd, struct drm_xe_engine_class_instance *eci,igt_assert (n_engines <=MAX_N_ENGINES );igt_assert (n_execs >0 ); - vm =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_BIND_OPS ,0 ); + vm =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_DEFAULT ,0 ); bo_size =sizeof (*data) * n_execs; bo_size =ALIGN (bo_size +xe_cs_prefetch_size (fd),xe_get_default_alignment (fd));
tests/xe/xe_huc_copy.c +1 -1
@@ -118,7 +118,7 @@ test_huc_copy(int fd) { .addr = ADDR_BATCH, .size = SIZE_BATCH }, // batch }; - vm =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_BIND_OPS ,0 ); + vm =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_DEFAULT ,0 ); engine =xe_engine_create_class (fd, vm,DRM_XE_ENGINE_CLASS_VIDEO_DECODE ); sync.flags =DRM_XE_SYNC_SYNCOBJ |DRM_XE_SYNC_SIGNAL ; sync.handle =syncobj_create (fd,0 );
tests/xe/xe_intel_bb.c +1 -1
@@ -195,7 +195,7 @@ static void simple_bb(struct buf_ops *bops, bool new_context)intel_bb_reset (ibb, true);if (new_context) { - ctx =xe_vm_create (xe,DRM_XE_VM_CREATE_ASYNC_BIND_OPS ,0 ); + ctx =xe_vm_create (xe,DRM_XE_VM_CREATE_ASYNC_DEFAULT ,0 );intel_bb_destroy (ibb); ibb =intel_bb_create_with_context (xe, ctx,NULL ,PAGE_SIZE );intel_bb_out (ibb,MI_BATCH_BUFFER_END );
tests/xe/xe_pm.c +1 -1
@@ -238,7 +238,7 @@ test_exec(device_t device, struct drm_xe_engine_class_instance *eci,if (check_rpm)igt_assert (in_d3 (device, d_state)); - vm =xe_vm_create (device.fd_xe,DRM_XE_VM_CREATE_ASYNC_BIND_OPS ,0 ); + vm =xe_vm_create (device.fd_xe,DRM_XE_VM_CREATE_ASYNC_DEFAULT ,0 );if (check_rpm)igt_assert (out_of_d3 (device, d_state));
tests/xe/xe_vm.c +37 -131
256 lines not on this page · raw message
tests/xe/xe_waitfence.c +2 -16
@@ -34,7 +34,7 @@ static void do_bind(int fd, uint32_t vm, uint32_t bo, uint64_t offset, sync[0 ].addr =to_user_pointer (&wait_fence); sync[0 ].timeline_value = val; -xe_vm_bind (fd, vm, bo, offset, addr, size, sync,1 ); +xe_vm_bind_async (fd, vm,0 , bo, offset, addr, size, sync,1 ); }/**
@@ -52,7 +52,7 @@ test(int fd)uint32_t bo_6;uint32_t bo_7; -uint32_t vm =xe_vm_create (fd,0 ,0 ); +uint32_t vm =xe_vm_create (fd,DRM_XE_VM_CREATE_ASYNC_DEFAULT ,0 ); bo_1 =xe_bo_create_flags (fd, vm,0x40000 ,MY_FLAG );do_bind (fd, vm, bo_1,0 ,0x200000 ,0x40000 ,1 ); bo_2 =xe_bo_create_flags (fd, vm,0x40000 ,MY_FLAG );
@@ -68,20 +68,6 @@ test(int fd) bo_7 =xe_bo_create_flags (fd, vm,0x10000 ,MY_FLAG );do_bind (fd, vm, bo_7,0 ,0xeffff0000 ,0x10000 ,7 );xe_wait_ufence (fd, &wait_fence,7 ,NULL ,2000 ); -xe_vm_unbind_sync (fd, vm,0 ,0x200000 ,0x40000 ); -xe_vm_unbind_sync (fd, vm,0 ,0xc0000000 ,0x40000 ); -xe_vm_unbind_sync (fd, vm,0 ,0x180000000 ,0x40000 ); -xe_vm_unbind_sync (fd, vm,0 ,0x140000000 ,0x10000 ); -xe_vm_unbind_sync (fd, vm,0 ,0x100000000 ,0x100000 ); -xe_vm_unbind_sync (fd, vm,0 ,0xc0040000 ,0x1c0000 ); -xe_vm_unbind_sync (fd, vm,0 ,0xeffff0000 ,0x10000 ); -gem_close (fd, bo_7); -gem_close (fd, bo_6); -gem_close (fd, bo_5); -gem_close (fd, bo_4); -gem_close (fd, bo_3); -gem_close (fd, bo_2); -gem_close (fd, bo_1); } igt_main
Signature · 1 line
2.34.1
Patchwork
CI[igt-dev] ✗ Fi.CI.BUILD: failure for IGT updates for upstreaming prep
In reply to Matthew Brost== Series Details ==
== Series Details ==
Series: IGT updates for upstreaming prep
URL : https://patchwork.freedesktop.org/series/117177/
State : failure
== Summary ==
IGT patchset build failed on latest successful build 1cb3507f3ff28d11bd5cfabcde576fe78ddab571 intel-ci/xe-fast-feedback.testlist: Add eviction tests
ninja: Entering directory `/opt/igt/build'
[1/436] Generating version.h with a custom command.
[2/4] Generating xe_tests.rst with a custom command.
FAILED: docs/testplan/xe_tests.rst/usr/src/igt-gpu-tools/scripts/igt_doc.py --config /usr/src/igt-gpu-tools/tests/xe/xe_test_config.json --rest docs/testplan/xe_tests.rst --check-testlist --igt-build-path /opt/igt/build
Warning: Missing documentation for igt@xe_exec_threads@threads-hang-rebind-err
Warning: Missing documentation for igt@xe_exec_threads@threads-hang-userptr-rebind-err
Warning: Missing documentation for igt@xe_exec_threads@threads-rebind-err
Warning: Missing documentation for igt@xe_exec_threads@threads-userptr-rebind-err
ninja: build stopped: subcommand failed.