Madvise Tests in IGT
cover Madvise Tests in IGT
From: Nishit Sharma <[email protected]>
Revision 1:
Added madvise tests in IGT which validate different features related to attributes passed. Madvise tests related to Atomic operation, Preferred Loc have been added and validated. Madvise tests are called as part of different struct section and available as madvise-<test-name> in list of subtests.
ver2:
- added back subtest which was deleted due to rebasing
Nishit Sharma (5):
include/drm-uap: Added drm_xe_madvise structure
lib/xe: Added xe_vm_madvise ioctl support
lib/xe: Helper added to get memory attributes tests/intel/xe_exec_system_allocator: Added preferred_loc_smem test tests/intel/xe_exec_system_allocator: Added atomic_batch test in IGT
include/drm-uapi/xe_drm.h | 289 +++++++++- lib/xe/xe_ioctl.c | 148 +++++ lib/xe/xe_ioctl.h | 9 +- tests/intel/xe_exec_system_allocator.c | 716 ++++++++++++++++++++++--- 4 files changed, 1075 insertions(+), 87 deletions(-)
Signature · 1 line
2.43.0
PATCH 1/5 include/drm-uap: Added drm_xe_madvise structure
From: Nishit Sharma <[email protected]>
Defined IOCTL number for madvise operation. Added drm_xe_madvise which is passed as Input to MADVISE IOCTL.
Note: DON'T MERGE
--- include/drm-uapi/xe_drm.h | 289 ++++++++++++++++++++++++++++++++++++-- 1 file changed, 281 insertions(+), 8 deletions(-)
include/drm-uapi/xe_drm.h +281 -8
@@ -3,8 +3,8 @@* Copyright © 2023 Intel Corporation */ -#ifndef _XE_DRM_H_ -#define _XE_DRM_H_ +#ifndef _UAPI_XE_DRM_H_ +#define _UAPI_XE_DRM_H_#include "drm.h"
@@ -81,6 +81,8 @@ extern "C" {* - &DRM_IOCTL_XE_EXEC * - &DRM_IOCTL_XE_WAIT_USER_FENCE * - &DRM_IOCTL_XE_OBSERVATION +* - &DRM_IOCTL_XE_MADVISE +* - &DRM_IOCTL_XE_VM_QUERY_MEM_RANGE_ATTRS */ /*
@@ -102,6 +104,8 @@ extern "C" {#define DRM_XE_EXEC 0x09#define DRM_XE_WAIT_USER_FENCE 0x0a#define DRM_XE_OBSERVATION 0x0b +#define DRM_XE_MADVISE 0x0c +#define DRM_XE_VM_QUERY_MEM_REGION_ATTRS 0x0d/* Must be kept compact -- no holes */
@@ -117,6 +121,8 @@ extern "C" {#define DRM_IOCTL_XE_EXEC DRM_IOW(DRM_COMMAND_BASE + DRM_XE_EXEC, struct drm_xe_exec)#define DRM_IOCTL_XE_WAIT_USER_FENCE DRM_IOWR(DRM_COMMAND_BASE + DRM_XE_WAIT_USER_FENCE, struct drm_xe_wait_user_fence)#define DRM_IOCTL_XE_OBSERVATION DRM_IOW(DRM_COMMAND_BASE + DRM_XE_OBSERVATION, struct drm_xe_observation_param) +#define DRM_IOCTL_XE_MADVISE DRM_IOW(DRM_COMMAND_BASE + DRM_XE_MADVISE, struct drm_xe_madvise) +#define DRM_IOCTL_XE_VM_QUERY_MEM_RANGE_ATTRS DRM_IOWR(DRM_COMMAND_BASE + DRM_XE_VM_QUERY_MEM_REGION_ATTRS, struct drm_xe_vm_query_mem_range_attr)/** * DOC: Xe IOCTL Extensions
@@ -134,7 +140,7 @@ extern "C" {* redefine the interface more easily than an ever growing struct of * increasing complexity, and for large parts of that interface to be * entirely optional. The downside is more pointer chasing; chasing across -* the boundary with pointers encapsulated inside u64. +* the __user boundary with pointers encapsulated inside u64. * * Example chaining: *
@@ -925,9 +931,9 @@ struct drm_xe_gem_mmap_offset {* - %DRM_XE_VM_CREATE_FLAG_LR_MODE - An LR, or Long Running VM accepts * exec submissions to its exec_queues that don't have an upper time * limit on the job execution time. But exec submissions to these -* don't allow any of the flags DRM_XE_SYNC_FLAG_SYNCOBJ, -* DRM_XE_SYNC_FLAG_TIMELINE_SYNCOBJ, DRM_XE_SYNC_FLAG_DMA_BUF, -* used as out-syncobjs, that is, together with DRM_XE_SYNC_FLAG_SIGNAL. +* don't allow any of the sync types DRM_XE_SYNC_TYPE_SYNCOBJ, +* DRM_XE_SYNC_TYPE_TIMELINE_SYNCOBJ, used as out-syncobjs, that is, +* together with sync flag DRM_XE_SYNC_FLAG_SIGNAL. * LR VMs can be created in recoverable page-fault mode using * DRM_XE_VM_CREATE_FLAG_FAULT_MODE, if the device supports it. * If that flag is omitted, the UMD can not rely on the slightly
@@ -1003,6 +1009,10 @@ struct drm_xe_vm_destroy {* valid on VMs with DRM_XE_VM_CREATE_FLAG_FAULT_MODE set. The CPU address * mirror flag are only valid for DRM_XE_VM_BIND_OP_MAP operations, the BO * handle MBZ, and the BO offset MBZ. +* +* The @prefetch_mem_region_instance for %DRM_XE_VM_BIND_OP_PREFETCH can also be: +* - %DRM_XE_CONSULT_MEM_ADVISE_PREF_LOC, which ensures prefetching occurs in +* the memory region advised by madvise. */ struct drm_xe_vm_bind_op {/** @extensions: Pointer to the first extension struct, if any */
@@ -1108,6 +1118,7 @@ struct drm_xe_vm_bind_op {/** @flags: Bind flags */ __u32 flags; +#define DRM_XE_CONSULT_MEM_ADVISE_PREF_LOC -1/** * @prefetch_mem_region_instance: Memory region to prefetch VMA to. * It is a region instance, not a mask.
@@ -1394,7 +1405,7 @@ struct drm_xe_sync {/** * @timeline_value: Input for the timeline sync object. Needs to be -* different than 0 when used with %DRM_XE_SYNC_FLAG_TIMELINE_SYNCOBJ. +* different than 0 when used with %DRM_XE_SYNC_TYPE_TIMELINE_SYNCOBJ. */ __u64 timeline_value;
@@ -1974,8 +1985,270 @@ struct drm_xe_query_eu_stall {__u64 sampling_rates[]; }; +/** +* struct drm_xe_madvise - Input of &DRM_IOCTL_XE_MADVISE +* +* This structure is used to set memory attributes for a virtual address range +* in a VM. The type of attribute is specified by @type, and the corresponding +* union member is used to provide additional parameters for @type. +* +* Supported attribute types: +* - DRM_XE_MEM_RANGE_ATTR_PREFERRED_LOC: Set preferred memory location. +* - DRM_XE_MEM_RANGE_ATTR_ATOMIC: Set atomic access policy. +* - DRM_XE_MEM_RANGE_ATTR_PAT: Set page attribute table index. +* +* Example: +* +* .. code-block:: C +* +* struct drm_xe_madvise madvise = { +* .vm_id = vm_id, +* .start = 0x100000, +* .range = 0x2000, +* .type = DRM_XE_MEM_RANGE_ATTR_ATOMIC, +* .atomic_val = DRM_XE_ATOMIC_DEVICE, +* .pad = 0, +* }; +* +* ioctl(fd, DRM_IOCTL_XE_MADVISE, &madvise); +* +*/ +struct drm_xe_madvise { +/** @extensions: Pointer to the first extension struct, if any */ +__u64 extensions; + +/** @start: start of the virtual address range */ +__u64 start; + +/** @range: size of the virtual address range */ +__u64 range; + +/** @vm_id: vm_id of the virtual range */ +__u32 vm_id; + +#define DRM_XE_MEM_RANGE_ATTR_PREFERRED_LOC 0 +#define DRM_XE_MEM_RANGE_ATTR_ATOMIC 1 +#define DRM_XE_MEM_RANGE_ATTR_PAT 2 +/** @type: type of attribute */ +__u32 type; + +union { +/** +* @preferred_mem_loc: preferred memory location +* +* Used when @type == DRM_XE_MEM_RANGE_ATTR_PREFERRED_LOC +* +* Supported values for @preferred_mem_loc.devmem_fd: +* - DRM_XE_PREFERRED_LOC_DEFAULT_DEVICE: set vram of faulting tile as preferred loc +* - DRM_XE_PREFERRED_LOC_DEFAULT_SYSTEM: set smem as preferred loc +* +* Supported values for @preferred_mem_loc.migration_policy: +* - DRM_XE_MIGRATE_ALL_PAGES +* - DRM_XE_MIGRATE_ONLY_SYSTEM_PAGES +*/ +struct { +#define DRM_XE_PREFERRED_LOC_DEFAULT_DEVICE 0 +#define DRM_XE_PREFERRED_LOC_DEFAULT_SYSTEM -1 +/** @preferred_mem_loc.devmem_fd: fd for preferred loc */ +__u32 devmem_fd; + +#define DRM_XE_MIGRATE_ALL_PAGES 0 +#define DRM_XE_MIGRATE_ONLY_SYSTEM_PAGES 1 +/** @preferred_mem_loc.migration_policy: Page migration policy */ +__u16 migration_policy; + +/** @preferred_mem_loc.pad : MBZ */ +__u16 pad; + +/** @preferred_mem_loc.reserved : Reserved */ +__u64 reserved; + } preferred_mem_loc; + +/** +* @atomic: Atomic access policy +* +* Used when @type == DRM_XE_MEM_RANGE_ATTR_ATOMIC. +* +* Supported values for @atomic.val: +* - DRM_XE_ATOMIC_UNDEFINED: Undefined or default behaviour +* Support both GPU and CPU atomic operations for system allocator +* Support GPU atomic operations for normal(bo) allocator +* - DRM_XE_ATOMIC_DEVICE: Support GPU atomic operations +* - DRM_XE_ATOMIC_GLOBAL: Support both GPU and CPU atomic operations +* - DRM_XE_ATOMIC_CPU: Support CPU atomic +*/ +struct { +#define DRM_XE_ATOMIC_UNDEFINED 0 +#define DRM_XE_ATOMIC_DEVICE 1 +#define DRM_XE_ATOMIC_GLOBAL 2 +#define DRM_XE_ATOMIC_CPU 3 +/** @atomic.val: value of atomic operation */ +__u32 val; + +/** @atomic.pad: MBZ */ +__u32 pad; + +/** @atomic.reserved: Reserved */ +__u64 reserved; + } atomic; + +/** +* @pat_index: Page attribute table index +* +* Used when @type == DRM_XE_MEM_RANGE_ATTR_PAT. +*/ +struct { +/** @pat_index.val: PAT index value */ +__u32 val; + +/** @pat_index.pad: MBZ */ +__u32 pad; + +/** @pat_index.reserved: Reserved */ +__u64 reserved; + } pat_index; + }; + +/** @reserved: Reserved */ +__u64 reserved[2 ]; +}; + +/** +* struct drm_xe_mem_range_attr - Output of &DRM_IOCTL_XE_VM_QUERY_MEM_RANGES_ATTRS +* +* This structure is provided by userspace and filled by KMD in response to the +* DRM_IOCTL_XE_VM_QUERY_MEM_RANGES_ATTRS ioctl. It describes memory attributes of +* a memory ranges within a user specified address range in a VM. +* +* The structure includes information such as atomic access policy, +* page attribute table (PAT) index, and preferred memory location. +* Userspace allocates an array of these structures and passes a pointer to the +* ioctl to retrieve attributes for each memory ranges +* +* @extensions: Pointer to the first extension struct, if any +* @start: Start address of the memory range +* @end: End address of the virtual memory range +* +*/ +struct drm_xe_mem_range_attr { +/** @extensions: Pointer to the first extension struct, if any */ +__u64 extensions; + +/** @start: start of the memory range */ +__u64 start; + +/** @end: end of the memory range */ +__u64 end; + +/** @preferred_mem_loc: preferred memory location */ +struct { +/** @preferred_mem_loc.devmem_fd: fd for preferred loc */ +__u32 devmem_fd; + +/** @preferred_mem_loc.migration_policy: Page migration policy */ +__u32 migration_policy; + } preferred_mem_loc; + +struct { +/** @atomic.val: atomic attribute */ +__u32 val; + +/** @atomic.reserved: Reserved */ +__u32 reserved; + } atomic; + +struct { +/** @pat_index.val: PAT index */ +__u32 val; + +/** @pat_index.reserved: Reserved */ +__u32 reserved; + } pat_index; + +/** @reserved: Reserved */ +__u64 reserved[2 ]; +}; + +/** +* struct drm_xe_vm_query_mem_range_attr - Input of &DRM_IOCTL_XE_VM_QUERY_MEM_ATTRIBUTES +* +* This structure is used to query memory attributes of memory regions +* within a user specified address range in a VM. It provides detailed +* information about each memory range, including atomic access policy, +* page attribute table (PAT) index, and preferred memory location. +* +* Userspace first calls the ioctl with @num_mem_ranges = 0, +* @sizeof_mem_ranges_attr = 0 and @vector_of_vma_mem_attr = NULL to retrieve +* the number of memory regions and size of each memory range attribute. +* Then, it allocates a buffer of that size and calls the ioctl again to fill +* the buffer with memory range attributes. +* +* If second call fails with -ENOSPC, it means memory ranges changed between +* first call and now, retry IOCTL again with @num_mem_ranges = 0, +* @sizeof_mem_ranges_attr = 0 and @vector_of_vma_mem_attr = NULL followed by +* Second ioctl call. +* +* Example: +* +* .. code-block:: C +* struct drm_xe_vm_query_mem_range_attr query = { +* .vm_id = vm_id, +* .start = 0x100000, +* .range = 0x2000, +* }; +* +* // First ioctl call to get num of mem regions and sizeof each attribute +* ioctl(fd, DRM_IOCTL_XE_VM_QUERY_MEM_RANGE_ATTRS, &query); +* +* // Allocate buffer for the memory region attributes +* void *ptr = malloc(query.num_mem_ranges * query.sizeof_mem_range_attr); +* +* query.vector_of_mem_attr = (uintptr_t)ptr; +* +* // Second ioctl call to actually fill the memory attributes +* ioctl(fd, DRM_IOCTL_XE_VM_QUERY_MEM_RANGE_ATTRS, &query); +* +* // Iterate over the returned memory region attributes +* for (unsigned int i = 0; i < query.num_mem_ranges; ++i) { +* struct drm_xe_mem_range_attr *attr = (struct drm_xe_mem_range_attr *)ptr; +* +* // Do something with attr +* +* // Move pointer by one entry +* ptr += query.sizeof_mem_range_attr; +* } +* +* free(ptr); +*/ +struct drm_xe_vm_query_mem_range_attr { +/** @extensions: Pointer to the first extension struct, if any */ +__u64 extensions; + +/** @vm_id: vm_id of the virtual range */ +__u32 vm_id; + +/** @num_mem_ranges: number of mem_ranges in range */ +__u32 num_mem_ranges; + +/** @start: start of the virtual address range */ +__u64 start; + +/** @range: size of the virtual address range */ +__u64 range; + +/** @sizeof_mem_range_attr: size of struct drm_xe_mem_range_attr */ +__u64 sizeof_mem_range_attr; + +/** @vector_of_ops: userptr to array of struct drm_xe_mem_range_attr */ +__u64 vector_of_mem_attr; + +/** @reserved: Reserved */ +__u64 reserved[2 ]; + +}; +#if defined(__cplusplus) }#endif -#endif /* _XE_DRM_H_ */ +#endif /* _UAPI_XE_DRM_H_ */
Signature · 1 line
2.43.0
PATCH 2/5 lib/xe: Added xe_vm_madvise ioctl support
In reply to [email protected]From: Nishit Sharma <[email protected]>
Added xe_vm_madvise() which issues madvise ioctl DRM_IOCTL_XE_MADVISE for VM region advising the driver about expected usage or memory policy for specified address range. MADVISE ioctl requires pointer to drm_xe_madvise structure as one of the inputs. Depending upon type of madvise operation like Atomic, Preferred LOC or PAT required members of drm_xe_madvise structure are initialized and passed in MADVISE ioctl.
--- lib/xe/xe_ioctl.c | 56 +++++++++++++++++++++++++++++++++++++++++++++++ lib/xe/xe_ioctl.h | 5 ++++- 2 files changed, 60 insertions(+), 1 deletion(-)
lib/xe/xe_ioctl.c +56 -0
@@ -585,3 +585,59 @@ int64_t xe_wait_ufence(int fd, uint64_t *addr, uint64_t value,igt_assert_eq (__xe_wait_ufence (fd, addr, value, exec_queue, &timeout),0 );return timeout; } + +int __xe_vm_madvise (int fd,uint32_t vm,uint64_t addr,uint64_t range, +uint64_t ext,uint32_t type,uint32_t op_val,uint16_t policy) +{ +struct drm_xe_madvise madvise = {}; + + madvise.extensions = ext; + madvise.vm_id = vm; + madvise.start = addr; + madvise.range = range; + +if (type ==DRM_XE_MEM_RANGE_ATTR_ATOMIC ) { + madvise.type =DRM_XE_MEM_RANGE_ATTR_ATOMIC ; + madvise.atomic.val = op_val; + }else if (type ==DRM_XE_MEM_RANGE_ATTR_PREFERRED_LOC ) { + madvise.type =DRM_XE_MEM_RANGE_ATTR_PREFERRED_LOC ; + madvise.preferred_mem_loc.devmem_fd = op_val; + madvise.preferred_mem_loc.migration_policy = policy; +igt_debug ("madvise.preferred_mem_loc.devmem_fd = %d\n" , + madvise.preferred_mem_loc.devmem_fd); + }else if (type ==DRM_XE_MEM_RANGE_ATTR_PAT ) { + madvise.type =DRM_XE_MEM_RANGE_ATTR_PAT ; + madvise.pat_index.val = op_val; + }else { +igt_warn ("Unknown attribute\n" ); +return -EINVAL ; + } + +if (igt_ioctl (fd,DRM_IOCTL_XE_MADVISE , &madvise)) +return -errno; + +return 0 ; +} + +/** +* xe_vm_madvise: +* @fd: xe device fd +* @vm: vm_id of the virtual range +* @addr: start of the virtual address range +* @range: size of the virtual address range +* @ext: Pointer to the first extension struct, if any +* @type: type of attribute +* @op_val: fd/atomic value/pat index, depending upon type of operation +* @policy: Page migration policy +* +* Function initializes different members of struct drm_xe_madvise and calls +* MADVISE IOCTL . +* +* Returns 0 if success and asserts otherwise. +*/ +int xe_vm_madvise (int fd,uint32_t vm,uint64_t addr,uint64_t range, +uint64_t ext,uint32_t type,uint32_t op_val,uint16_t policy) +{ +igt_assert_eq (__xe_vm_madvise (fd, vm, addr, range, ext, type, op_val, policy),0 ); +return 0 ; +}
lib/xe/xe_ioctl.h +4 -1
@@ -99,5 +99,8 @@ int __xe_wait_ufence(int fd, uint64_t *addr, uint64_t value, uint32_t exec_queue, int64_t *timeout);int64_t xe_wait_ufence (int fd,uint64_t *addr,uint64_t value,uint32_t exec_queue,int64_t timeout); - +int __xe_vm_madvise (int fd,uint32_t vm,uint64_t addr,uint64_t range,uint64_t ext, +uint32_t type,uint32_t op_val,uint16_t policy); +int xe_vm_madvise (int fd,uint32_t vm,uint64_t addr,uint64_t range,uint64_t ext, +uint32_t type,uint32_t op_val,uint16_t policy);#endif /* XE_IOCTL_H */
Signature · 1 line
2.43.0
PATCH 3/5 lib/xe: Helper added to get memory attributes
In reply to [email protected]From: Nishit Sharma <[email protected]>
xe_vm_print_mem_attr_values_in_range() function added which calls QUERY_MEM_RANGES_ATTRS ioctl to get different memory attributes from KMD and then prints memory attributes returned by KMD for different access policies like atomic access, preferred loc and pat index.
--- lib/xe/xe_ioctl.c | 92 +++++++++++++++++++++++++++++++++++++++++++++++ lib/xe/xe_ioctl.h | 4 +++ 2 files changed, 96 insertions(+)
lib/xe/xe_ioctl.c +92 -0
@@ -57,6 +57,98 @@ uint64_t xe_bb_size(int fd, uint64_t reqsize) xe_get_default_alignment(fd)); } +int xe_vm_number_vmas_in_range (int fd,struct drm_xe_vm_query_mem_range_attr *vmas_attr) +{ +if (igt_ioctl (fd,DRM_IOCTL_XE_VM_QUERY_MEM_RANGE_ATTRS , vmas_attr)) +return -errno; +return 0 ; +} + +int xe_vm_vma_attrs (int fd,struct drm_xe_vm_query_mem_range_attr *vmas_attr, +struct drm_xe_mem_range_attr *mem_attr) +{ +if (!mem_attr) +return -EINVAL ; + + vmas_attr->vector_of_mem_attr = (uintptr_t )mem_attr; + +if (igt_ioctl (fd,DRM_IOCTL_XE_VM_QUERY_MEM_RANGE_ATTRS , vmas_attr)) +return -errno; + +return 0 ; +} + +/** +* xe_vm_print_mem_attr_values_in_range: +* @fd: xe device fd +* @vm: vm_id of the virtual range +* @start: start of the virtual address range +* @range: size of the virtual address range +* +* Calls QUERY_MEM_RANGES_ATTRS ioctl to get memory attributes for different +* memory ranges from KMD. prints memory attributes as returned by KMD for +* atomic, prefrred loc and pat index types. +* +* Returns 0 for success or error for failure +*/ + +int xe_vm_print_mem_attr_values_in_range (int fd,uint32_t vm,uint64_t start,uint64_t range) +{ + +void *ptr_start, *ptr; +int err; +struct drm_xe_vm_query_mem_range_attr query = { + .vm_id = vm, + .start = start, + .range = range, + .num_mem_ranges =0 , + .sizeof_mem_range_attr =0 , + .vector_of_mem_attr = (uintptr_t )NULL , + }; + +igt_debug ("mem_attr_values_in_range called start = %" PRIu64"\n range = %" PRIu64"\n" , + start, range); + + err =xe_vm_number_vmas_in_range (fd, &query); +if (err || !query.num_mem_ranges || !query.sizeof_mem_range_attr) { +igt_warn ("ioctl failed for xe_vm_number_vmas_in_range\n" ); +igt_debug ("vmas_in_range err = %d query.num_mem_ranges = %u query.sizeof_mem_range_attr=%lld\n" , + err, query.num_mem_ranges, query.sizeof_mem_range_attr); +return err; + } + +/* Allocate buffer for the memory region attributes */ + ptr =malloc (query.num_mem_ranges * query.sizeof_mem_range_attr); + ptr_start = ptr; + +if (!ptr) +return -ENOMEM ; + + err =xe_vm_vma_attrs (fd, &query, ptr); +if (err) { +igt_warn ("ioctl failed for vma_attrs err = %d\n" , err); +return err; + } + +/* Iterate over the returned memory region attributes */ +for (unsigned int i =0 ; i < query.num_mem_ranges; ++i) { +struct drm_xe_mem_range_attr *mem_attrs = (struct drm_xe_mem_range_attr *)ptr; + +igt_info ("vma_id = %d\nvma_start = 0x%016llx\nvma_end = 0x%016llx\n" +"vma:atomic = %d\nvma:pat_index = %d\nvma:preferred_loc_region = %d\n" +"vma:preferred_loc_devmem_fd = %d\n\n\n" , i, mem_attrs->start, + mem_attrs->end, + mem_attrs->atomic.val, mem_attrs->pat_index.val, + mem_attrs->preferred_mem_loc.migration_policy, + mem_attrs->preferred_mem_loc.devmem_fd); + + ptr += query.sizeof_mem_range_attr; + } + +free (ptr_start); +return 0 ; +} +uint32_t xe_vm_create (int fd,uint32_t flags,uint64_t ext) {struct drm_xe_vm_create create = {
lib/xe/xe_ioctl.h +4 -0
@@ -103,4 +103,8 @@ int __xe_vm_madvise(int fd, uint32_t vm, uint64_t addr, uint64_t range, uint64_t uint32_t type, uint32_t op_val, uint16_t policy);int xe_vm_madvise (int fd,uint32_t vm,uint64_t addr,uint64_t range,uint64_t ext,uint32_t type,uint32_t op_val,uint16_t policy); +int xe_vm_number_vmas_in_range (int fd,struct drm_xe_vm_query_mem_range_attr *vmas_attr); +int xe_vm_vma_attrs (int fd,struct drm_xe_vm_query_mem_range_attr *vmas_attr, +struct drm_xe_mem_range_attr *mem_attr); +int xe_vm_print_mem_attr_values_in_range (int fd,uint32_t vm,uint64_t start,uint64_t range);#endif /* XE_IOCTL_H */
Signature · 1 line
2.43.0
PATCH 4/5 tests/intel/xe_exec_system_allocator: Added preferred_loc_smem test
In reply to [email protected]From: Nishit Sharma <[email protected]>
Added preferred-loc-smem test which is called in combination with other tests as well. In this test the buffer object preferred location is system memory. MADVISE ioctl is called with preferred_loc attribute and default_system system memory as preferred location.
--- tests/intel/xe_exec_system_allocator.c | 227 ++++++++++++++++++++----- 1 file changed, 188 insertions(+), 39 deletions(-)
tests/intel/xe_exec_system_allocator.c +188 -39
@@ -138,7 +138,6 @@ static void signal_pdata(struct process_data *pdata)#define CPU_FAULT_THREADS (0x1 << 2)#define CPU_FAULT_PROCESS (0x1 << 3)#define CPU_FAULT_SAME_PAGE (0x1 << 4) -static void process_check (void *ptr,uint64_t alloc_size,uint64_t stride,unsigned int flags) {
@@ -406,6 +405,39 @@ static void __aligned_partial_free(struct aligned_alloc_type *aligned_alloc_typ aligned_alloc_type->__size - aligned_alloc_type->size - begin_size); } +#define MAX_N_EXEC_QUEUES 16 + +#define MMAP (0x1 << 0) +#define NEW (0x1 << 1) +#define BO_UNMAP (0x1 << 2) +#define FREE (0x1 << 3) +#define BUSY (0x1 << 4) +#define BO_MAP (0x1 << 5) +#define RACE (0x1 << 6) +#define SKIP_MEMSET (0x1 << 7) +#define FAULT (0x1 << 8) +#define FILE_BACKED (0x1 << 9) +#define LOCK (0x1 << 10) +#define MMAP_SHARED (0x1 << 11) +#define HUGE_PAGE (0x1 << 12) +#define SHARED_ALLOC (0x1 << 13) +#define FORK_READ (0x1 << 14) +#define FORK_READ_AFTER (0x1 << 15) +#define MREMAP (0x1 << 16) +#define DONTUNMAP (0x1 << 17) +#define READ_ONLY_REMAP (0x1 << 18) +#define SYNC_EXEC (0x1 << 19) +#define EVERY_OTHER_CHECK (0x1 << 20) +#define MULTI_FAULT (0x1 << 21) +#define PREFETCH (0x1 << 22) +#define THREADS (0x1 << 23) +#define PROCESSES (0x1 << 24) +#define PREFETCH_BENCHMARK (0x1 << 25) +#define PREFETCH_SYS_BENCHMARK (0x1 << 26) +#define PREFERRED_LOC_SMEM (0x1 << 27) + +#define N_MULTI_FAULT 4 +/** * SUBTEST: unaligned-alloc * Description: allocate unaligned sizes of memory
@@ -460,7 +492,7 @@ many_allocs(int fd, struct drm_xe_engine_class_instance *eci,uint32_t *bos =NULL ;struct timespec tv = {};uint64_t submit, read, elapsed; -int i; +int i, err; vm =xe_vm_create (fd,DRM_XE_VM_CREATE_FLAG_LR_MODE |DRM_XE_VM_CREATE_FLAG_FAULT_MODE ,0 );
@@ -500,6 +532,15 @@ many_allocs(int fd, struct drm_xe_engine_class_instance *eci, alloc.ptr =aligned_alloc (SZ_2M , alloc_size);igt_assert (alloc.ptr); } + +if (flags &PREFERRED_LOC_SMEM ) { + err =xe_vm_madvise (fd, vm,to_user_pointer (alloc.ptr), alloc_size,0 , +DRM_XE_MEM_RANGE_ATTR_PREFERRED_LOC , +DRM_XE_PREFERRED_LOC_DEFAULT_SYSTEM ,0 ); +if (err) +igt_warn ("MADVISE_FAILURE err = %s, vm =%u data=%" PRIu64" alloc_size =%" PRIu64"\n" , +strerror (errno), vm,to_user_pointer (alloc.ptr), alloc_size); + } allocs[i] = alloc; touch_all_pages(fd, exec_queue, allocs[i].ptr, alloc_size, stride,
@@ -662,7 +703,7 @@ partial(int fd, struct drm_xe_engine_class_instance *eci, unsigned int flags)size_t bo_size =SZ_2M , unmap_offset =0 ;uint32_t vm, exec_queue;u64 *exec_ufence =NULL ; -int i; +int i, err;void *old, *new =NULL ;struct aligned_alloc_type alloc;
@@ -688,6 +729,15 @@ partial(int fd, struct drm_xe_engine_class_instance *eci, unsigned int flags)xe_wait_ufence (fd, &data[0 ].vm_sync,USER_FENCE_VALUE ,0 ,FIVE_SEC ); data[0 ].vm_sync =0 ; +if (flags &PREFERRED_LOC_SMEM ) { + err =xe_vm_madvise (fd, vm,to_user_pointer (data), bo_size,0 , +DRM_XE_MEM_RANGE_ATTR_PREFERRED_LOC , +DRM_XE_PREFERRED_LOC_DEFAULT_SYSTEM ,0 ); +if (err) +igt_warn ("MADVISE_FAILURE err = %s, vm =%u data=%" PRIu64" alloc_size = %zu\n" , +strerror (errno), vm,to_user_pointer (data), bo_size); + } + exec_ufence =mmap (NULL ,SZ_4K ,PROT_READ |PROT_WRITE ,MAP_SHARED |MAP_ANONYMOUS ,-1 ,0 );
@@ -747,38 +797,6 @@ partial(int fd, struct drm_xe_engine_class_instance *eci, unsigned int flags)xe_vm_destroy (fd, vm); } -#define MAX_N_EXEC_QUEUES 16 - -#define MMAP (0x1 << 0) -#define NEW (0x1 << 1) -#define BO_UNMAP (0x1 << 2) -#define FREE (0x1 << 3) -#define BUSY (0x1 << 4) -#define BO_MAP (0x1 << 5) -#define RACE (0x1 << 6) -#define SKIP_MEMSET (0x1 << 7) -#define FAULT (0x1 << 8) -#define FILE_BACKED (0x1 << 9) -#define LOCK (0x1 << 10) -#define MMAP_SHARED (0x1 << 11) -#define HUGE_PAGE (0x1 << 12) -#define SHARED_ALLOC (0x1 << 13) -#define FORK_READ (0x1 << 14) -#define FORK_READ_AFTER (0x1 << 15) -#define MREMAP (0x1 << 16) -#define DONTUNMAP (0x1 << 17) -#define READ_ONLY_REMAP (0x1 << 18) -#define SYNC_EXEC (0x1 << 19) -#define EVERY_OTHER_CHECK (0x1 << 20) -#define MULTI_FAULT (0x1 << 21) -#define PREFETCH (0x1 << 22) -#define THREADS (0x1 << 23) -#define PROCESSES (0x1 << 24) -#define PREFETCH_BENCHMARK (0x1 << 25) -#define PREFETCH_SYS_BENCHMARK (0x1 << 26) - -#define N_MULTI_FAULT 4 -/** * SUBTEST: once-%s * Description: Run %arg[1] system allocator test only once
@@ -951,6 +969,80 @@ partial(int fd, struct drm_xe_engine_class_instance *eci, unsigned int flags)* @mmap-new-nomemset: mmap a new buffer for each exec, skip memset of buffers * @mmap-new-huge-nomemset: mmap huge page new buffer for each exec, skip memset of buffers * @mmap-new-race-nomemset: mmap a new buffer for each exec with race between cpu and gpu access, skip memset of buffers +* @free-nomemset-preferred-loc-smem: malloc and free buffer for each exec and perform preferred loc madvise operation +* @free-preferred-loc-smem: free buffer for each exec and perform preferred loc madvise operation +* @free-race-nomemset-preferred-loc-smem: free buffer for each exec with race between cpu and gpu access and perform madvise operation skipping memset +* @free-race-preferred-loc-smem: free buffer for each exec with race between cpu and gpu access and perform madvise operation +* @malloc-bo-unmap-nomemset-preferred-loc-smem: malloc single buffer for all execs, bind and unbind a BO to same address, skip memset and perform madvise operation +* @malloc-busy-nomemset-preferred-loc-smem: malloc single buffer for all execs, try to unbind while buffer valid, skip memset of buffers and perform madvise operation +* @malloc-busy-preferred-loc-smem: malloc single buffer for all execs, try to unbind while buffer valid and perform madvise operation +* @malloc-fork-read-after-preferred-loc-smem: malloc single buffer for all execs, fork a process to read test output, perform madvise operation +* @malloc-fork-read-preferred-loc-smem: malloc single buffer for all execs, fork a process to read test output, perform madvise operation +* @malloc-mlock-nomemset-preferred-loc-smem: malloc and mlock single buffer for all execs, skip memset of buffers, perform madvise operation +* @malloc-mlock-preferred-loc-smem: malloc and mlock single buffer for all execs, perform madvise operation +* @malloc-multi-fault-preferred-loc-smem: malloc single buffer for all execs and perform madvise operation +* @malloc-nomemset-preferred-loc-smem: malloc single buffer for all execs, skip memset of buffers and perform madvise operation +* @malloc-preferred-loc-smem: malloc single buffer for all execs, issue a command which will trigger multiple faults, perform madvise operation +* @malloc-prefetch-preferred-loc-smem: malloc single buffer for all execs, prefetch buffer before each exec, perform madvise operation +* @malloc-prefetch-race-preferred-loc-smem: malloc single buffer for all execs, prefetch buffer before each exec, perform madvise operation +* @malloc-race-nomemset-preferred-loc-smem: malloc single buffer for all execs with race between cpu and gpu access, perform madvise operation +* @malloc-race-preferred-loc-smem: malloc single buffer for all execs with race between cpu and gpu access, perform madvise operation +* @free-race-nomemset-preferred-loc-smem: malloc and free buffer for each exec with race between cpu and gpu access, perform madvise operation +* @free-race-preferred-loc-smem: malloc and free buffer for each exec with race between cpu and gpu access, perform madvise operation +* @malloc-bo-unmap-nomemset-preferred-loc-smem: malloc single buffer for all execs, bind and unbind a BO to same address before execs, perform madvise operation +* @malloc-bo-unmap-preferred-loc-smem: malloc single buffer for all execs and perform madvise operation +* @malloc-busy-nomemset-preferred-loc-smem: malloc single buffer for all execs and perform madvise operation +* @malloc-busy-preferred-loc-smem: malloc single buffer for all execs and perform madvise +* @mmap-file-mlock-nomemset-preferred-loc-smem: mmap and mlock single buffer, with file backing, perform madvise +* @mmap-file-mlock-preferred-loc-smem: mmap and mlock single buffer, with file backing, perform madvise +* @mmap-file-nomemset-preferred-loc-smem: mmap single buffer, with file backing and perform madvise +* @mmap-file-preferred-loc-smem: mmap single buffer, with file backing and perform madvise +* @mmap-free-huge-nomemset-preferred-loc-smem: mmap huge page and free buffer for each exec and perform madvise +* @mmap-free-huge-preferred-loc-smem: mmap huge page and free buffer for each exec and perform madvise +* @mmap-free-nomemset-preferred-loc-smem: mmap and free buffer for each exec and perform madvise +* @mmap-free-preferred-loc-smem: mmap and free buffer for each exec and perform madvise +* @mmap-free-race-nomemset-preferred-loc-smem: +* @mmap-free-race-preferred-loc-smem: +* @mmap-huge-nomemset-preferred-loc-smem: +* @mmap-huge-preferred-loc-smem: +* @mmap-mlock-nomemset-preferred-loc-smem: +* @mmap-mlock-preferred-loc-smem: +* @mmap-new-huge-nomemset-preferred-loc-smem: +* @mmap-new-huge-preferred-loc-smem: +* @mmap-new-nomemset-preferred-loc-smem: +* @mmap-new-preferred-loc-smem: +* @mmap-new-race-nomemset-preferred-loc-smem: +* @mmap-new-race-preferred-loc-smem: +* @mmap-nomemset-preferred-loc-smem: +* @mmap-preferred-loc-smem: +* @mmap-prefetch-preferred-loc-smem: +* @mmap-prefetch-shared-preferred-loc-smem: +* @mmap-race-nomemset-preferred-loc-smem: +* @mmap-race-preferred-loc-smem: +* @mmap-remap-dontunmap-eocheck-preferred-loc-smem: +* @mmap-remap-dontunmap-preferred-loc-smem: +* @mmap-remap-eocheck-preferred-loc-smem: +* @mmap-remap-preferred-loc-smem: +* @mmap-remap-ro-dontunmap-eocheck-preferred-loc-smem: +* @mmap-remap-ro-dontunmap-preferred-loc-smem: +* @mmap-remap-ro-eocheck-preferred-loc-smem: +* @mmap-remap-ro-preferred-loc-smem: +* @mmap-shared-nomemset-preferred-loc-smem: +* @mmap-shared-preferred-loc-smem: +* @mmap-shared-remap-dontunmap-eocheck-preferred-loc-smem: +* @mmap-shared-remap-dontunmap-preferred-loc-smem: +* @mmap-shared-remap-eocheck-preferred-loc-smem: +* @mmap-shared-remap-preferred-loc-smem: +* @new-bo-map-nomemset-preferred-loc-smem: +* @new-bo-map-preferred-loc-smem: +* @new-busy-nomemset-preferred-loc-smem: +* @new-busy-preferred-loc-smem: +* @new-nomemset-preferred-loc-smem: +* @new-preferred-loc-smem: +* @new-prefetch-preferred-loc-smem: +* @new-race-nomemset-preferred-loc-smem: +* @new-race-preferred-loc-smem: +* @prefetch-benchmark: * * SUBTEST: prefetch-benchmark * Description: Prefetch a 64M buffer 128 times, measure bandwidth of prefetch
@@ -1020,14 +1112,14 @@ test_exec(int fd, struct drm_xe_engine_class_instance *eci,uint32_t bo =0 , bind_sync =0 ;void **pending_free;u64 *exec_ufence =NULL , *bind_ufence =NULL ; -int i, j, b, file_fd =-1 , prev_idx, pf_count; +int i, j, b, file_fd =-1 , prev_idx, pf_count, err;bool free_vm = false;size_t aligned_size = bo_size ?:xe_get_default_alignment (fd);size_t orig_size = bo_size;struct aligned_alloc_type aligned_alloc_type;uint32_t mem_region =vram_if_possible (fd, eci->gt_id);uint32_t region = mem_region &4 ?2 : mem_region &2 ?1 :0 ; -uint64_t prefetch_ns =0 , prefetch_sys_ns =0 ; +uint64_t prefetch_ns =0 ;const char *pf_count_stat ="svm_pagefault_count" ;if (flags &MULTI_FAULT ) {
@@ -1133,6 +1225,15 @@ test_exec(int fd, struct drm_xe_engine_class_instance *eci, addr =to_user_pointer (data); +if (flags &PREFERRED_LOC_SMEM ) { + err =xe_vm_madvise (fd, vm,to_user_pointer (data), bo_size,0 , +DRM_XE_MEM_RANGE_ATTR_PREFERRED_LOC , +DRM_XE_PREFERRED_LOC_DEFAULT_SYSTEM ,0 ); +if (err) +igt_warn ("MADVISE_FAILURE err = %s, vm =%u data=%" PRIu64" alloc_size = %zu\n" , +strerror (errno), vm,to_user_pointer (data), bo_size); + } +if (flags &BO_UNMAP ) { bo_flags =DRM_XE_GEM_CREATE_FLAG_NEEDS_VISIBLE_VRAM ; bo = xe_bo_create(fd, vm, bo_size,
@@ -1202,7 +1303,7 @@ test_exec(int fd, struct drm_xe_engine_class_instance *eci,uint64_t batch_addr = addr + batch_offset;uint64_t sdi_offset = (char *)&data[idx].data - (char *)data;uint64_t sdi_addr = addr + sdi_offset; -int e = i % n_exec_queues, err; +int e = i % n_exec_queues;bool fault_inject = (FAULT & flags) && i == n_execs /2 ;bool fault_injected = (FAULT & flags) && i > n_execs;
@@ -1232,6 +1333,16 @@ test_exec(int fd, struct drm_xe_engine_class_instance *eci, aligned_alloc_type =__aligned_alloc (aligned_size, bo_size); next_data = aligned_alloc_type.ptr;igt_assert (next_data); + +if (flags &PREFERRED_LOC_SMEM ) { + err =xe_vm_madvise (fd, vm,to_user_pointer (next_data), bo_size,0 , +DRM_XE_MEM_RANGE_ATTR_PREFERRED_LOC , +DRM_XE_PREFERRED_LOC_DEFAULT_SYSTEM ,0 ); +if (err) +igt_warn ("MADVISE_FAILURE err = %s, vm =%u data=%" PRIu64" alloc_size = %zu\n" , +strerror (errno), vm,to_user_pointer (next_data), bo_size); + } +__aligned_partial_free (&aligned_alloc_type); b =0 ;
@@ -1253,6 +1364,7 @@ test_exec(int fd, struct drm_xe_engine_class_instance *eci, sync[0 ].addr =to_user_pointer (bind_ufence); start =igt_nsec_elapsed (&tv); +xe_vm_prefetch_async (fd, vm,0 ,0 , addr, bo_size, sync,1 , region); end =igt_nsec_elapsed (&tv);
@@ -1355,6 +1467,7 @@ test_exec(int fd, struct drm_xe_engine_class_instance *eci, }else {igt_assert_eq (data[idx].data,READ_VALUE (&data[idx])); +if (flags &PREFETCH_SYS_BENCHMARK ) {struct timespec tv = {};u64 start, end;
@@ -1429,6 +1542,7 @@ test_exec(int fd, struct drm_xe_engine_class_instance *eci, aligned_alloc_type =__aligned_alloc (aligned_size, bo_size); data = aligned_alloc_type.ptr;igt_assert (data); +__aligned_partial_free (&aligned_alloc_type); bo_flags =DRM_XE_GEM_CREATE_FLAG_NEEDS_VISIBLE_VRAM ;
@@ -1450,6 +1564,7 @@ test_exec(int fd, struct drm_xe_engine_class_instance *eci, } bo =0 ; data =aligned_alloc (aligned_size, bo_size); +igt_assert (data); } addr =to_user_pointer (data);
@@ -1460,6 +1575,15 @@ test_exec(int fd, struct drm_xe_engine_class_instance *eci, prev_idx = idx; } +if (flags &PREFERRED_LOC_SMEM ) { + err =xe_vm_madvise (fd, vm,to_user_pointer (data), bo_size,0 , +DRM_XE_MEM_RANGE_ATTR_PREFERRED_LOC , +DRM_XE_PREFERRED_LOC_DEFAULT_SYSTEM ,0 ); +if (err) +igt_warn ("MADVISE_FAILURE err = %s, vm =%u data=%" PRIu64" alloc_size = %zu\n" , +strerror (errno), vm,to_user_pointer (data), bo_size); + } +if (flags &PREFETCH_BENCHMARK ) { igt_info("Prefetch VRAM execution took %.3fms, %.1f5 GB/s\n", 1e-6 * prefetch_ns,
@@ -1587,6 +1711,7 @@ threads(int fd, int n_exec_queues, int n_execs, size_t bo_size,uint32_t vm =0 ;bool go = false;void *alloc =NULL ; +int err;if ((FILE_BACKED |FORK_READ ) & flags)return ;
@@ -1614,6 +1739,15 @@ threads(int fd, int n_exec_queues, int n_execs, size_t bo_size, alloc =aligned_alloc (SZ_2M , alloc_size);igt_assert (alloc); +if (flags &PREFERRED_LOC_SMEM ) { + err =xe_vm_madvise (fd, vm,to_user_pointer (alloc), alloc_size,0 , +DRM_XE_MEM_RANGE_ATTR_PREFERRED_LOC , +DRM_XE_PREFERRED_LOC_DEFAULT_SYSTEM ,0 ); +if (err) +igt_warn ("MADVISE_FAILURE err = %s, vm =%u data=%" PRIu64" alloc_size = %zu\n" , +strerror (errno), vm,to_user_pointer (alloc), alloc_size); + } +memset (alloc,5 , alloc_size); flags &= ~SHARED_ALLOC ; }
@@ -1831,6 +1965,7 @@ igt_main { NULL }, };int fd; +int num_sections; igt_fixture {struct xe_device *xe;
@@ -1843,7 +1978,21 @@ igt_mainopen_sync_file (); } -for (const struct section *s = sections; s->name; s++) { + + num_sections =0 ; +for (const struct section *s = sections; s[num_sections].name; num_sections++) + ; + +for (int i =0 ; i < num_sections *2 ; i++) { +struct section *s = §ions[i % num_sections]; + +if (i/num_sections ==0 ) { +static char modified_name[256 ]; +snprintf (modified_name,sizeof (modified_name),"%s-preferred-loc-smem" , s->name); + s->name = modified_name; + s->flags |=PREFERRED_LOC_SMEM ; + } + igt_subtest_f("once-%s", s->name)xe_for_each_engine (fd, hwe) test_exec(fd, hwe, 1, 1, 0, 0, 0, NULL,
Signature · 1 line
2.43.0
PATCH 5/5 tests/intel/xe_exec_system_allocator: Added atomic_batch test in IGT
In reply to [email protected]From: Nishit Sharma <[email protected]>
ATOMIC_BATCH flag is introduced when true MI_ATOMIC | MI_ATOMIC_INC operation will be called. This will avoid writing another function which performs atomic increment operations. ATOMIC_BATCH flag is passed as argument in write_dword() if true then value will be written on passed address and incremented by ATOMIC_INC operation. For all memory operations this flag will be used to verify if ATOMIC operation is working or not.
--- tests/intel/xe_exec_system_allocator.c | 551 +++++++++++++++++++++---- 1 file changed, 481 insertions(+), 70 deletions(-)
tests/intel/xe_exec_system_allocator.c +481 -70
@@ -29,6 +29,14 @@#define USER_FENCE_VALUE 0xdeadbeefdeadbeefull#define QUARTER_SEC (NSEC_PER_SEC / 4)#define FIVE_SEC (5LL * NSEC_PER_SEC) +struct test_exec_data { +uint32_t batch[32 ]; +uint64_t pad; +uint64_t vm_sync; +uint64_t exec_sync; +uint32_t data; +uint32_t expected_data; +};struct batch_data {uint32_t batch[16 ];
@@ -37,6 +45,7 @@ struct batch_data {uint32_t expected_data; }; +#define VAL_ATOMIC_EXPECTED 56#define WRITE_VALUE (data__, i__) ({ \ if (!(data__)->expected_data) \ (data__)->expected_data = rand() << 12 | (i__); \
@@ -53,10 +62,19 @@ static void __write_dword(uint32_t *batch, uint64_t sdi_addr, uint32_t wdata, batch[(*idx)++] = wdata; } -static void write_dword (uint32_t *batch,uint64_t sdi_addr,uint32_t wdata, -int *idx) +static void write_dword (struct test_exec_data *data,uint64_t sdi_addr,uint32_t wdata, +int *idx,bool atomic) { -__write_dword (batch, sdi_addr, wdata, idx); +uint32_t *batch = data->batch; + +if (atomic) { + data->data =55 ; + batch[(*idx)++] =MI_ATOMIC |MI_ATOMIC_INC ; + batch[(*idx)++] = sdi_addr; + batch[(*idx)++] = sdi_addr >>32 ; + }else { +__write_dword (batch, sdi_addr, wdata, idx); + } batch[(*idx)++] =MI_BATCH_BUFFER_END ; }
@@ -271,7 +289,7 @@ check_all_pages_threads(void *ptr, uint64_t alloc_size, uint64_t stride,static void touch_all_pages (int fd,uint32_t exec_queue,void *ptr,uint64_t alloc_size,uint64_t stride, -struct timespec *tv,uint64_t *submit) +struct timespec *tv,uint64_t *submit,bool atomic) {struct drm_xe_sync sync[1 ] = { { .type =DRM_XE_SYNC_TYPE_USER_FENCE ,
@@ -302,7 +320,8 @@ static void touch_all_pages(int fd, uint32_t exec_queue, void *ptr,uint64_t sdi_addr = addr + sdi_offset;int b =0 ; -write_dword (data->batch, sdi_addr,WRITE_VALUE (data, i), &b); +write_dword ((struct test_exec_data *)data, sdi_addr,WRITE_VALUE (data, i), + &b, atomic ? true : false);igt_assert (b <=ARRAY_SIZE (data->batch)); }
@@ -407,34 +426,45 @@ static void __aligned_partial_free(struct aligned_alloc_type *aligned_alloc_typ#define MAX_N_EXEC_QUEUES 16 -#define MMAP (0x1 << 0) -#define NEW (0x1 << 1) -#define BO_UNMAP (0x1 << 2) -#define FREE (0x1 << 3) -#define BUSY (0x1 << 4) -#define BO_MAP (0x1 << 5) -#define RACE (0x1 << 6) -#define SKIP_MEMSET (0x1 << 7) -#define FAULT (0x1 << 8) -#define FILE_BACKED (0x1 << 9) -#define LOCK (0x1 << 10) -#define MMAP_SHARED (0x1 << 11) -#define HUGE_PAGE (0x1 << 12) -#define SHARED_ALLOC (0x1 << 13) -#define FORK_READ (0x1 << 14) -#define FORK_READ_AFTER (0x1 << 15) -#define MREMAP (0x1 << 16) -#define DONTUNMAP (0x1 << 17) -#define READ_ONLY_REMAP (0x1 << 18) -#define SYNC_EXEC (0x1 << 19) -#define EVERY_OTHER_CHECK (0x1 << 20) -#define MULTI_FAULT (0x1 << 21) -#define PREFETCH (0x1 << 22) -#define THREADS (0x1 << 23) -#define PROCESSES (0x1 << 24) -#define PREFETCH_BENCHMARK (0x1 << 25) -#define PREFETCH_SYS_BENCHMARK (0x1 << 26) -#define PREFERRED_LOC_SMEM (0x1 << 27) +#define MMAP (0x1 << 0) +#define NEW (0x1 << 1) +#define BO_UNMAP (0x1 << 2) +#define FREE (0x1 << 3) +#define BUSY (0x1 << 4) +#define BO_MAP (0x1 << 5) +#define RACE (0x1 << 6) +#define SKIP_MEMSET (0x1 << 7) +#define FAULT (0x1 << 8) +#define FILE_BACKED (0x1 << 9) +#define LOCK (0x1 << 10) +#define MMAP_SHARED (0x1 << 11) +#define HUGE_PAGE (0x1 << 12) +#define SHARED_ALLOC (0x1 << 13) +#define FORK_READ (0x1 << 14) +#define FORK_READ_AFTER (0x1 << 15) +#define MREMAP (0x1 << 16) +#define DONTUNMAP (0x1 << 17) +#define READ_ONLY_REMAP (0x1 << 18) +#define SYNC_EXEC (0x1 << 19) +#define EVERY_OTHER_CHECK (0x1 << 20) +#define MULTI_FAULT (0x1 << 21) +#define PREFETCH (0x1 << 22) +#define THREADS (0x1 << 23) +#define PROCESSES (0x1 << 24) +#define PREFETCH_BENCHMARK (0x1 << 25) +#define PREFETCH_SYS_BENCHMARK (0x1 << 26) +#define PREFERRED_LOC_SMEM (0x1 << 27) +#define ATOMIC_BATCH (0x1 << 28) +#define MIGRATE_ALL_PAGES (0x1 << 29) +#define PREFERRED_LOC_ATOMIC_DEVICE (0x1 << 30) +#define PREFERRED_LOC_ATOMIC_GL (0x1ull << 31) +#define PREFERRED_LOC_ATOMIC_CPU (0x1ull << 32) +#define MADVISE_MULTI_VMA (0x1ull << 33) +#define MADVISE_SPLIT_VMA (0x1ull << 34) +#define MADVISE_ATOMIC_VMA (0x1ull << 35) +#define PREFETCH_SPLIT_VMA (0x1ull << 36) +#define PREFETCH_CHANGE_ATTR (0x1ull << 37) +#define PREFETCH_SAME_ATTR (0x1ull << 38)#define N_MULTI_FAULT 4
@@ -478,6 +508,47 @@ static void __aligned_partial_free(struct aligned_alloc_type *aligned_alloc_typ* SUBTEST: processes-evict-malloc-mix-bo * Description: multi-process trigger eviction of VRAM allocated via malloc and BO create * Test category: stress test +* +* SUBTEST: madvise-multi-vma +* Description: performs multiple madvise operations on multiple virtual memory area using atomic device attributes +* Test category: functionality test +* +* SUBTEST: madvise-split-vma +* Description: perform madvise operations on multiple type VMAs (BO and CPU VMAs) +* Test category: perform madvise operations on multiple type VMAs (BO and CPU VMAs) +* +* SUBTEST: madvise-atomic-vma +* Description: perform madvise atomic operations on BO in VRAM/SMEM if atomic ATTR global/device +* Test category: functionality test +* +* SUBTEST: madvise-split-vma-with-mapping +* Description: performs prefetch and page migration +* Test category: functionality test +* +* SUBTEST: madvise-preffered-loc-atomic-vram +* Description: performs both atomic and preferred loc madvise operations atomic device attributes set +* Test category: functionality test +* +* SUBTEST: madvise-preffered-loc-atomic-gl +* Description: performs both atomic and preferred loc madvise operations with atomic global attributes set +* Test category: functionality test +* +* SUBTEST: madvise-preffered-loc-atomic-cpu +* Description: performs both atomic and preferred loc madvise operations with atomic cpu attributes set +* Test category: functionality test +* +* SUBTEST: madvise-preffered-loc-sram-migrate-pages +* Description: performs preferred loc madvise operations and migrating all pages in smem +* Test category: functionality test +* +* SUBTEST: madvise-no-range-invalidate-same-attr +* Description: performs atomic global madvise operation, prefetch and again madvise operation with same atomic attribute +* Test category: functionality test +* +* SUBTEST: madvise-range-invalidate-change-attr +* Description: performs atomic global madvise operation, prefetch and again madvise operation with different atomic attribute +* Test category: functionality test +* */ static void
@@ -544,7 +615,7 @@ many_allocs(int fd, struct drm_xe_engine_class_instance *eci, allocs[i] = alloc;touch_all_pages (fd, exec_queue, allocs[i].ptr, alloc_size, stride, - &tv, &submit); + &tv, &submit, flags &ATOMIC_BATCH ); }if (barrier)
@@ -692,7 +763,7 @@ partial(int fd, struct drm_xe_engine_class_instance *eci, unsigned int flags) .num_syncs = 1, .syncs = to_user_pointer(sync), }; -struct { +struct batch_data {uint32_t batch[16 ];uint64_t pad;uint64_t vm_sync;
@@ -750,7 +821,8 @@ partial(int fd, struct drm_xe_engine_class_instance *eci, unsigned int flags)uint64_t sdi_addr = addr + sdi_offset;int b =0 ; -write_dword (data[i].batch, sdi_addr,WRITE_VALUE (&data[i], i), &b); +write_dword ((struct test_exec_data *)&data[i], sdi_addr,WRITE_VALUE (&data[i], i), + &b,ATOMIC_BATCH ? true : false);igt_assert (b <=ARRAY_SIZE (data[i].batch));if (!i)
@@ -773,7 +845,10 @@ partial(int fd, struct drm_xe_engine_class_instance *eci, unsigned int flags)xe_wait_ufence (fd, new ?: exec_ufence,USER_FENCE_VALUE , exec_queue,FIVE_SEC );if (i || (flags &CPU_FAULT )) -igt_assert_eq (data[i].data,READ_VALUE (&data[i])); +igt_assert_eq (data[i].data, + flags &ATOMIC_BATCH + ?VAL_ATOMIC_EXPECTED + :READ_VALUE (&data[i])); exec_ufence[0 ] =0 ;if (!i) {
@@ -1072,16 +1147,6 @@ partial(int fd, struct drm_xe_engine_class_instance *eci, unsigned int flags)* Description: Create multiple threads with a faults on different hardware engines to same addresses, racing between CPU and GPU access * Test category: stress test */ - -struct test_exec_data { -uint32_t batch[32 ]; -uint64_t pad; -uint64_t vm_sync; -uint64_t exec_sync; -uint32_t data; -uint32_t expected_data; -}; -static void igt_require_hugepages (void ) { igt_skip_on_f(!igt_get_meminfo("HugePages_Total"),
@@ -1090,11 +1155,52 @@ static void igt_require_hugepages(void) "No huge pages available!\n"); } +static int +xe_vm_madvixe_pat_attr (int fd,uint32_t vm,uint64_t addr,uint64_t range, +int pat_index) +{ +int err; + + err =xe_vm_madvise (fd, vm, addr, range,0 , +DRM_XE_MEM_RANGE_ATTR_PAT , pat_index,0 ); +if (err) +return err; +return 0 ; +} + +static int +xe_vm_madvise_atomic_attr (int fd,uint32_t vm,uint64_t addr,uint64_t range, +int mem_attr) +{ +int err; + + err =xe_vm_madvise (fd, vm, addr, range,0 , +DRM_XE_MEM_RANGE_ATTR_ATOMIC , + mem_attr,0 ); +if (err) +return err; +return 0 ; +} + +static int +xe_vm_madvise_migrate_pages (int fd,uint32_t vm,uint64_t addr,uint64_t range) +{ +int err; + + err =xe_vm_madvise (fd, vm, addr, range,0 , +DRM_XE_MEM_RANGE_ATTR_PREFERRED_LOC , +DRM_XE_PREFERRED_LOC_DEFAULT_SYSTEM , +DRM_XE_MIGRATE_ALL_PAGES ); +if (err) +return err; +return 0 ; +} +static void test_exec (int fd,struct drm_xe_engine_class_instance *eci,int n_exec_queues,int n_execs,size_t bo_size,size_t stride,uint32_t vm,void *alloc,pthread_barrier_t *barrier, -unsigned int flags) +unsigned long long flags) {uint64_t addr;struct drm_xe_sync sync[1 ] = {
@@ -1107,9 +1213,10 @@ test_exec(int fd, struct drm_xe_engine_class_instance *eci, .syncs = to_user_pointer(sync), };uint32_t exec_queues[MAX_N_EXEC_QUEUES ]; -struct test_exec_data *data, *next_data =NULL ; +struct test_exec_data *data, *next_data =NULL , *org_data;uint32_t bo_flags;uint32_t bo =0 , bind_sync =0 ; +uint32_t val;void **pending_free;u64 *exec_ufence =NULL , *bind_ufence =NULL ;int i, j, b, file_fd =-1 , prev_idx, pf_count, err;
@@ -1234,6 +1341,133 @@ test_exec(int fd, struct drm_xe_engine_class_instance *eci, strerror(errno), vm, to_user_pointer(data), bo_size); } +if (flags &PREFERRED_LOC_SMEM ) { +if (flags &MIGRATE_ALL_PAGES ) { + err =xe_vm_madvise_migrate_pages (fd, vm,to_user_pointer (data), bo_size); +if (err) +igt_warn ("failure in page migration err = %s, vm =%u, data=%" PRIu64"\n" , +strerror (errno), vm,to_user_pointer (data)); + }else { + err =xe_vm_madvise (fd, vm,to_user_pointer (data), bo_size,0 , +DRM_XE_MEM_RANGE_ATTR_PREFERRED_LOC , +DRM_XE_PREFERRED_LOC_DEFAULT_SYSTEM ,0 ); + } + } +if (flags &PREFERRED_LOC_ATOMIC_DEVICE ) { + err =xe_vm_madvise_migrate_pages (fd, vm,to_user_pointer (data), bo_size); +if (err) +igt_warn ("failure in page migration err = %s, vm =%u data=%" PRIu64"\n" , +strerror (errno), vm,to_user_pointer (data)); + + err =xe_vm_madvise_atomic_attr (fd, vm,to_user_pointer (data), bo_size, +DRM_XE_ATOMIC_DEVICE ); +if (err) +igt_warn ("failure in atomic device attr err = %s, vm =%u data=%" PRIu64"\n" , +strerror (errno), vm,to_user_pointer (data)); + } +if (flags &PREFERRED_LOC_ATOMIC_GL ) { + err =xe_vm_madvise_migrate_pages (fd, vm,to_user_pointer (data), bo_size); +if (err) +igt_warn ("failure in page migration err = %s, vm =%u data=%" PRIu64"\n" , +strerror (errno), vm,to_user_pointer (data)); + err =xe_vm_madvise_atomic_attr (fd, vm,to_user_pointer (data), bo_size, +DRM_XE_ATOMIC_GLOBAL ); +if (err) +igt_warn ("failure in atomic global attr err = %s, vm =%u data=%" PRIu64"\n" , +strerror (errno), vm,to_user_pointer (data)); + } +if (flags &PREFERRED_LOC_ATOMIC_CPU ) { + err =xe_vm_madvise_migrate_pages (fd, vm,to_user_pointer (data), bo_size); +if (err) +igt_warn ("failure in page migration err = %s, vm =%u data=%" PRIu64"\n" , +strerror (errno), vm,to_user_pointer (data)); + + err =xe_vm_madvise_atomic_attr (fd, vm,to_user_pointer (data), bo_size, +DRM_XE_ATOMIC_CPU ); +if (err) +igt_warn ("failure in atomic cpu attr err = %s, vm =%u data=%" PRIu64"\n" , +strerror (errno), vm,to_user_pointer (data)); + } +if (flags &MADVISE_MULTI_VMA ) { +if (bo_size) + bo_size =ALIGN (bo_size,SZ_4K ); + err =xe_vm_madvise_atomic_attr (fd, vm,to_user_pointer (data) + bo_size/2 , + bo_size/2 ,DRM_XE_ATOMIC_DEVICE ); +if (err) +igt_warn ("failure in atomic device attr err = %s, vm =%u data=%" PRIu64"\n" , +strerror (errno), vm,to_user_pointer (data)); + + err =xe_vm_madvise_atomic_attr (fd, vm,to_user_pointer (data) + bo_size, + bo_size,DRM_XE_ATOMIC_DEVICE ); +if (err) +igt_warn ("failure in atomic multi_vma err = %s, vm =%u data=%" PRIu64"\n" , +strerror (errno), vm,to_user_pointer (data)); + + err =xe_vm_madvixe_pat_attr (fd, vm,to_user_pointer (data), bo_size,4 ); +if (err) +igt_warn ("failure in pat attr index 4 err = %s, vm =%u data=%" PRIu64"\n" , +strerror (errno), vm,to_user_pointer (data)); + + err =xe_vm_madvixe_pat_attr (fd, vm,to_user_pointer (data), bo_size,3 ); +if (err) +igt_warn ("failure in pat attr index 3 err = %s, vm =%u data=%" PRIu64"\n" , +strerror (errno), vm,to_user_pointer (data)); + + err =xe_vm_madvixe_pat_attr (fd, vm,to_user_pointer (data) + bo_size/2 , bo_size/2 , +8 ); +if (err) +igt_warn ("failure in pat attr index 8 err = %s, vm =%u data=%" PRIu64"\n" , +strerror (errno), vm,to_user_pointer (data)); + } +if (flags &MADVISE_SPLIT_VMA ) { +if (bo_size) + bo_size =ALIGN (bo_size,SZ_4K ); + + bo_flags =DRM_XE_GEM_CREATE_FLAG_NEEDS_VISIBLE_VRAM ; + bo =xe_bo_create (fd, vm, bo_size, +vram_if_possible (fd, eci->gt_id), bo_flags); +xe_vm_bind_async (fd, vm,0 , bo,0 ,to_user_pointer (data) + bo_size/2 , + bo_size/2 ,0 ,0 ); + +__xe_vm_bind_assert (fd, vm,0 ,0 ,0 ,to_user_pointer (data) + bo_size/2 , + bo_size/2 ,DRM_XE_VM_BIND_OP_MAP , +DRM_XE_VM_BIND_FLAG_CPU_ADDR_MIRROR , sync, +1 ,0 ,0 ); +xe_wait_ufence (fd, &data[0 ].vm_sync,USER_FENCE_VALUE ,0 ,FIVE_SEC ); + data[0 ].vm_sync =0 ; +gem_close (fd, bo); + bo =0 ; + + err =xe_vm_madvise_atomic_attr (fd, vm,to_user_pointer (data), + bo_size/2 ,DRM_XE_ATOMIC_DEVICE ); +if (err) +igt_warn ("failure in split atomic device attr err = %s, vm =%u data=%" PRIu64"\n" , +strerror (errno), vm,to_user_pointer (data)); + } +if (flags &MADVISE_ATOMIC_VMA ) { +if (bo_size) + bo_size =ALIGN (bo_size,SZ_4K ); + + bo_flags =DRM_XE_GEM_CREATE_FLAG_NEEDS_VISIBLE_VRAM ; + bo =xe_bo_create (fd, vm, bo_size,vram_if_possible (fd, eci->gt_id), bo_flags); +xe_vm_bind_async (fd, vm,0 , bo,0 ,to_user_pointer (data), bo_size,0 ,0 ); + +__xe_vm_bind_assert (fd, vm,0 ,0 ,0 ,to_user_pointer (data), bo_size, +DRM_XE_VM_BIND_OP_MAP , +DRM_XE_VM_BIND_FLAG_CPU_ADDR_MIRROR , sync, +1 ,0 ,0 ); +xe_wait_ufence (fd, &data[0 ].vm_sync,USER_FENCE_VALUE ,0 ,FIVE_SEC ); + data[0 ].vm_sync =0 ; +gem_close (fd, bo); + bo =0 ; + + err =xe_vm_madvise_atomic_attr (fd, vm,to_user_pointer (data), bo_size/2 , +DRM_XE_ATOMIC_GLOBAL ); +if (err) +igt_warn ("failure in atomic vma err = %s, vm =%u data=%" PRIu64"\n" , +strerror (errno), vm,to_user_pointer (data)); + } +if (flags &BO_UNMAP ) { bo_flags =DRM_XE_GEM_CREATE_FLAG_NEEDS_VISIBLE_VRAM ; bo = xe_bo_create(fd, vm, bo_size,
@@ -1316,18 +1550,74 @@ test_exec(int fd, struct drm_xe_engine_class_instance *eci,__write_dword (data[idx].batch, sdi_addr + j * orig_size,WRITE_VALUE (&data[idx], idx), &b); -write_dword (data[idx].batch, sdi_addr + j * orig_size, -WRITE_VALUE (&data[idx], idx), &b); +write_dword (&data[idx], sdi_addr + j * orig_size, +WRITE_VALUE (&data[idx], idx), &b, + flags &ATOMIC_BATCH ? true : false);igt_assert (b <=ARRAY_SIZE (data[idx].batch)); }else if (!(flags &EVERY_OTHER_CHECK )) { b =0 ; -write_dword (data[idx].batch, sdi_addr, -WRITE_VALUE (&data[idx], idx), &b); +write_dword (&data[idx], sdi_addr, +WRITE_VALUE (&data[idx], idx), &b, + flags &ATOMIC_BATCH ? true : false);igt_assert (b <=ARRAY_SIZE (data[idx].batch)); +if (flags &PREFETCH ) { +if (flags &PREFETCH_SPLIT_VMA ) { + bo_size =ALIGN (bo_size,SZ_4K ); + +xe_vm_prefetch_async (fd, vm,0 ,0 , addr, bo_size,NULL ,0 ,0 ); + +igt_info ("num_vmas before madvise = %d \n" , val); + + val =xe_vm_print_mem_attr_values_in_range (fd, vm, addr, bo_size); + + err =xe_vm_madvise_migrate_pages (fd, vm,to_user_pointer (data), bo_size/2 ); +if (err) +igt_warn ("failure in prefetch split vma err = %s, vm =%u data=%" PRIu64"\n" , +strerror (errno), vm,to_user_pointer (data)); +igt_info ("num_vmas after madvise= %d \n" , val); + val =xe_vm_print_mem_attr_values_in_range (fd, vm, addr, bo_size); + }else if (flags &PREFETCH_SAME_ATTR ) { + err =xe_vm_madvise_atomic_attr (fd, vm,to_user_pointer (data), bo_size, +DRM_XE_ATOMIC_GLOBAL ); +if (err) +igt_warn ("failure in prefetch same attr err = %s, vm =%u data=%" PRIu64"\n" , +strerror (errno), vm,to_user_pointer (data)); + val =xe_vm_print_mem_attr_values_in_range (fd, vm, addr, bo_size); +xe_vm_prefetch_async (fd, vm,0 ,0 , addr, bo_size,NULL ,0 ,1 ); + err =xe_vm_madvise_atomic_attr (fd, vm,to_user_pointer (data), bo_size/2 , +DRM_XE_ATOMIC_GLOBAL ); +if (err) +igt_warn ("failure in prefetch atomic attr err = %s, vm =%u data=%" PRIu64"\n" , +strerror (errno), vm,to_user_pointer (data)); + }else if (flags &PREFETCH_CHANGE_ATTR ) { + err =xe_vm_madvise_atomic_attr (fd, vm,to_user_pointer (data), bo_size, +DRM_XE_ATOMIC_GLOBAL ); +if (err) +igt_warn ("failure in prefetch atomic attr err = %s, vm =%u data=%" PRIu64"\n" , +strerror (errno), vm,to_user_pointer (data)); + val =xe_vm_print_mem_attr_values_in_range (fd, vm, addr, bo_size); + +xe_vm_prefetch_async (fd, vm,0 ,0 , addr, bo_size,NULL ,0 ,1 ); + + err =xe_vm_madvise_atomic_attr (fd, vm,to_user_pointer (data), bo_size, +DRM_XE_ATOMIC_DEVICE ); +if (err) +igt_warn ("failure in prefetch change attr err = %s, vm =%u data=%" PRIu64"\n" , +strerror (errno), vm,to_user_pointer (data)); + val =xe_vm_print_mem_attr_values_in_range (fd, vm, addr, bo_size); + } + }else { + b =0 ; +write_dword ((struct test_exec_data *)&data[idx], sdi_addr, +WRITE_VALUE (&data[idx], idx), &b, + flags &ATOMIC_BATCH ? true : false); +igt_assert (b <=ARRAY_SIZE (data[idx].batch)); + } }else if (flags &EVERY_OTHER_CHECK && !odd (i)) { b =0 ; -write_dword (data[idx].batch, sdi_addr, -WRITE_VALUE (&data[idx], idx), &b); +write_dword (&data[idx], sdi_addr, +WRITE_VALUE (&data[idx], idx), &b, + flags &ATOMIC_BATCH ? true : false);igt_assert (b <=ARRAY_SIZE (data[idx].batch)); aligned_alloc_type =__aligned_alloc (aligned_size, bo_size);
@@ -1346,10 +1636,11 @@ test_exec(int fd, struct drm_xe_engine_class_instance *eci,__aligned_partial_free (&aligned_alloc_type); b =0 ; -write_dword (data[next_idx].batch, +write_dword (&data[next_idx],to_user_pointer (next_data) + (char *)&data[next_idx].data - (char *)data, -WRITE_VALUE (&data[next_idx], next_idx), &b); +WRITE_VALUE (&data[next_idx], next_idx), &b, + flags &ATOMIC_BATCH ? true : false);igt_assert (b <=ARRAY_SIZE (data[next_idx].batch)); }
@@ -1404,9 +1695,18 @@ test_exec(int fd, struct drm_xe_engine_class_instance *eci, exec_queues[e], &timeout);igt_assert (err == -ETIME || err == -EIO ); }else { -xe_wait_ufence (fd, exec_ufence ? exec_ufence : - &data[idx].exec_sync,USER_FENCE_VALUE , - exec_queues[e],FIVE_SEC ); +if (flags &PREFERRED_LOC_ATOMIC_CPU ) { +int64_t timeout =QUARTER_SEC ; + err =__xe_wait_ufence (fd, exec_ufence ? exec_ufence : + &data[idx].exec_sync, +USER_FENCE_VALUE , + exec_queues[e], &timeout); +if (err) + goto cleanup; + }else +xe_wait_ufence (fd, exec_ufence ? exec_ufence : + &data[idx].exec_sync,USER_FENCE_VALUE , + exec_queues[e],FIVE_SEC );if (flags &LOCK && !i)munlock (data, bo_size);
@@ -1456,17 +1756,17 @@ test_exec(int fd, struct drm_xe_engine_class_instance *eci,if (flags &FORK_READ ) { igt_fork(child, 1)igt_assert_eq (data[idx].data, -READ_VALUE (&data[idx])); + flags &ATOMIC_BATCH ?VAL_ATOMIC_EXPECTED :READ_VALUE (&data[idx]));if (!(flags &FORK_READ_AFTER ))igt_assert_eq (data[idx].data, -READ_VALUE (&data[idx])); + flags &ATOMIC_BATCH ?VAL_ATOMIC_EXPECTED :READ_VALUE (&data[idx]));igt_waitchildren ();if (flags &FORK_READ_AFTER )igt_assert_eq (data[idx].data, -READ_VALUE (&data[idx])); + flags &ATOMIC_BATCH ?VAL_ATOMIC_EXPECTED :READ_VALUE (&data[idx])); }else {igt_assert_eq (data[idx].data, -READ_VALUE (&data[idx])); + flags &ATOMIC_BATCH ?VAL_ATOMIC_EXPECTED :READ_VALUE (&data[idx]));if (flags &PREFETCH_SYS_BENCHMARK ) {struct timespec tv = {};
@@ -1494,13 +1794,13 @@ test_exec(int fd, struct drm_xe_engine_class_instance *eci, ((void *)data) + j * orig_size;igt_assert_eq (__data[idx].data, -READ_VALUE (&data[idx])); + flags &ATOMIC_BATCH ?VAL_ATOMIC_EXPECTED :READ_VALUE (&data[idx])); } } }if (flags &EVERY_OTHER_CHECK )igt_assert_eq (data[prev_idx].data, -READ_VALUE (&data[prev_idx])); + flags &ATOMIC_BATCH ?VAL_ATOMIC_EXPECTED :READ_VALUE (&data[prev_idx])); } }
@@ -1521,6 +1821,88 @@ test_exec(int fd, struct drm_xe_engine_class_instance *eci,gem_close (fd, bo); } +if (flags &MADVISE_MULTI_VMA ) { + addr = addr + bo_size/2 ; + org_data = data; + data =from_user_pointer (addr); + + idx = !stride ? i : i * stride; + batch_offset = (char *)&(data[idx].batch) - (char *)data; + batch_addr = addr + batch_offset; + sdi_offset = (char *)&(data[idx].data) - (char *)data; + sdi_addr = addr + sdi_offset; + + e = i % n_exec_queues; + fault_inject = (FAULT & flags) && i == n_execs /2 ; + fault_injected = (FAULT & flags) && i > n_execs; + + b =0 ; +write_dword ((struct test_exec_data *)&data[idx], sdi_addr,WRITE_VALUE (data, i), + &b, flags &ATOMIC_BATCH ? true : false); +igt_assert (b <=ARRAY_SIZE (data[idx].batch)); + +#if USE_UFENCE + sync[0 ].addr = exec_ufence ?to_user_pointer (exec_ufence) : + addr + (char *)&data[idx].exec_sync - (char *)data; +#else + exec_ufence =NULL ; + sync[0 ].addr = addr + (char *)&data[idx].exec_sync - (char *)data; +#endif +if (!exec_ufence) + data[idx].exec_sync =0 ; + + exec.exec_queue_id = exec_queues[e]; +if (fault_inject) + exec.address = batch_addr *2 ; +else + exec.address = batch_addr; + +if (fault_injected) { + err =__xe_exec (fd, &exec); +igt_assert (err == -ENOENT ); + }else { +xe_exec (fd, &exec); + } +if (fault_inject || fault_injected) { +int64_t timeout =QUARTER_SEC ; + + err =__xe_wait_ufence (fd, exec_ufence ? exec_ufence : + &data[idx].exec_sync, +USER_FENCE_VALUE , + exec_queues[e], &timeout); +igt_assert (err == -ETIME || err == -EIO ); + }else { +if (flags &PREFERRED_LOC_ATOMIC_CPU ) { +int64_t timeout =QUARTER_SEC ; + err =__xe_wait_ufence (fd, exec_ufence ? exec_ufence : + &data[idx].exec_sync, +USER_FENCE_VALUE , + exec_queues[e], &timeout); +if (err) + goto cleanup; + }else +xe_wait_ufence (fd, exec_ufence ? exec_ufence : + &data[idx].exec_sync,USER_FENCE_VALUE , + exec_queues[e],FIVE_SEC ); + } +igt_assert_eq (data[i].data, + flags &ATOMIC_BATCH ?VAL_ATOMIC_EXPECTED :READ_VALUE (&data[i])); + +if (exec_ufence) + exec_ufence[0 ] =0 ; + +if (bo) { +__xe_vm_bind_assert (fd, vm,0 , +0 ,0 , addr, bo_size, +DRM_XE_VM_BIND_OP_MAP , +DRM_XE_VM_BIND_FLAG_CPU_ADDR_MIRROR , +NULL ,0 ,0 ,0 ); +munmap (data, bo_size); +gem_close (fd, bo); + } + data = org_data; + } +if (flags &NEW ) {if (flags &MMAP ) {if (flags &FREE )
@@ -1610,6 +1992,7 @@ test_exec(int fd, struct drm_xe_engine_class_instance *eci, pf_count, pf_count_after); } +cleanup:if (bo) { sync[0 ].addr =to_user_pointer (bind_ufence); __xe_vm_bind_assert(fd, vm, 0,
@@ -1864,7 +2247,7 @@ processes(int fd, int n_exec_queues, int n_execs, size_t bo_size,struct section {const char *name; -unsigned int flags; +unsigned long long flags; }; igt_main
@@ -1964,6 +2347,19 @@ igt_main { "malloc-mix-bo", MIX_BO_ALLOC }, { NULL }, }; +const struct section msections[] = { + {"preffered-loc-sram-migrate-pages" ,PREFERRED_LOC_SMEM |MIGRATE_ALL_PAGES |ATOMIC_BATCH }, + {"preffered-loc-atomic-vram" ,PREFERRED_LOC_ATOMIC_DEVICE |ATOMIC_BATCH }, + {"preffered-loc-atomic-gl" ,PREFERRED_LOC_ATOMIC_GL |ATOMIC_BATCH }, + {"preffered-loc-atomic-cpu" ,PREFERRED_LOC_ATOMIC_CPU |ATOMIC_BATCH }, + {"multi-vma" ,MADVISE_MULTI_VMA |ATOMIC_BATCH }, + {"split-vma" ,MADVISE_SPLIT_VMA |ATOMIC_BATCH }, + {"atomic-vma" ,MADVISE_ATOMIC_VMA |ATOMIC_BATCH }, + {"split-vma-with-mapping" ,PREFETCH |PREFETCH_SPLIT_VMA |ATOMIC_BATCH }, + {"range-invalidate-change-attr" ,PREFETCH |PREFETCH_CHANGE_ATTR |ATOMIC_BATCH }, + {"no-range-invalidate-same-attr" ,PREFETCH |PREFETCH_SAME_ATTR |ATOMIC_BATCH }, + {NULL }, + };int fd;int num_sections;
@@ -1983,15 +2379,22 @@ igt_mainfor (const struct section *s = sections; s[num_sections].name; num_sections++) ; -for (int i =0 ; i < num_sections *2 ; i++) { -struct section *s = §ions[i % num_sections]; +for (int i =0 ; i < num_sections *3 ; i++) { +struct section p = sections[i % num_sections]; +struct section *s = &p; -if (i/num_sections ==0 ) { +if (i/num_sections ==1 ) {static char modified_name[256 ];snprintf (modified_name,sizeof (modified_name),"%s-preferred-loc-smem" , s->name); s->name = modified_name; s->flags |=PREFERRED_LOC_SMEM ; } +if (i/num_sections ==2 ) { +static char modified_name[256 ]; +snprintf (modified_name,sizeof (modified_name),"atomic-batch-%s" , s->name); + s->name = modified_name; + s->flags |=ATOMIC_BATCH ; + }igt_subtest_f ("once-%s" , s->name) xe_for_each_engine(fd, hwe)
@@ -2175,6 +2578,14 @@ igt_mainprocesses_evict (fd,SZ_8M ,SZ_1M , s->flags); } +for (const struct section *s = msections; s->name; s++) { + igt_subtest_f("madvise-%s", s->name) { +xe_for_each_engine (fd, hwe) +test_exec (fd, hwe,1 ,1 ,SZ_64K ,0 ,0 ,NULL , +NULL , s->flags); + } + } + igt_fixture {xe_device_put (fd);drm_close_driver (fd);
Signature · 1 line
2.43.0