AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Changes for RDMA with VMM (#801)

Browse Source 
* rocr: Add support for VMM and RDMA

Add extra CPU mapping so that kernel-mode drivers can look up the memory
mapping by virtual address.

* Update projects/rocr-runtime/runtime/hsa-runtime/core/runtime/runtime.cpp

Co-authored-by: Yiannis Papadopoulos <102817138+ypapadop-amd@users.noreply.github.com>

* Update projects/rocr-runtime/runtime/hsa-runtime/core/inc/runtime.h

Co-authored-by: Yiannis Papadopoulos <102817138+ypapadop-amd@users.noreply.github.com>

* rocr: Honor uncache flag in memory_lock_to_pool()

Also, combined several flag options used in apis into a
single integer.

Signed-off-by: Chris Freehill <cfreehil@amd.com>

* rocr: Fix hsa_amd_pointer_info on CPU agents

Fix hsa_amd_pointer_info query returning allowd on VMM pointers for CPU
agents when CPU mapping was mapped with PROT_NONE.

---------

Signed-off-by: Chris Freehill <cfreehil@amd.com>
Co-authored-by: Yiannis Papadopoulos <102817138+ypapadop-amd@users.noreply.github.com>
Co-authored-by: Chris Freehill <cfreehil@amd.com>
Co-authored-by: cfreeamd <166262151+cfreeamd@users.noreply.github.com>
This commit is contained in:
David Yat Sin
2025-10-21 12:19:02 -04:00
committed by GitHub
parent 65d4ff9d04
commit e2f3bd2429
8 changed files with 76 additions and 47 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/rocr-runtime/libhsakmt/src/fmm.c
+11 -21
View File
@@ -1061,9 +1061,7 @@ static HsaMemFlags fmm_translate_ioc_to_hsa_flags(uint32_t ioc_flags)
}
static HSAKMT_STATUS fmm_register_mem_svm_api(void *address,
uint64_t size,
bool coarse_grain,
bool ext_coherent)
uint64_t size, HsaMemFlags flags)
{
struct kfd_ioctl_svm_args *args;
size_t s_attr;
@@ -1080,10 +1078,11 @@ static HSAKMT_STATUS fmm_register_mem_svm_api(void *address,
args->size = aligned_size;
args->op = KFD_IOCTL_SVM_OP_SET_ATTR;
args->nattr = 2;
args->attrs[0].type = coarse_grain ?
args->attrs[0].type = flags.ui32.CoarseGrain ?
HSA_SVM_ATTR_CLR_FLAGS : HSA_SVM_ATTR_SET_FLAGS;
args->attrs[0].value = HSA_SVM_FLAG_COHERENT;
args->attrs[1].type = ext_coherent ? HSA_SVM_ATTR_SET_FLAGS : HSA_SVM_ATTR_CLR_FLAGS ;
args->attrs[1].type = flags.ui32.ExtendedCoherent ?
HSA_SVM_ATTR_SET_FLAGS : HSA_SVM_ATTR_CLR_FLAGS;
args->attrs[1].value = HSA_SVM_FLAG_EXT_COHERENT;
pr_debug("Registering to SVM %p size: %ld\n", (void*)aligned_addr,
aligned_size);
@@ -3748,8 +3747,7 @@ bool hsakmt_fmm_get_handle(void *address, uint64_t *handle, uint64_t *size_offse
static HSAKMT_STATUS fmm_register_user_memory(void *addr,
HSAuint64 size,
vm_object_t **obj_ret,
bool coarse_grain,
bool ext_coherent)
HsaMemFlags flags)
{
manageable_aperture_t *aperture = svm.dgpu_aperture;
HSAuint32 page_offset = (HSAuint64)addr & (PAGE_SIZE-1);
@@ -3774,8 +3772,8 @@ static HSAKMT_STATUS fmm_register_user_memory(void *addr,
&aligned_addr, KFD_IOC_ALLOC_MEM_FLAGS_USERPTR |
KFD_IOC_ALLOC_MEM_FLAGS_WRITABLE |
KFD_IOC_ALLOC_MEM_FLAGS_EXECUTABLE |
(coarse_grain ? 0 : KFD_IOC_ALLOC_MEM_FLAGS_COHERENT) |
(ext_coherent ? KFD_IOC_ALLOC_MEM_FLAGS_EXT_COHERENT : 0),
(flags.ui32.CoarseGrain ? 0 : KFD_IOC_ALLOC_MEM_FLAGS_COHERENT) |
(flags.ui32.ExtendedCoherent ? KFD_IOC_ALLOC_MEM_FLAGS_EXT_COHERENT : 0),
0,
&obj);
if (!svm_addr)
@@ -3813,8 +3811,7 @@ static HSAKMT_STATUS fmm_register_user_memory(void *addr,
HSAKMT_STATUS hsakmt_fmm_register_memory(void *address, uint64_t size_in_bytes,
uint32_t *gpu_id_array,
uint32_t gpu_id_array_size,
bool coarse_grain,
bool ext_coherent)
HsaMemFlags flags)
{
manageable_aperture_t *aperture = NULL;
vm_object_t *object = NULL;
@@ -3823,7 +3820,7 @@ HSAKMT_STATUS hsakmt_fmm_register_memory(void *address, uint64_t size_in_bytes,
if (gpu_id_array_size > 0 && !gpu_id_array)
return HSAKMT_STATUS_INVALID_PARAMETER;
if (coarse_grain && ext_coherent)
if (flags.ui32.CoarseGrain && flags.ui32.ExtendedCoherent)
return HSAKMT_STATUS_INVALID_PARAMETER;
object = vm_find_object(address, size_in_bytes, &aperture);
@@ -3834,19 +3831,12 @@ HSAKMT_STATUS hsakmt_fmm_register_memory(void *address, uint64_t size_in_bytes,
/* Register a new user ptr */
if (hsakmt_is_svm_api_supported) {
ret = fmm_register_mem_svm_api(address,
size_in_bytes,
coarse_grain,
ext_coherent);
ret = fmm_register_mem_svm_api(address, size_in_bytes, flags);
if (ret == HSAKMT_STATUS_SUCCESS)
return ret;
pr_debug("SVM failed, falling back to old registration\n");
}
ret = fmm_register_user_memory(address,
size_in_bytes,
&object,
coarse_grain,
ext_coherent);
ret = fmm_register_user_memory(address, size_in_bytes, &object, flags);
if (ret != HSAKMT_STATUS_SUCCESS)
return ret;
projects/rocr-runtime/libhsakmt/src/fmm.h
+1 -2
View File
@@ -75,8 +75,7 @@ HSAKMT_STATUS hsakmt_fmm_get_aperture_base_and_limit(aperture_type_e aperture_ty
HSAKMT_STATUS hsakmt_fmm_register_memory(void *address, uint64_t size_in_bytes,
uint32_t *gpu_id_array,
uint32_t gpu_id_array_size,
bool coarse_grain,
bool ext_coherent);
HsaMemFlags flags);
HSAKMT_STATUS hsakmt_fmm_register_graphics_handle(HSAuint64 GraphicsResourceHandle,
HsaGraphicsResourceInfo *GraphicsResourceInfo,
uint32_t *gpu_id_array,
projects/rocr-runtime/libhsakmt/src/memory.c
+10 -3
View File
@@ -268,8 +268,11 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtRegisterMemory(void *MemoryAddress,
/* TODO: support mixed APU and dGPU configurations */
return HSAKMT_STATUS_SUCCESS;
HsaMemFlags flags;
flags.ui32.CoarseGrain = 1;
flags.ui32.ExtendedCoherent = 0;
return hsakmt_fmm_register_memory(MemoryAddress, MemorySizeInBytes,
NULL, 0, true, false);
NULL, 0, flags);
}
HSAKMT_STATUS HSAKMTAPI hsaKmtRegisterMemoryToNodes(void *MemoryAddress,
@@ -292,10 +295,14 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtRegisterMemoryToNodes(void *MemoryAddress,
NumberOfNodes, NodeArray);
if (ret == HSAKMT_STATUS_SUCCESS) {
HsaMemFlags flags;
flags.ui32.CoarseGrain = 1;
flags.ui32.ExtendedCoherent = 0;
ret = hsakmt_fmm_register_memory(MemoryAddress, MemorySizeInBytes,
gpu_id_array,
NumberOfNodes*sizeof(uint32_t),
true, false);
flags);
if (ret != HSAKMT_STATUS_SUCCESS)
free(gpu_id_array);
}
@@ -325,7 +332,7 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtRegisterMemoryWithFlags(void *MemoryAddress,
return HSAKMT_STATUS_NOT_SUPPORTED;
ret = hsakmt_fmm_register_memory(MemoryAddress, MemorySizeInBytes,
NULL, 0, MemFlags.ui32.CoarseGrain, MemFlags.ui32.ExtendedCoherent);
NULL, 0, MemFlags);
return ret;
}