1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
|
From eace5958a8fdac8d2650d1a60c9c253a8bc85fff Mon Sep 17 00:00:00 2001
From: Felix Kuehling <Felix.Kuehling@amd.com>
Date: Tue, 20 Nov 2018 21:44:27 -0500
Subject: [PATCH 2727/2940] drm/amdkfd: Add support for doorbell BOs
This allows user mode to map doorbell pages into GPUVM address space.
That way GPUs can submit to user mode queues (self-dispatch).
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
.../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 59 +++++++++++++++++--
drivers/gpu/drm/amd/amdkfd/kfd_chardev.c | 6 ++
.../gpu/drm/amd/include/kgd_kfd_interface.h | 4 +-
3 files changed, 62 insertions(+), 7 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
index 0b105783e3b3..b1cc59490f96 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
@@ -878,6 +878,24 @@ static int map_bo_to_gpuvm(struct amdgpu_device *adev,
return ret;
}
+static struct sg_table *create_doorbell_sg(uint64_t addr, uint32_t size)
+{
+ struct sg_table *sg = kmalloc(sizeof(*sg), GFP_KERNEL);
+
+ if (!sg)
+ return NULL;
+ if (sg_alloc_table(sg, 1, GFP_KERNEL)) {
+ kfree(sg);
+ return NULL;
+ }
+ sg->sgl->dma_address = addr;
+ sg->sgl->length = size;
+#ifdef CONFIG_NEED_SG_DMA_LENGTH
+ sg->sgl->dma_length = size;
+#endif
+ return sg;
+}
+
static int process_validate_vms(struct amdkfd_process_info *process_info)
{
struct amdgpu_vm *peer_vm;
@@ -1161,6 +1179,8 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
{
struct amdgpu_device *adev = get_amdgpu_device(kgd);
struct amdgpu_vm *avm = (struct amdgpu_vm *)vm;
+ enum ttm_bo_type bo_type = ttm_bo_type_device;
+ struct sg_table *sg = NULL;
uint64_t user_addr = 0;
struct amdgpu_bo *bo;
struct amdgpu_bo_param bp;
@@ -1189,13 +1209,25 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
if (!offset || !*offset)
return -EINVAL;
user_addr = *offset;
+ } else if (flags & ALLOC_MEM_FLAGS_DOORBELL) {
+ domain = AMDGPU_GEM_DOMAIN_GTT;
+ alloc_domain = AMDGPU_GEM_DOMAIN_CPU;
+ bo_type = ttm_bo_type_sg;
+ alloc_flags = 0;
+ if (size > UINT_MAX)
+ return -EINVAL;
+ sg = create_doorbell_sg(*offset, size);
+ if (!sg)
+ return -ENOMEM;
} else {
return -EINVAL;
}
*mem = kzalloc(sizeof(struct kgd_mem), GFP_KERNEL);
- if (!*mem)
- return -ENOMEM;
+ if (!*mem) {
+ ret = -ENOMEM;
+ goto err;
+ }
INIT_LIST_HEAD(&(*mem)->bo_va_list);
mutex_init(&(*mem)->lock);
(*mem)->aql_queue = !!(flags & ALLOC_MEM_FLAGS_AQL_QUEUE_MEM);
@@ -1228,7 +1260,7 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
amdgpu_sync_create(&(*mem)->sync);
- ret = amdgpu_amdkfd_reserve_mem_limit(adev, size, alloc_domain, false);
+ ret = amdgpu_amdkfd_reserve_mem_limit(adev, size, alloc_domain, !!sg);
if (ret) {
pr_debug("Insufficient system memory\n");
goto err_reserve_limit;
@@ -1242,7 +1274,7 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
bp.byte_align = byte_align;
bp.domain = alloc_domain;
bp.flags = alloc_flags;
- bp.type = ttm_bo_type_device;
+ bp.type = bo_type;
bp.resv = NULL;
ret = amdgpu_bo_create(adev, &bp, &bo);
if (ret) {
@@ -1250,6 +1282,10 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
domain_string(alloc_domain), ret);
goto err_bo_create;
}
+ if (bo_type == ttm_bo_type_sg) {
+ bo->tbo.sg = sg;
+ bo->tbo.ttm->sg = sg;
+ }
bo->kfd_bo = *mem;
(*mem)->bo = bo;
if (user_addr)
@@ -1281,10 +1317,15 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
/* Don't unreserve system mem limit twice */
goto err_reserve_limit;
err_bo_create:
- unreserve_mem_limit(adev, size, alloc_domain, false);
+ unreserve_mem_limit(adev, size, alloc_domain, !!sg);
err_reserve_limit:
mutex_destroy(&(*mem)->lock);
kfree(*mem);
+err:
+ if (sg) {
+ sg_free_table(sg);
+ kfree(sg);
+ }
return ret;
}
@@ -1354,6 +1395,14 @@ int amdgpu_amdkfd_gpuvm_free_memory_of_gpu(
/* Free the sync object */
amdgpu_sync_free(&mem->sync);
+ /* If the SG is not NULL, it's one we created for a doorbell
+ * BO. We need to free it.
+ */
+ if (mem->bo->tbo.sg) {
+ sg_free_table(mem->bo->tbo.sg);
+ kfree(mem->bo->tbo.sg);
+ }
+
/* Free the BO*/
amdgpu_bo_unref(&mem->bo);
mutex_destroy(&mem->lock);
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
index ae3ae0fb2602..3623538baf6f 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
@@ -1274,6 +1274,12 @@ static int kfd_ioctl_alloc_memory_of_gpu(struct file *filep,
return -EINVAL;
}
+ if (flags & KFD_IOC_ALLOC_MEM_FLAGS_DOORBELL) {
+ if (args->size != kfd_doorbell_process_slice(dev))
+ return -EINVAL;
+ offset = kfd_get_process_doorbells(dev, p);
+ }
+
mutex_lock(&p->mutex);
pdd = kfd_bind_process_to_device(dev, p);
diff --git a/drivers/gpu/drm/amd/include/kgd_kfd_interface.h b/drivers/gpu/drm/amd/include/kgd_kfd_interface.h
index 8fbd7a52a49d..0285e1a71fa4 100644
--- a/drivers/gpu/drm/amd/include/kgd_kfd_interface.h
+++ b/drivers/gpu/drm/amd/include/kgd_kfd_interface.h
@@ -160,8 +160,8 @@ struct tile_config {
*/
#define ALLOC_MEM_FLAGS_VRAM (1 << 0)
#define ALLOC_MEM_FLAGS_GTT (1 << 1)
-#define ALLOC_MEM_FLAGS_USERPTR (1 << 2) /* TODO */
-#define ALLOC_MEM_FLAGS_DOORBELL (1 << 3) /* TODO */
+#define ALLOC_MEM_FLAGS_USERPTR (1 << 2)
+#define ALLOC_MEM_FLAGS_DOORBELL (1 << 3)
/*
* Allocation flags attributes/access options.
--
2.17.1
|
