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
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
|
/*
* Copyright 2014 Advanced Micro Devices, Inc.
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
* OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
* ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
* OTHER DEALINGS IN THE SOFTWARE.
*/
#include "kfd_priv.h"
#include <linux/mm.h>
#include <linux/mman.h>
#include <linux/slab.h>
#include <linux/io.h>
#include <linux/idr.h>
/*
* This extension supports a kernel level doorbells management for the
* kernel queues using the first doorbell page reserved for the kernel.
*/
static DEFINE_IDA(doorbell_ida);
static unsigned int max_doorbell_slices;
/*
* Each device exposes a doorbell aperture, a PCI MMIO aperture that
* receives 32-bit writes that are passed to queues as wptr values.
* The doorbells are intended to be written by applications as part
* of queueing work on user-mode queues.
* We assign doorbells to applications in PAGE_SIZE-sized and aligned chunks.
* We map the doorbell address space into user-mode when a process creates
* its first queue on each device.
* Although the mapping is done by KFD, it is equivalent to an mmap of
* the /dev/kfd with the particular device encoded in the mmap offset.
* There will be other uses for mmap of /dev/kfd, so only a range of
* offsets (KFD_MMAP_DOORBELL_START-END) is used for doorbells.
*/
/* # of doorbell bytes allocated for each process. */
static size_t kfd_doorbell_process_slice(struct kfd_dev *kfd)
{
return roundup(kfd->device_info->doorbell_size *
KFD_MAX_NUM_OF_QUEUES_PER_PROCESS,
PAGE_SIZE);
}
/* Doorbell calculations for device init. */
int kfd_doorbell_init(struct kfd_dev *kfd)
{
size_t doorbell_start_offset;
size_t doorbell_aperture_size;
size_t doorbell_process_limit;
/*
* We start with calculations in bytes because the input data might
* only be byte-aligned.
* Only after we have done the rounding can we assume any alignment.
*/
doorbell_start_offset =
roundup(kfd->shared_resources.doorbell_start_offset,
kfd_doorbell_process_slice(kfd));
doorbell_aperture_size =
rounddown(kfd->shared_resources.doorbell_aperture_size,
kfd_doorbell_process_slice(kfd));
if (doorbell_aperture_size > doorbell_start_offset)
doorbell_process_limit =
(doorbell_aperture_size - doorbell_start_offset) /
kfd_doorbell_process_slice(kfd);
else
return -ENOSPC;
if (!max_doorbell_slices ||
doorbell_process_limit < max_doorbell_slices)
max_doorbell_slices = doorbell_process_limit;
kfd->doorbell_base = kfd->shared_resources.doorbell_physical_address +
doorbell_start_offset;
kfd->doorbell_id_offset = doorbell_start_offset / sizeof(u32);
kfd->doorbell_kernel_ptr = ioremap(kfd->doorbell_base,
kfd_doorbell_process_slice(kfd));
if (!kfd->doorbell_kernel_ptr)
return -ENOMEM;
pr_debug("Doorbell initialization:\n");
pr_debug("doorbell base == 0x%08lX\n",
(uintptr_t)kfd->doorbell_base);
pr_debug("doorbell_id_offset == 0x%08lX\n",
kfd->doorbell_id_offset);
pr_debug("doorbell_process_limit == 0x%08lX\n",
doorbell_process_limit);
pr_debug("doorbell_kernel_offset == 0x%08lX\n",
(uintptr_t)kfd->doorbell_base);
pr_debug("doorbell aperture size == 0x%08lX\n",
kfd->shared_resources.doorbell_aperture_size);
pr_debug("doorbell kernel address == %p\n", kfd->doorbell_kernel_ptr);
return 0;
}
void kfd_doorbell_fini(struct kfd_dev *kfd)
{
if (kfd->doorbell_kernel_ptr)
iounmap(kfd->doorbell_kernel_ptr);
}
int kfd_doorbell_mmap(struct kfd_process *process, struct vm_area_struct *vma)
{
phys_addr_t address;
struct kfd_dev *dev;
/* Find kfd device according to gpu id */
dev = kfd_device_by_id(vma->vm_pgoff);
if (!dev)
return -EINVAL;
/*
* For simplicitly we only allow mapping of the entire doorbell
* allocation of a single device & process.
*/
if (vma->vm_end - vma->vm_start != kfd_doorbell_process_slice(dev))
return -EINVAL;
/* Calculate physical address of doorbell */
address = kfd_get_process_doorbells(dev, process);
vma->vm_flags |= VM_IO | VM_DONTCOPY | VM_DONTEXPAND | VM_NORESERVE |
VM_DONTDUMP | VM_PFNMAP;
vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
pr_debug("Mapping doorbell page\n"
" target user address == 0x%08llX\n"
" physical address == 0x%08llX\n"
" vm_flags == 0x%04lX\n"
" size == 0x%04lX\n",
(unsigned long long) vma->vm_start, address, vma->vm_flags,
kfd_doorbell_process_slice(dev));
return io_remap_pfn_range(vma,
vma->vm_start,
address >> PAGE_SHIFT,
kfd_doorbell_process_slice(dev),
vma->vm_page_prot);
}
/* get kernel iomem pointer for a doorbell */
void __iomem *kfd_get_kernel_doorbell(struct kfd_dev *kfd,
unsigned int *doorbell_off)
{
u32 inx;
mutex_lock(&kfd->doorbell_mutex);
inx = find_first_zero_bit(kfd->doorbell_available_index,
KFD_MAX_NUM_OF_QUEUES_PER_PROCESS);
__set_bit(inx, kfd->doorbell_available_index);
mutex_unlock(&kfd->doorbell_mutex);
if (inx >= KFD_MAX_NUM_OF_QUEUES_PER_PROCESS)
return NULL;
inx *= kfd->device_info->doorbell_size / sizeof(u32);
/*
* Calculating the kernel doorbell offset using the first
* doorbell page.
*/
*doorbell_off = kfd->doorbell_id_offset + inx;
pr_debug("Get kernel queue doorbell\n"
" doorbell offset == 0x%08X\n"
" kernel address == %p\n",
*doorbell_off, (kfd->doorbell_kernel_ptr + inx));
return kfd->doorbell_kernel_ptr + inx;
}
void kfd_release_kernel_doorbell(struct kfd_dev *kfd, u32 __iomem *db_addr)
{
unsigned int inx;
inx = (unsigned int)(db_addr - kfd->doorbell_kernel_ptr);
mutex_lock(&kfd->doorbell_mutex);
__clear_bit(inx, kfd->doorbell_available_index);
mutex_unlock(&kfd->doorbell_mutex);
}
void write_kernel_doorbell(void __iomem *db, u32 value)
{
if (db) {
writel(value, db);
pr_debug("Writing %d to doorbell address %p\n", value, db);
}
}
/*
* queue_ids are in the range [0,MAX_PROCESS_QUEUES) and are mapped 1:1
* to doorbells with the process's doorbell page
*/
unsigned int kfd_queue_id_to_doorbell(struct kfd_dev *kfd,
struct kfd_process *process,
unsigned int queue_id)
{
/*
* doorbell_id_offset accounts for doorbells taken by KGD.
* index * kfd_doorbell_process_slice/sizeof(u32) adjusts to
* the process's doorbells. The offset returned is in dword
* units regardless of the ASIC-dependent doorbell size.
*/
return kfd->doorbell_id_offset +
process->doorbell_index
* kfd_doorbell_process_slice(kfd) / sizeof(u32) +
queue_id * kfd->device_info->doorbell_size / sizeof(u32);
}
uint64_t kfd_get_number_elems(struct kfd_dev *kfd)
{
uint64_t num_of_elems = (kfd->shared_resources.doorbell_aperture_size -
kfd->shared_resources.doorbell_start_offset) /
kfd_doorbell_process_slice(kfd) + 1;
return num_of_elems;
}
phys_addr_t kfd_get_process_doorbells(struct kfd_dev *dev,
struct kfd_process *process)
{
return dev->doorbell_base +
process->doorbell_index * kfd_doorbell_process_slice(dev);
}
int kfd_alloc_process_doorbells(struct kfd_process *process)
{
int r = ida_simple_get(&doorbell_ida, 1, max_doorbell_slices,
GFP_KERNEL);
if (r > 0)
process->doorbell_index = r;
return r;
}
void kfd_free_process_doorbells(struct kfd_process *process)
{
if (process->doorbell_index)
ida_simple_remove(&doorbell_ida, process->doorbell_index);
}
|