Rev 5078 | Go to most recent revision | Details | Compare with Previous | Last modification | View Log | RSS feed
Rev | Author | Line No. | Line |
---|---|---|---|
5078 | serge | 1 | /* |
2 | * Copyright 2013 Advanced Micro Devices, Inc. |
||
3 | * |
||
4 | * Permission is hereby granted, free of charge, to any person obtaining a |
||
5 | * copy of this software and associated documentation files (the "Software"), |
||
6 | * to deal in the Software without restriction, including without limitation |
||
7 | * the rights to use, copy, modify, merge, publish, distribute, sublicense, |
||
8 | * and/or sell copies of the Software, and to permit persons to whom the |
||
9 | * Software is furnished to do so, subject to the following conditions: |
||
10 | * |
||
11 | * The above copyright notice and this permission notice shall be included in |
||
12 | * all copies or substantial portions of the Software. |
||
13 | * |
||
14 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
||
15 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
||
16 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL |
||
17 | * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR |
||
18 | * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, |
||
19 | * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR |
||
20 | * OTHER DEALINGS IN THE SOFTWARE. |
||
21 | * |
||
22 | * Authors: Alex Deucher |
||
23 | */ |
||
24 | #include |
||
25 | #include "radeon.h" |
||
26 | #include "radeon_asic.h" |
||
27 | #include "radeon_trace.h" |
||
28 | #include "sid.h" |
||
29 | |||
30 | u32 si_gpu_check_soft_reset(struct radeon_device *rdev); |
||
31 | |||
32 | /** |
||
33 | * si_dma_is_lockup - Check if the DMA engine is locked up |
||
34 | * |
||
35 | * @rdev: radeon_device pointer |
||
36 | * @ring: radeon_ring structure holding ring information |
||
37 | * |
||
38 | * Check if the async DMA engine is locked up. |
||
39 | * Returns true if the engine appears to be locked up, false if not. |
||
40 | */ |
||
41 | bool si_dma_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring) |
||
42 | { |
||
43 | u32 reset_mask = si_gpu_check_soft_reset(rdev); |
||
44 | u32 mask; |
||
45 | |||
46 | if (ring->idx == R600_RING_TYPE_DMA_INDEX) |
||
47 | mask = RADEON_RESET_DMA; |
||
48 | else |
||
49 | mask = RADEON_RESET_DMA1; |
||
50 | |||
51 | if (!(reset_mask & mask)) { |
||
52 | radeon_ring_lockup_update(rdev, ring); |
||
53 | return false; |
||
54 | } |
||
55 | return radeon_ring_test_lockup(rdev, ring); |
||
56 | } |
||
57 | |||
58 | /** |
||
59 | * si_dma_vm_copy_pages - update PTEs by copying them from the GART |
||
60 | * |
||
61 | * @rdev: radeon_device pointer |
||
62 | * @ib: indirect buffer to fill with commands |
||
63 | * @pe: addr of the page entry |
||
64 | * @src: src addr where to copy from |
||
65 | * @count: number of page entries to update |
||
66 | * |
||
67 | * Update PTEs by copying them from the GART using the DMA (SI). |
||
68 | */ |
||
69 | void si_dma_vm_copy_pages(struct radeon_device *rdev, |
||
70 | struct radeon_ib *ib, |
||
71 | uint64_t pe, uint64_t src, |
||
72 | unsigned count) |
||
73 | { |
||
74 | while (count) { |
||
75 | unsigned bytes = count * 8; |
||
76 | if (bytes > 0xFFFF8) |
||
77 | bytes = 0xFFFF8; |
||
78 | |||
79 | ib->ptr[ib->length_dw++] = DMA_PACKET(DMA_PACKET_COPY, |
||
80 | 1, 0, 0, bytes); |
||
81 | ib->ptr[ib->length_dw++] = lower_32_bits(pe); |
||
82 | ib->ptr[ib->length_dw++] = lower_32_bits(src); |
||
83 | ib->ptr[ib->length_dw++] = upper_32_bits(pe) & 0xff; |
||
84 | ib->ptr[ib->length_dw++] = upper_32_bits(src) & 0xff; |
||
85 | |||
86 | pe += bytes; |
||
87 | src += bytes; |
||
88 | count -= bytes / 8; |
||
89 | } |
||
90 | } |
||
91 | |||
92 | /** |
||
93 | * si_dma_vm_write_pages - update PTEs by writing them manually |
||
94 | * |
||
95 | * @rdev: radeon_device pointer |
||
96 | * @ib: indirect buffer to fill with commands |
||
97 | * @pe: addr of the page entry |
||
98 | * @addr: dst addr to write into pe |
||
99 | * @count: number of page entries to update |
||
100 | * @incr: increase next addr by incr bytes |
||
101 | * @flags: access flags |
||
102 | * |
||
103 | * Update PTEs by writing them manually using the DMA (SI). |
||
104 | */ |
||
105 | void si_dma_vm_write_pages(struct radeon_device *rdev, |
||
106 | struct radeon_ib *ib, |
||
107 | uint64_t pe, |
||
108 | uint64_t addr, unsigned count, |
||
109 | uint32_t incr, uint32_t flags) |
||
110 | { |
||
111 | uint64_t value; |
||
112 | unsigned ndw; |
||
113 | |||
114 | while (count) { |
||
115 | ndw = count * 2; |
||
116 | if (ndw > 0xFFFFE) |
||
117 | ndw = 0xFFFFE; |
||
118 | |||
119 | /* for non-physically contiguous pages (system) */ |
||
120 | ib->ptr[ib->length_dw++] = DMA_PACKET(DMA_PACKET_WRITE, 0, 0, 0, ndw); |
||
121 | ib->ptr[ib->length_dw++] = pe; |
||
122 | ib->ptr[ib->length_dw++] = upper_32_bits(pe) & 0xff; |
||
123 | for (; ndw > 0; ndw -= 2, --count, pe += 8) { |
||
124 | if (flags & R600_PTE_SYSTEM) { |
||
125 | value = radeon_vm_map_gart(rdev, addr); |
||
126 | value &= 0xFFFFFFFFFFFFF000ULL; |
||
127 | } else if (flags & R600_PTE_VALID) { |
||
128 | value = addr; |
||
129 | } else { |
||
130 | value = 0; |
||
131 | } |
||
132 | addr += incr; |
||
133 | value |= flags; |
||
134 | ib->ptr[ib->length_dw++] = value; |
||
135 | ib->ptr[ib->length_dw++] = upper_32_bits(value); |
||
136 | } |
||
137 | } |
||
138 | } |
||
139 | |||
140 | /** |
||
141 | * si_dma_vm_set_pages - update the page tables using the DMA |
||
142 | * |
||
143 | * @rdev: radeon_device pointer |
||
144 | * @ib: indirect buffer to fill with commands |
||
145 | * @pe: addr of the page entry |
||
146 | * @addr: dst addr to write into pe |
||
147 | * @count: number of page entries to update |
||
148 | * @incr: increase next addr by incr bytes |
||
149 | * @flags: access flags |
||
150 | * |
||
151 | * Update the page tables using the DMA (SI). |
||
152 | */ |
||
153 | void si_dma_vm_set_pages(struct radeon_device *rdev, |
||
154 | struct radeon_ib *ib, |
||
155 | uint64_t pe, |
||
156 | uint64_t addr, unsigned count, |
||
157 | uint32_t incr, uint32_t flags) |
||
158 | { |
||
159 | uint64_t value; |
||
160 | unsigned ndw; |
||
161 | |||
162 | while (count) { |
||
163 | ndw = count * 2; |
||
164 | if (ndw > 0xFFFFE) |
||
165 | ndw = 0xFFFFE; |
||
166 | |||
167 | if (flags & R600_PTE_VALID) |
||
168 | value = addr; |
||
169 | else |
||
170 | value = 0; |
||
171 | |||
172 | /* for physically contiguous pages (vram) */ |
||
173 | ib->ptr[ib->length_dw++] = DMA_PTE_PDE_PACKET(ndw); |
||
174 | ib->ptr[ib->length_dw++] = pe; /* dst addr */ |
||
175 | ib->ptr[ib->length_dw++] = upper_32_bits(pe) & 0xff; |
||
176 | ib->ptr[ib->length_dw++] = flags; /* mask */ |
||
177 | ib->ptr[ib->length_dw++] = 0; |
||
178 | ib->ptr[ib->length_dw++] = value; /* value */ |
||
179 | ib->ptr[ib->length_dw++] = upper_32_bits(value); |
||
180 | ib->ptr[ib->length_dw++] = incr; /* increment size */ |
||
181 | ib->ptr[ib->length_dw++] = 0; |
||
182 | pe += ndw * 4; |
||
183 | addr += (ndw / 2) * incr; |
||
184 | count -= ndw / 2; |
||
185 | } |
||
186 | } |
||
187 | |||
5271 | serge | 188 | void si_dma_vm_flush(struct radeon_device *rdev, struct radeon_ring *ring, |
189 | unsigned vm_id, uint64_t pd_addr) |
||
190 | |||
5078 | serge | 191 | { |
192 | radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_SRBM_WRITE, 0, 0, 0, 0)); |
||
5271 | serge | 193 | if (vm_id < 8) { |
194 | radeon_ring_write(ring, (0xf << 16) | ((VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (vm_id << 2)) >> 2)); |
||
5078 | serge | 195 | } else { |
5271 | serge | 196 | radeon_ring_write(ring, (0xf << 16) | ((VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + ((vm_id - 8) << 2)) >> 2)); |
5078 | serge | 197 | } |
5271 | serge | 198 | radeon_ring_write(ring, pd_addr >> 12); |
5078 | serge | 199 | |
200 | /* flush hdp cache */ |
||
201 | radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_SRBM_WRITE, 0, 0, 0, 0)); |
||
202 | radeon_ring_write(ring, (0xf << 16) | (HDP_MEM_COHERENCY_FLUSH_CNTL >> 2)); |
||
203 | radeon_ring_write(ring, 1); |
||
204 | |||
205 | /* bits 0-7 are the VM contexts0-7 */ |
||
206 | radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_SRBM_WRITE, 0, 0, 0, 0)); |
||
207 | radeon_ring_write(ring, (0xf << 16) | (VM_INVALIDATE_REQUEST >> 2)); |
||
5271 | serge | 208 | radeon_ring_write(ring, 1 << vm_id); |
5078 | serge | 209 | } |
210 | |||
211 | /** |
||
212 | * si_copy_dma - copy pages using the DMA engine |
||
213 | * |
||
214 | * @rdev: radeon_device pointer |
||
215 | * @src_offset: src GPU address |
||
216 | * @dst_offset: dst GPU address |
||
217 | * @num_gpu_pages: number of GPU pages to xfer |
||
5271 | serge | 218 | * @resv: reservation object to sync to |
5078 | serge | 219 | * |
220 | * Copy GPU paging using the DMA engine (SI). |
||
221 | * Used by the radeon ttm implementation to move pages if |
||
222 | * registered as the asic copy callback. |
||
223 | */ |
||
5271 | serge | 224 | struct radeon_fence *si_copy_dma(struct radeon_device *rdev, |
5078 | serge | 225 | uint64_t src_offset, uint64_t dst_offset, |
226 | unsigned num_gpu_pages, |
||
5271 | serge | 227 | struct reservation_object *resv) |
5078 | serge | 228 | { |
5271 | serge | 229 | struct radeon_fence *fence; |
230 | struct radeon_sync sync; |
||
5078 | serge | 231 | int ring_index = rdev->asic->copy.dma_ring_index; |
232 | struct radeon_ring *ring = &rdev->ring[ring_index]; |
||
233 | u32 size_in_bytes, cur_size_in_bytes; |
||
234 | int i, num_loops; |
||
235 | int r = 0; |
||
236 | |||
5271 | serge | 237 | radeon_sync_create(&sync); |
5078 | serge | 238 | |
239 | size_in_bytes = (num_gpu_pages << RADEON_GPU_PAGE_SHIFT); |
||
240 | num_loops = DIV_ROUND_UP(size_in_bytes, 0xfffff); |
||
241 | r = radeon_ring_lock(rdev, ring, num_loops * 5 + 11); |
||
242 | if (r) { |
||
243 | DRM_ERROR("radeon: moving bo (%d).\n", r); |
||
5271 | serge | 244 | radeon_sync_free(rdev, &sync, NULL); |
245 | return ERR_PTR(r); |
||
5078 | serge | 246 | } |
247 | |||
5271 | serge | 248 | radeon_sync_resv(rdev, &sync, resv, false); |
249 | radeon_sync_rings(rdev, &sync, ring->idx); |
||
5078 | serge | 250 | |
251 | for (i = 0; i < num_loops; i++) { |
||
252 | cur_size_in_bytes = size_in_bytes; |
||
253 | if (cur_size_in_bytes > 0xFFFFF) |
||
254 | cur_size_in_bytes = 0xFFFFF; |
||
255 | size_in_bytes -= cur_size_in_bytes; |
||
256 | radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_COPY, 1, 0, 0, cur_size_in_bytes)); |
||
257 | radeon_ring_write(ring, lower_32_bits(dst_offset)); |
||
258 | radeon_ring_write(ring, lower_32_bits(src_offset)); |
||
259 | radeon_ring_write(ring, upper_32_bits(dst_offset) & 0xff); |
||
260 | radeon_ring_write(ring, upper_32_bits(src_offset) & 0xff); |
||
261 | src_offset += cur_size_in_bytes; |
||
262 | dst_offset += cur_size_in_bytes; |
||
263 | } |
||
264 | |||
5271 | serge | 265 | r = radeon_fence_emit(rdev, &fence, ring->idx); |
5078 | serge | 266 | if (r) { |
267 | radeon_ring_unlock_undo(rdev, ring); |
||
5271 | serge | 268 | radeon_sync_free(rdev, &sync, NULL); |
269 | return ERR_PTR(r); |
||
5078 | serge | 270 | } |
271 | |||
272 | radeon_ring_unlock_commit(rdev, ring, false); |
||
5271 | serge | 273 | radeon_sync_free(rdev, &sync, fence); |
5078 | serge | 274 | |
5271 | serge | 275 | return fence; |
5078 | serge | 276 | }>><>><>><>><>><>><>><>><>> |
277 |