File: | dev/pci/drm/i915/gt/intel_ggtt.c |
Warning: | line 962, column 18 Value stored to 'pdev' during its initialization is never read |
Press '?' to see keyboard shortcuts
Keyboard shortcuts:
1 | // SPDX-License-Identifier: MIT |
2 | /* |
3 | * Copyright © 2020 Intel Corporation |
4 | */ |
5 | |
6 | #include <asm/set_memory.h> |
7 | #include <asm/smp.h> |
8 | #include <linux/types.h> |
9 | #include <linux/stop_machine.h> |
10 | |
11 | #include <drm/i915_drm.h> |
12 | #include <drm/intel-gtt.h> |
13 | |
14 | #include "gem/i915_gem_lmem.h" |
15 | |
16 | #include "intel_ggtt_gmch.h" |
17 | #include "intel_gt.h" |
18 | #include "intel_gt_regs.h" |
19 | #include "intel_pci_config.h" |
20 | #include "i915_drv.h" |
21 | #include "i915_pci.h" |
22 | #include "i915_scatterlist.h" |
23 | #include "i915_utils.h" |
24 | #include "i915_vgpu.h" |
25 | |
26 | #include "intel_gtt.h" |
27 | #include "gen8_ppgtt.h" |
28 | |
29 | #include <dev/pci/pcivar.h> |
30 | #include <dev/pci/agpvar.h> |
31 | |
32 | static inline bool_Bool suspend_retains_ptes(struct i915_address_space *vm) |
33 | { |
34 | return GRAPHICS_VER(vm->i915)((&(vm->i915)->__runtime)->graphics.ip.ver) >= 8 && |
35 | !HAS_LMEM(vm->i915)((&(vm->i915)->__runtime)->memory_regions & ( (1UL << (INTEL_REGION_LMEM_0)))) && |
36 | vm->is_ggtt; |
37 | } |
38 | |
39 | static void i915_ggtt_color_adjust(const struct drm_mm_node *node, |
40 | unsigned long color, |
41 | u64 *start, |
42 | u64 *end) |
43 | { |
44 | if (i915_node_color_differs(node, color)) |
45 | *start += I915_GTT_PAGE_SIZE(1ULL << (12)); |
46 | |
47 | /* |
48 | * Also leave a space between the unallocated reserved node after the |
49 | * GTT and any objects within the GTT, i.e. we use the color adjustment |
50 | * to insert a guard page to prevent prefetches crossing over the |
51 | * GTT boundary. |
52 | */ |
53 | node = list_next_entry(node, node_list)({ const __typeof( ((typeof(*(node)) *)0)->node_list ) *__mptr = (((node)->node_list.next)); (typeof(*(node)) *)( (char * )__mptr - __builtin_offsetof(typeof(*(node)), node_list) );}); |
54 | if (node->color != color) |
55 | *end -= I915_GTT_PAGE_SIZE(1ULL << (12)); |
56 | } |
57 | |
58 | static int ggtt_init_hw(struct i915_ggtt *ggtt) |
59 | { |
60 | struct drm_i915_privateinteldrm_softc *i915 = ggtt->vm.i915; |
61 | int i; |
62 | |
63 | i915_address_space_init(&ggtt->vm, VM_CLASS_GGTT0); |
64 | |
65 | ggtt->vm.is_ggtt = true1; |
66 | |
67 | /* Only VLV supports read-only GGTT mappings */ |
68 | ggtt->vm.has_read_only = IS_VALLEYVIEW(i915)IS_PLATFORM(i915, INTEL_VALLEYVIEW); |
69 | |
70 | if (!HAS_LLC(i915)((&(i915)->__info)->has_llc) && !HAS_PPGTT(i915)(((&(i915)->__runtime)->ppgtt_type) != INTEL_PPGTT_NONE )) |
71 | ggtt->vm.mm.color_adjust = i915_ggtt_color_adjust; |
72 | |
73 | if (ggtt->mappable_end) { |
74 | #ifdef __linux__ |
75 | if (!io_mapping_init_wc(&ggtt->iomap, |
76 | ggtt->gmadr.start, |
77 | ggtt->mappable_end)) { |
78 | ggtt->vm.cleanup(&ggtt->vm); |
79 | return -EIO5; |
80 | } |
81 | |
82 | ggtt->mtrr = arch_phys_wc_add(ggtt->gmadr.start, |
83 | ggtt->mappable_end); |
84 | #else |
85 | /* XXX would be a lot nicer to get agp info before now */ |
86 | uvm_page_physload(atop(ggtt->gmadr.start)((ggtt->gmadr.start) >> 12), |
87 | atop(ggtt->gmadr.start + ggtt->mappable_end)((ggtt->gmadr.start + ggtt->mappable_end) >> 12), |
88 | atop(ggtt->gmadr.start)((ggtt->gmadr.start) >> 12), |
89 | atop(ggtt->gmadr.start + ggtt->mappable_end)((ggtt->gmadr.start + ggtt->mappable_end) >> 12), |
90 | PHYSLOAD_DEVICE0x01); |
91 | /* array of vm pages that physload introduced. */ |
92 | i915->pgs = PHYS_TO_VM_PAGE(ggtt->gmadr.start); |
93 | KASSERT(i915->pgs != NULL)((i915->pgs != ((void *)0)) ? (void)0 : __assert("diagnostic " , "/usr/src/sys/dev/pci/drm/i915/gt/intel_ggtt.c", 93, "i915->pgs != NULL" )); |
94 | /* |
95 | * XXX mark all pages write combining so user mmaps get the |
96 | * right bits. We really need a proper MI api for doing this, |
97 | * but for now this allows us to use PAT where available. |
98 | */ |
99 | for (i = 0; i < atop(ggtt->mappable_end)((ggtt->mappable_end) >> 12); i++) |
100 | atomic_setbits_intx86_atomic_setbits_u32(&(i915->pgs[i].pg_flags), |
101 | PG_PMAP_WC0x04000000); |
102 | if (agp_init_map(i915->bst, ggtt->gmadr.start, |
103 | ggtt->mappable_end, |
104 | BUS_SPACE_MAP_LINEAR0x0002 | BUS_SPACE_MAP_PREFETCHABLE0x0008, |
105 | &i915->agph)) |
106 | panic("can't map aperture"); |
107 | #endif |
108 | } |
109 | |
110 | intel_ggtt_init_fences(ggtt); |
111 | |
112 | return 0; |
113 | } |
114 | |
115 | /** |
116 | * i915_ggtt_init_hw - Initialize GGTT hardware |
117 | * @i915: i915 device |
118 | */ |
119 | int i915_ggtt_init_hw(struct drm_i915_privateinteldrm_softc *i915) |
120 | { |
121 | int ret; |
122 | |
123 | /* |
124 | * Note that we use page colouring to enforce a guard page at the |
125 | * end of the address space. This is required as the CS may prefetch |
126 | * beyond the end of the batch buffer, across the page boundary, |
127 | * and beyond the end of the GTT if we do not provide a guard. |
128 | */ |
129 | ret = ggtt_init_hw(to_gt(i915)->ggtt); |
130 | if (ret) |
131 | return ret; |
132 | |
133 | return 0; |
134 | } |
135 | |
136 | /* |
137 | * Return the value of the last GGTT pte cast to an u64, if |
138 | * the system is supposed to retain ptes across resume. 0 otherwise. |
139 | */ |
140 | static u64 read_last_pte(struct i915_address_space *vm) |
141 | { |
142 | struct i915_ggtt *ggtt = i915_vm_to_ggtt(vm); |
143 | gen8_pte_t __iomem *ptep; |
144 | |
145 | if (!suspend_retains_ptes(vm)) |
146 | return 0; |
147 | |
148 | GEM_BUG_ON(GRAPHICS_VER(vm->i915) < 8)((void)0); |
149 | ptep = (typeof(ptep))ggtt->gsm + (ggtt_total_entries(ggtt)((ggtt)->vm.total >> 12) - 1); |
150 | return readq(ptep)ioread64(ptep); |
151 | } |
152 | |
153 | /** |
154 | * i915_ggtt_suspend_vm - Suspend the memory mappings for a GGTT or DPT VM |
155 | * @vm: The VM to suspend the mappings for |
156 | * |
157 | * Suspend the memory mappings for all objects mapped to HW via the GGTT or a |
158 | * DPT page table. |
159 | */ |
160 | void i915_ggtt_suspend_vm(struct i915_address_space *vm) |
161 | { |
162 | struct i915_vma *vma, *vn; |
163 | int save_skip_rewrite; |
164 | |
165 | drm_WARN_ON(&vm->i915->drm, !vm->is_ggtt && !vm->is_dpt)({ int __ret = !!((!vm->is_ggtt && !vm->is_dpt) ); if (__ret) printf("%s %s: " "%s", dev_driver_string(((& vm->i915->drm))->dev), "", "drm_WARN_ON(" "!vm->is_ggtt && !vm->is_dpt" ")"); __builtin_expect(!!(__ret), 0); }); |
166 | |
167 | retry: |
168 | i915_gem_drain_freed_objects(vm->i915); |
169 | |
170 | mutex_lock(&vm->mutex)rw_enter_write(&vm->mutex); |
171 | |
172 | /* |
173 | * Skip rewriting PTE on VMA unbind. |
174 | * FIXME: Use an argument to i915_vma_unbind() instead? |
175 | */ |
176 | save_skip_rewrite = vm->skip_pte_rewrite; |
177 | vm->skip_pte_rewrite = true1; |
178 | |
179 | list_for_each_entry_safe(vma, vn, &vm->bound_list, vm_link)for (vma = ({ const __typeof( ((__typeof(*vma) *)0)->vm_link ) *__mptr = ((&vm->bound_list)->next); (__typeof(* vma) *)( (char *)__mptr - __builtin_offsetof(__typeof(*vma), vm_link ) );}), vn = ({ const __typeof( ((__typeof(*vma) *)0)->vm_link ) *__mptr = (vma->vm_link.next); (__typeof(*vma) *)( (char *)__mptr - __builtin_offsetof(__typeof(*vma), vm_link) );}); &vma->vm_link != (&vm->bound_list); vma = vn, vn = ({ const __typeof( ((__typeof(*vn) *)0)->vm_link ) *__mptr = (vn->vm_link.next); (__typeof(*vn) *)( (char *)__mptr - __builtin_offsetof(__typeof(*vn), vm_link) );})) { |
180 | struct drm_i915_gem_object *obj = vma->obj; |
181 | |
182 | GEM_BUG_ON(!drm_mm_node_allocated(&vma->node))((void)0); |
183 | |
184 | if (i915_vma_is_pinned(vma) || !i915_vma_is_bound(vma, I915_VMA_GLOBAL_BIND((int)(1UL << (10))))) |
185 | continue; |
186 | |
187 | /* unlikely to race when GPU is idle, so no worry about slowpath.. */ |
188 | if (WARN_ON(!i915_gem_object_trylock(obj, NULL))({ int __ret = !!(!i915_gem_object_trylock(obj, ((void *)0))) ; if (__ret) printf("WARNING %s failed at %s:%d\n", "!i915_gem_object_trylock(obj, ((void *)0))" , "/usr/src/sys/dev/pci/drm/i915/gt/intel_ggtt.c", 188); __builtin_expect (!!(__ret), 0); })) { |
189 | /* |
190 | * No dead objects should appear here, GPU should be |
191 | * completely idle, and userspace suspended |
192 | */ |
193 | i915_gem_object_get(obj); |
194 | |
195 | mutex_unlock(&vm->mutex)rw_exit_write(&vm->mutex); |
196 | |
197 | i915_gem_object_lock(obj, NULL((void *)0)); |
198 | GEM_WARN_ON(i915_vma_unbind(vma))({ __builtin_expect(!!(!!(i915_vma_unbind(vma))), 0); }); |
199 | i915_gem_object_unlock(obj); |
200 | i915_gem_object_put(obj); |
201 | |
202 | vm->skip_pte_rewrite = save_skip_rewrite; |
203 | goto retry; |
204 | } |
205 | |
206 | if (!i915_vma_is_bound(vma, I915_VMA_GLOBAL_BIND((int)(1UL << (10))))) { |
207 | i915_vma_wait_for_bind(vma); |
208 | |
209 | __i915_vma_evict(vma, false0); |
210 | drm_mm_remove_node(&vma->node); |
211 | } |
212 | |
213 | i915_gem_object_unlock(obj); |
214 | } |
215 | |
216 | if (!suspend_retains_ptes(vm)) |
217 | vm->clear_range(vm, 0, vm->total); |
218 | else |
219 | i915_vm_to_ggtt(vm)->probed_pte = read_last_pte(vm); |
220 | |
221 | vm->skip_pte_rewrite = save_skip_rewrite; |
222 | |
223 | mutex_unlock(&vm->mutex)rw_exit_write(&vm->mutex); |
224 | } |
225 | |
226 | void i915_ggtt_suspend(struct i915_ggtt *ggtt) |
227 | { |
228 | i915_ggtt_suspend_vm(&ggtt->vm); |
229 | ggtt->invalidate(ggtt); |
230 | |
231 | intel_gt_check_and_clear_faults(ggtt->vm.gt); |
232 | } |
233 | |
234 | void gen6_ggtt_invalidate(struct i915_ggtt *ggtt) |
235 | { |
236 | struct intel_uncore *uncore = ggtt->vm.gt->uncore; |
237 | |
238 | spin_lock_irq(&uncore->lock)mtx_enter(&uncore->lock); |
239 | intel_uncore_write_fw(uncore, GFX_FLSH_CNTL_GEN6, GFX_FLSH_CNTL_EN)__raw_uncore_write32(uncore, ((const i915_reg_t){ .reg = (0x101008 ) }), (1 << 0)); |
240 | intel_uncore_read_fw(uncore, GFX_FLSH_CNTL_GEN6)__raw_uncore_read32(uncore, ((const i915_reg_t){ .reg = (0x101008 ) })); |
241 | spin_unlock_irq(&uncore->lock)mtx_leave(&uncore->lock); |
242 | } |
243 | |
244 | static void gen8_ggtt_invalidate(struct i915_ggtt *ggtt) |
245 | { |
246 | struct intel_uncore *uncore = ggtt->vm.gt->uncore; |
247 | |
248 | /* |
249 | * Note that as an uncached mmio write, this will flush the |
250 | * WCB of the writes into the GGTT before it triggers the invalidate. |
251 | */ |
252 | intel_uncore_write_fw(uncore, GFX_FLSH_CNTL_GEN6, GFX_FLSH_CNTL_EN)__raw_uncore_write32(uncore, ((const i915_reg_t){ .reg = (0x101008 ) }), (1 << 0)); |
253 | } |
254 | |
255 | static void guc_ggtt_invalidate(struct i915_ggtt *ggtt) |
256 | { |
257 | struct intel_uncore *uncore = ggtt->vm.gt->uncore; |
258 | struct drm_i915_privateinteldrm_softc *i915 = ggtt->vm.i915; |
259 | |
260 | gen8_ggtt_invalidate(ggtt); |
261 | |
262 | if (GRAPHICS_VER(i915)((&(i915)->__runtime)->graphics.ip.ver) >= 12) |
263 | intel_uncore_write_fw(uncore, GEN12_GUC_TLB_INV_CR,__raw_uncore_write32(uncore, ((const i915_reg_t){ .reg = (0xcee8 ) }), (1 << 0)) |
264 | GEN12_GUC_TLB_INV_CR_INVALIDATE)__raw_uncore_write32(uncore, ((const i915_reg_t){ .reg = (0xcee8 ) }), (1 << 0)); |
265 | else |
266 | intel_uncore_write_fw(uncore, GEN8_GTCR, GEN8_GTCR_INVALIDATE)__raw_uncore_write32(uncore, ((const i915_reg_t){ .reg = (0x4274 ) }), (1<<0)); |
267 | } |
268 | |
269 | u64 gen8_ggtt_pte_encode(dma_addr_t addr, |
270 | enum i915_cache_level level, |
271 | u32 flags) |
272 | { |
273 | gen8_pte_t pte = addr | GEN8_PAGE_PRESENT(1ULL << (0)); |
274 | |
275 | if (flags & PTE_LM(1UL << (1))) |
276 | pte |= GEN12_GGTT_PTE_LM(1ULL << (1)); |
277 | |
278 | return pte; |
279 | } |
280 | |
281 | static void gen8_set_pte(void __iomem *addr, gen8_pte_t pte) |
282 | { |
283 | writeq(pte, addr)iowrite64(pte, addr); |
284 | } |
285 | |
286 | static void gen8_ggtt_insert_page(struct i915_address_space *vm, |
287 | dma_addr_t addr, |
288 | u64 offset, |
289 | enum i915_cache_level level, |
290 | u32 flags) |
291 | { |
292 | struct i915_ggtt *ggtt = i915_vm_to_ggtt(vm); |
293 | gen8_pte_t __iomem *pte = |
294 | (gen8_pte_t __iomem *)ggtt->gsm + offset / I915_GTT_PAGE_SIZE(1ULL << (12)); |
295 | |
296 | gen8_set_pte(pte, gen8_ggtt_pte_encode(addr, level, flags)); |
297 | |
298 | ggtt->invalidate(ggtt); |
299 | } |
300 | |
301 | static void gen8_ggtt_insert_entries(struct i915_address_space *vm, |
302 | struct i915_vma_resource *vma_res, |
303 | enum i915_cache_level level, |
304 | u32 flags) |
305 | { |
306 | const gen8_pte_t pte_encode = gen8_ggtt_pte_encode(0, level, flags); |
307 | struct i915_ggtt *ggtt = i915_vm_to_ggtt(vm); |
308 | gen8_pte_t __iomem *gte; |
309 | gen8_pte_t __iomem *end; |
310 | struct sgt_iter iter; |
311 | dma_addr_t addr; |
312 | |
313 | /* |
314 | * Note that we ignore PTE_READ_ONLY here. The caller must be careful |
315 | * not to allow the user to override access to a read only page. |
316 | */ |
317 | |
318 | gte = (gen8_pte_t __iomem *)ggtt->gsm; |
319 | gte += vma_res->start / I915_GTT_PAGE_SIZE(1ULL << (12)); |
320 | end = gte + vma_res->node_size / I915_GTT_PAGE_SIZE(1ULL << (12)); |
321 | |
322 | for_each_sgt_daddr(addr, iter, vma_res->bi.pages)for ((iter) = __sgt_iter((vma_res->bi.pages)->sgl, 1); ( (addr) = (iter).dma + (iter).curr), (iter).sgp; (((iter).curr += ((1ULL << (12)))) >= (iter).max) ? (iter) = __sgt_iter (__sg_next((iter).sgp), 1), 0 : 0) |
323 | gen8_set_pte(gte++, pte_encode | addr); |
324 | GEM_BUG_ON(gte > end)((void)0); |
325 | |
326 | /* Fill the allocated but "unused" space beyond the end of the buffer */ |
327 | while (gte < end) |
328 | gen8_set_pte(gte++, vm->scratch[0]->encode); |
329 | |
330 | /* |
331 | * We want to flush the TLBs only after we're certain all the PTE |
332 | * updates have finished. |
333 | */ |
334 | ggtt->invalidate(ggtt); |
335 | } |
336 | |
337 | static void gen6_ggtt_insert_page(struct i915_address_space *vm, |
338 | dma_addr_t addr, |
339 | u64 offset, |
340 | enum i915_cache_level level, |
341 | u32 flags) |
342 | { |
343 | struct i915_ggtt *ggtt = i915_vm_to_ggtt(vm); |
344 | gen6_pte_t __iomem *pte = |
345 | (gen6_pte_t __iomem *)ggtt->gsm + offset / I915_GTT_PAGE_SIZE(1ULL << (12)); |
346 | |
347 | iowrite32(vm->pte_encode(addr, level, flags), pte); |
348 | |
349 | ggtt->invalidate(ggtt); |
350 | } |
351 | |
352 | /* |
353 | * Binds an object into the global gtt with the specified cache level. |
354 | * The object will be accessible to the GPU via commands whose operands |
355 | * reference offsets within the global GTT as well as accessible by the GPU |
356 | * through the GMADR mapped BAR (i915->mm.gtt->gtt). |
357 | */ |
358 | static void gen6_ggtt_insert_entries(struct i915_address_space *vm, |
359 | struct i915_vma_resource *vma_res, |
360 | enum i915_cache_level level, |
361 | u32 flags) |
362 | { |
363 | struct i915_ggtt *ggtt = i915_vm_to_ggtt(vm); |
364 | gen6_pte_t __iomem *gte; |
365 | gen6_pte_t __iomem *end; |
366 | struct sgt_iter iter; |
367 | dma_addr_t addr; |
368 | |
369 | gte = (gen6_pte_t __iomem *)ggtt->gsm; |
370 | gte += vma_res->start / I915_GTT_PAGE_SIZE(1ULL << (12)); |
371 | end = gte + vma_res->node_size / I915_GTT_PAGE_SIZE(1ULL << (12)); |
372 | |
373 | for_each_sgt_daddr(addr, iter, vma_res->bi.pages)for ((iter) = __sgt_iter((vma_res->bi.pages)->sgl, 1); ( (addr) = (iter).dma + (iter).curr), (iter).sgp; (((iter).curr += ((1ULL << (12)))) >= (iter).max) ? (iter) = __sgt_iter (__sg_next((iter).sgp), 1), 0 : 0) |
374 | iowrite32(vm->pte_encode(addr, level, flags), gte++); |
375 | GEM_BUG_ON(gte > end)((void)0); |
376 | |
377 | /* Fill the allocated but "unused" space beyond the end of the buffer */ |
378 | while (gte < end) |
379 | iowrite32(vm->scratch[0]->encode, gte++); |
380 | |
381 | /* |
382 | * We want to flush the TLBs only after we're certain all the PTE |
383 | * updates have finished. |
384 | */ |
385 | ggtt->invalidate(ggtt); |
386 | } |
387 | |
388 | static void nop_clear_range(struct i915_address_space *vm, |
389 | u64 start, u64 length) |
390 | { |
391 | } |
392 | |
393 | static void gen8_ggtt_clear_range(struct i915_address_space *vm, |
394 | u64 start, u64 length) |
395 | { |
396 | struct i915_ggtt *ggtt = i915_vm_to_ggtt(vm); |
397 | unsigned int first_entry = start / I915_GTT_PAGE_SIZE(1ULL << (12)); |
398 | unsigned int num_entries = length / I915_GTT_PAGE_SIZE(1ULL << (12)); |
399 | const gen8_pte_t scratch_pte = vm->scratch[0]->encode; |
400 | gen8_pte_t __iomem *gtt_base = |
401 | (gen8_pte_t __iomem *)ggtt->gsm + first_entry; |
402 | const int max_entries = ggtt_total_entries(ggtt)((ggtt)->vm.total >> 12) - first_entry; |
403 | int i; |
404 | |
405 | if (WARN(num_entries > max_entries,({ int __ret = !!(num_entries > max_entries); if (__ret) printf ("First entry = %d; Num entries = %d (max=%d)\n", first_entry , num_entries, max_entries); __builtin_expect(!!(__ret), 0); } ) |
406 | "First entry = %d; Num entries = %d (max=%d)\n",({ int __ret = !!(num_entries > max_entries); if (__ret) printf ("First entry = %d; Num entries = %d (max=%d)\n", first_entry , num_entries, max_entries); __builtin_expect(!!(__ret), 0); } ) |
407 | first_entry, num_entries, max_entries)({ int __ret = !!(num_entries > max_entries); if (__ret) printf ("First entry = %d; Num entries = %d (max=%d)\n", first_entry , num_entries, max_entries); __builtin_expect(!!(__ret), 0); } )) |
408 | num_entries = max_entries; |
409 | |
410 | for (i = 0; i < num_entries; i++) |
411 | gen8_set_pte(>t_base[i], scratch_pte); |
412 | } |
413 | |
414 | static void bxt_vtd_ggtt_wa(struct i915_address_space *vm) |
415 | { |
416 | /* |
417 | * Make sure the internal GAM fifo has been cleared of all GTT |
418 | * writes before exiting stop_machine(). This guarantees that |
419 | * any aperture accesses waiting to start in another process |
420 | * cannot back up behind the GTT writes causing a hang. |
421 | * The register can be any arbitrary GAM register. |
422 | */ |
423 | intel_uncore_posting_read_fw(vm->gt->uncore, GFX_FLSH_CNTL_GEN6)((void)__raw_uncore_read32(vm->gt->uncore, ((const i915_reg_t ){ .reg = (0x101008) }))); |
424 | } |
425 | |
426 | struct insert_page { |
427 | struct i915_address_space *vm; |
428 | dma_addr_t addr; |
429 | u64 offset; |
430 | enum i915_cache_level level; |
431 | }; |
432 | |
433 | static int bxt_vtd_ggtt_insert_page__cb(void *_arg) |
434 | { |
435 | struct insert_page *arg = _arg; |
436 | |
437 | gen8_ggtt_insert_page(arg->vm, arg->addr, arg->offset, arg->level, 0); |
438 | bxt_vtd_ggtt_wa(arg->vm); |
439 | |
440 | return 0; |
441 | } |
442 | |
443 | static void bxt_vtd_ggtt_insert_page__BKL(struct i915_address_space *vm, |
444 | dma_addr_t addr, |
445 | u64 offset, |
446 | enum i915_cache_level level, |
447 | u32 unused) |
448 | { |
449 | struct insert_page arg = { vm, addr, offset, level }; |
450 | |
451 | stop_machine(bxt_vtd_ggtt_insert_page__cb, &arg, NULL((void *)0)); |
452 | } |
453 | |
454 | struct insert_entries { |
455 | struct i915_address_space *vm; |
456 | struct i915_vma_resource *vma_res; |
457 | enum i915_cache_level level; |
458 | u32 flags; |
459 | }; |
460 | |
461 | static int bxt_vtd_ggtt_insert_entries__cb(void *_arg) |
462 | { |
463 | struct insert_entries *arg = _arg; |
464 | |
465 | gen8_ggtt_insert_entries(arg->vm, arg->vma_res, arg->level, arg->flags); |
466 | bxt_vtd_ggtt_wa(arg->vm); |
467 | |
468 | return 0; |
469 | } |
470 | |
471 | static void bxt_vtd_ggtt_insert_entries__BKL(struct i915_address_space *vm, |
472 | struct i915_vma_resource *vma_res, |
473 | enum i915_cache_level level, |
474 | u32 flags) |
475 | { |
476 | struct insert_entries arg = { vm, vma_res, level, flags }; |
477 | |
478 | stop_machine(bxt_vtd_ggtt_insert_entries__cb, &arg, NULL((void *)0)); |
479 | } |
480 | |
481 | static void gen6_ggtt_clear_range(struct i915_address_space *vm, |
482 | u64 start, u64 length) |
483 | { |
484 | struct i915_ggtt *ggtt = i915_vm_to_ggtt(vm); |
485 | unsigned int first_entry = start / I915_GTT_PAGE_SIZE(1ULL << (12)); |
486 | unsigned int num_entries = length / I915_GTT_PAGE_SIZE(1ULL << (12)); |
487 | gen6_pte_t scratch_pte, __iomem *gtt_base = |
488 | (gen6_pte_t __iomem *)ggtt->gsm + first_entry; |
489 | const int max_entries = ggtt_total_entries(ggtt)((ggtt)->vm.total >> 12) - first_entry; |
490 | int i; |
491 | |
492 | if (WARN(num_entries > max_entries,({ int __ret = !!(num_entries > max_entries); if (__ret) printf ("First entry = %d; Num entries = %d (max=%d)\n", first_entry , num_entries, max_entries); __builtin_expect(!!(__ret), 0); } ) |
493 | "First entry = %d; Num entries = %d (max=%d)\n",({ int __ret = !!(num_entries > max_entries); if (__ret) printf ("First entry = %d; Num entries = %d (max=%d)\n", first_entry , num_entries, max_entries); __builtin_expect(!!(__ret), 0); } ) |
494 | first_entry, num_entries, max_entries)({ int __ret = !!(num_entries > max_entries); if (__ret) printf ("First entry = %d; Num entries = %d (max=%d)\n", first_entry , num_entries, max_entries); __builtin_expect(!!(__ret), 0); } )) |
495 | num_entries = max_entries; |
496 | |
497 | scratch_pte = vm->scratch[0]->encode; |
498 | for (i = 0; i < num_entries; i++) |
499 | iowrite32(scratch_pte, >t_base[i]); |
500 | } |
501 | |
502 | void intel_ggtt_bind_vma(struct i915_address_space *vm, |
503 | struct i915_vm_pt_stash *stash, |
504 | struct i915_vma_resource *vma_res, |
505 | enum i915_cache_level cache_level, |
506 | u32 flags) |
507 | { |
508 | u32 pte_flags; |
509 | |
510 | if (vma_res->bound_flags & (~flags & I915_VMA_BIND_MASK(((int)(1UL << (10))) | ((int)(1UL << (11)))))) |
511 | return; |
512 | |
513 | vma_res->bound_flags |= flags; |
514 | |
515 | /* Applicable to VLV (gen8+ do not support RO in the GGTT) */ |
516 | pte_flags = 0; |
517 | if (vma_res->bi.readonly) |
518 | pte_flags |= PTE_READ_ONLY(1UL << (0)); |
519 | if (vma_res->bi.lmem) |
520 | pte_flags |= PTE_LM(1UL << (1)); |
521 | |
522 | vm->insert_entries(vm, vma_res, cache_level, pte_flags); |
523 | vma_res->page_sizes_gtt = I915_GTT_PAGE_SIZE(1ULL << (12)); |
524 | } |
525 | |
526 | void intel_ggtt_unbind_vma(struct i915_address_space *vm, |
527 | struct i915_vma_resource *vma_res) |
528 | { |
529 | vm->clear_range(vm, vma_res->start, vma_res->vma_size); |
530 | } |
531 | |
532 | /* |
533 | * Reserve the top of the GuC address space for firmware images. Addresses |
534 | * beyond GUC_GGTT_TOP in the GuC address space are inaccessible by GuC, |
535 | * which makes for a suitable range to hold GuC/HuC firmware images if the |
536 | * size of the GGTT is 4G. However, on a 32-bit platform the size of the GGTT |
537 | * is limited to 2G, which is less than GUC_GGTT_TOP, but we reserve a chunk |
538 | * of the same size anyway, which is far more than needed, to keep the logic |
539 | * in uc_fw_ggtt_offset() simple. |
540 | */ |
541 | #define GUC_TOP_RESERVE_SIZE((4ULL << 30) - 0xFEE00000) (SZ_4G(4ULL << 30) - GUC_GGTT_TOP0xFEE00000) |
542 | |
543 | static int ggtt_reserve_guc_top(struct i915_ggtt *ggtt) |
544 | { |
545 | u64 offset; |
546 | int ret; |
547 | |
548 | if (!intel_uc_uses_guc(&ggtt->vm.gt->uc)) |
549 | return 0; |
550 | |
551 | GEM_BUG_ON(ggtt->vm.total <= GUC_TOP_RESERVE_SIZE)((void)0); |
552 | offset = ggtt->vm.total - GUC_TOP_RESERVE_SIZE((4ULL << 30) - 0xFEE00000); |
553 | |
554 | ret = i915_gem_gtt_reserve(&ggtt->vm, NULL((void *)0), &ggtt->uc_fw, |
555 | GUC_TOP_RESERVE_SIZE((4ULL << 30) - 0xFEE00000), offset, |
556 | I915_COLOR_UNEVICTABLE(-1), PIN_NOEVICT(1ULL << (0))); |
557 | if (ret) |
558 | drm_dbg(&ggtt->vm.i915->drm,__drm_dev_dbg(((void *)0), (&ggtt->vm.i915->drm) ? ( &ggtt->vm.i915->drm)->dev : ((void *)0), DRM_UT_DRIVER , "Failed to reserve top of GGTT for GuC\n") |
559 | "Failed to reserve top of GGTT for GuC\n")__drm_dev_dbg(((void *)0), (&ggtt->vm.i915->drm) ? ( &ggtt->vm.i915->drm)->dev : ((void *)0), DRM_UT_DRIVER , "Failed to reserve top of GGTT for GuC\n"); |
560 | |
561 | return ret; |
562 | } |
563 | |
564 | static void ggtt_release_guc_top(struct i915_ggtt *ggtt) |
565 | { |
566 | if (drm_mm_node_allocated(&ggtt->uc_fw)) |
567 | drm_mm_remove_node(&ggtt->uc_fw); |
568 | } |
569 | |
570 | static void cleanup_init_ggtt(struct i915_ggtt *ggtt) |
571 | { |
572 | ggtt_release_guc_top(ggtt); |
573 | if (drm_mm_node_allocated(&ggtt->error_capture)) |
574 | drm_mm_remove_node(&ggtt->error_capture); |
575 | mutex_destroy(&ggtt->error_mutex); |
576 | } |
577 | |
578 | static int init_ggtt(struct i915_ggtt *ggtt) |
579 | { |
580 | /* |
581 | * Let GEM Manage all of the aperture. |
582 | * |
583 | * However, leave one page at the end still bound to the scratch page. |
584 | * There are a number of places where the hardware apparently prefetches |
585 | * past the end of the object, and we've seen multiple hangs with the |
586 | * GPU head pointer stuck in a batchbuffer bound at the last page of the |
587 | * aperture. One page should be enough to keep any prefetching inside |
588 | * of the aperture. |
589 | */ |
590 | unsigned long hole_start, hole_end; |
591 | struct drm_mm_node *entry; |
592 | int ret; |
593 | |
594 | ggtt->pte_lost = true1; |
595 | |
596 | /* |
597 | * GuC requires all resources that we're sharing with it to be placed in |
598 | * non-WOPCM memory. If GuC is not present or not in use we still need a |
599 | * small bias as ring wraparound at offset 0 sometimes hangs. No idea |
600 | * why. |
601 | */ |
602 | ggtt->pin_bias = max_t(u32, I915_GTT_PAGE_SIZE,({ u32 __max_a = ((1ULL << (12))); u32 __max_b = (intel_wopcm_guc_size (&ggtt->vm.i915->wopcm)); __max_a > __max_b ? __max_a : __max_b; }) |
603 | intel_wopcm_guc_size(&ggtt->vm.i915->wopcm))({ u32 __max_a = ((1ULL << (12))); u32 __max_b = (intel_wopcm_guc_size (&ggtt->vm.i915->wopcm)); __max_a > __max_b ? __max_a : __max_b; }); |
604 | |
605 | ret = intel_vgt_balloon(ggtt); |
606 | if (ret) |
607 | return ret; |
608 | |
609 | rw_init(&ggtt->error_mutex, "ggtter")_rw_init_flags(&ggtt->error_mutex, "ggtter", 0, ((void *)0)); |
610 | if (ggtt->mappable_end) { |
611 | /* |
612 | * Reserve a mappable slot for our lockless error capture. |
613 | * |
614 | * We strongly prefer taking address 0x0 in order to protect |
615 | * other critical buffers against accidental overwrites, |
616 | * as writing to address 0 is a very common mistake. |
617 | * |
618 | * Since 0 may already be in use by the system (e.g. the BIOS |
619 | * framebuffer), we let the reservation fail quietly and hope |
620 | * 0 remains reserved always. |
621 | * |
622 | * If we fail to reserve 0, and then fail to find any space |
623 | * for an error-capture, remain silent. We can afford not |
624 | * to reserve an error_capture node as we have fallback |
625 | * paths, and we trust that 0 will remain reserved. However, |
626 | * the only likely reason for failure to insert is a driver |
627 | * bug, which we expect to cause other failures... |
628 | */ |
629 | ggtt->error_capture.size = I915_GTT_PAGE_SIZE(1ULL << (12)); |
630 | ggtt->error_capture.color = I915_COLOR_UNEVICTABLE(-1); |
631 | if (drm_mm_reserve_node(&ggtt->vm.mm, &ggtt->error_capture)) |
632 | drm_mm_insert_node_in_range(&ggtt->vm.mm, |
633 | &ggtt->error_capture, |
634 | ggtt->error_capture.size, 0, |
635 | ggtt->error_capture.color, |
636 | 0, ggtt->mappable_end, |
637 | DRM_MM_INSERT_LOW); |
638 | } |
639 | if (drm_mm_node_allocated(&ggtt->error_capture)) |
640 | drm_dbg(&ggtt->vm.i915->drm,__drm_dev_dbg(((void *)0), (&ggtt->vm.i915->drm) ? ( &ggtt->vm.i915->drm)->dev : ((void *)0), DRM_UT_DRIVER , "Reserved GGTT:[%llx, %llx] for use by error capture\n", ggtt ->error_capture.start, ggtt->error_capture.start + ggtt ->error_capture.size) |
641 | "Reserved GGTT:[%llx, %llx] for use by error capture\n",__drm_dev_dbg(((void *)0), (&ggtt->vm.i915->drm) ? ( &ggtt->vm.i915->drm)->dev : ((void *)0), DRM_UT_DRIVER , "Reserved GGTT:[%llx, %llx] for use by error capture\n", ggtt ->error_capture.start, ggtt->error_capture.start + ggtt ->error_capture.size) |
642 | ggtt->error_capture.start,__drm_dev_dbg(((void *)0), (&ggtt->vm.i915->drm) ? ( &ggtt->vm.i915->drm)->dev : ((void *)0), DRM_UT_DRIVER , "Reserved GGTT:[%llx, %llx] for use by error capture\n", ggtt ->error_capture.start, ggtt->error_capture.start + ggtt ->error_capture.size) |
643 | ggtt->error_capture.start + ggtt->error_capture.size)__drm_dev_dbg(((void *)0), (&ggtt->vm.i915->drm) ? ( &ggtt->vm.i915->drm)->dev : ((void *)0), DRM_UT_DRIVER , "Reserved GGTT:[%llx, %llx] for use by error capture\n", ggtt ->error_capture.start, ggtt->error_capture.start + ggtt ->error_capture.size); |
644 | |
645 | /* |
646 | * The upper portion of the GuC address space has a sizeable hole |
647 | * (several MB) that is inaccessible by GuC. Reserve this range within |
648 | * GGTT as it can comfortably hold GuC/HuC firmware images. |
649 | */ |
650 | ret = ggtt_reserve_guc_top(ggtt); |
651 | if (ret) |
652 | goto err; |
653 | |
654 | /* Clear any non-preallocated blocks */ |
655 | drm_mm_for_each_hole(entry, &ggtt->vm.mm, hole_start, hole_end)for (entry = ({ const __typeof( ((typeof(*entry) *)0)->hole_stack ) *__mptr = ((&(&ggtt->vm.mm)->hole_stack)-> next); (typeof(*entry) *)( (char *)__mptr - __builtin_offsetof (typeof(*entry), hole_stack) );}); &entry->hole_stack != &(&ggtt->vm.mm)->hole_stack ? hole_start = drm_mm_hole_node_start (entry), hole_end = hole_start + entry->hole_size, 1 : 0; entry = ({ const __typeof( ((typeof(*(entry)) *)0)->hole_stack ) *__mptr = (((entry)->hole_stack.next)); (typeof(*(entry)) *)( (char *)__mptr - __builtin_offsetof(typeof(*(entry)), hole_stack ) );})) { |
656 | drm_dbg(&ggtt->vm.i915->drm,__drm_dev_dbg(((void *)0), (&ggtt->vm.i915->drm) ? ( &ggtt->vm.i915->drm)->dev : ((void *)0), DRM_UT_DRIVER , "clearing unused GTT space: [%lx, %lx]\n", hole_start, hole_end ) |
657 | "clearing unused GTT space: [%lx, %lx]\n",__drm_dev_dbg(((void *)0), (&ggtt->vm.i915->drm) ? ( &ggtt->vm.i915->drm)->dev : ((void *)0), DRM_UT_DRIVER , "clearing unused GTT space: [%lx, %lx]\n", hole_start, hole_end ) |
658 | hole_start, hole_end)__drm_dev_dbg(((void *)0), (&ggtt->vm.i915->drm) ? ( &ggtt->vm.i915->drm)->dev : ((void *)0), DRM_UT_DRIVER , "clearing unused GTT space: [%lx, %lx]\n", hole_start, hole_end ); |
659 | ggtt->vm.clear_range(&ggtt->vm, hole_start, |
660 | hole_end - hole_start); |
661 | } |
662 | |
663 | /* And finally clear the reserved guard page */ |
664 | ggtt->vm.clear_range(&ggtt->vm, ggtt->vm.total - PAGE_SIZE(1 << 12), PAGE_SIZE(1 << 12)); |
665 | |
666 | return 0; |
667 | |
668 | err: |
669 | cleanup_init_ggtt(ggtt); |
670 | return ret; |
671 | } |
672 | |
673 | static void aliasing_gtt_bind_vma(struct i915_address_space *vm, |
674 | struct i915_vm_pt_stash *stash, |
675 | struct i915_vma_resource *vma_res, |
676 | enum i915_cache_level cache_level, |
677 | u32 flags) |
678 | { |
679 | u32 pte_flags; |
680 | |
681 | /* Currently applicable only to VLV */ |
682 | pte_flags = 0; |
683 | if (vma_res->bi.readonly) |
684 | pte_flags |= PTE_READ_ONLY(1UL << (0)); |
685 | |
686 | if (flags & I915_VMA_LOCAL_BIND((int)(1UL << (11)))) |
687 | ppgtt_bind_vma(&i915_vm_to_ggtt(vm)->alias->vm, |
688 | stash, vma_res, cache_level, flags); |
689 | |
690 | if (flags & I915_VMA_GLOBAL_BIND((int)(1UL << (10)))) |
691 | vm->insert_entries(vm, vma_res, cache_level, pte_flags); |
692 | |
693 | vma_res->bound_flags |= flags; |
694 | } |
695 | |
696 | static void aliasing_gtt_unbind_vma(struct i915_address_space *vm, |
697 | struct i915_vma_resource *vma_res) |
698 | { |
699 | if (vma_res->bound_flags & I915_VMA_GLOBAL_BIND((int)(1UL << (10)))) |
700 | vm->clear_range(vm, vma_res->start, vma_res->vma_size); |
701 | |
702 | if (vma_res->bound_flags & I915_VMA_LOCAL_BIND((int)(1UL << (11)))) |
703 | ppgtt_unbind_vma(&i915_vm_to_ggtt(vm)->alias->vm, vma_res); |
704 | } |
705 | |
706 | static int init_aliasing_ppgtt(struct i915_ggtt *ggtt) |
707 | { |
708 | struct i915_vm_pt_stash stash = {}; |
709 | struct i915_ppgtt *ppgtt; |
710 | int err; |
711 | |
712 | ppgtt = i915_ppgtt_create(ggtt->vm.gt, 0); |
713 | if (IS_ERR(ppgtt)) |
714 | return PTR_ERR(ppgtt); |
715 | |
716 | if (GEM_WARN_ON(ppgtt->vm.total < ggtt->vm.total)({ __builtin_expect(!!(!!(ppgtt->vm.total < ggtt->vm .total)), 0); })) { |
717 | err = -ENODEV19; |
718 | goto err_ppgtt; |
719 | } |
720 | |
721 | err = i915_vm_alloc_pt_stash(&ppgtt->vm, &stash, ggtt->vm.total); |
722 | if (err) |
723 | goto err_ppgtt; |
724 | |
725 | i915_gem_object_lock(ppgtt->vm.scratch[0], NULL((void *)0)); |
726 | err = i915_vm_map_pt_stash(&ppgtt->vm, &stash); |
727 | i915_gem_object_unlock(ppgtt->vm.scratch[0]); |
728 | if (err) |
729 | goto err_stash; |
730 | |
731 | /* |
732 | * Note we only pre-allocate as far as the end of the global |
733 | * GTT. On 48b / 4-level page-tables, the difference is very, |
734 | * very significant! We have to preallocate as GVT/vgpu does |
735 | * not like the page directory disappearing. |
736 | */ |
737 | ppgtt->vm.allocate_va_range(&ppgtt->vm, &stash, 0, ggtt->vm.total); |
738 | |
739 | ggtt->alias = ppgtt; |
740 | ggtt->vm.bind_async_flags |= ppgtt->vm.bind_async_flags; |
741 | |
742 | GEM_BUG_ON(ggtt->vm.vma_ops.bind_vma != intel_ggtt_bind_vma)((void)0); |
743 | ggtt->vm.vma_ops.bind_vma = aliasing_gtt_bind_vma; |
744 | |
745 | GEM_BUG_ON(ggtt->vm.vma_ops.unbind_vma != intel_ggtt_unbind_vma)((void)0); |
746 | ggtt->vm.vma_ops.unbind_vma = aliasing_gtt_unbind_vma; |
747 | |
748 | i915_vm_free_pt_stash(&ppgtt->vm, &stash); |
749 | return 0; |
750 | |
751 | err_stash: |
752 | i915_vm_free_pt_stash(&ppgtt->vm, &stash); |
753 | err_ppgtt: |
754 | i915_vm_put(&ppgtt->vm); |
755 | return err; |
756 | } |
757 | |
758 | static void fini_aliasing_ppgtt(struct i915_ggtt *ggtt) |
759 | { |
760 | struct i915_ppgtt *ppgtt; |
761 | |
762 | ppgtt = fetch_and_zero(&ggtt->alias)({ typeof(*&ggtt->alias) __T = *(&ggtt->alias); *(&ggtt->alias) = (typeof(*&ggtt->alias))0; __T ; }); |
763 | if (!ppgtt) |
764 | return; |
765 | |
766 | i915_vm_put(&ppgtt->vm); |
767 | |
768 | ggtt->vm.vma_ops.bind_vma = intel_ggtt_bind_vma; |
769 | ggtt->vm.vma_ops.unbind_vma = intel_ggtt_unbind_vma; |
770 | } |
771 | |
772 | int i915_init_ggtt(struct drm_i915_privateinteldrm_softc *i915) |
773 | { |
774 | int ret; |
775 | |
776 | ret = init_ggtt(to_gt(i915)->ggtt); |
777 | if (ret) |
778 | return ret; |
779 | |
780 | if (INTEL_PPGTT(i915)((&(i915)->__runtime)->ppgtt_type) == INTEL_PPGTT_ALIASING) { |
781 | ret = init_aliasing_ppgtt(to_gt(i915)->ggtt); |
782 | if (ret) |
783 | cleanup_init_ggtt(to_gt(i915)->ggtt); |
784 | } |
785 | |
786 | return 0; |
787 | } |
788 | |
789 | static void ggtt_cleanup_hw(struct i915_ggtt *ggtt) |
790 | { |
791 | struct i915_vma *vma, *vn; |
792 | |
793 | flush_workqueue(ggtt->vm.i915->wq); |
794 | i915_gem_drain_freed_objects(ggtt->vm.i915); |
795 | |
796 | mutex_lock(&ggtt->vm.mutex)rw_enter_write(&ggtt->vm.mutex); |
797 | |
798 | ggtt->vm.skip_pte_rewrite = true1; |
799 | |
800 | list_for_each_entry_safe(vma, vn, &ggtt->vm.bound_list, vm_link)for (vma = ({ const __typeof( ((__typeof(*vma) *)0)->vm_link ) *__mptr = ((&ggtt->vm.bound_list)->next); (__typeof (*vma) *)( (char *)__mptr - __builtin_offsetof(__typeof(*vma) , vm_link) );}), vn = ({ const __typeof( ((__typeof(*vma) *)0 )->vm_link ) *__mptr = (vma->vm_link.next); (__typeof(* vma) *)( (char *)__mptr - __builtin_offsetof(__typeof(*vma), vm_link ) );}); &vma->vm_link != (&ggtt->vm.bound_list) ; vma = vn, vn = ({ const __typeof( ((__typeof(*vn) *)0)-> vm_link ) *__mptr = (vn->vm_link.next); (__typeof(*vn) *)( (char *)__mptr - __builtin_offsetof(__typeof(*vn), vm_link) ) ;})) { |
801 | struct drm_i915_gem_object *obj = vma->obj; |
802 | bool_Bool trylock; |
803 | |
804 | trylock = i915_gem_object_trylock(obj, NULL((void *)0)); |
805 | WARN_ON(!trylock)({ int __ret = !!(!trylock); if (__ret) printf("WARNING %s failed at %s:%d\n" , "!trylock", "/usr/src/sys/dev/pci/drm/i915/gt/intel_ggtt.c" , 805); __builtin_expect(!!(__ret), 0); }); |
806 | |
807 | WARN_ON(__i915_vma_unbind(vma))({ int __ret = !!(__i915_vma_unbind(vma)); if (__ret) printf( "WARNING %s failed at %s:%d\n", "__i915_vma_unbind(vma)", "/usr/src/sys/dev/pci/drm/i915/gt/intel_ggtt.c" , 807); __builtin_expect(!!(__ret), 0); }); |
808 | if (trylock) |
809 | i915_gem_object_unlock(obj); |
810 | } |
811 | |
812 | if (drm_mm_node_allocated(&ggtt->error_capture)) |
813 | drm_mm_remove_node(&ggtt->error_capture); |
814 | mutex_destroy(&ggtt->error_mutex); |
815 | |
816 | ggtt_release_guc_top(ggtt); |
817 | intel_vgt_deballoon(ggtt); |
818 | |
819 | ggtt->vm.cleanup(&ggtt->vm); |
820 | |
821 | mutex_unlock(&ggtt->vm.mutex)rw_exit_write(&ggtt->vm.mutex); |
822 | i915_address_space_fini(&ggtt->vm); |
823 | |
824 | #ifdef notyet |
825 | arch_phys_wc_del(ggtt->mtrr); |
826 | |
827 | if (ggtt->iomap.size) |
828 | io_mapping_fini(&ggtt->iomap); |
829 | #endif |
830 | } |
831 | |
832 | /** |
833 | * i915_ggtt_driver_release - Clean up GGTT hardware initialization |
834 | * @i915: i915 device |
835 | */ |
836 | void i915_ggtt_driver_release(struct drm_i915_privateinteldrm_softc *i915) |
837 | { |
838 | struct i915_ggtt *ggtt = to_gt(i915)->ggtt; |
839 | |
840 | fini_aliasing_ppgtt(ggtt); |
841 | |
842 | intel_ggtt_fini_fences(ggtt); |
843 | ggtt_cleanup_hw(ggtt); |
844 | } |
845 | |
846 | /** |
847 | * i915_ggtt_driver_late_release - Cleanup of GGTT that needs to be done after |
848 | * all free objects have been drained. |
849 | * @i915: i915 device |
850 | */ |
851 | void i915_ggtt_driver_late_release(struct drm_i915_privateinteldrm_softc *i915) |
852 | { |
853 | struct i915_ggtt *ggtt = to_gt(i915)->ggtt; |
854 | |
855 | GEM_WARN_ON(kref_read(&ggtt->vm.resv_ref) != 1)({ __builtin_expect(!!(!!(kref_read(&ggtt->vm.resv_ref ) != 1)), 0); }); |
856 | dma_resv_fini(&ggtt->vm._resv); |
857 | } |
858 | |
859 | static unsigned int gen6_get_total_gtt_size(u16 snb_gmch_ctl) |
860 | { |
861 | snb_gmch_ctl >>= SNB_GMCH_GGMS_SHIFT8; |
862 | snb_gmch_ctl &= SNB_GMCH_GGMS_MASK0x3; |
863 | return snb_gmch_ctl << 20; |
864 | } |
865 | |
866 | static unsigned int gen8_get_total_gtt_size(u16 bdw_gmch_ctl) |
867 | { |
868 | bdw_gmch_ctl >>= BDW_GMCH_GGMS_SHIFT6; |
869 | bdw_gmch_ctl &= BDW_GMCH_GGMS_MASK0x3; |
870 | if (bdw_gmch_ctl) |
871 | bdw_gmch_ctl = 1 << bdw_gmch_ctl; |
872 | |
873 | #ifdef CONFIG_X86_32 |
874 | /* Limit 32b platforms to a 2GB GGTT: 4 << 20 / pte size * I915_GTT_PAGE_SIZE */ |
875 | if (bdw_gmch_ctl > 4) |
876 | bdw_gmch_ctl = 4; |
877 | #endif |
878 | |
879 | return bdw_gmch_ctl << 20; |
880 | } |
881 | |
882 | static unsigned int chv_get_total_gtt_size(u16 gmch_ctrl) |
883 | { |
884 | gmch_ctrl >>= SNB_GMCH_GGMS_SHIFT8; |
885 | gmch_ctrl &= SNB_GMCH_GGMS_MASK0x3; |
886 | |
887 | if (gmch_ctrl) |
888 | return 1 << (20 + gmch_ctrl); |
889 | |
890 | return 0; |
891 | } |
892 | |
893 | static unsigned int gen6_gttmmadr_size(struct drm_i915_privateinteldrm_softc *i915) |
894 | { |
895 | /* |
896 | * GEN6: GTTMMADR size is 4MB and GTTADR starts at 2MB offset |
897 | * GEN8: GTTMMADR size is 16MB and GTTADR starts at 8MB offset |
898 | */ |
899 | GEM_BUG_ON(GRAPHICS_VER(i915) < 6)((void)0); |
900 | return (GRAPHICS_VER(i915)((&(i915)->__runtime)->graphics.ip.ver) < 8) ? SZ_4M(4 << 20) : SZ_16M(16 << 20); |
901 | } |
902 | |
903 | static unsigned int gen6_gttadr_offset(struct drm_i915_privateinteldrm_softc *i915) |
904 | { |
905 | return gen6_gttmmadr_size(i915) / 2; |
906 | } |
907 | |
908 | #ifdef __linux__ |
909 | |
910 | static int ggtt_probe_common(struct i915_ggtt *ggtt, u64 size) |
911 | { |
912 | struct drm_i915_privateinteldrm_softc *i915 = ggtt->vm.i915; |
913 | struct pci_dev *pdev = to_pci_dev(i915->drm.dev); |
914 | phys_addr_t phys_addr; |
915 | u32 pte_flags; |
916 | int ret; |
917 | |
918 | GEM_WARN_ON(pci_resource_len(pdev, GTTMMADR_BAR) != gen6_gttmmadr_size(i915))({ __builtin_expect(!!(!!(pci_resource_len(pdev, 0) != gen6_gttmmadr_size (i915))), 0); }); |
919 | phys_addr = pci_resource_start(pdev, GTTMMADR_BAR0) + gen6_gttadr_offset(i915); |
920 | |
921 | /* |
922 | * On BXT+/ICL+ writes larger than 64 bit to the GTT pagetable range |
923 | * will be dropped. For WC mappings in general we have 64 byte burst |
924 | * writes when the WC buffer is flushed, so we can't use it, but have to |
925 | * resort to an uncached mapping. The WC issue is easily caught by the |
926 | * readback check when writing GTT PTE entries. |
927 | */ |
928 | if (IS_GEN9_LP(i915)(((&(i915)->__runtime)->graphics.ip.ver) == 9 && ((&(i915)->__info)->is_lp)) || GRAPHICS_VER(i915)((&(i915)->__runtime)->graphics.ip.ver) >= 11) |
929 | ggtt->gsm = ioremap(phys_addr, size); |
930 | else |
931 | ggtt->gsm = ioremap_wc(phys_addr, size); |
932 | if (!ggtt->gsm) { |
933 | drm_err(&i915->drm, "Failed to map the ggtt page table\n")printf("drm:pid%d:%s *ERROR* " "[drm] " "*ERROR* " "Failed to map the ggtt page table\n" , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__); |
934 | return -ENOMEM12; |
935 | } |
936 | |
937 | kref_init(&ggtt->vm.resv_ref); |
938 | ret = setup_scratch_page(&ggtt->vm); |
939 | if (ret) { |
940 | drm_err(&i915->drm, "Scratch setup failed\n")printf("drm:pid%d:%s *ERROR* " "[drm] " "*ERROR* " "Scratch setup failed\n" , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__); |
941 | /* iounmap will also get called at remove, but meh */ |
942 | iounmap(ggtt->gsm); |
943 | return ret; |
944 | } |
945 | |
946 | pte_flags = 0; |
947 | if (i915_gem_object_is_lmem(ggtt->vm.scratch[0])) |
948 | pte_flags |= PTE_LM(1UL << (1)); |
949 | |
950 | ggtt->vm.scratch[0]->encode = |
951 | ggtt->vm.pte_encode(px_dma(ggtt->vm.scratch[0])(__px_dma(__builtin_choose_expr( __builtin_types_compatible_p (typeof(ggtt->vm.scratch[0]), struct drm_i915_gem_object * ) || __builtin_types_compatible_p(typeof(ggtt->vm.scratch[ 0]), const struct drm_i915_gem_object *), ({ struct drm_i915_gem_object * __x = (struct drm_i915_gem_object *)(ggtt->vm.scratch[0 ]); __x; }), __builtin_choose_expr( __builtin_types_compatible_p (typeof(ggtt->vm.scratch[0]), struct i915_page_table *) || __builtin_types_compatible_p(typeof(ggtt->vm.scratch[0]), const struct i915_page_table *), ({ struct i915_page_table * __x = (struct i915_page_table *)(ggtt->vm.scratch[0]); __x ->base; }), __builtin_choose_expr( __builtin_types_compatible_p (typeof(ggtt->vm.scratch[0]), struct i915_page_directory * ) || __builtin_types_compatible_p(typeof(ggtt->vm.scratch[ 0]), const struct i915_page_directory *), ({ struct i915_page_directory * __x = (struct i915_page_directory *)(ggtt->vm.scratch[0 ]); __x->pt.base; }), (void)0))))), |
952 | I915_CACHE_NONE, pte_flags); |
953 | |
954 | return 0; |
955 | } |
956 | |
957 | #else |
958 | |
959 | static int ggtt_probe_common(struct i915_ggtt *ggtt, u64 size) |
960 | { |
961 | struct drm_i915_privateinteldrm_softc *i915 = ggtt->vm.i915; |
962 | struct pci_dev *pdev = i915->drm.pdev; |
Value stored to 'pdev' during its initialization is never read | |
963 | phys_addr_t phys_addr; |
964 | bus_addr_t addr; |
965 | bus_size_t len; |
966 | pcireg_t type; |
967 | int flags; |
968 | u32 pte_flags; |
969 | int ret; |
970 | |
971 | /* For Modern GENs the PTEs and register space are split in the BAR */ |
972 | type = pci_mapreg_type(i915->pc, i915->tag, 0x10); |
973 | ret = -pci_mapreg_info(i915->pc, i915->tag, 0x10, type, |
974 | &addr, &len, NULL((void *)0)); |
975 | if (ret) |
976 | return ret; |
977 | |
978 | /* |
979 | * On BXT+/ICL+ writes larger than 64 bit to the GTT pagetable range |
980 | * will be dropped. For WC mappings in general we have 64 byte burst |
981 | * writes when the WC buffer is flushed, so we can't use it, but have to |
982 | * resort to an uncached mapping. The WC issue is easily caught by the |
983 | * readback check when writing GTT PTE entries. |
984 | */ |
985 | if (IS_GEN9_LP(i915)(((&(i915)->__runtime)->graphics.ip.ver) == 9 && ((&(i915)->__info)->is_lp)) || GRAPHICS_VER(i915)((&(i915)->__runtime)->graphics.ip.ver) >= 11) |
986 | flags = 0; |
987 | else |
988 | flags = BUS_SPACE_MAP_PREFETCHABLE0x0008; |
989 | ret = -bus_space_map(i915->bst, addr + len / 2, size, |
990 | flags | BUS_SPACE_MAP_LINEAR0x0002, &ggtt->gsm_bsh); |
991 | if (ret) { |
992 | drm_err(&i915->drm, "Failed to map the ggtt page table\n")printf("drm:pid%d:%s *ERROR* " "[drm] " "*ERROR* " "Failed to map the ggtt page table\n" , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__); |
993 | return ret; |
994 | } |
995 | ggtt->gsm = bus_space_vaddr(i915->bst, ggtt->gsm_bsh)((i915->bst)->vaddr((ggtt->gsm_bsh))); |
996 | ggtt->gsm_size = size; |
997 | if (!ggtt->gsm) { |
998 | DRM_ERROR("Failed to map the ggtt page table\n")__drm_err("Failed to map the ggtt page table\n"); |
999 | return -ENOMEM12; |
1000 | } |
1001 | |
1002 | kref_init(&ggtt->vm.resv_ref); |
1003 | ret = setup_scratch_page(&ggtt->vm); |
1004 | if (ret) { |
1005 | drm_err(&i915->drm, "Scratch setup failed\n")printf("drm:pid%d:%s *ERROR* " "[drm] " "*ERROR* " "Scratch setup failed\n" , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__); |
1006 | /* iounmap will also get called at remove, but meh */ |
1007 | bus_space_unmap(i915->bst, ggtt->gsm_bsh, size); |
1008 | return ret; |
1009 | } |
1010 | |
1011 | pte_flags = 0; |
1012 | if (i915_gem_object_is_lmem(ggtt->vm.scratch[0])) |
1013 | pte_flags |= PTE_LM(1UL << (1)); |
1014 | |
1015 | ggtt->vm.scratch[0]->encode = |
1016 | ggtt->vm.pte_encode(px_dma(ggtt->vm.scratch[0])(__px_dma(__builtin_choose_expr( __builtin_types_compatible_p (typeof(ggtt->vm.scratch[0]), struct drm_i915_gem_object * ) || __builtin_types_compatible_p(typeof(ggtt->vm.scratch[ 0]), const struct drm_i915_gem_object *), ({ struct drm_i915_gem_object * __x = (struct drm_i915_gem_object *)(ggtt->vm.scratch[0 ]); __x; }), __builtin_choose_expr( __builtin_types_compatible_p (typeof(ggtt->vm.scratch[0]), struct i915_page_table *) || __builtin_types_compatible_p(typeof(ggtt->vm.scratch[0]), const struct i915_page_table *), ({ struct i915_page_table * __x = (struct i915_page_table *)(ggtt->vm.scratch[0]); __x ->base; }), __builtin_choose_expr( __builtin_types_compatible_p (typeof(ggtt->vm.scratch[0]), struct i915_page_directory * ) || __builtin_types_compatible_p(typeof(ggtt->vm.scratch[ 0]), const struct i915_page_directory *), ({ struct i915_page_directory * __x = (struct i915_page_directory *)(ggtt->vm.scratch[0 ]); __x->pt.base; }), (void)0))))), |
1017 | I915_CACHE_NONE, pte_flags); |
1018 | |
1019 | return 0; |
1020 | } |
1021 | |
1022 | #endif |
1023 | |
1024 | static void gen6_gmch_remove(struct i915_address_space *vm) |
1025 | { |
1026 | struct i915_ggtt *ggtt = i915_vm_to_ggtt(vm); |
1027 | |
1028 | #ifdef __linux__ |
1029 | iounmap(ggtt->gsm); |
1030 | #else |
1031 | bus_space_unmap(vm->i915->bst, ggtt->gsm_bsh, ggtt->gsm_size); |
1032 | #endif |
1033 | free_scratch(vm); |
1034 | } |
1035 | |
1036 | #ifdef __linux__ |
1037 | static struct resource pci_resource(struct pci_dev *pdev, int bar) |
1038 | { |
1039 | return (struct resource)DEFINE_RES_MEM(pci_resource_start(pdev, bar),{ .start = (pci_resource_start(pdev, bar)), .end = (pci_resource_start (pdev, bar)) + (pci_resource_len(pdev, bar)) - 1, } |
1040 | pci_resource_len(pdev, bar)){ .start = (pci_resource_start(pdev, bar)), .end = (pci_resource_start (pdev, bar)) + (pci_resource_len(pdev, bar)) - 1, }; |
1041 | } |
1042 | #endif |
1043 | |
1044 | static int gen8_gmch_probe(struct i915_ggtt *ggtt) |
1045 | { |
1046 | struct drm_i915_privateinteldrm_softc *i915 = ggtt->vm.i915; |
1047 | struct pci_dev *pdev = i915->drm.pdev; |
1048 | unsigned int size; |
1049 | u16 snb_gmch_ctl; |
1050 | |
1051 | if (!HAS_LMEM(i915)((&(i915)->__runtime)->memory_regions & ((1UL << (INTEL_REGION_LMEM_0))))) { |
1052 | #ifdef __linux__ |
1053 | if (!i915_pci_resource_valid(pdev, GTT_APERTURE_BAR2)) |
1054 | return -ENXIO6; |
1055 | |
1056 | ggtt->gmadr = pci_resource(pdev, GTT_APERTURE_BAR2); |
1057 | ggtt->mappable_end = resource_size(&ggtt->gmadr); |
1058 | #else |
1059 | bus_addr_t base; |
1060 | bus_size_t sz; |
1061 | pcireg_t type; |
1062 | int err; |
1063 | |
1064 | type = pci_mapreg_type(i915->pc, i915->tag, 0x18); |
1065 | err = -pci_mapreg_info(i915->pc, i915->tag, 0x18, type, |
1066 | &base, &sz, NULL((void *)0)); |
1067 | if (err) |
1068 | return err; |
1069 | ggtt->gmadr.start = base; |
1070 | ggtt->mappable_end = sz; |
1071 | #endif |
1072 | } |
1073 | |
1074 | pci_read_config_word(pdev, SNB_GMCH_CTRL0x50, &snb_gmch_ctl); |
1075 | if (IS_CHERRYVIEW(i915)IS_PLATFORM(i915, INTEL_CHERRYVIEW)) |
1076 | size = chv_get_total_gtt_size(snb_gmch_ctl); |
1077 | else |
1078 | size = gen8_get_total_gtt_size(snb_gmch_ctl); |
1079 | |
1080 | ggtt->vm.alloc_pt_dma = alloc_pt_dma; |
1081 | ggtt->vm.alloc_scratch_dma = alloc_pt_dma; |
1082 | ggtt->vm.lmem_pt_obj_flags = I915_BO_ALLOC_PM_EARLY(1UL << (5)); |
1083 | |
1084 | ggtt->vm.total = (size / sizeof(gen8_pte_t)) * I915_GTT_PAGE_SIZE(1ULL << (12)); |
1085 | ggtt->vm.cleanup = gen6_gmch_remove; |
1086 | ggtt->vm.insert_page = gen8_ggtt_insert_page; |
1087 | ggtt->vm.clear_range = nop_clear_range; |
1088 | if (intel_scanout_needs_vtd_wa(i915)) |
1089 | ggtt->vm.clear_range = gen8_ggtt_clear_range; |
1090 | |
1091 | ggtt->vm.insert_entries = gen8_ggtt_insert_entries; |
1092 | |
1093 | /* |
1094 | * Serialize GTT updates with aperture access on BXT if VT-d is on, |
1095 | * and always on CHV. |
1096 | */ |
1097 | if (intel_vm_no_concurrent_access_wa(i915)) { |
1098 | ggtt->vm.insert_entries = bxt_vtd_ggtt_insert_entries__BKL; |
1099 | ggtt->vm.insert_page = bxt_vtd_ggtt_insert_page__BKL; |
1100 | |
1101 | /* |
1102 | * Calling stop_machine() version of GGTT update function |
1103 | * at error capture/reset path will raise lockdep warning. |
1104 | * Allow calling gen8_ggtt_insert_* directly at reset path |
1105 | * which is safe from parallel GGTT updates. |
1106 | */ |
1107 | ggtt->vm.raw_insert_page = gen8_ggtt_insert_page; |
1108 | ggtt->vm.raw_insert_entries = gen8_ggtt_insert_entries; |
1109 | |
1110 | ggtt->vm.bind_async_flags = |
1111 | I915_VMA_GLOBAL_BIND((int)(1UL << (10))) | I915_VMA_LOCAL_BIND((int)(1UL << (11))); |
1112 | } |
1113 | |
1114 | ggtt->invalidate = gen8_ggtt_invalidate; |
1115 | |
1116 | ggtt->vm.vma_ops.bind_vma = intel_ggtt_bind_vma; |
1117 | ggtt->vm.vma_ops.unbind_vma = intel_ggtt_unbind_vma; |
1118 | |
1119 | ggtt->vm.pte_encode = gen8_ggtt_pte_encode; |
1120 | |
1121 | setup_private_pat(ggtt->vm.gt->uncore); |
1122 | |
1123 | return ggtt_probe_common(ggtt, size); |
1124 | } |
1125 | |
1126 | static u64 snb_pte_encode(dma_addr_t addr, |
1127 | enum i915_cache_level level, |
1128 | u32 flags) |
1129 | { |
1130 | gen6_pte_t pte = GEN6_PTE_ADDR_ENCODE(addr)((addr) | (((addr) >> 28) & 0xff0)) | GEN6_PTE_VALID((u32)((1UL << (0)) + 0)); |
1131 | |
1132 | switch (level) { |
1133 | case I915_CACHE_L3_LLC: |
1134 | case I915_CACHE_LLC: |
1135 | pte |= GEN6_PTE_CACHE_LLC(2 << 1); |
1136 | break; |
1137 | case I915_CACHE_NONE: |
1138 | pte |= GEN6_PTE_UNCACHED(1 << 1); |
1139 | break; |
1140 | default: |
1141 | MISSING_CASE(level)({ int __ret = !!(1); if (__ret) printf("Missing case (%s == %ld)\n" , "level", (long)(level)); __builtin_expect(!!(__ret), 0); }); |
1142 | } |
1143 | |
1144 | return pte; |
1145 | } |
1146 | |
1147 | static u64 ivb_pte_encode(dma_addr_t addr, |
1148 | enum i915_cache_level level, |
1149 | u32 flags) |
1150 | { |
1151 | gen6_pte_t pte = GEN6_PTE_ADDR_ENCODE(addr)((addr) | (((addr) >> 28) & 0xff0)) | GEN6_PTE_VALID((u32)((1UL << (0)) + 0)); |
1152 | |
1153 | switch (level) { |
1154 | case I915_CACHE_L3_LLC: |
1155 | pte |= GEN7_PTE_CACHE_L3_LLC(3 << 1); |
1156 | break; |
1157 | case I915_CACHE_LLC: |
1158 | pte |= GEN6_PTE_CACHE_LLC(2 << 1); |
1159 | break; |
1160 | case I915_CACHE_NONE: |
1161 | pte |= GEN6_PTE_UNCACHED(1 << 1); |
1162 | break; |
1163 | default: |
1164 | MISSING_CASE(level)({ int __ret = !!(1); if (__ret) printf("Missing case (%s == %ld)\n" , "level", (long)(level)); __builtin_expect(!!(__ret), 0); }); |
1165 | } |
1166 | |
1167 | return pte; |
1168 | } |
1169 | |
1170 | static u64 byt_pte_encode(dma_addr_t addr, |
1171 | enum i915_cache_level level, |
1172 | u32 flags) |
1173 | { |
1174 | gen6_pte_t pte = GEN6_PTE_ADDR_ENCODE(addr)((addr) | (((addr) >> 28) & 0xff0)) | GEN6_PTE_VALID((u32)((1UL << (0)) + 0)); |
1175 | |
1176 | if (!(flags & PTE_READ_ONLY(1UL << (0)))) |
1177 | pte |= BYT_PTE_WRITEABLE((u32)((1UL << (1)) + 0)); |
1178 | |
1179 | if (level != I915_CACHE_NONE) |
1180 | pte |= BYT_PTE_SNOOPED_BY_CPU_CACHES((u32)((1UL << (2)) + 0)); |
1181 | |
1182 | return pte; |
1183 | } |
1184 | |
1185 | static u64 hsw_pte_encode(dma_addr_t addr, |
1186 | enum i915_cache_level level, |
1187 | u32 flags) |
1188 | { |
1189 | gen6_pte_t pte = HSW_PTE_ADDR_ENCODE(addr)((addr) | (((addr) >> 28) & 0x7f0)) | GEN6_PTE_VALID((u32)((1UL << (0)) + 0)); |
1190 | |
1191 | if (level != I915_CACHE_NONE) |
1192 | pte |= HSW_WB_LLC_AGE3((((0x2) & 0x7) << 1) | (((0x2) & 0x8) << (11 - 3))); |
1193 | |
1194 | return pte; |
1195 | } |
1196 | |
1197 | static u64 iris_pte_encode(dma_addr_t addr, |
1198 | enum i915_cache_level level, |
1199 | u32 flags) |
1200 | { |
1201 | gen6_pte_t pte = HSW_PTE_ADDR_ENCODE(addr)((addr) | (((addr) >> 28) & 0x7f0)) | GEN6_PTE_VALID((u32)((1UL << (0)) + 0)); |
1202 | |
1203 | switch (level) { |
1204 | case I915_CACHE_NONE: |
1205 | break; |
1206 | case I915_CACHE_WT: |
1207 | pte |= HSW_WT_ELLC_LLC_AGE3((((0x7) & 0x7) << 1) | (((0x7) & 0x8) << (11 - 3))); |
1208 | break; |
1209 | default: |
1210 | pte |= HSW_WB_ELLC_LLC_AGE3((((0x8) & 0x7) << 1) | (((0x8) & 0x8) << (11 - 3))); |
1211 | break; |
1212 | } |
1213 | |
1214 | return pte; |
1215 | } |
1216 | |
1217 | static int gen6_gmch_probe(struct i915_ggtt *ggtt) |
1218 | { |
1219 | struct drm_i915_privateinteldrm_softc *i915 = ggtt->vm.i915; |
1220 | struct pci_dev *pdev = i915->drm.pdev; |
1221 | unsigned int size; |
1222 | u16 snb_gmch_ctl; |
1223 | |
1224 | #ifdef __linux__ |
1225 | if (!i915_pci_resource_valid(pdev, GTT_APERTURE_BAR2)) |
1226 | return -ENXIO6; |
1227 | |
1228 | ggtt->gmadr = pci_resource(pdev, GTT_APERTURE_BAR2); |
1229 | ggtt->mappable_end = resource_size(&ggtt->gmadr); |
1230 | #else |
1231 | { |
1232 | bus_addr_t base; |
1233 | bus_size_t sz; |
1234 | pcireg_t type; |
1235 | int err; |
1236 | |
1237 | type = pci_mapreg_type(i915->pc, i915->tag, 0x18); |
1238 | err = -pci_mapreg_info(i915->pc, i915->tag, 0x18, type, |
1239 | &base, &sz, NULL((void *)0)); |
1240 | if (err) |
1241 | return err; |
1242 | ggtt->gmadr.start = base; |
1243 | ggtt->mappable_end = sz; |
1244 | } |
1245 | #endif |
1246 | |
1247 | /* |
1248 | * 64/512MB is the current min/max we actually know of, but this is |
1249 | * just a coarse sanity check. |
1250 | */ |
1251 | if (ggtt->mappable_end < (64 << 20) || |
1252 | ggtt->mappable_end > (512 << 20)) { |
1253 | drm_err(&i915->drm, "Unknown GMADR size (%pa)\n",printf("drm:pid%d:%s *ERROR* " "[drm] " "*ERROR* " "Unknown GMADR size (%pa)\n" , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__ , &ggtt ->mappable_end) |
1254 | &ggtt->mappable_end)printf("drm:pid%d:%s *ERROR* " "[drm] " "*ERROR* " "Unknown GMADR size (%pa)\n" , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__ , &ggtt ->mappable_end); |
1255 | return -ENXIO6; |
1256 | } |
1257 | |
1258 | pci_read_config_word(pdev, SNB_GMCH_CTRL0x50, &snb_gmch_ctl); |
1259 | |
1260 | size = gen6_get_total_gtt_size(snb_gmch_ctl); |
1261 | ggtt->vm.total = (size / sizeof(gen6_pte_t)) * I915_GTT_PAGE_SIZE(1ULL << (12)); |
1262 | |
1263 | ggtt->vm.alloc_pt_dma = alloc_pt_dma; |
1264 | ggtt->vm.alloc_scratch_dma = alloc_pt_dma; |
1265 | |
1266 | ggtt->vm.clear_range = nop_clear_range; |
1267 | if (!HAS_FULL_PPGTT(i915)(((&(i915)->__runtime)->ppgtt_type) >= INTEL_PPGTT_FULL ) || intel_scanout_needs_vtd_wa(i915)) |
1268 | ggtt->vm.clear_range = gen6_ggtt_clear_range; |
1269 | ggtt->vm.insert_page = gen6_ggtt_insert_page; |
1270 | ggtt->vm.insert_entries = gen6_ggtt_insert_entries; |
1271 | ggtt->vm.cleanup = gen6_gmch_remove; |
1272 | |
1273 | ggtt->invalidate = gen6_ggtt_invalidate; |
1274 | |
1275 | if (HAS_EDRAM(i915)((i915)->edram_size_mb)) |
1276 | ggtt->vm.pte_encode = iris_pte_encode; |
1277 | else if (IS_HASWELL(i915)IS_PLATFORM(i915, INTEL_HASWELL)) |
1278 | ggtt->vm.pte_encode = hsw_pte_encode; |
1279 | else if (IS_VALLEYVIEW(i915)IS_PLATFORM(i915, INTEL_VALLEYVIEW)) |
1280 | ggtt->vm.pte_encode = byt_pte_encode; |
1281 | else if (GRAPHICS_VER(i915)((&(i915)->__runtime)->graphics.ip.ver) >= 7) |
1282 | ggtt->vm.pte_encode = ivb_pte_encode; |
1283 | else |
1284 | ggtt->vm.pte_encode = snb_pte_encode; |
1285 | |
1286 | ggtt->vm.vma_ops.bind_vma = intel_ggtt_bind_vma; |
1287 | ggtt->vm.vma_ops.unbind_vma = intel_ggtt_unbind_vma; |
1288 | |
1289 | return ggtt_probe_common(ggtt, size); |
1290 | } |
1291 | |
1292 | static int ggtt_probe_hw(struct i915_ggtt *ggtt, struct intel_gt *gt) |
1293 | { |
1294 | struct drm_i915_privateinteldrm_softc *i915 = gt->i915; |
1295 | int ret; |
1296 | |
1297 | ggtt->vm.gt = gt; |
1298 | ggtt->vm.i915 = i915; |
1299 | #ifdef notyet |
1300 | ggtt->vm.dma = i915->drm.dev; |
1301 | #endif |
1302 | dma_resv_init(&ggtt->vm._resv); |
1303 | |
1304 | if (GRAPHICS_VER(i915)((&(i915)->__runtime)->graphics.ip.ver) >= 8) |
1305 | ret = gen8_gmch_probe(ggtt); |
1306 | else if (GRAPHICS_VER(i915)((&(i915)->__runtime)->graphics.ip.ver) >= 6) |
1307 | ret = gen6_gmch_probe(ggtt); |
1308 | else |
1309 | ret = intel_ggtt_gmch_probe(ggtt); |
1310 | |
1311 | if (ret) { |
1312 | dma_resv_fini(&ggtt->vm._resv); |
1313 | return ret; |
1314 | } |
1315 | |
1316 | if ((ggtt->vm.total - 1) >> 32) { |
1317 | drm_err(&i915->drm,printf("drm:pid%d:%s *ERROR* " "[drm] " "*ERROR* " "We never expected a Global GTT with more than 32bits" " of address space! Found %lldM!\n", ({struct cpu_info *__ci ; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof (struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p-> ps_pid, __func__ , ggtt->vm.total >> 20) |
1318 | "We never expected a Global GTT with more than 32bits"printf("drm:pid%d:%s *ERROR* " "[drm] " "*ERROR* " "We never expected a Global GTT with more than 32bits" " of address space! Found %lldM!\n", ({struct cpu_info *__ci ; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof (struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p-> ps_pid, __func__ , ggtt->vm.total >> 20) |
1319 | " of address space! Found %lldM!\n",printf("drm:pid%d:%s *ERROR* " "[drm] " "*ERROR* " "We never expected a Global GTT with more than 32bits" " of address space! Found %lldM!\n", ({struct cpu_info *__ci ; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof (struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p-> ps_pid, __func__ , ggtt->vm.total >> 20) |
1320 | ggtt->vm.total >> 20)printf("drm:pid%d:%s *ERROR* " "[drm] " "*ERROR* " "We never expected a Global GTT with more than 32bits" " of address space! Found %lldM!\n", ({struct cpu_info *__ci ; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof (struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p-> ps_pid, __func__ , ggtt->vm.total >> 20); |
1321 | ggtt->vm.total = 1ULL << 32; |
1322 | ggtt->mappable_end = |
1323 | min_t(u64, ggtt->mappable_end, ggtt->vm.total)({ u64 __min_a = (ggtt->mappable_end); u64 __min_b = (ggtt ->vm.total); __min_a < __min_b ? __min_a : __min_b; }); |
1324 | } |
1325 | |
1326 | if (ggtt->mappable_end > ggtt->vm.total) { |
1327 | drm_err(&i915->drm,printf("drm:pid%d:%s *ERROR* " "[drm] " "*ERROR* " "mappable aperture extends past end of GGTT," " aperture=%pa, total=%llx\n", ({struct cpu_info *__ci; asm volatile ("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid , __func__ , &ggtt->mappable_end, ggtt->vm.total) |
1328 | "mappable aperture extends past end of GGTT,"printf("drm:pid%d:%s *ERROR* " "[drm] " "*ERROR* " "mappable aperture extends past end of GGTT," " aperture=%pa, total=%llx\n", ({struct cpu_info *__ci; asm volatile ("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid , __func__ , &ggtt->mappable_end, ggtt->vm.total) |
1329 | " aperture=%pa, total=%llx\n",printf("drm:pid%d:%s *ERROR* " "[drm] " "*ERROR* " "mappable aperture extends past end of GGTT," " aperture=%pa, total=%llx\n", ({struct cpu_info *__ci; asm volatile ("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid , __func__ , &ggtt->mappable_end, ggtt->vm.total) |
1330 | &ggtt->mappable_end, ggtt->vm.total)printf("drm:pid%d:%s *ERROR* " "[drm] " "*ERROR* " "mappable aperture extends past end of GGTT," " aperture=%pa, total=%llx\n", ({struct cpu_info *__ci; asm volatile ("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid , __func__ , &ggtt->mappable_end, ggtt->vm.total); |
1331 | ggtt->mappable_end = ggtt->vm.total; |
1332 | } |
1333 | |
1334 | /* GMADR is the PCI mmio aperture into the global GTT. */ |
1335 | drm_dbg(&i915->drm, "GGTT size = %lluM\n", ggtt->vm.total >> 20)__drm_dev_dbg(((void *)0), (&i915->drm) ? (&i915-> drm)->dev : ((void *)0), DRM_UT_DRIVER, "GGTT size = %lluM\n" , ggtt->vm.total >> 20); |
1336 | drm_dbg(&i915->drm, "GMADR size = %lluM\n",__drm_dev_dbg(((void *)0), (&i915->drm) ? (&i915-> drm)->dev : ((void *)0), DRM_UT_DRIVER, "GMADR size = %lluM\n" , (u64)ggtt->mappable_end >> 20) |
1337 | (u64)ggtt->mappable_end >> 20)__drm_dev_dbg(((void *)0), (&i915->drm) ? (&i915-> drm)->dev : ((void *)0), DRM_UT_DRIVER, "GMADR size = %lluM\n" , (u64)ggtt->mappable_end >> 20); |
1338 | drm_dbg(&i915->drm, "DSM size = %lluM\n",__drm_dev_dbg(((void *)0), (&i915->drm) ? (&i915-> drm)->dev : ((void *)0), DRM_UT_DRIVER, "DSM size = %lluM\n" , (u64)resource_size(&intel_graphics_stolen_res) >> 20) |
1339 | (u64)resource_size(&intel_graphics_stolen_res) >> 20)__drm_dev_dbg(((void *)0), (&i915->drm) ? (&i915-> drm)->dev : ((void *)0), DRM_UT_DRIVER, "DSM size = %lluM\n" , (u64)resource_size(&intel_graphics_stolen_res) >> 20); |
1340 | |
1341 | return 0; |
1342 | } |
1343 | |
1344 | /** |
1345 | * i915_ggtt_probe_hw - Probe GGTT hardware location |
1346 | * @i915: i915 device |
1347 | */ |
1348 | int i915_ggtt_probe_hw(struct drm_i915_privateinteldrm_softc *i915) |
1349 | { |
1350 | int ret; |
1351 | |
1352 | ret = ggtt_probe_hw(to_gt(i915)->ggtt, to_gt(i915)); |
1353 | if (ret) |
1354 | return ret; |
1355 | |
1356 | if (i915_vtd_active(i915)) |
1357 | drm_info(&i915->drm, "VT-d active for gfx access\n")do { } while(0); |
1358 | |
1359 | return 0; |
1360 | } |
1361 | |
1362 | int i915_ggtt_enable_hw(struct drm_i915_privateinteldrm_softc *i915) |
1363 | { |
1364 | if (GRAPHICS_VER(i915)((&(i915)->__runtime)->graphics.ip.ver) < 6) |
1365 | return intel_ggtt_gmch_enable_hw(i915); |
1366 | |
1367 | return 0; |
1368 | } |
1369 | |
1370 | void i915_ggtt_enable_guc(struct i915_ggtt *ggtt) |
1371 | { |
1372 | GEM_BUG_ON(ggtt->invalidate != gen8_ggtt_invalidate)((void)0); |
1373 | |
1374 | ggtt->invalidate = guc_ggtt_invalidate; |
1375 | |
1376 | ggtt->invalidate(ggtt); |
1377 | } |
1378 | |
1379 | void i915_ggtt_disable_guc(struct i915_ggtt *ggtt) |
1380 | { |
1381 | /* XXX Temporary pardon for error unload */ |
1382 | if (ggtt->invalidate == gen8_ggtt_invalidate) |
1383 | return; |
1384 | |
1385 | /* We should only be called after i915_ggtt_enable_guc() */ |
1386 | GEM_BUG_ON(ggtt->invalidate != guc_ggtt_invalidate)((void)0); |
1387 | |
1388 | ggtt->invalidate = gen8_ggtt_invalidate; |
1389 | |
1390 | ggtt->invalidate(ggtt); |
1391 | } |
1392 | |
1393 | /** |
1394 | * i915_ggtt_resume_vm - Restore the memory mappings for a GGTT or DPT VM |
1395 | * @vm: The VM to restore the mappings for |
1396 | * |
1397 | * Restore the memory mappings for all objects mapped to HW via the GGTT or a |
1398 | * DPT page table. |
1399 | * |
1400 | * Returns %true if restoring the mapping for any object that was in a write |
1401 | * domain before suspend. |
1402 | */ |
1403 | bool_Bool i915_ggtt_resume_vm(struct i915_address_space *vm) |
1404 | { |
1405 | struct i915_vma *vma; |
1406 | bool_Bool write_domain_objs = false0; |
1407 | bool_Bool retained_ptes; |
1408 | |
1409 | drm_WARN_ON(&vm->i915->drm, !vm->is_ggtt && !vm->is_dpt)({ int __ret = !!((!vm->is_ggtt && !vm->is_dpt) ); if (__ret) printf("%s %s: " "%s", dev_driver_string(((& vm->i915->drm))->dev), "", "drm_WARN_ON(" "!vm->is_ggtt && !vm->is_dpt" ")"); __builtin_expect(!!(__ret), 0); }); |
1410 | |
1411 | /* |
1412 | * First fill our portion of the GTT with scratch pages if |
1413 | * they were not retained across suspend. |
1414 | */ |
1415 | retained_ptes = suspend_retains_ptes(vm) && |
1416 | !i915_vm_to_ggtt(vm)->pte_lost && |
1417 | !GEM_WARN_ON(i915_vm_to_ggtt(vm)->probed_pte != read_last_pte(vm))({ __builtin_expect(!!(!!(i915_vm_to_ggtt(vm)->probed_pte != read_last_pte(vm))), 0); }); |
1418 | |
1419 | if (!retained_ptes) |
1420 | vm->clear_range(vm, 0, vm->total); |
1421 | |
1422 | /* clflush objects bound into the GGTT and rebind them. */ |
1423 | list_for_each_entry(vma, &vm->bound_list, vm_link)for (vma = ({ const __typeof( ((__typeof(*vma) *)0)->vm_link ) *__mptr = ((&vm->bound_list)->next); (__typeof(* vma) *)( (char *)__mptr - __builtin_offsetof(__typeof(*vma), vm_link ) );}); &vma->vm_link != (&vm->bound_list); vma = ({ const __typeof( ((__typeof(*vma) *)0)->vm_link ) *__mptr = (vma->vm_link.next); (__typeof(*vma) *)( (char *)__mptr - __builtin_offsetof(__typeof(*vma), vm_link) );})) { |
1424 | struct drm_i915_gem_object *obj = vma->obj; |
1425 | unsigned int was_bound = |
1426 | atomic_read(&vma->flags)({ typeof(*(&vma->flags)) __tmp = *(volatile typeof(*( &vma->flags)) *)&(*(&vma->flags)); membar_datadep_consumer (); __tmp; }) & I915_VMA_BIND_MASK(((int)(1UL << (10))) | ((int)(1UL << (11)))); |
1427 | |
1428 | GEM_BUG_ON(!was_bound)((void)0); |
1429 | if (!retained_ptes) { |
1430 | /* |
1431 | * Clear the bound flags of the vma resource to allow |
1432 | * ptes to be repopulated. |
1433 | */ |
1434 | vma->resource->bound_flags = 0; |
1435 | vma->ops->bind_vma(vm, NULL((void *)0), vma->resource, |
1436 | obj ? obj->cache_level : 0, |
1437 | was_bound); |
1438 | } |
1439 | if (obj) { /* only used during resume => exclusive access */ |
1440 | write_domain_objs |= fetch_and_zero(&obj->write_domain)({ typeof(*&obj->write_domain) __T = *(&obj->write_domain ); *(&obj->write_domain) = (typeof(*&obj->write_domain ))0; __T; }); |
1441 | obj->read_domains |= I915_GEM_DOMAIN_GTT0x00000040; |
1442 | } |
1443 | } |
1444 | |
1445 | return write_domain_objs; |
1446 | } |
1447 | |
1448 | void i915_ggtt_resume(struct i915_ggtt *ggtt) |
1449 | { |
1450 | bool_Bool flush; |
1451 | |
1452 | intel_gt_check_and_clear_faults(ggtt->vm.gt); |
1453 | |
1454 | flush = i915_ggtt_resume_vm(&ggtt->vm); |
1455 | |
1456 | ggtt->invalidate(ggtt); |
1457 | |
1458 | if (flush) |
1459 | wbinvd_on_all_cpus(); |
1460 | |
1461 | if (GRAPHICS_VER(ggtt->vm.i915)((&(ggtt->vm.i915)->__runtime)->graphics.ip.ver) >= 8) |
1462 | setup_private_pat(ggtt->vm.gt->uncore); |
1463 | |
1464 | intel_ggtt_restore_fences(ggtt); |
1465 | } |
1466 | |
1467 | void i915_ggtt_mark_pte_lost(struct drm_i915_privateinteldrm_softc *i915, bool_Bool val) |
1468 | { |
1469 | to_gt(i915)->ggtt->pte_lost = val; |
1470 | } |