2 * Copyright 2016 Advanced Micro Devices, Inc.
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:
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
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.
25 #include "amdgpu_ih.h"
28 #include "oss/osssys_4_0_offset.h"
29 #include "oss/osssys_4_0_sh_mask.h"
31 #include "soc15_common.h"
32 #include "vega10_ih.h"
36 static void vega10_ih_set_interrupt_funcs(struct amdgpu_device *adev);
39 * vega10_ih_enable_interrupts - Enable the interrupt ring buffer
41 * @adev: amdgpu_device pointer
43 * Enable the interrupt ring buffer (VEGA10).
45 static void vega10_ih_enable_interrupts(struct amdgpu_device *adev)
47 u32 ih_rb_cntl = RREG32_SOC15(OSSSYS, 0, mmIH_RB_CNTL);
49 ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, RB_ENABLE, 1);
50 ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, ENABLE_INTR, 1);
51 WREG32_SOC15(OSSSYS, 0, mmIH_RB_CNTL, ih_rb_cntl);
52 adev->irq.ih.enabled = true;
56 * vega10_ih_disable_interrupts - Disable the interrupt ring buffer
58 * @adev: amdgpu_device pointer
60 * Disable the interrupt ring buffer (VEGA10).
62 static void vega10_ih_disable_interrupts(struct amdgpu_device *adev)
64 u32 ih_rb_cntl = RREG32_SOC15(OSSSYS, 0, mmIH_RB_CNTL);
66 ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, RB_ENABLE, 0);
67 ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, ENABLE_INTR, 0);
68 WREG32_SOC15(OSSSYS, 0, mmIH_RB_CNTL, ih_rb_cntl);
69 /* set rptr, wptr to 0 */
70 WREG32_SOC15(OSSSYS, 0, mmIH_RB_RPTR, 0);
71 WREG32_SOC15(OSSSYS, 0, mmIH_RB_WPTR, 0);
72 adev->irq.ih.enabled = false;
73 adev->irq.ih.rptr = 0;
77 * vega10_ih_irq_init - init and enable the interrupt ring
79 * @adev: amdgpu_device pointer
81 * Allocate a ring buffer for the interrupt controller,
82 * enable the RLC, disable interrupts, enable the IH
83 * ring buffer and enable it (VI).
84 * Called at device load and reume.
85 * Returns 0 for success, errors for failure.
87 static int vega10_ih_irq_init(struct amdgpu_device *adev)
91 u32 ih_rb_cntl, ih_doorbell_rtpr;
96 vega10_ih_disable_interrupts(adev);
98 adev->nbio_funcs->ih_control(adev);
100 ih_rb_cntl = RREG32_SOC15(OSSSYS, 0, mmIH_RB_CNTL);
101 /* Ring Buffer base. [39:8] of 40-bit address of the beginning of the ring buffer*/
102 if (adev->irq.ih.use_bus_addr) {
103 WREG32_SOC15(OSSSYS, 0, mmIH_RB_BASE, adev->irq.ih.rb_dma_addr >> 8);
104 WREG32_SOC15(OSSSYS, 0, mmIH_RB_BASE_HI, ((u64)adev->irq.ih.rb_dma_addr >> 40) & 0xff);
105 ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, MC_SPACE, 1);
107 WREG32_SOC15(OSSSYS, 0, mmIH_RB_BASE, adev->irq.ih.gpu_addr >> 8);
108 WREG32_SOC15(OSSSYS, 0, mmIH_RB_BASE_HI, (adev->irq.ih.gpu_addr >> 40) & 0xff);
109 ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, MC_SPACE, 4);
111 rb_bufsz = order_base_2(adev->irq.ih.ring_size / 4);
112 ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, WPTR_OVERFLOW_CLEAR, 1);
113 ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, WPTR_OVERFLOW_ENABLE, 1);
114 ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, RB_SIZE, rb_bufsz);
115 /* Ring Buffer write pointer writeback. If enabled, IH_RB_WPTR register value is written to memory */
116 ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, WPTR_WRITEBACK_ENABLE, 1);
117 ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, MC_SNOOP, 1);
118 ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, MC_RO, 0);
119 ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, MC_VMID, 0);
121 if (adev->irq.msi_enabled)
122 ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, RPTR_REARM, 1);
124 WREG32_SOC15(OSSSYS, 0, mmIH_RB_CNTL, ih_rb_cntl);
126 /* set the writeback address whether it's enabled or not */
127 if (adev->irq.ih.use_bus_addr)
128 wptr_off = adev->irq.ih.rb_dma_addr + (adev->irq.ih.wptr_offs * 4);
130 wptr_off = adev->wb.gpu_addr + (adev->irq.ih.wptr_offs * 4);
131 WREG32_SOC15(OSSSYS, 0, mmIH_RB_WPTR_ADDR_LO, lower_32_bits(wptr_off));
132 WREG32_SOC15(OSSSYS, 0, mmIH_RB_WPTR_ADDR_HI, upper_32_bits(wptr_off) & 0xFF);
134 /* set rptr, wptr to 0 */
135 WREG32_SOC15(OSSSYS, 0, mmIH_RB_RPTR, 0);
136 WREG32_SOC15(OSSSYS, 0, mmIH_RB_WPTR, 0);
138 ih_doorbell_rtpr = RREG32_SOC15(OSSSYS, 0, mmIH_DOORBELL_RPTR);
139 if (adev->irq.ih.use_doorbell) {
140 ih_doorbell_rtpr = REG_SET_FIELD(ih_doorbell_rtpr, IH_DOORBELL_RPTR,
141 OFFSET, adev->irq.ih.doorbell_index);
142 ih_doorbell_rtpr = REG_SET_FIELD(ih_doorbell_rtpr, IH_DOORBELL_RPTR,
145 ih_doorbell_rtpr = REG_SET_FIELD(ih_doorbell_rtpr, IH_DOORBELL_RPTR,
148 WREG32_SOC15(OSSSYS, 0, mmIH_DOORBELL_RPTR, ih_doorbell_rtpr);
149 adev->nbio_funcs->ih_doorbell_range(adev, adev->irq.ih.use_doorbell,
150 adev->irq.ih.doorbell_index);
152 tmp = RREG32_SOC15(OSSSYS, 0, mmIH_STORM_CLIENT_LIST_CNTL);
153 tmp = REG_SET_FIELD(tmp, IH_STORM_CLIENT_LIST_CNTL,
154 CLIENT18_IS_STORM_CLIENT, 1);
155 WREG32_SOC15(OSSSYS, 0, mmIH_STORM_CLIENT_LIST_CNTL, tmp);
157 tmp = RREG32_SOC15(OSSSYS, 0, mmIH_INT_FLOOD_CNTL);
158 tmp = REG_SET_FIELD(tmp, IH_INT_FLOOD_CNTL, FLOOD_CNTL_ENABLE, 1);
159 WREG32_SOC15(OSSSYS, 0, mmIH_INT_FLOOD_CNTL, tmp);
161 pci_set_master(adev->pdev);
163 /* enable interrupts */
164 vega10_ih_enable_interrupts(adev);
170 * vega10_ih_irq_disable - disable interrupts
172 * @adev: amdgpu_device pointer
174 * Disable interrupts on the hw (VEGA10).
176 static void vega10_ih_irq_disable(struct amdgpu_device *adev)
178 vega10_ih_disable_interrupts(adev);
180 /* Wait and acknowledge irq */
185 * vega10_ih_get_wptr - get the IH ring buffer wptr
187 * @adev: amdgpu_device pointer
189 * Get the IH ring buffer wptr from either the register
190 * or the writeback memory buffer (VEGA10). Also check for
191 * ring buffer overflow and deal with it.
192 * Returns the value of the wptr.
194 static u32 vega10_ih_get_wptr(struct amdgpu_device *adev)
198 if (adev->irq.ih.use_bus_addr)
199 wptr = le32_to_cpu(adev->irq.ih.ring[adev->irq.ih.wptr_offs]);
201 wptr = le32_to_cpu(adev->wb.wb[adev->irq.ih.wptr_offs]);
203 if (REG_GET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW)) {
204 wptr = REG_SET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW, 0);
206 /* When a ring buffer overflow happen start parsing interrupt
207 * from the last not overwritten vector (wptr + 32). Hopefully
208 * this should allow us to catchup.
210 tmp = (wptr + 32) & adev->irq.ih.ptr_mask;
211 dev_warn(adev->dev, "IH ring buffer overflow (0x%08X, 0x%08X, 0x%08X)\n",
212 wptr, adev->irq.ih.rptr, tmp);
213 adev->irq.ih.rptr = tmp;
215 tmp = RREG32_NO_KIQ(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_CNTL));
216 tmp = REG_SET_FIELD(tmp, IH_RB_CNTL, WPTR_OVERFLOW_CLEAR, 1);
217 WREG32_NO_KIQ(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_CNTL), tmp);
219 return (wptr & adev->irq.ih.ptr_mask);
223 * vega10_ih_prescreen_iv - prescreen an interrupt vector
225 * @adev: amdgpu_device pointer
227 * Returns true if the interrupt vector should be further processed.
229 static bool vega10_ih_prescreen_iv(struct amdgpu_device *adev)
231 u32 ring_index = adev->irq.ih.rptr >> 2;
232 u32 dw0, dw3, dw4, dw5;
235 struct amdgpu_vm *vm;
238 dw0 = le32_to_cpu(adev->irq.ih.ring[ring_index + 0]);
239 dw3 = le32_to_cpu(adev->irq.ih.ring[ring_index + 3]);
240 dw4 = le32_to_cpu(adev->irq.ih.ring[ring_index + 4]);
241 dw5 = le32_to_cpu(adev->irq.ih.ring[ring_index + 5]);
243 /* Filter retry page faults, let only the first one pass. If
244 * there are too many outstanding faults, ignore them until
245 * some faults get cleared.
247 switch (dw0 & 0xff) {
248 case AMDGPU_IH_CLIENTID_VMC:
249 case AMDGPU_IH_CLIENTID_UTCL2:
256 pasid = dw3 & 0xffff;
257 /* No PASID, can't identify faulting process */
261 /* Not a retry fault, check fault credit */
263 if (!amdgpu_vm_pasid_fault_credit(adev, pasid))
268 addr = ((u64)(dw5 & 0xf) << 44) | ((u64)dw4 << 12);
269 key = AMDGPU_VM_FAULT(pasid, addr);
270 r = amdgpu_ih_add_fault(adev, key);
272 /* Hash table is full or the fault is already being processed,
273 * ignore further page faults
278 /* Track retry faults in per-VM fault FIFO. */
279 spin_lock(&adev->vm_manager.pasid_lock);
280 vm = idr_find(&adev->vm_manager.pasid_idr, pasid);
281 spin_unlock(&adev->vm_manager.pasid_lock);
282 if (WARN_ON_ONCE(!vm)) {
283 /* VM not found, process it normally */
284 amdgpu_ih_clear_fault(adev, key);
287 /* No locking required with single writer and single reader */
288 r = kfifo_put(&vm->faults, key);
290 /* FIFO is full. Ignore it until there is space */
291 amdgpu_ih_clear_fault(adev, key);
295 /* It's the first fault for this address, process it normally */
299 adev->irq.ih.rptr += 32;
304 * vega10_ih_decode_iv - decode an interrupt vector
306 * @adev: amdgpu_device pointer
308 * Decodes the interrupt vector at the current rptr
309 * position and also advance the position.
311 static void vega10_ih_decode_iv(struct amdgpu_device *adev,
312 struct amdgpu_iv_entry *entry)
314 /* wptr/rptr are in bytes! */
315 u32 ring_index = adev->irq.ih.rptr >> 2;
318 dw[0] = le32_to_cpu(adev->irq.ih.ring[ring_index + 0]);
319 dw[1] = le32_to_cpu(adev->irq.ih.ring[ring_index + 1]);
320 dw[2] = le32_to_cpu(adev->irq.ih.ring[ring_index + 2]);
321 dw[3] = le32_to_cpu(adev->irq.ih.ring[ring_index + 3]);
322 dw[4] = le32_to_cpu(adev->irq.ih.ring[ring_index + 4]);
323 dw[5] = le32_to_cpu(adev->irq.ih.ring[ring_index + 5]);
324 dw[6] = le32_to_cpu(adev->irq.ih.ring[ring_index + 6]);
325 dw[7] = le32_to_cpu(adev->irq.ih.ring[ring_index + 7]);
327 entry->client_id = dw[0] & 0xff;
328 entry->src_id = (dw[0] >> 8) & 0xff;
329 entry->ring_id = (dw[0] >> 16) & 0xff;
330 entry->vmid = (dw[0] >> 24) & 0xf;
331 entry->vmid_src = (dw[0] >> 31);
332 entry->timestamp = dw[1] | ((u64)(dw[2] & 0xffff) << 32);
333 entry->timestamp_src = dw[2] >> 31;
334 entry->pas_id = dw[3] & 0xffff;
335 entry->pasid_src = dw[3] >> 31;
336 entry->src_data[0] = dw[4];
337 entry->src_data[1] = dw[5];
338 entry->src_data[2] = dw[6];
339 entry->src_data[3] = dw[7];
342 /* wptr/rptr are in bytes! */
343 adev->irq.ih.rptr += 32;
347 * vega10_ih_set_rptr - set the IH ring buffer rptr
349 * @adev: amdgpu_device pointer
351 * Set the IH ring buffer rptr.
353 static void vega10_ih_set_rptr(struct amdgpu_device *adev)
355 if (adev->irq.ih.use_doorbell) {
356 /* XXX check if swapping is necessary on BE */
357 if (adev->irq.ih.use_bus_addr)
358 adev->irq.ih.ring[adev->irq.ih.rptr_offs] = adev->irq.ih.rptr;
360 adev->wb.wb[adev->irq.ih.rptr_offs] = adev->irq.ih.rptr;
361 WDOORBELL32(adev->irq.ih.doorbell_index, adev->irq.ih.rptr);
363 WREG32_SOC15(OSSSYS, 0, mmIH_RB_RPTR, adev->irq.ih.rptr);
367 static int vega10_ih_early_init(void *handle)
369 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
371 vega10_ih_set_interrupt_funcs(adev);
375 static int vega10_ih_sw_init(void *handle)
378 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
380 r = amdgpu_ih_ring_init(adev, 256 * 1024, true);
384 adev->irq.ih.use_doorbell = true;
385 adev->irq.ih.doorbell_index = AMDGPU_DOORBELL64_IH << 1;
387 adev->irq.ih.faults = kmalloc(sizeof(*adev->irq.ih.faults), GFP_KERNEL);
388 if (!adev->irq.ih.faults)
390 INIT_CHASH_TABLE(adev->irq.ih.faults->hash,
391 AMDGPU_PAGEFAULT_HASH_BITS, 8, 0);
392 spin_lock_init(&adev->irq.ih.faults->lock);
393 adev->irq.ih.faults->count = 0;
395 r = amdgpu_irq_init(adev);
400 static int vega10_ih_sw_fini(void *handle)
402 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
404 amdgpu_irq_fini(adev);
405 amdgpu_ih_ring_fini(adev);
407 kfree(adev->irq.ih.faults);
408 adev->irq.ih.faults = NULL;
413 static int vega10_ih_hw_init(void *handle)
416 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
418 r = vega10_ih_irq_init(adev);
425 static int vega10_ih_hw_fini(void *handle)
427 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
429 vega10_ih_irq_disable(adev);
434 static int vega10_ih_suspend(void *handle)
436 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
438 return vega10_ih_hw_fini(adev);
441 static int vega10_ih_resume(void *handle)
443 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
445 return vega10_ih_hw_init(adev);
448 static bool vega10_ih_is_idle(void *handle)
454 static int vega10_ih_wait_for_idle(void *handle)
460 static int vega10_ih_soft_reset(void *handle)
467 static int vega10_ih_set_clockgating_state(void *handle,
468 enum amd_clockgating_state state)
473 static int vega10_ih_set_powergating_state(void *handle,
474 enum amd_powergating_state state)
479 const struct amd_ip_funcs vega10_ih_ip_funcs = {
481 .early_init = vega10_ih_early_init,
483 .sw_init = vega10_ih_sw_init,
484 .sw_fini = vega10_ih_sw_fini,
485 .hw_init = vega10_ih_hw_init,
486 .hw_fini = vega10_ih_hw_fini,
487 .suspend = vega10_ih_suspend,
488 .resume = vega10_ih_resume,
489 .is_idle = vega10_ih_is_idle,
490 .wait_for_idle = vega10_ih_wait_for_idle,
491 .soft_reset = vega10_ih_soft_reset,
492 .set_clockgating_state = vega10_ih_set_clockgating_state,
493 .set_powergating_state = vega10_ih_set_powergating_state,
496 static const struct amdgpu_ih_funcs vega10_ih_funcs = {
497 .get_wptr = vega10_ih_get_wptr,
498 .prescreen_iv = vega10_ih_prescreen_iv,
499 .decode_iv = vega10_ih_decode_iv,
500 .set_rptr = vega10_ih_set_rptr
503 static void vega10_ih_set_interrupt_funcs(struct amdgpu_device *adev)
505 if (adev->irq.ih_funcs == NULL)
506 adev->irq.ih_funcs = &vega10_ih_funcs;
509 const struct amdgpu_ip_block_version vega10_ih_ip_block =
511 .type = AMD_IP_BLOCK_TYPE_IH,
515 .funcs = &vega10_ih_ip_funcs,