xref: /linux/drivers/gpu/drm/nouveau/nvkm/subdev/fault/gv100.c (revision 7ae9fb1b7ecbb5d85d07857943f677fd1a559b18)
1 /*
2  * Copyright 2018 Red Hat 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 #include "priv.h"
23 
24 #include <core/memory.h>
25 #include <subdev/mmu.h>
26 #include <engine/fifo.h>
27 
28 #include <nvif/class.h>
29 
30 void
gv100_fault_buffer_process(struct work_struct * work)31 gv100_fault_buffer_process(struct work_struct *work)
32 {
33 	struct nvkm_fault *fault = container_of(work, typeof(*fault), nrpfb_work);
34 	struct nvkm_fault_buffer *buffer = fault->buffer[0];
35 	struct nvkm_device *device = fault->subdev.device;
36 	struct nvkm_memory *mem = buffer->mem;
37 	u32 get = nvkm_rd32(device, buffer->get);
38 	u32 put = nvkm_rd32(device, buffer->put);
39 	if (put == get)
40 		return;
41 
42 	nvkm_kmap(mem);
43 	while (get != put) {
44 		const u32   base = get * buffer->fault->func->buffer.entry_size;
45 		const u32 instlo = nvkm_ro32(mem, base + 0x00);
46 		const u32 insthi = nvkm_ro32(mem, base + 0x04);
47 		const u32 addrlo = nvkm_ro32(mem, base + 0x08);
48 		const u32 addrhi = nvkm_ro32(mem, base + 0x0c);
49 		const u32 timelo = nvkm_ro32(mem, base + 0x10);
50 		const u32 timehi = nvkm_ro32(mem, base + 0x14);
51 		const u32  info0 = nvkm_ro32(mem, base + 0x18);
52 		const u32  info1 = nvkm_ro32(mem, base + 0x1c);
53 		struct nvkm_fault_data info;
54 
55 		if (++get == buffer->entries)
56 			get = 0;
57 		nvkm_wr32(device, buffer->get, get);
58 
59 		info.addr   = ((u64)addrhi << 32) | addrlo;
60 		info.inst   = ((u64)insthi << 32) | instlo;
61 		info.time   = ((u64)timehi << 32) | timelo;
62 		info.engine = (info0 & 0x000000ff);
63 		info.valid  = (info1 & 0x80000000) >> 31;
64 		info.gpc    = (info1 & 0x1f000000) >> 24;
65 		info.hub    = (info1 & 0x00100000) >> 20;
66 		info.access = (info1 & 0x000f0000) >> 16;
67 		info.client = (info1 & 0x00007f00) >> 8;
68 		info.reason = (info1 & 0x0000001f);
69 
70 		nvkm_fifo_fault(device->fifo, &info);
71 	}
72 	nvkm_done(mem);
73 }
74 
75 static void
gv100_fault_buffer_intr(struct nvkm_fault_buffer * buffer,bool enable)76 gv100_fault_buffer_intr(struct nvkm_fault_buffer *buffer, bool enable)
77 {
78 	struct nvkm_device *device = buffer->fault->subdev.device;
79 	const u32 intr = buffer->id ? 0x08000000 : 0x20000000;
80 	if (enable)
81 		nvkm_mask(device, 0x100a2c, intr, intr);
82 	else
83 		nvkm_mask(device, 0x100a34, intr, intr);
84 }
85 
86 static void
gv100_fault_buffer_fini(struct nvkm_fault_buffer * buffer)87 gv100_fault_buffer_fini(struct nvkm_fault_buffer *buffer)
88 {
89 	struct nvkm_device *device = buffer->fault->subdev.device;
90 	const u32 foff = buffer->id * 0x14;
91 	nvkm_mask(device, 0x100e34 + foff, 0x80000000, 0x00000000);
92 }
93 
94 static void
gv100_fault_buffer_init(struct nvkm_fault_buffer * buffer)95 gv100_fault_buffer_init(struct nvkm_fault_buffer *buffer)
96 {
97 	struct nvkm_device *device = buffer->fault->subdev.device;
98 	const u32 foff = buffer->id * 0x14;
99 
100 	nvkm_mask(device, 0x100e34 + foff, 0xc0000000, 0x40000000);
101 	nvkm_wr32(device, 0x100e28 + foff, upper_32_bits(buffer->addr));
102 	nvkm_wr32(device, 0x100e24 + foff, lower_32_bits(buffer->addr));
103 	nvkm_mask(device, 0x100e34 + foff, 0x80000000, 0x80000000);
104 }
105 
106 static void
gv100_fault_buffer_info(struct nvkm_fault_buffer * buffer)107 gv100_fault_buffer_info(struct nvkm_fault_buffer *buffer)
108 {
109 	struct nvkm_device *device = buffer->fault->subdev.device;
110 	const u32 foff = buffer->id * 0x14;
111 
112 	nvkm_mask(device, 0x100e34 + foff, 0x40000000, 0x40000000);
113 
114 	buffer->entries = nvkm_rd32(device, 0x100e34 + foff) & 0x000fffff;
115 	buffer->get = 0x100e2c + foff;
116 	buffer->put = 0x100e30 + foff;
117 }
118 
119 static int
gv100_fault_ntfy_nrpfb(struct nvkm_event_ntfy * ntfy,u32 bits)120 gv100_fault_ntfy_nrpfb(struct nvkm_event_ntfy *ntfy, u32 bits)
121 {
122 	struct nvkm_fault *fault = container_of(ntfy, typeof(*fault), nrpfb);
123 
124 	schedule_work(&fault->nrpfb_work);
125 	return NVKM_EVENT_KEEP;
126 }
127 
128 static void
gv100_fault_intr_fault(struct nvkm_fault * fault)129 gv100_fault_intr_fault(struct nvkm_fault *fault)
130 {
131 	struct nvkm_subdev *subdev = &fault->subdev;
132 	struct nvkm_device *device = subdev->device;
133 	struct nvkm_fault_data info;
134 	const u32 addrlo = nvkm_rd32(device, 0x100e4c);
135 	const u32 addrhi = nvkm_rd32(device, 0x100e50);
136 	const u32  info0 = nvkm_rd32(device, 0x100e54);
137 	const u32 insthi = nvkm_rd32(device, 0x100e58);
138 	const u32  info1 = nvkm_rd32(device, 0x100e5c);
139 
140 	info.addr = ((u64)addrhi << 32) | addrlo;
141 	info.inst = ((u64)insthi << 32) | (info0 & 0xfffff000);
142 	info.time = 0;
143 	info.engine = (info0 & 0x000000ff);
144 	info.valid  = (info1 & 0x80000000) >> 31;
145 	info.gpc    = (info1 & 0x1f000000) >> 24;
146 	info.hub    = (info1 & 0x00100000) >> 20;
147 	info.access = (info1 & 0x000f0000) >> 16;
148 	info.client = (info1 & 0x00007f00) >> 8;
149 	info.reason = (info1 & 0x0000001f);
150 
151 	nvkm_fifo_fault(device->fifo, &info);
152 }
153 
154 static void
gv100_fault_intr(struct nvkm_fault * fault)155 gv100_fault_intr(struct nvkm_fault *fault)
156 {
157 	struct nvkm_subdev *subdev = &fault->subdev;
158 	struct nvkm_device *device = subdev->device;
159 	u32 stat = nvkm_rd32(device, 0x100a20);
160 
161 	if (stat & 0x80000000) {
162 		gv100_fault_intr_fault(fault);
163 		nvkm_wr32(device, 0x100e60, 0x80000000);
164 		stat &= ~0x80000000;
165 	}
166 
167 	if (stat & 0x20000000) {
168 		if (fault->buffer[0]) {
169 			nvkm_event_ntfy(&fault->event, 0, NVKM_FAULT_BUFFER_EVENT_PENDING);
170 			stat &= ~0x20000000;
171 		}
172 	}
173 
174 	if (stat & 0x08000000) {
175 		if (fault->buffer[1]) {
176 			nvkm_event_ntfy(&fault->event, 1, NVKM_FAULT_BUFFER_EVENT_PENDING);
177 			stat &= ~0x08000000;
178 		}
179 	}
180 
181 	if (stat) {
182 		nvkm_debug(subdev, "intr %08x\n", stat);
183 	}
184 }
185 
186 static void
gv100_fault_fini(struct nvkm_fault * fault)187 gv100_fault_fini(struct nvkm_fault *fault)
188 {
189 	nvkm_event_ntfy_block(&fault->nrpfb);
190 	flush_work(&fault->nrpfb_work);
191 
192 	if (fault->buffer[0])
193 		fault->func->buffer.fini(fault->buffer[0]);
194 
195 	nvkm_mask(fault->subdev.device, 0x100a34, 0x80000000, 0x80000000);
196 }
197 
198 static void
gv100_fault_init(struct nvkm_fault * fault)199 gv100_fault_init(struct nvkm_fault *fault)
200 {
201 	nvkm_mask(fault->subdev.device, 0x100a2c, 0x80000000, 0x80000000);
202 	fault->func->buffer.init(fault->buffer[0]);
203 	nvkm_event_ntfy_allow(&fault->nrpfb);
204 }
205 
206 int
gv100_fault_oneinit(struct nvkm_fault * fault)207 gv100_fault_oneinit(struct nvkm_fault *fault)
208 {
209 	nvkm_event_ntfy_add(&fault->event, 0, NVKM_FAULT_BUFFER_EVENT_PENDING, true,
210 			    gv100_fault_ntfy_nrpfb, &fault->nrpfb);
211 	return 0;
212 }
213 
214 static const struct nvkm_fault_func
215 gv100_fault = {
216 	.oneinit = gv100_fault_oneinit,
217 	.init = gv100_fault_init,
218 	.fini = gv100_fault_fini,
219 	.intr = gv100_fault_intr,
220 	.buffer.nr = 2,
221 	.buffer.entry_size = 32,
222 	.buffer.info = gv100_fault_buffer_info,
223 	.buffer.pin = gp100_fault_buffer_pin,
224 	.buffer.init = gv100_fault_buffer_init,
225 	.buffer.fini = gv100_fault_buffer_fini,
226 	.buffer.intr = gv100_fault_buffer_intr,
227 	/*TODO: Figure out how to expose non-replayable fault buffer, which,
228 	 *      for some reason, is where recoverable CE faults appear...
229 	 *
230 	 * 	It's a bit tricky, as both NVKM and SVM will need access to
231 	 * 	the non-replayable fault buffer.
232 	 */
233 	.user = { { 0, 0, VOLTA_FAULT_BUFFER_A }, 1 },
234 };
235 
236 int
gv100_fault_new(struct nvkm_device * device,enum nvkm_subdev_type type,int inst,struct nvkm_fault ** pfault)237 gv100_fault_new(struct nvkm_device *device, enum nvkm_subdev_type type, int inst,
238 		struct nvkm_fault **pfault)
239 {
240 	int ret = nvkm_fault_new_(&gv100_fault, device, type, inst, pfault);
241 	if (ret)
242 		return ret;
243 
244 	INIT_WORK(&(*pfault)->nrpfb_work, gv100_fault_buffer_process);
245 	return 0;
246 }
247