1 /*
2  * Copyright 2014 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  */
23 
24 #include <drm/drmP.h>
25 #include "amdgpu.h"
26 #include "amdgpu_ih.h"
27 #include "amdgpu_amdkfd.h"
28 
29 /**
30  * amdgpu_ih_ring_alloc - allocate memory for the IH ring
31  *
32  * @adev: amdgpu_device pointer
33  *
34  * Allocate a ring buffer for the interrupt controller.
35  * Returns 0 for success, errors for failure.
36  */
37 static int amdgpu_ih_ring_alloc(struct amdgpu_device *adev)
38 {
39 	int r;
40 
41 	/* Allocate ring buffer */
42 	if (adev->irq.ih.ring_obj == NULL) {
43 		r = amdgpu_bo_create(adev, adev->irq.ih.ring_size,
44 				     PAGE_SIZE, true,
45 				     AMDGPU_GEM_DOMAIN_GTT, 0,
46 				     NULL, &adev->irq.ih.ring_obj);
47 		if (r) {
48 			DRM_ERROR("amdgpu: failed to create ih ring buffer (%d).\n", r);
49 			return r;
50 		}
51 		r = amdgpu_bo_reserve(adev->irq.ih.ring_obj, false);
52 		if (unlikely(r != 0))
53 			return r;
54 		r = amdgpu_bo_pin(adev->irq.ih.ring_obj,
55 				  AMDGPU_GEM_DOMAIN_GTT,
56 				  &adev->irq.ih.gpu_addr);
57 		if (r) {
58 			amdgpu_bo_unreserve(adev->irq.ih.ring_obj);
59 			DRM_ERROR("amdgpu: failed to pin ih ring buffer (%d).\n", r);
60 			return r;
61 		}
62 		r = amdgpu_bo_kmap(adev->irq.ih.ring_obj,
63 				   (void **)&adev->irq.ih.ring);
64 		amdgpu_bo_unreserve(adev->irq.ih.ring_obj);
65 		if (r) {
66 			DRM_ERROR("amdgpu: failed to map ih ring buffer (%d).\n", r);
67 			return r;
68 		}
69 	}
70 	return 0;
71 }
72 
73 /**
74  * amdgpu_ih_ring_init - initialize the IH state
75  *
76  * @adev: amdgpu_device pointer
77  *
78  * Initializes the IH state and allocates a buffer
79  * for the IH ring buffer.
80  * Returns 0 for success, errors for failure.
81  */
82 int amdgpu_ih_ring_init(struct amdgpu_device *adev, unsigned ring_size,
83 			bool use_bus_addr)
84 {
85 	u32 rb_bufsz;
86 	int r;
87 
88 	/* Align ring size */
89 	rb_bufsz = order_base_2(ring_size / 4);
90 	ring_size = (1 << rb_bufsz) * 4;
91 	adev->irq.ih.ring_size = ring_size;
92 	adev->irq.ih.ptr_mask = adev->irq.ih.ring_size - 1;
93 	adev->irq.ih.rptr = 0;
94 	adev->irq.ih.use_bus_addr = use_bus_addr;
95 
96 	if (adev->irq.ih.use_bus_addr) {
97 		if (!adev->irq.ih.ring) {
98 			/* add 8 bytes for the rptr/wptr shadows and
99 			 * add them to the end of the ring allocation.
100 			 */
101 			adev->irq.ih.ring = kzalloc(adev->irq.ih.ring_size + 8, GFP_KERNEL);
102 			if (adev->irq.ih.ring == NULL)
103 				return -ENOMEM;
104 			adev->irq.ih.rb_dma_addr = pci_map_single(adev->pdev,
105 								  (void *)adev->irq.ih.ring,
106 								  adev->irq.ih.ring_size,
107 								  PCI_DMA_BIDIRECTIONAL);
108 			if (pci_dma_mapping_error(adev->pdev, adev->irq.ih.rb_dma_addr)) {
109 				dev_err(&adev->pdev->dev, "Failed to DMA MAP the IH RB page\n");
110 				kfree((void *)adev->irq.ih.ring);
111 				return -ENOMEM;
112 			}
113 			adev->irq.ih.wptr_offs = (adev->irq.ih.ring_size / 4) + 0;
114 			adev->irq.ih.rptr_offs = (adev->irq.ih.ring_size / 4) + 1;
115 		}
116 		return 0;
117 	} else {
118 		r = amdgpu_wb_get(adev, &adev->irq.ih.wptr_offs);
119 		if (r) {
120 			dev_err(adev->dev, "(%d) ih wptr_offs wb alloc failed\n", r);
121 			return r;
122 		}
123 
124 		r = amdgpu_wb_get(adev, &adev->irq.ih.rptr_offs);
125 		if (r) {
126 			amdgpu_wb_free(adev, adev->irq.ih.wptr_offs);
127 			dev_err(adev->dev, "(%d) ih rptr_offs wb alloc failed\n", r);
128 			return r;
129 		}
130 
131 		return amdgpu_ih_ring_alloc(adev);
132 	}
133 }
134 
135 /**
136  * amdgpu_ih_ring_fini - tear down the IH state
137  *
138  * @adev: amdgpu_device pointer
139  *
140  * Tears down the IH state and frees buffer
141  * used for the IH ring buffer.
142  */
143 void amdgpu_ih_ring_fini(struct amdgpu_device *adev)
144 {
145 	int r;
146 
147 	if (adev->irq.ih.use_bus_addr) {
148 		if (adev->irq.ih.ring) {
149 			/* add 8 bytes for the rptr/wptr shadows and
150 			 * add them to the end of the ring allocation.
151 			 */
152 			pci_unmap_single(adev->pdev, adev->irq.ih.rb_dma_addr,
153 					 adev->irq.ih.ring_size + 8, PCI_DMA_BIDIRECTIONAL);
154 			kfree((void *)adev->irq.ih.ring);
155 			adev->irq.ih.ring = NULL;
156 		}
157 	} else {
158 		if (adev->irq.ih.ring_obj) {
159 			r = amdgpu_bo_reserve(adev->irq.ih.ring_obj, false);
160 			if (likely(r == 0)) {
161 				amdgpu_bo_kunmap(adev->irq.ih.ring_obj);
162 				amdgpu_bo_unpin(adev->irq.ih.ring_obj);
163 				amdgpu_bo_unreserve(adev->irq.ih.ring_obj);
164 			}
165 			amdgpu_bo_unref(&adev->irq.ih.ring_obj);
166 			adev->irq.ih.ring = NULL;
167 			adev->irq.ih.ring_obj = NULL;
168 		}
169 		amdgpu_wb_free(adev, adev->irq.ih.wptr_offs);
170 		amdgpu_wb_free(adev, adev->irq.ih.rptr_offs);
171 	}
172 }
173 
174 /**
175  * amdgpu_ih_process - interrupt handler
176  *
177  * @adev: amdgpu_device pointer
178  *
179  * Interrupt hander (VI), walk the IH ring.
180  * Returns irq process return code.
181  */
182 int amdgpu_ih_process(struct amdgpu_device *adev)
183 {
184 	struct amdgpu_iv_entry entry;
185 	u32 wptr;
186 
187 	if (!adev->irq.ih.enabled || adev->shutdown)
188 		return IRQ_NONE;
189 
190 	wptr = amdgpu_ih_get_wptr(adev);
191 
192 restart_ih:
193 	/* is somebody else already processing irqs? */
194 	if (atomic_xchg(&adev->irq.ih.lock, 1))
195 		return IRQ_NONE;
196 
197 	DRM_DEBUG("%s: rptr %d, wptr %d\n", __func__, adev->irq.ih.rptr, wptr);
198 
199 	/* Order reading of wptr vs. reading of IH ring data */
200 	rmb();
201 
202 	while (adev->irq.ih.rptr != wptr) {
203 		u32 ring_index = adev->irq.ih.rptr >> 2;
204 
205 		/* Before dispatching irq to IP blocks, send it to amdkfd */
206 		amdgpu_amdkfd_interrupt(adev,
207 				(const void *) &adev->irq.ih.ring[ring_index]);
208 
209 		amdgpu_ih_decode_iv(adev, &entry);
210 		adev->irq.ih.rptr &= adev->irq.ih.ptr_mask;
211 
212 		amdgpu_irq_dispatch(adev, &entry);
213 	}
214 	amdgpu_ih_set_rptr(adev);
215 	atomic_set(&adev->irq.ih.lock, 0);
216 
217 	/* make sure wptr hasn't changed while processing */
218 	wptr = amdgpu_ih_get_wptr(adev);
219 	if (wptr != adev->irq.ih.rptr)
220 		goto restart_ih;
221 
222 	return IRQ_HANDLED;
223 }
224