diff options
author | Christian König <christian.koenig@amd.com> | 2018-09-16 14:13:21 -0400 |
---|---|---|
committer | Alex Deucher <alexander.deucher@amd.com> | 2018-09-26 22:09:21 -0400 |
commit | 425c31437f26436e17bdb3041a26e9864d18ba13 (patch) | |
tree | e58d951076960cd2ab75d9fe9fbf1270c1d604e3 /drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c | |
parent | f54b30d70bc606f7a154edba5883c7fa23838e9f (diff) |
drm/amdgpu: cleanup amdgpu_ih.c
Cleanup amdgpu_ih.c to be able to handle multiple interrupt rings.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Huang Rui <ray.huang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c')
-rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c | 152 |
1 files changed, 67 insertions, 85 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c index 4ed86218cef3..15fb0f9738ab 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c | |||
@@ -27,43 +27,19 @@ | |||
27 | #include "amdgpu_amdkfd.h" | 27 | #include "amdgpu_amdkfd.h" |
28 | 28 | ||
29 | /** | 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_kernel(adev, adev->irq.ih.ring_size, | ||
44 | PAGE_SIZE, AMDGPU_GEM_DOMAIN_GTT, | ||
45 | &adev->irq.ih.ring_obj, | ||
46 | &adev->irq.ih.gpu_addr, | ||
47 | (void **)&adev->irq.ih.ring); | ||
48 | if (r) { | ||
49 | DRM_ERROR("amdgpu: failed to create ih ring buffer (%d).\n", r); | ||
50 | return r; | ||
51 | } | ||
52 | } | ||
53 | return 0; | ||
54 | } | ||
55 | |||
56 | /** | ||
57 | * amdgpu_ih_ring_init - initialize the IH state | 30 | * amdgpu_ih_ring_init - initialize the IH state |
58 | * | 31 | * |
59 | * @adev: amdgpu_device pointer | 32 | * @adev: amdgpu_device pointer |
33 | * @ih: ih ring to initialize | ||
34 | * @ring_size: ring size to allocate | ||
35 | * @use_bus_addr: true when we can use dma_alloc_coherent | ||
60 | * | 36 | * |
61 | * Initializes the IH state and allocates a buffer | 37 | * Initializes the IH state and allocates a buffer |
62 | * for the IH ring buffer. | 38 | * for the IH ring buffer. |
63 | * Returns 0 for success, errors for failure. | 39 | * Returns 0 for success, errors for failure. |
64 | */ | 40 | */ |
65 | int amdgpu_ih_ring_init(struct amdgpu_device *adev, unsigned ring_size, | 41 | int amdgpu_ih_ring_init(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih, |
66 | bool use_bus_addr) | 42 | unsigned ring_size, bool use_bus_addr) |
67 | { | 43 | { |
68 | u32 rb_bufsz; | 44 | u32 rb_bufsz; |
69 | int r; | 45 | int r; |
@@ -71,70 +47,76 @@ int amdgpu_ih_ring_init(struct amdgpu_device *adev, unsigned ring_size, | |||
71 | /* Align ring size */ | 47 | /* Align ring size */ |
72 | rb_bufsz = order_base_2(ring_size / 4); | 48 | rb_bufsz = order_base_2(ring_size / 4); |
73 | ring_size = (1 << rb_bufsz) * 4; | 49 | ring_size = (1 << rb_bufsz) * 4; |
74 | adev->irq.ih.ring_size = ring_size; | 50 | ih->ring_size = ring_size; |
75 | adev->irq.ih.ptr_mask = adev->irq.ih.ring_size - 1; | 51 | ih->ptr_mask = ih->ring_size - 1; |
76 | adev->irq.ih.rptr = 0; | 52 | ih->rptr = 0; |
77 | adev->irq.ih.use_bus_addr = use_bus_addr; | 53 | ih->use_bus_addr = use_bus_addr; |
78 | 54 | ||
79 | if (adev->irq.ih.use_bus_addr) { | 55 | if (use_bus_addr) { |
80 | if (!adev->irq.ih.ring) { | 56 | if (ih->ring) |
81 | /* add 8 bytes for the rptr/wptr shadows and | 57 | return 0; |
82 | * add them to the end of the ring allocation. | 58 | |
83 | */ | 59 | /* add 8 bytes for the rptr/wptr shadows and |
84 | adev->irq.ih.ring = pci_alloc_consistent(adev->pdev, | 60 | * add them to the end of the ring allocation. |
85 | adev->irq.ih.ring_size + 8, | 61 | */ |
86 | &adev->irq.ih.rb_dma_addr); | 62 | ih->ring = dma_alloc_coherent(adev->dev, ih->ring_size + 8, |
87 | if (adev->irq.ih.ring == NULL) | 63 | &ih->rb_dma_addr, GFP_KERNEL); |
88 | return -ENOMEM; | 64 | if (ih->ring == NULL) |
89 | memset((void *)adev->irq.ih.ring, 0, adev->irq.ih.ring_size + 8); | 65 | return -ENOMEM; |
90 | adev->irq.ih.wptr_offs = (adev->irq.ih.ring_size / 4) + 0; | 66 | |
91 | adev->irq.ih.rptr_offs = (adev->irq.ih.ring_size / 4) + 1; | 67 | memset((void *)ih->ring, 0, ih->ring_size + 8); |
92 | } | 68 | ih->wptr_offs = (ih->ring_size / 4) + 0; |
93 | return 0; | 69 | ih->rptr_offs = (ih->ring_size / 4) + 1; |
94 | } else { | 70 | } else { |
95 | r = amdgpu_device_wb_get(adev, &adev->irq.ih.wptr_offs); | 71 | r = amdgpu_device_wb_get(adev, &ih->wptr_offs); |
72 | if (r) | ||
73 | return r; | ||
74 | |||
75 | r = amdgpu_device_wb_get(adev, &ih->rptr_offs); | ||
96 | if (r) { | 76 | if (r) { |
97 | dev_err(adev->dev, "(%d) ih wptr_offs wb alloc failed\n", r); | 77 | amdgpu_device_wb_free(adev, ih->wptr_offs); |
98 | return r; | 78 | return r; |
99 | } | 79 | } |
100 | 80 | ||
101 | r = amdgpu_device_wb_get(adev, &adev->irq.ih.rptr_offs); | 81 | r = amdgpu_bo_create_kernel(adev, ih->ring_size, PAGE_SIZE, |
82 | AMDGPU_GEM_DOMAIN_GTT, | ||
83 | &ih->ring_obj, &ih->gpu_addr, | ||
84 | (void **)&ih->ring); | ||
102 | if (r) { | 85 | if (r) { |
103 | amdgpu_device_wb_free(adev, adev->irq.ih.wptr_offs); | 86 | amdgpu_device_wb_free(adev, ih->rptr_offs); |
104 | dev_err(adev->dev, "(%d) ih rptr_offs wb alloc failed\n", r); | 87 | amdgpu_device_wb_free(adev, ih->wptr_offs); |
105 | return r; | 88 | return r; |
106 | } | 89 | } |
107 | |||
108 | return amdgpu_ih_ring_alloc(adev); | ||
109 | } | 90 | } |
91 | return 0; | ||
110 | } | 92 | } |
111 | 93 | ||
112 | /** | 94 | /** |
113 | * amdgpu_ih_ring_fini - tear down the IH state | 95 | * amdgpu_ih_ring_fini - tear down the IH state |
114 | * | 96 | * |
115 | * @adev: amdgpu_device pointer | 97 | * @adev: amdgpu_device pointer |
98 | * @ih: ih ring to tear down | ||
116 | * | 99 | * |
117 | * Tears down the IH state and frees buffer | 100 | * Tears down the IH state and frees buffer |
118 | * used for the IH ring buffer. | 101 | * used for the IH ring buffer. |
119 | */ | 102 | */ |
120 | void amdgpu_ih_ring_fini(struct amdgpu_device *adev) | 103 | void amdgpu_ih_ring_fini(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih) |
121 | { | 104 | { |
122 | if (adev->irq.ih.use_bus_addr) { | 105 | if (ih->use_bus_addr) { |
123 | if (adev->irq.ih.ring) { | 106 | if (!ih->ring) |
124 | /* add 8 bytes for the rptr/wptr shadows and | 107 | return; |
125 | * add them to the end of the ring allocation. | 108 | |
126 | */ | 109 | /* add 8 bytes for the rptr/wptr shadows and |
127 | pci_free_consistent(adev->pdev, adev->irq.ih.ring_size + 8, | 110 | * add them to the end of the ring allocation. |
128 | (void *)adev->irq.ih.ring, | 111 | */ |
129 | adev->irq.ih.rb_dma_addr); | 112 | dma_free_coherent(adev->dev, ih->ring_size + 8, |
130 | adev->irq.ih.ring = NULL; | 113 | (void *)ih->ring, ih->rb_dma_addr); |
131 | } | 114 | ih->ring = NULL; |
132 | } else { | 115 | } else { |
133 | amdgpu_bo_free_kernel(&adev->irq.ih.ring_obj, | 116 | amdgpu_bo_free_kernel(&ih->ring_obj, &ih->gpu_addr, |
134 | &adev->irq.ih.gpu_addr, | 117 | (void **)&ih->ring); |
135 | (void **)&adev->irq.ih.ring); | 118 | amdgpu_device_wb_free(adev, ih->wptr_offs); |
136 | amdgpu_device_wb_free(adev, adev->irq.ih.wptr_offs); | 119 | amdgpu_device_wb_free(adev, ih->rptr_offs); |
137 | amdgpu_device_wb_free(adev, adev->irq.ih.rptr_offs); | ||
138 | } | 120 | } |
139 | } | 121 | } |
140 | 122 | ||
@@ -142,56 +124,56 @@ void amdgpu_ih_ring_fini(struct amdgpu_device *adev) | |||
142 | * amdgpu_ih_process - interrupt handler | 124 | * amdgpu_ih_process - interrupt handler |
143 | * | 125 | * |
144 | * @adev: amdgpu_device pointer | 126 | * @adev: amdgpu_device pointer |
127 | * @ih: ih ring to process | ||
145 | * | 128 | * |
146 | * Interrupt hander (VI), walk the IH ring. | 129 | * Interrupt hander (VI), walk the IH ring. |
147 | * Returns irq process return code. | 130 | * Returns irq process return code. |
148 | */ | 131 | */ |
149 | int amdgpu_ih_process(struct amdgpu_device *adev) | 132 | int amdgpu_ih_process(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih) |
150 | { | 133 | { |
151 | struct amdgpu_iv_entry entry; | 134 | struct amdgpu_iv_entry entry; |
152 | u32 wptr; | 135 | u32 wptr; |
153 | 136 | ||
154 | if (!adev->irq.ih.enabled || adev->shutdown) | 137 | if (!ih->enabled || adev->shutdown) |
155 | return IRQ_NONE; | 138 | return IRQ_NONE; |
156 | 139 | ||
157 | wptr = amdgpu_ih_get_wptr(adev); | 140 | wptr = amdgpu_ih_get_wptr(adev); |
158 | 141 | ||
159 | restart_ih: | 142 | restart_ih: |
160 | /* is somebody else already processing irqs? */ | 143 | /* is somebody else already processing irqs? */ |
161 | if (atomic_xchg(&adev->irq.ih.lock, 1)) | 144 | if (atomic_xchg(&ih->lock, 1)) |
162 | return IRQ_NONE; | 145 | return IRQ_NONE; |
163 | 146 | ||
164 | DRM_DEBUG("%s: rptr %d, wptr %d\n", __func__, adev->irq.ih.rptr, wptr); | 147 | DRM_DEBUG("%s: rptr %d, wptr %d\n", __func__, ih->rptr, wptr); |
165 | 148 | ||
166 | /* Order reading of wptr vs. reading of IH ring data */ | 149 | /* Order reading of wptr vs. reading of IH ring data */ |
167 | rmb(); | 150 | rmb(); |
168 | 151 | ||
169 | while (adev->irq.ih.rptr != wptr) { | 152 | while (ih->rptr != wptr) { |
170 | u32 ring_index = adev->irq.ih.rptr >> 2; | 153 | u32 ring_index = ih->rptr >> 2; |
171 | 154 | ||
172 | /* Prescreening of high-frequency interrupts */ | 155 | /* Prescreening of high-frequency interrupts */ |
173 | if (!amdgpu_ih_prescreen_iv(adev)) { | 156 | if (!amdgpu_ih_prescreen_iv(adev)) { |
174 | adev->irq.ih.rptr &= adev->irq.ih.ptr_mask; | 157 | ih->rptr &= ih->ptr_mask; |
175 | continue; | 158 | continue; |
176 | } | 159 | } |
177 | 160 | ||
178 | /* Before dispatching irq to IP blocks, send it to amdkfd */ | 161 | /* Before dispatching irq to IP blocks, send it to amdkfd */ |
179 | amdgpu_amdkfd_interrupt(adev, | 162 | amdgpu_amdkfd_interrupt(adev, |
180 | (const void *) &adev->irq.ih.ring[ring_index]); | 163 | (const void *) &ih->ring[ring_index]); |
181 | 164 | ||
182 | entry.iv_entry = (const uint32_t *) | 165 | entry.iv_entry = (const uint32_t *)&ih->ring[ring_index]; |
183 | &adev->irq.ih.ring[ring_index]; | ||
184 | amdgpu_ih_decode_iv(adev, &entry); | 166 | amdgpu_ih_decode_iv(adev, &entry); |
185 | adev->irq.ih.rptr &= adev->irq.ih.ptr_mask; | 167 | ih->rptr &= ih->ptr_mask; |
186 | 168 | ||
187 | amdgpu_irq_dispatch(adev, &entry); | 169 | amdgpu_irq_dispatch(adev, &entry); |
188 | } | 170 | } |
189 | amdgpu_ih_set_rptr(adev); | 171 | amdgpu_ih_set_rptr(adev); |
190 | atomic_set(&adev->irq.ih.lock, 0); | 172 | atomic_set(&ih->lock, 0); |
191 | 173 | ||
192 | /* make sure wptr hasn't changed while processing */ | 174 | /* make sure wptr hasn't changed while processing */ |
193 | wptr = amdgpu_ih_get_wptr(adev); | 175 | wptr = amdgpu_ih_get_wptr(adev); |
194 | if (wptr != adev->irq.ih.rptr) | 176 | if (wptr != ih->rptr) |
195 | goto restart_ih; | 177 | goto restart_ih; |
196 | 178 | ||
197 | return IRQ_HANDLED; | 179 | return IRQ_HANDLED; |