Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
drm/panfrost: Expose the selected coherency protocol to the UMD
Will be needed if we want to skip CPU cache maintenance operations when
the GPU can snoop CPU caches.
v2:
- New commit
v3:
- Fix the coherency values (enum instead of bitmask)
v4:
- Fix init/test on coherency_features
v5:
- No changes
v6:
- Collect R-b
v7:
- No changes
v8:
- No changes
Reviewed-by: Steven Price <steven.price@arm.com>
Link: https://patch.msgid.link/20251208100841.730527-10-boris.brezillon@collabora.com
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
+40
-5
+1
drivers/gpu/drm/panfrost/panfrost_device.h
+1
drivers/gpu/drm/panfrost/panfrost_device.h
+1
drivers/gpu/drm/panfrost/panfrost_drv.c
+1
drivers/gpu/drm/panfrost/panfrost_drv.c
···
95
95
PANFROST_FEATURE_ARRAY(JS_FEATURES, js_features, 15);
96
96
PANFROST_FEATURE(NR_CORE_GROUPS, nr_core_groups);
97
97
PANFROST_FEATURE(THREAD_TLS_ALLOC, thread_tls_alloc);
98
+
PANFROST_FEATURE(SELECTED_COHERENCY, selected_coherency);
98
99
99
100
case DRM_PANFROST_PARAM_SYSTEM_TIMESTAMP:
100
101
ret = panfrost_ioctl_query_timestamp(pfdev, ¶m->value);
+23
-3
drivers/gpu/drm/panfrost/panfrost_gpu.c
+23
-3
drivers/gpu/drm/panfrost/panfrost_gpu.c
···
159
159
pfdev->features.revision >= 0x2000)
160
160
quirks |= JM_MAX_JOB_THROTTLE_LIMIT << JM_JOB_THROTTLE_LIMIT_SHIFT;
161
161
else if (panfrost_model_eq(pfdev, 0x6000) &&
162
-
pfdev->features.coherency_features == COHERENCY_ACE)
163
-
quirks |= (COHERENCY_ACE_LITE | COHERENCY_ACE) <<
162
+
pfdev->features.coherency_features == BIT(COHERENCY_ACE))
163
+
quirks |= (BIT(COHERENCY_ACE_LITE) | BIT(COHERENCY_ACE)) <<
164
164
JM_FORCE_COHERENCY_FEATURES_SHIFT;
165
165
166
166
if (panfrost_has_hw_feature(pfdev, HW_FEATURE_IDVS_GROUP_SIZE))
···
263
263
pfdev->features.max_threads = gpu_read(pfdev, GPU_THREAD_MAX_THREADS);
264
264
pfdev->features.thread_max_workgroup_sz = gpu_read(pfdev, GPU_THREAD_MAX_WORKGROUP_SIZE);
265
265
pfdev->features.thread_max_barrier_sz = gpu_read(pfdev, GPU_THREAD_MAX_BARRIER_SIZE);
266
-
pfdev->features.coherency_features = gpu_read(pfdev, GPU_COHERENCY_FEATURES);
266
+
267
+
if (panfrost_has_hw_feature(pfdev, HW_FEATURE_COHERENCY_REG))
268
+
pfdev->features.coherency_features = gpu_read(pfdev, GPU_COHERENCY_FEATURES);
269
+
else
270
+
pfdev->features.coherency_features = BIT(COHERENCY_ACE_LITE);
271
+
272
+
BUILD_BUG_ON(COHERENCY_ACE_LITE != DRM_PANFROST_GPU_COHERENCY_ACE_LITE);
273
+
BUILD_BUG_ON(COHERENCY_ACE != DRM_PANFROST_GPU_COHERENCY_ACE);
274
+
BUILD_BUG_ON(COHERENCY_NONE != DRM_PANFROST_GPU_COHERENCY_NONE);
275
+
276
+
if (!pfdev->coherent) {
277
+
pfdev->features.selected_coherency = COHERENCY_NONE;
278
+
} else if (pfdev->features.coherency_features & BIT(COHERENCY_ACE)) {
279
+
pfdev->features.selected_coherency = COHERENCY_ACE;
280
+
} else if (pfdev->features.coherency_features & BIT(COHERENCY_ACE_LITE)) {
281
+
pfdev->features.selected_coherency = COHERENCY_ACE_LITE;
282
+
} else {
283
+
drm_WARN(&pfdev->base, true, "No known coherency protocol supported");
284
+
pfdev->features.selected_coherency = COHERENCY_NONE;
285
+
}
286
+
267
287
pfdev->features.afbc_features = gpu_read(pfdev, GPU_AFBC_FEATURES);
268
288
for (i = 0; i < 4; i++)
269
289
pfdev->features.texture_features[i] = gpu_read(pfdev, GPU_TEXTURE_FEATURES(i));
+8
-2
drivers/gpu/drm/panfrost/panfrost_regs.h
+8
-2
drivers/gpu/drm/panfrost/panfrost_regs.h
···
102
102
#define GPU_L2_PRESENT_LO 0x120 /* (RO) Level 2 cache present bitmap, low word */
103
103
#define GPU_L2_PRESENT_HI 0x124 /* (RO) Level 2 cache present bitmap, high word */
104
104
105
+
/* GPU_COHERENCY_FEATURES is a bitmask of BIT(COHERENCY_xxx) values encoding the
106
+
* set of supported coherency protocols. GPU_COHERENCY_ENABLE is passed a
107
+
* COHERENCY_xxx value.
108
+
*/
105
109
#define GPU_COHERENCY_FEATURES 0x300 /* (RO) Coherency features present */
106
-
#define COHERENCY_ACE_LITE BIT(0)
107
-
#define COHERENCY_ACE BIT(1)
110
+
#define GPU_COHERENCY_ENABLE 0x304 /* (RW) Coherency protocol selection */
111
+
#define COHERENCY_ACE_LITE 0
112
+
#define COHERENCY_ACE 1
113
+
#define COHERENCY_NONE 31
108
114
109
115
#define GPU_STACK_PRESENT_LO 0xE00 /* (RO) Core stack present bitmap, low word */
110
116
#define GPU_STACK_PRESENT_HI 0xE04 /* (RO) Core stack present bitmap, high word */
+7
include/uapi/drm/panfrost_drm.h
+7
include/uapi/drm/panfrost_drm.h
···
228
228
DRM_PANFROST_PARAM_SYSTEM_TIMESTAMP,
229
229
DRM_PANFROST_PARAM_SYSTEM_TIMESTAMP_FREQUENCY,
230
230
DRM_PANFROST_PARAM_ALLOWED_JM_CTX_PRIORITIES,
231
+
DRM_PANFROST_PARAM_SELECTED_COHERENCY,
232
+
};
233
+
234
+
enum drm_panfrost_gpu_coherency {
235
+
DRM_PANFROST_GPU_COHERENCY_ACE_LITE = 0,
236
+
DRM_PANFROST_GPU_COHERENCY_ACE = 1,
237
+
DRM_PANFROST_GPU_COHERENCY_NONE = 31,
231
238
};
232
239
233
240
struct drm_panfrost_get_param {