2 * Copyright (C) 2013 Red Hat
3 * Author: Rob Clark <robdclark@gmail.com>
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 as published by
7 * the Free Software Foundation.
9 * This program is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
14 * You should have received a copy of the GNU General Public License along with
15 * this program. If not, see <http://www.gnu.org/licenses/>.
18 #ifdef CONFIG_MSM_OCMEM
19 # include <mach/ocmem.h>
24 #define A3XX_INT0_MASK \
25 (A3XX_INT0_RBBM_AHB_ERROR | \
26 A3XX_INT0_RBBM_ATB_BUS_OVERFLOW | \
27 A3XX_INT0_CP_T0_PACKET_IN_IB | \
28 A3XX_INT0_CP_OPCODE_ERROR | \
29 A3XX_INT0_CP_RESERVED_BIT_ERROR | \
30 A3XX_INT0_CP_HW_FAULT | \
31 A3XX_INT0_CP_IB1_INT | \
32 A3XX_INT0_CP_IB2_INT | \
33 A3XX_INT0_CP_RB_INT | \
34 A3XX_INT0_CP_REG_PROTECT_FAULT | \
35 A3XX_INT0_CP_AHB_ERROR_HALT | \
36 A3XX_INT0_UCHE_OOB_ACCESS)
39 static bool hang_debug = false;
40 MODULE_PARM_DESC(hang_debug, "Dump registers when hang is detected (can be slow!)");
41 module_param_named(hang_debug, hang_debug, bool, 0600);
42 static void a3xx_dump(struct msm_gpu *gpu);
44 static struct platform_device *a3xx_pdev;
46 static void a3xx_me_init(struct msm_gpu *gpu)
48 struct msm_ringbuffer *ring = gpu->rb;
50 OUT_PKT3(ring, CP_ME_INIT, 17);
51 OUT_RING(ring, 0x000003f7);
52 OUT_RING(ring, 0x00000000);
53 OUT_RING(ring, 0x00000000);
54 OUT_RING(ring, 0x00000000);
55 OUT_RING(ring, 0x00000080);
56 OUT_RING(ring, 0x00000100);
57 OUT_RING(ring, 0x00000180);
58 OUT_RING(ring, 0x00006600);
59 OUT_RING(ring, 0x00000150);
60 OUT_RING(ring, 0x0000014e);
61 OUT_RING(ring, 0x00000154);
62 OUT_RING(ring, 0x00000001);
63 OUT_RING(ring, 0x00000000);
64 OUT_RING(ring, 0x00000000);
65 OUT_RING(ring, 0x00000000);
66 OUT_RING(ring, 0x00000000);
67 OUT_RING(ring, 0x00000000);
69 gpu->funcs->flush(gpu);
70 gpu->funcs->idle(gpu);
73 static int a3xx_hw_init(struct msm_gpu *gpu)
75 struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
76 struct a3xx_gpu *a3xx_gpu = to_a3xx_gpu(adreno_gpu);
82 if (adreno_is_a305(adreno_gpu)) {
83 /* Set up 16 deep read/write request queues: */
84 gpu_write(gpu, REG_A3XX_VBIF_IN_RD_LIM_CONF0, 0x10101010);
85 gpu_write(gpu, REG_A3XX_VBIF_IN_RD_LIM_CONF1, 0x10101010);
86 gpu_write(gpu, REG_A3XX_VBIF_OUT_RD_LIM_CONF0, 0x10101010);
87 gpu_write(gpu, REG_A3XX_VBIF_OUT_WR_LIM_CONF0, 0x10101010);
88 gpu_write(gpu, REG_A3XX_VBIF_DDR_OUT_MAX_BURST, 0x0000303);
89 gpu_write(gpu, REG_A3XX_VBIF_IN_WR_LIM_CONF0, 0x10101010);
90 gpu_write(gpu, REG_A3XX_VBIF_IN_WR_LIM_CONF1, 0x10101010);
92 gpu_write(gpu, REG_A3XX_VBIF_GATE_OFF_WRREQ_EN, 0x0000ff);
93 /* Set up round robin arbitration between both AXI ports: */
94 gpu_write(gpu, REG_A3XX_VBIF_ARB_CTL, 0x00000030);
96 gpu_write(gpu, REG_A3XX_VBIF_OUT_AXI_AOOO_EN, 0x0000003c);
97 gpu_write(gpu, REG_A3XX_VBIF_OUT_AXI_AOOO, 0x003c003c);
99 } else if (adreno_is_a320(adreno_gpu)) {
100 /* Set up 16 deep read/write request queues: */
101 gpu_write(gpu, REG_A3XX_VBIF_IN_RD_LIM_CONF0, 0x10101010);
102 gpu_write(gpu, REG_A3XX_VBIF_IN_RD_LIM_CONF1, 0x10101010);
103 gpu_write(gpu, REG_A3XX_VBIF_OUT_RD_LIM_CONF0, 0x10101010);
104 gpu_write(gpu, REG_A3XX_VBIF_OUT_WR_LIM_CONF0, 0x10101010);
105 gpu_write(gpu, REG_A3XX_VBIF_DDR_OUT_MAX_BURST, 0x0000303);
106 gpu_write(gpu, REG_A3XX_VBIF_IN_WR_LIM_CONF0, 0x10101010);
107 gpu_write(gpu, REG_A3XX_VBIF_IN_WR_LIM_CONF1, 0x10101010);
109 gpu_write(gpu, REG_A3XX_VBIF_GATE_OFF_WRREQ_EN, 0x0000ff);
110 /* Set up round robin arbitration between both AXI ports: */
111 gpu_write(gpu, REG_A3XX_VBIF_ARB_CTL, 0x00000030);
113 gpu_write(gpu, REG_A3XX_VBIF_OUT_AXI_AOOO_EN, 0x0000003c);
114 gpu_write(gpu, REG_A3XX_VBIF_OUT_AXI_AOOO, 0x003c003c);
115 /* Enable 1K sort: */
116 gpu_write(gpu, REG_A3XX_VBIF_ABIT_SORT, 0x000000ff);
117 gpu_write(gpu, REG_A3XX_VBIF_ABIT_SORT_CONF, 0x000000a4);
119 } else if (adreno_is_a330v2(adreno_gpu)) {
121 * Most of the VBIF registers on 8974v2 have the correct
122 * values at power on, so we won't modify those if we don't
125 /* Enable 1k sort: */
126 gpu_write(gpu, REG_A3XX_VBIF_ABIT_SORT, 0x0001003f);
127 gpu_write(gpu, REG_A3XX_VBIF_ABIT_SORT_CONF, 0x000000a4);
129 gpu_write(gpu, REG_A3XX_VBIF_GATE_OFF_WRREQ_EN, 0x00003f);
130 gpu_write(gpu, REG_A3XX_VBIF_DDR_OUT_MAX_BURST, 0x0000303);
131 /* Set up VBIF_ROUND_ROBIN_QOS_ARB: */
132 gpu_write(gpu, REG_A3XX_VBIF_ROUND_ROBIN_QOS_ARB, 0x0003);
134 } else if (adreno_is_a330(adreno_gpu)) {
135 /* Set up 16 deep read/write request queues: */
136 gpu_write(gpu, REG_A3XX_VBIF_IN_RD_LIM_CONF0, 0x18181818);
137 gpu_write(gpu, REG_A3XX_VBIF_IN_RD_LIM_CONF1, 0x18181818);
138 gpu_write(gpu, REG_A3XX_VBIF_OUT_RD_LIM_CONF0, 0x18181818);
139 gpu_write(gpu, REG_A3XX_VBIF_OUT_WR_LIM_CONF0, 0x18181818);
140 gpu_write(gpu, REG_A3XX_VBIF_DDR_OUT_MAX_BURST, 0x0000303);
141 gpu_write(gpu, REG_A3XX_VBIF_IN_WR_LIM_CONF0, 0x18181818);
142 gpu_write(gpu, REG_A3XX_VBIF_IN_WR_LIM_CONF1, 0x18181818);
144 gpu_write(gpu, REG_A3XX_VBIF_GATE_OFF_WRREQ_EN, 0x00003f);
145 /* Set up round robin arbitration between both AXI ports: */
146 gpu_write(gpu, REG_A3XX_VBIF_ARB_CTL, 0x00000030);
147 /* Set up VBIF_ROUND_ROBIN_QOS_ARB: */
148 gpu_write(gpu, REG_A3XX_VBIF_ROUND_ROBIN_QOS_ARB, 0x0001);
150 gpu_write(gpu, REG_A3XX_VBIF_OUT_AXI_AOOO_EN, 0x0000003f);
151 gpu_write(gpu, REG_A3XX_VBIF_OUT_AXI_AOOO, 0x003f003f);
152 /* Enable 1K sort: */
153 gpu_write(gpu, REG_A3XX_VBIF_ABIT_SORT, 0x0001003f);
154 gpu_write(gpu, REG_A3XX_VBIF_ABIT_SORT_CONF, 0x000000a4);
155 /* Disable VBIF clock gating. This is to enable AXI running
156 * higher frequency than GPU:
158 gpu_write(gpu, REG_A3XX_VBIF_CLKON, 0x00000001);
164 /* Make all blocks contribute to the GPU BUSY perf counter: */
165 gpu_write(gpu, REG_A3XX_RBBM_GPU_BUSY_MASKED, 0xffffffff);
167 /* Tune the hystersis counters for SP and CP idle detection: */
168 gpu_write(gpu, REG_A3XX_RBBM_SP_HYST_CNT, 0x10);
169 gpu_write(gpu, REG_A3XX_RBBM_WAIT_IDLE_CLOCKS_CTL, 0x10);
171 /* Enable the RBBM error reporting bits. This lets us get
172 * useful information on failure:
174 gpu_write(gpu, REG_A3XX_RBBM_AHB_CTL0, 0x00000001);
176 /* Enable AHB error reporting: */
177 gpu_write(gpu, REG_A3XX_RBBM_AHB_CTL1, 0xa6ffffff);
179 /* Turn on the power counters: */
180 gpu_write(gpu, REG_A3XX_RBBM_RBBM_CTL, 0x00030000);
182 /* Turn on hang detection - this spews a lot of useful information
183 * into the RBBM registers on a hang:
185 gpu_write(gpu, REG_A3XX_RBBM_INTERFACE_HANG_INT_CTL, 0x00010fff);
187 /* Enable 64-byte cacheline size. HW Default is 32-byte (0x000000E0): */
188 gpu_write(gpu, REG_A3XX_UCHE_CACHE_MODE_CONTROL_REG, 0x00000001);
190 /* Enable Clock gating: */
191 if (adreno_is_a320(adreno_gpu))
192 gpu_write(gpu, REG_A3XX_RBBM_CLOCK_CTL, 0xbfffffff);
193 else if (adreno_is_a330v2(adreno_gpu))
194 gpu_write(gpu, REG_A3XX_RBBM_CLOCK_CTL, 0xaaaaaaaa);
195 else if (adreno_is_a330(adreno_gpu))
196 gpu_write(gpu, REG_A3XX_RBBM_CLOCK_CTL, 0xbffcffff);
198 if (adreno_is_a330v2(adreno_gpu))
199 gpu_write(gpu, REG_A3XX_RBBM_GPR0_CTL, 0x05515455);
200 else if (adreno_is_a330(adreno_gpu))
201 gpu_write(gpu, REG_A3XX_RBBM_GPR0_CTL, 0x00000000);
203 /* Set the OCMEM base address for A330, etc */
204 if (a3xx_gpu->ocmem_hdl) {
205 gpu_write(gpu, REG_A3XX_RB_GMEM_BASE_ADDR,
206 (unsigned int)(a3xx_gpu->ocmem_base >> 14));
209 /* Turn on performance counters: */
210 gpu_write(gpu, REG_A3XX_RBBM_PERFCTR_CTL, 0x01);
212 /* Set SP perfcounter 7 to count SP_FS_FULL_ALU_INSTRUCTIONS
213 * we will use this to augment our hang detection:
215 gpu_write(gpu, REG_A3XX_SP_PERFCOUNTER7_SELECT,
216 SP_FS_FULL_ALU_INSTRUCTIONS);
218 gpu_write(gpu, REG_A3XX_RBBM_INT_0_MASK, A3XX_INT0_MASK);
220 ret = adreno_hw_init(gpu);
224 /* setup access protection: */
225 gpu_write(gpu, REG_A3XX_CP_PROTECT_CTRL, 0x00000007);
228 gpu_write(gpu, REG_A3XX_CP_PROTECT(0), 0x63000040);
229 gpu_write(gpu, REG_A3XX_CP_PROTECT(1), 0x62000080);
230 gpu_write(gpu, REG_A3XX_CP_PROTECT(2), 0x600000cc);
231 gpu_write(gpu, REG_A3XX_CP_PROTECT(3), 0x60000108);
232 gpu_write(gpu, REG_A3XX_CP_PROTECT(4), 0x64000140);
233 gpu_write(gpu, REG_A3XX_CP_PROTECT(5), 0x66000400);
236 gpu_write(gpu, REG_A3XX_CP_PROTECT(6), 0x65000700);
237 gpu_write(gpu, REG_A3XX_CP_PROTECT(7), 0x610007d8);
238 gpu_write(gpu, REG_A3XX_CP_PROTECT(8), 0x620007e0);
239 gpu_write(gpu, REG_A3XX_CP_PROTECT(9), 0x61001178);
240 gpu_write(gpu, REG_A3XX_CP_PROTECT(10), 0x64001180);
243 gpu_write(gpu, REG_A3XX_CP_PROTECT(11), 0x60003300);
246 gpu_write(gpu, REG_A3XX_CP_PROTECT(12), 0x6b00c000);
248 /* NOTE: PM4/micro-engine firmware registers look to be the same
249 * for a2xx and a3xx.. we could possibly push that part down to
250 * adreno_gpu base class. Or push both PM4 and PFP but
251 * parameterize the pfp ucode addr/data registers..
255 ptr = (uint32_t *)(adreno_gpu->pm4->data);
256 len = adreno_gpu->pm4->size / 4;
257 DBG("loading PM4 ucode version: %x", ptr[1]);
259 gpu_write(gpu, REG_AXXX_CP_DEBUG,
260 AXXX_CP_DEBUG_DYNAMIC_CLK_DISABLE |
261 AXXX_CP_DEBUG_MIU_128BIT_WRITE_ENABLE);
262 gpu_write(gpu, REG_AXXX_CP_ME_RAM_WADDR, 0);
263 for (i = 1; i < len; i++)
264 gpu_write(gpu, REG_AXXX_CP_ME_RAM_DATA, ptr[i]);
267 ptr = (uint32_t *)(adreno_gpu->pfp->data);
268 len = adreno_gpu->pfp->size / 4;
269 DBG("loading PFP ucode version: %x", ptr[5]);
271 gpu_write(gpu, REG_A3XX_CP_PFP_UCODE_ADDR, 0);
272 for (i = 1; i < len; i++)
273 gpu_write(gpu, REG_A3XX_CP_PFP_UCODE_DATA, ptr[i]);
275 /* CP ROQ queue sizes (bytes) - RB:16, ST:16, IB1:32, IB2:64 */
276 if (adreno_is_a305(adreno_gpu) || adreno_is_a320(adreno_gpu)) {
277 gpu_write(gpu, REG_AXXX_CP_QUEUE_THRESHOLDS,
278 AXXX_CP_QUEUE_THRESHOLDS_CSQ_IB1_START(2) |
279 AXXX_CP_QUEUE_THRESHOLDS_CSQ_IB2_START(6) |
280 AXXX_CP_QUEUE_THRESHOLDS_CSQ_ST_START(14));
281 } else if (adreno_is_a330(adreno_gpu)) {
282 /* NOTE: this (value take from downstream android driver)
283 * includes some bits outside of the known bitfields. But
284 * A330 has this "MERCIU queue" thing too, which might
285 * explain a new bitfield or reshuffling:
287 gpu_write(gpu, REG_AXXX_CP_QUEUE_THRESHOLDS, 0x003e2008);
290 /* clear ME_HALT to start micro engine */
291 gpu_write(gpu, REG_AXXX_CP_ME_CNTL, 0);
298 static void a3xx_recover(struct msm_gpu *gpu)
300 /* dump registers before resetting gpu, if enabled: */
303 gpu_write(gpu, REG_A3XX_RBBM_SW_RESET_CMD, 1);
304 gpu_read(gpu, REG_A3XX_RBBM_SW_RESET_CMD);
305 gpu_write(gpu, REG_A3XX_RBBM_SW_RESET_CMD, 0);
309 static void a3xx_destroy(struct msm_gpu *gpu)
311 struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
312 struct a3xx_gpu *a3xx_gpu = to_a3xx_gpu(adreno_gpu);
314 DBG("%s", gpu->name);
316 adreno_gpu_cleanup(adreno_gpu);
318 #ifdef CONFIG_MSM_OCMEM
319 if (a3xx_gpu->ocmem_base)
320 ocmem_free(OCMEM_GRAPHICS, a3xx_gpu->ocmem_hdl);
323 put_device(&a3xx_gpu->pdev->dev);
327 static void a3xx_idle(struct msm_gpu *gpu)
329 /* wait for ringbuffer to drain: */
332 /* then wait for GPU to finish: */
333 if (spin_until(!(gpu_read(gpu, REG_A3XX_RBBM_STATUS) &
334 A3XX_RBBM_STATUS_GPU_BUSY)))
335 DRM_ERROR("%s: timeout waiting for GPU to idle!\n", gpu->name);
337 /* TODO maybe we need to reset GPU here to recover from hang? */
340 static irqreturn_t a3xx_irq(struct msm_gpu *gpu)
344 status = gpu_read(gpu, REG_A3XX_RBBM_INT_0_STATUS);
345 DBG("%s: %08x", gpu->name, status);
349 gpu_write(gpu, REG_A3XX_RBBM_INT_CLEAR_CMD, status);
356 static const unsigned int a3xx_registers[] = {
357 0x0000, 0x0002, 0x0010, 0x0012, 0x0018, 0x0018, 0x0020, 0x0027,
358 0x0029, 0x002b, 0x002e, 0x0033, 0x0040, 0x0042, 0x0050, 0x005c,
359 0x0060, 0x006c, 0x0080, 0x0082, 0x0084, 0x0088, 0x0090, 0x00e5,
360 0x00ea, 0x00ed, 0x0100, 0x0100, 0x0110, 0x0123, 0x01c0, 0x01c1,
361 0x01c3, 0x01c5, 0x01c7, 0x01c7, 0x01d5, 0x01d9, 0x01dc, 0x01dd,
362 0x01ea, 0x01ea, 0x01ee, 0x01f1, 0x01f5, 0x01f5, 0x01fc, 0x01ff,
363 0x0440, 0x0440, 0x0443, 0x0443, 0x0445, 0x0445, 0x044d, 0x044f,
364 0x0452, 0x0452, 0x0454, 0x046f, 0x047c, 0x047c, 0x047f, 0x047f,
365 0x0578, 0x057f, 0x0600, 0x0602, 0x0605, 0x0607, 0x060a, 0x060e,
366 0x0612, 0x0614, 0x0c01, 0x0c02, 0x0c06, 0x0c1d, 0x0c3d, 0x0c3f,
367 0x0c48, 0x0c4b, 0x0c80, 0x0c80, 0x0c88, 0x0c8b, 0x0ca0, 0x0cb7,
368 0x0cc0, 0x0cc1, 0x0cc6, 0x0cc7, 0x0ce4, 0x0ce5, 0x0e00, 0x0e05,
369 0x0e0c, 0x0e0c, 0x0e22, 0x0e23, 0x0e41, 0x0e45, 0x0e64, 0x0e65,
370 0x0e80, 0x0e82, 0x0e84, 0x0e89, 0x0ea0, 0x0ea1, 0x0ea4, 0x0ea7,
371 0x0ec4, 0x0ecb, 0x0ee0, 0x0ee0, 0x0f00, 0x0f01, 0x0f03, 0x0f09,
372 0x2040, 0x2040, 0x2044, 0x2044, 0x2048, 0x204d, 0x2068, 0x2069,
373 0x206c, 0x206d, 0x2070, 0x2070, 0x2072, 0x2072, 0x2074, 0x2075,
374 0x2079, 0x207a, 0x20c0, 0x20d3, 0x20e4, 0x20ef, 0x2100, 0x2109,
375 0x210c, 0x210c, 0x210e, 0x210e, 0x2110, 0x2111, 0x2114, 0x2115,
376 0x21e4, 0x21e4, 0x21ea, 0x21ea, 0x21ec, 0x21ed, 0x21f0, 0x21f0,
377 0x2200, 0x2212, 0x2214, 0x2217, 0x221a, 0x221a, 0x2240, 0x227e,
378 0x2280, 0x228b, 0x22c0, 0x22c0, 0x22c4, 0x22ce, 0x22d0, 0x22d8,
379 0x22df, 0x22e6, 0x22e8, 0x22e9, 0x22ec, 0x22ec, 0x22f0, 0x22f7,
380 0x22ff, 0x22ff, 0x2340, 0x2343, 0x2348, 0x2349, 0x2350, 0x2356,
381 0x2360, 0x2360, 0x2440, 0x2440, 0x2444, 0x2444, 0x2448, 0x244d,
382 0x2468, 0x2469, 0x246c, 0x246d, 0x2470, 0x2470, 0x2472, 0x2472,
383 0x2474, 0x2475, 0x2479, 0x247a, 0x24c0, 0x24d3, 0x24e4, 0x24ef,
384 0x2500, 0x2509, 0x250c, 0x250c, 0x250e, 0x250e, 0x2510, 0x2511,
385 0x2514, 0x2515, 0x25e4, 0x25e4, 0x25ea, 0x25ea, 0x25ec, 0x25ed,
386 0x25f0, 0x25f0, 0x2600, 0x2612, 0x2614, 0x2617, 0x261a, 0x261a,
387 0x2640, 0x267e, 0x2680, 0x268b, 0x26c0, 0x26c0, 0x26c4, 0x26ce,
388 0x26d0, 0x26d8, 0x26df, 0x26e6, 0x26e8, 0x26e9, 0x26ec, 0x26ec,
389 0x26f0, 0x26f7, 0x26ff, 0x26ff, 0x2740, 0x2743, 0x2748, 0x2749,
390 0x2750, 0x2756, 0x2760, 0x2760, 0x300c, 0x300e, 0x301c, 0x301d,
391 0x302a, 0x302a, 0x302c, 0x302d, 0x3030, 0x3031, 0x3034, 0x3036,
392 0x303c, 0x303c, 0x305e, 0x305f,
395 #ifdef CONFIG_DEBUG_FS
396 static void a3xx_show(struct msm_gpu *gpu, struct seq_file *m)
398 struct drm_device *dev = gpu->dev;
403 mutex_lock(&dev->struct_mutex);
405 gpu->funcs->pm_resume(gpu);
407 seq_printf(m, "status: %08x\n",
408 gpu_read(gpu, REG_A3XX_RBBM_STATUS));
410 /* dump these out in a form that can be parsed by demsm: */
411 seq_printf(m, "IO:region %s 00000000 00020000\n", gpu->name);
412 for (i = 0; i < ARRAY_SIZE(a3xx_registers); i += 2) {
413 uint32_t start = a3xx_registers[i];
414 uint32_t end = a3xx_registers[i+1];
417 for (addr = start; addr <= end; addr++) {
418 uint32_t val = gpu_read(gpu, addr);
419 seq_printf(m, "IO:R %08x %08x\n", addr<<2, val);
423 gpu->funcs->pm_suspend(gpu);
425 mutex_unlock(&dev->struct_mutex);
429 /* would be nice to not have to duplicate the _show() stuff with printk(): */
430 static void a3xx_dump(struct msm_gpu *gpu)
435 printk("status: %08x\n",
436 gpu_read(gpu, REG_A3XX_RBBM_STATUS));
438 /* dump these out in a form that can be parsed by demsm: */
439 printk("IO:region %s 00000000 00020000\n", gpu->name);
440 for (i = 0; i < ARRAY_SIZE(a3xx_registers); i += 2) {
441 uint32_t start = a3xx_registers[i];
442 uint32_t end = a3xx_registers[i+1];
445 for (addr = start; addr <= end; addr++) {
446 uint32_t val = gpu_read(gpu, addr);
447 printk("IO:R %08x %08x\n", addr<<2, val);
452 static const struct adreno_gpu_funcs funcs = {
454 .get_param = adreno_get_param,
455 .hw_init = a3xx_hw_init,
456 .pm_suspend = msm_gpu_pm_suspend,
457 .pm_resume = msm_gpu_pm_resume,
458 .recover = a3xx_recover,
459 .last_fence = adreno_last_fence,
460 .submit = adreno_submit,
461 .flush = adreno_flush,
464 .destroy = a3xx_destroy,
465 #ifdef CONFIG_DEBUG_FS
471 struct msm_gpu *a3xx_gpu_init(struct drm_device *dev)
473 struct a3xx_gpu *a3xx_gpu = NULL;
474 struct adreno_gpu *adreno_gpu;
476 struct platform_device *pdev = a3xx_pdev;
477 struct adreno_platform_config *config;
481 dev_err(dev->dev, "no a3xx device\n");
486 config = pdev->dev.platform_data;
488 a3xx_gpu = kzalloc(sizeof(*a3xx_gpu), GFP_KERNEL);
494 adreno_gpu = &a3xx_gpu->base;
495 gpu = &adreno_gpu->base;
497 get_device(&pdev->dev);
498 a3xx_gpu->pdev = pdev;
500 gpu->fast_rate = config->fast_rate;
501 gpu->slow_rate = config->slow_rate;
502 gpu->bus_freq = config->bus_freq;
503 #ifdef CONFIG_MSM_BUS_SCALING
504 gpu->bus_scale_table = config->bus_scale_table;
507 DBG("fast_rate=%u, slow_rate=%u, bus_freq=%u",
508 gpu->fast_rate, gpu->slow_rate, gpu->bus_freq);
510 ret = adreno_gpu_init(dev, pdev, adreno_gpu, &funcs, config->rev);
514 /* if needed, allocate gmem: */
515 if (adreno_is_a330(adreno_gpu)) {
516 #ifdef CONFIG_MSM_OCMEM
517 /* TODO this is different/missing upstream: */
518 struct ocmem_buf *ocmem_hdl =
519 ocmem_allocate(OCMEM_GRAPHICS, adreno_gpu->gmem);
521 a3xx_gpu->ocmem_hdl = ocmem_hdl;
522 a3xx_gpu->ocmem_base = ocmem_hdl->addr;
523 adreno_gpu->gmem = ocmem_hdl->len;
524 DBG("using %dK of OCMEM at 0x%08x", adreno_gpu->gmem / 1024,
525 a3xx_gpu->ocmem_base);
530 /* TODO we think it is possible to configure the GPU to
531 * restrict access to VRAM carveout. But the required
532 * registers are unknown. For now just bail out and
533 * limp along with just modesetting. If it turns out
534 * to not be possible to restrict access, then we must
535 * implement a cmdstream validator.
537 dev_err(dev->dev, "No memory protection without IOMMU\n");
546 a3xx_destroy(&a3xx_gpu->base.base);
555 #if defined(CONFIG_MSM_BUS_SCALING) && !defined(CONFIG_OF)
556 # include <mach/kgsl.h>
559 static int a3xx_probe(struct platform_device *pdev)
561 static struct adreno_platform_config config = {};
563 struct device_node *child, *node = pdev->dev.of_node;
567 ret = of_property_read_u32(node, "qcom,chipid", &val);
569 dev_err(&pdev->dev, "could not find chipid: %d\n", ret);
573 config.rev = ADRENO_REV((val >> 24) & 0xff,
574 (val >> 16) & 0xff, (val >> 8) & 0xff, val & 0xff);
576 /* find clock rates: */
577 config.fast_rate = 0;
578 config.slow_rate = ~0;
579 for_each_child_of_node(node, child) {
580 if (of_device_is_compatible(child, "qcom,gpu-pwrlevels")) {
581 struct device_node *pwrlvl;
582 for_each_child_of_node(child, pwrlvl) {
583 ret = of_property_read_u32(pwrlvl, "qcom,gpu-freq", &val);
585 dev_err(&pdev->dev, "could not find gpu-freq: %d\n", ret);
588 config.fast_rate = max(config.fast_rate, val);
589 config.slow_rate = min(config.slow_rate, val);
594 if (!config.fast_rate) {
595 dev_err(&pdev->dev, "could not find clk rates\n");
600 struct kgsl_device_platform_data *pdata = pdev->dev.platform_data;
601 uint32_t version = socinfo_get_version();
602 if (cpu_is_apq8064ab()) {
603 config.fast_rate = 450000000;
604 config.slow_rate = 27000000;
606 config.rev = ADRENO_REV(3, 2, 1, 0);
607 } else if (cpu_is_apq8064()) {
608 config.fast_rate = 400000000;
609 config.slow_rate = 27000000;
612 if (SOCINFO_VERSION_MAJOR(version) == 2)
613 config.rev = ADRENO_REV(3, 2, 0, 2);
614 else if ((SOCINFO_VERSION_MAJOR(version) == 1) &&
615 (SOCINFO_VERSION_MINOR(version) == 1))
616 config.rev = ADRENO_REV(3, 2, 0, 1);
618 config.rev = ADRENO_REV(3, 2, 0, 0);
620 } else if (cpu_is_msm8960ab()) {
621 config.fast_rate = 400000000;
622 config.slow_rate = 320000000;
625 if (SOCINFO_VERSION_MINOR(version) == 0)
626 config.rev = ADRENO_REV(3, 2, 1, 0);
628 config.rev = ADRENO_REV(3, 2, 1, 1);
630 } else if (cpu_is_msm8930()) {
631 config.fast_rate = 400000000;
632 config.slow_rate = 27000000;
635 if ((SOCINFO_VERSION_MAJOR(version) == 1) &&
636 (SOCINFO_VERSION_MINOR(version) == 2))
637 config.rev = ADRENO_REV(3, 0, 5, 2);
639 config.rev = ADRENO_REV(3, 0, 5, 0);
642 # ifdef CONFIG_MSM_BUS_SCALING
643 config.bus_scale_table = pdata->bus_scale_table;
646 pdev->dev.platform_data = &config;
651 static int a3xx_remove(struct platform_device *pdev)
657 static const struct of_device_id dt_match[] = {
658 { .compatible = "qcom,kgsl-3d0" },
661 MODULE_DEVICE_TABLE(of, dt_match);
663 static struct platform_driver a3xx_driver = {
665 .remove = a3xx_remove,
668 .of_match_table = dt_match,
672 void __init a3xx_register(void)
674 platform_driver_register(&a3xx_driver);
677 void __exit a3xx_unregister(void)
679 platform_driver_unregister(&a3xx_driver);