static int a3xx_hw_init(struct msm_gpu *gpu)
{
struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
struct a3xx_gpu *a3xx_gpu = to_a3xx_gpu(adreno_gpu);
uint32_t *ptr, len;
int i, ret;
DBG("%s", gpu->name);
if (adreno_is_a305(adreno_gpu)) {
/* Set up 16 deep read/write request queues: */
gpu_write(gpu, REG_A3XX_VBIF_IN_RD_LIM_CONF0, 0x10101010);
gpu_write(gpu, REG_A3XX_VBIF_IN_RD_LIM_CONF1, 0x10101010);
gpu_write(gpu, REG_A3XX_VBIF_OUT_RD_LIM_CONF0, 0x10101010);
gpu_write(gpu, REG_A3XX_VBIF_OUT_WR_LIM_CONF0, 0x10101010);
gpu_write(gpu, REG_A3XX_VBIF_DDR_OUT_MAX_BURST, 0x0000303);
gpu_write(gpu, REG_A3XX_VBIF_IN_WR_LIM_CONF0, 0x10101010);
gpu_write(gpu, REG_A3XX_VBIF_IN_WR_LIM_CONF1, 0x10101010);
/* Enable WR-REQ: */
gpu_write(gpu, REG_A3XX_VBIF_GATE_OFF_WRREQ_EN, 0x0000ff);
/* Set up round robin arbitration between both AXI ports: */
gpu_write(gpu, REG_A3XX_VBIF_ARB_CTL, 0x00000030);
/* Set up AOOO: */
gpu_write(gpu, REG_A3XX_VBIF_OUT_AXI_AOOO_EN, 0x0000003c);
gpu_write(gpu, REG_A3XX_VBIF_OUT_AXI_AOOO, 0x003c003c);
} else if (adreno_is_a320(adreno_gpu)) {
/* Set up 16 deep read/write request queues: */
gpu_write(gpu, REG_A3XX_VBIF_IN_RD_LIM_CONF0, 0x10101010);
gpu_write(gpu, REG_A3XX_VBIF_IN_RD_LIM_CONF1, 0x10101010);
gpu_write(gpu, REG_A3XX_VBIF_OUT_RD_LIM_CONF0, 0x10101010);
gpu_write(gpu, REG_A3XX_VBIF_OUT_WR_LIM_CONF0, 0x10101010);
gpu_write(gpu, REG_A3XX_VBIF_DDR_OUT_MAX_BURST, 0x0000303);
gpu_write(gpu, REG_A3XX_VBIF_IN_WR_LIM_CONF0, 0x10101010);
gpu_write(gpu, REG_A3XX_VBIF_IN_WR_LIM_CONF1, 0x10101010);
/* Enable WR-REQ: */
gpu_write(gpu, REG_A3XX_VBIF_GATE_OFF_WRREQ_EN, 0x0000ff);
/* Set up round robin arbitration between both AXI ports: */
gpu_write(gpu, REG_A3XX_VBIF_ARB_CTL, 0x00000030);
/* Set up AOOO: */
gpu_write(gpu, REG_A3XX_VBIF_OUT_AXI_AOOO_EN, 0x0000003c);
gpu_write(gpu, REG_A3XX_VBIF_OUT_AXI_AOOO, 0x003c003c);
/* Enable 1K sort: */
gpu_write(gpu, REG_A3XX_VBIF_ABIT_SORT, 0x000000ff);
gpu_write(gpu, REG_A3XX_VBIF_ABIT_SORT_CONF, 0x000000a4);
} else if (adreno_is_a330v2(adreno_gpu)) {
/*
* Most of the VBIF registers on 8974v2 have the correct
* values at power on, so we won't modify those if we don't
* need to
*/
/* Enable 1k sort: */
gpu_write(gpu, REG_A3XX_VBIF_ABIT_SORT, 0x0001003f);
gpu_write(gpu, REG_A3XX_VBIF_ABIT_SORT_CONF, 0x000000a4);
/* Enable WR-REQ: */
gpu_write(gpu, REG_A3XX_VBIF_GATE_OFF_WRREQ_EN, 0x00003f);
gpu_write(gpu, REG_A3XX_VBIF_DDR_OUT_MAX_BURST, 0x0000303);
/* Set up VBIF_ROUND_ROBIN_QOS_ARB: */
gpu_write(gpu, REG_A3XX_VBIF_ROUND_ROBIN_QOS_ARB, 0x0003);
} else if (adreno_is_a330(adreno_gpu)) {
/* Set up 16 deep read/write request queues: */
gpu_write(gpu, REG_A3XX_VBIF_IN_RD_LIM_CONF0, 0x18181818);
gpu_write(gpu, REG_A3XX_VBIF_IN_RD_LIM_CONF1, 0x18181818);
gpu_write(gpu, REG_A3XX_VBIF_OUT_RD_LIM_CONF0, 0x18181818);
gpu_write(gpu, REG_A3XX_VBIF_OUT_WR_LIM_CONF0, 0x18181818);
gpu_write(gpu, REG_A3XX_VBIF_DDR_OUT_MAX_BURST, 0x0000303);
gpu_write(gpu, REG_A3XX_VBIF_IN_WR_LIM_CONF0, 0x18181818);
gpu_write(gpu, REG_A3XX_VBIF_IN_WR_LIM_CONF1, 0x18181818);
/* Enable WR-REQ: */
gpu_write(gpu, REG_A3XX_VBIF_GATE_OFF_WRREQ_EN, 0x00003f);
/* Set up round robin arbitration between both AXI ports: */
gpu_write(gpu, REG_A3XX_VBIF_ARB_CTL, 0x00000030);
/* Set up VBIF_ROUND_ROBIN_QOS_ARB: */
gpu_write(gpu, REG_A3XX_VBIF_ROUND_ROBIN_QOS_ARB, 0x0001);
/* Set up AOOO: */
gpu_write(gpu, REG_A3XX_VBIF_OUT_AXI_AOOO_EN, 0x0000003f);
gpu_write(gpu, REG_A3XX_VBIF_OUT_AXI_AOOO, 0x003f003f);
/* Enable 1K sort: */
gpu_write(gpu, REG_A3XX_VBIF_ABIT_SORT, 0x0001003f);
gpu_write(gpu, REG_A3XX_VBIF_ABIT_SORT_CONF, 0x000000a4);
/* Disable VBIF clock gating. This is to enable AXI running
* higher frequency than GPU:
*/
gpu_write(gpu, REG_A3XX_VBIF_CLKON, 0x00000001);
} else {
BUG();
}
/* Make all blocks contribute to the GPU BUSY perf counter: */
gpu_write(gpu, REG_A3XX_RBBM_GPU_BUSY_MASKED, 0xffffffff);
/* Tune the hystersis counters for SP and CP idle detection: */
gpu_write(gpu, REG_A3XX_RBBM_SP_HYST_CNT, 0x10);
gpu_write(gpu, REG_A3XX_RBBM_WAIT_IDLE_CLOCKS_CTL, 0x10);
/* Enable the RBBM error reporting bits. This lets us get
* useful information on failure:
*/
gpu_write(gpu, REG_A3XX_RBBM_AHB_CTL0, 0x00000001);
/* Enable AHB error reporting: */
gpu_write(gpu, REG_A3XX_RBBM_AHB_CTL1, 0xa6ffffff);
/* Turn on the power counters: */
gpu_write(gpu, REG_A3XX_RBBM_RBBM_CTL, 0x00030000);
/* Turn on hang detection - this spews a lot of useful information
* into the RBBM registers on a hang:
*/
gpu_write(gpu, REG_A3XX_RBBM_INTERFACE_HANG_INT_CTL, 0x00010fff);
/* Enable 64-byte cacheline size. HW Default is 32-byte (0x000000E0): */
gpu_write(gpu, REG_A3XX_UCHE_CACHE_MODE_CONTROL_REG, 0x00000001);
/* Enable Clock gating: */
if (adreno_is_a320(adreno_gpu))
gpu_write(gpu, REG_A3XX_RBBM_CLOCK_CTL, 0xbfffffff);
else if (adreno_is_a330v2(adreno_gpu))
gpu_write(gpu, REG_A3XX_RBBM_CLOCK_CTL, 0xaaaaaaaa);
else if (adreno_is_a330(adreno_gpu))
gpu_write(gpu, REG_A3XX_RBBM_CLOCK_CTL, 0xbffcffff);
if (adreno_is_a330v2(adreno_gpu))
gpu_write(gpu, REG_A3XX_RBBM_GPR0_CTL, 0x05515455);
else if (adreno_is_a330(adreno_gpu))
gpu_write(gpu, REG_A3XX_RBBM_GPR0_CTL, 0x00000000);
/* Set the OCMEM base address for A330, etc */
if (a3xx_gpu->ocmem_hdl) {
gpu_write(gpu, REG_A3XX_RB_GMEM_BASE_ADDR,
(unsigned int)(a3xx_gpu->ocmem_base >> 14));
}
int adreno_ringbuffer_start(struct adreno_ringbuffer *rb, unsigned int init_ram)
{
int status;
/*cp_rb_cntl_u cp_rb_cntl; */
union reg_cp_rb_cntl cp_rb_cntl;
unsigned int rb_cntl;
struct kgsl_device *device = rb->device;
struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
if (rb->flags & KGSL_FLAGS_STARTED)
return 0;
if (init_ram)
rb->timestamp[KGSL_MEMSTORE_GLOBAL] = 0;
kgsl_sharedmem_set(&rb->memptrs_desc, 0, 0,
sizeof(struct kgsl_rbmemptrs));
kgsl_sharedmem_set(&rb->buffer_desc, 0, 0xAA,
(rb->sizedwords << 2));
if (adreno_is_a2xx(adreno_dev)) {
adreno_regwrite(device, REG_CP_RB_WPTR_BASE,
(rb->memptrs_desc.gpuaddr
+ GSL_RB_MEMPTRS_WPTRPOLL_OFFSET));
/* setup WPTR delay */
adreno_regwrite(device, REG_CP_RB_WPTR_DELAY,
0 /*0x70000010 */);
}
/*setup REG_CP_RB_CNTL */
adreno_regread(device, REG_CP_RB_CNTL, &rb_cntl);
cp_rb_cntl.val = rb_cntl;
/*
* The size of the ringbuffer in the hardware is the log2
* representation of the size in quadwords (sizedwords / 2)
*/
cp_rb_cntl.f.rb_bufsz = ilog2(rb->sizedwords >> 1);
/*
* Specify the quadwords to read before updating mem RPTR.
* Like above, pass the log2 representation of the blocksize
* in quadwords.
*/
cp_rb_cntl.f.rb_blksz = ilog2(KGSL_RB_BLKSIZE >> 3);
if (adreno_is_a2xx(adreno_dev)) {
/* WPTR polling */
cp_rb_cntl.f.rb_poll_en = GSL_RB_CNTL_POLL_EN;
}
/* mem RPTR writebacks */
cp_rb_cntl.f.rb_no_update = GSL_RB_CNTL_NO_UPDATE;
adreno_regwrite(device, REG_CP_RB_CNTL, cp_rb_cntl.val);
adreno_regwrite(device, REG_CP_RB_BASE, rb->buffer_desc.gpuaddr);
adreno_regwrite(device, REG_CP_RB_RPTR_ADDR,
rb->memptrs_desc.gpuaddr +
GSL_RB_MEMPTRS_RPTR_OFFSET);
if (adreno_is_a3xx(adreno_dev)) {
/* enable access protection to privileged registers */
adreno_regwrite(device, A3XX_CP_PROTECT_CTRL, 0x00000007);
/* RBBM registers */
adreno_regwrite(device, A3XX_CP_PROTECT_REG_0, 0x63000040);
adreno_regwrite(device, A3XX_CP_PROTECT_REG_1, 0x62000080);
adreno_regwrite(device, A3XX_CP_PROTECT_REG_2, 0x600000CC);
adreno_regwrite(device, A3XX_CP_PROTECT_REG_3, 0x60000108);
adreno_regwrite(device, A3XX_CP_PROTECT_REG_4, 0x64000140);
adreno_regwrite(device, A3XX_CP_PROTECT_REG_5, 0x66000400);
/* CP registers */
adreno_regwrite(device, A3XX_CP_PROTECT_REG_6, 0x65000700);
adreno_regwrite(device, A3XX_CP_PROTECT_REG_7, 0x610007D8);
adreno_regwrite(device, A3XX_CP_PROTECT_REG_8, 0x620007E0);
adreno_regwrite(device, A3XX_CP_PROTECT_REG_9, 0x61001178);
adreno_regwrite(device, A3XX_CP_PROTECT_REG_A, 0x64001180);
/* RB registers */
adreno_regwrite(device, A3XX_CP_PROTECT_REG_B, 0x60003300);
/* VBIF registers */
adreno_regwrite(device, A3XX_CP_PROTECT_REG_C, 0x6B00C000);
}
if (adreno_is_a2xx(adreno_dev)) {
/* explicitly clear all cp interrupts */
adreno_regwrite(device, REG_CP_INT_ACK, 0xFFFFFFFF);
}
/* setup scratch/timestamp */
adreno_regwrite(device, REG_SCRATCH_ADDR, device->memstore.gpuaddr +
KGSL_MEMSTORE_OFFSET(KGSL_MEMSTORE_GLOBAL,
soptimestamp));
adreno_regwrite(device, REG_SCRATCH_UMSK,
GSL_RB_MEMPTRS_SCRATCH_MASK);
/* load the CP ucode */
status = adreno_ringbuffer_load_pm4_ucode(device);
if (status != 0)
return status;
/* load the prefetch parser ucode */
status = adreno_ringbuffer_load_pfp_ucode(device);
if (status != 0)
return status;
if (adreno_is_a305(adreno_dev) || adreno_is_a320(adreno_dev))
adreno_regwrite(device, REG_CP_QUEUE_THRESHOLDS, 0x000F0602);
rb->rptr = 0;
rb->wptr = 0;
/* clear ME_HALT to start micro engine */
adreno_regwrite(device, REG_CP_ME_CNTL, 0);
/* ME init is GPU specific, so jump into the sub-function */
adreno_dev->gpudev->rb_init(adreno_dev, rb);
/* idle device to validate ME INIT */
status = adreno_idle(device, KGSL_TIMEOUT_DEFAULT);
if (status == 0)
rb->flags |= KGSL_FLAGS_STARTED;
return status;
}
int adreno_ringbuffer_start(struct adreno_ringbuffer *rb, unsigned int init_ram)
{
int status;
union reg_cp_rb_cntl cp_rb_cntl;
unsigned int rb_cntl;
struct kgsl_device *device = rb->device;
struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
if (rb->flags & KGSL_FLAGS_STARTED)
return 0;
if (init_ram)
rb->timestamp[KGSL_MEMSTORE_GLOBAL] = 0;
kgsl_sharedmem_set(&rb->memptrs_desc, 0, 0,
sizeof(struct kgsl_rbmemptrs));
kgsl_sharedmem_set(&rb->buffer_desc, 0, 0xAA,
(rb->sizedwords << 2));
if (adreno_is_a2xx(adreno_dev)) {
adreno_regwrite(device, REG_CP_RB_WPTR_BASE,
(rb->memptrs_desc.gpuaddr
+ GSL_RB_MEMPTRS_WPTRPOLL_OFFSET));
adreno_regwrite(device, REG_CP_RB_WPTR_DELAY,
0 );
}
adreno_regread(device, REG_CP_RB_CNTL, &rb_cntl);
cp_rb_cntl.val = rb_cntl;
cp_rb_cntl.f.rb_bufsz = ilog2(rb->sizedwords >> 1);
cp_rb_cntl.f.rb_blksz = ilog2(KGSL_RB_BLKSIZE >> 3);
if (adreno_is_a2xx(adreno_dev)) {
cp_rb_cntl.f.rb_poll_en = GSL_RB_CNTL_POLL_EN;
}
cp_rb_cntl.f.rb_no_update = GSL_RB_CNTL_NO_UPDATE;
adreno_regwrite(device, REG_CP_RB_CNTL, cp_rb_cntl.val);
adreno_regwrite(device, REG_CP_RB_BASE, rb->buffer_desc.gpuaddr);
adreno_regwrite(device, REG_CP_RB_RPTR_ADDR,
rb->memptrs_desc.gpuaddr +
GSL_RB_MEMPTRS_RPTR_OFFSET);
if (adreno_is_a3xx(adreno_dev)) {
adreno_regwrite(device, A3XX_CP_PROTECT_CTRL, 0x00000007);
adreno_regwrite(device, A3XX_CP_PROTECT_REG_0, 0x63000040);
adreno_regwrite(device, A3XX_CP_PROTECT_REG_1, 0x62000080);
adreno_regwrite(device, A3XX_CP_PROTECT_REG_2, 0x600000CC);
adreno_regwrite(device, A3XX_CP_PROTECT_REG_3, 0x60000108);
adreno_regwrite(device, A3XX_CP_PROTECT_REG_4, 0x64000140);
adreno_regwrite(device, A3XX_CP_PROTECT_REG_5, 0x66000400);
adreno_regwrite(device, A3XX_CP_PROTECT_REG_6, 0x65000700);
adreno_regwrite(device, A3XX_CP_PROTECT_REG_7, 0x610007D8);
adreno_regwrite(device, A3XX_CP_PROTECT_REG_8, 0x620007E0);
adreno_regwrite(device, A3XX_CP_PROTECT_REG_9, 0x61001178);
adreno_regwrite(device, A3XX_CP_PROTECT_REG_A, 0x64001180);
adreno_regwrite(device, A3XX_CP_PROTECT_REG_B, 0x60003300);
adreno_regwrite(device, A3XX_CP_PROTECT_REG_C, 0x6B00C000);
}
if (adreno_is_a2xx(adreno_dev)) {
adreno_regwrite(device, REG_CP_INT_ACK, 0xFFFFFFFF);
}
adreno_regwrite(device, REG_SCRATCH_ADDR, device->memstore.gpuaddr +
KGSL_MEMSTORE_OFFSET(KGSL_MEMSTORE_GLOBAL,
soptimestamp));
adreno_regwrite(device, REG_SCRATCH_UMSK,
GSL_RB_MEMPTRS_SCRATCH_MASK);
status = adreno_ringbuffer_load_pm4_ucode(device);
if (status != 0)
return status;
status = adreno_ringbuffer_load_pfp_ucode(device);
if (status != 0)
return status;
if (adreno_is_a305(adreno_dev) || adreno_is_a320(adreno_dev))
adreno_regwrite(device, REG_CP_QUEUE_THRESHOLDS, 0x000E0602);
rb->rptr = 0;
rb->wptr = 0;
adreno_regwrite(device, REG_CP_ME_CNTL, 0);
adreno_dev->gpudev->rb_init(adreno_dev, rb);
status = adreno_idle(device, KGSL_TIMEOUT_DEFAULT);
if (status == 0)
rb->flags |= KGSL_FLAGS_STARTED;
return status;
}
