static int a4xx_hw_init(struct msm_gpu *gpu)
{
struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
struct a4xx_gpu *a4xx_gpu = to_a4xx_gpu(adreno_gpu);
uint32_t *ptr, len;
int i, ret;
if (adreno_is_a420(adreno_gpu)) {
gpu_write(gpu, REG_A4XX_VBIF_ABIT_SORT, 0x0001001F);
gpu_write(gpu, REG_A4XX_VBIF_ABIT_SORT_CONF, 0x000000A4);
gpu_write(gpu, REG_A4XX_VBIF_GATE_OFF_WRREQ_EN, 0x00000001);
gpu_write(gpu, REG_A4XX_VBIF_IN_RD_LIM_CONF0, 0x18181818);
gpu_write(gpu, REG_A4XX_VBIF_IN_RD_LIM_CONF1, 0x00000018);
gpu_write(gpu, REG_A4XX_VBIF_IN_WR_LIM_CONF0, 0x18181818);
gpu_write(gpu, REG_A4XX_VBIF_IN_WR_LIM_CONF1, 0x00000018);
gpu_write(gpu, REG_A4XX_VBIF_ROUND_ROBIN_QOS_ARB, 0x00000003);
} else if (adreno_is_a430(adreno_gpu)) {
gpu_write(gpu, REG_A4XX_VBIF_GATE_OFF_WRREQ_EN, 0x00000001);
gpu_write(gpu, REG_A4XX_VBIF_IN_RD_LIM_CONF0, 0x18181818);
gpu_write(gpu, REG_A4XX_VBIF_IN_RD_LIM_CONF1, 0x00000018);
gpu_write(gpu, REG_A4XX_VBIF_IN_WR_LIM_CONF0, 0x18181818);
gpu_write(gpu, REG_A4XX_VBIF_IN_WR_LIM_CONF1, 0x00000018);
gpu_write(gpu, REG_A4XX_VBIF_ROUND_ROBIN_QOS_ARB, 0x00000003);
} else {
BUG();
}
/* Make all blocks contribute to the GPU BUSY perf counter */
gpu_write(gpu, REG_A4XX_RBBM_GPU_BUSY_MASKED, 0xffffffff);
/* Tune the hystersis counters for SP and CP idle detection */
gpu_write(gpu, REG_A4XX_RBBM_SP_HYST_CNT, 0x10);
gpu_write(gpu, REG_A4XX_RBBM_WAIT_IDLE_CLOCKS_CTL, 0x10);
if (adreno_is_a430(adreno_gpu)) {
gpu_write(gpu, REG_A4XX_RBBM_WAIT_IDLE_CLOCKS_CTL2, 0x30);
}
/* Enable the RBBM error reporting bits */
gpu_write(gpu, REG_A4XX_RBBM_AHB_CTL0, 0x00000001);
/* Enable AHB error reporting*/
gpu_write(gpu, REG_A4XX_RBBM_AHB_CTL1, 0xa6ffffff);
/* Enable power counters*/
gpu_write(gpu, REG_A4XX_RBBM_RBBM_CTL, 0x00000030);
/*
* Turn on hang detection - this spews a lot of useful information
* into the RBBM registers on a hang:
*/
gpu_write(gpu, REG_A4XX_RBBM_INTERFACE_HANG_INT_CTL,
(1 << 30) | 0xFFFF);
gpu_write(gpu, REG_A4XX_RB_GMEM_BASE_ADDR,
(unsigned int)(a4xx_gpu->ocmem_base >> 14));
/* Turn on performance counters: */
gpu_write(gpu, REG_A4XX_RBBM_PERFCTR_CTL, 0x01);
/* use the first CP counter for timestamp queries.. userspace may set
* this as well but it selects the same counter/countable:
*/
gpu_write(gpu, REG_A4XX_CP_PERFCTR_CP_SEL_0, CP_ALWAYS_COUNT);
if (adreno_is_a430(adreno_gpu))
gpu_write(gpu, REG_A4XX_UCHE_CACHE_WAYS_VFD, 0x07);
/* Disable L2 bypass to avoid UCHE out of bounds errors */
gpu_write(gpu, REG_A4XX_UCHE_TRAP_BASE_LO, 0xffff0000);
gpu_write(gpu, REG_A4XX_UCHE_TRAP_BASE_HI, 0xffff0000);
gpu_write(gpu, REG_A4XX_CP_DEBUG, (1 << 25) |
(adreno_is_a420(adreno_gpu) ? (1 << 29) : 0));
/* On A430 enable SP regfile sleep for power savings */
/* TODO downstream does this for !420, so maybe applies for 405 too? */
if (!adreno_is_a420(adreno_gpu)) {
gpu_write(gpu, REG_A4XX_RBBM_SP_REGFILE_SLEEP_CNTL_0,
0x00000441);
gpu_write(gpu, REG_A4XX_RBBM_SP_REGFILE_SLEEP_CNTL_1,
0x00000441);
}
a4xx_enable_hwcg(gpu);
/*
* For A420 set RBBM_CLOCK_DELAY_HLSQ.CGC_HLSQ_TP_EARLY_CYC >= 2
* due to timing issue with HLSQ_TP_CLK_EN
*/
if (adreno_is_a420(adreno_gpu)) {
unsigned int val;
val = gpu_read(gpu, REG_A4XX_RBBM_CLOCK_DELAY_HLSQ);
val &= ~A4XX_CGC_HLSQ_EARLY_CYC__MASK;
val |= 2 << A4XX_CGC_HLSQ_EARLY_CYC__SHIFT;
gpu_write(gpu, REG_A4XX_RBBM_CLOCK_DELAY_HLSQ, val);
}
/* setup access protection: */
gpu_write(gpu, REG_A4XX_CP_PROTECT_CTRL, 0x00000007);
/* RBBM registers */
gpu_write(gpu, REG_A4XX_CP_PROTECT(0), 0x62000010);
gpu_write(gpu, REG_A4XX_CP_PROTECT(1), 0x63000020);
gpu_write(gpu, REG_A4XX_CP_PROTECT(2), 0x64000040);
gpu_write(gpu, REG_A4XX_CP_PROTECT(3), 0x65000080);
gpu_write(gpu, REG_A4XX_CP_PROTECT(4), 0x66000100);
gpu_write(gpu, REG_A4XX_CP_PROTECT(5), 0x64000200);
/* CP registers */
gpu_write(gpu, REG_A4XX_CP_PROTECT(6), 0x67000800);
gpu_write(gpu, REG_A4XX_CP_PROTECT(7), 0x64001600);
/* RB registers */
gpu_write(gpu, REG_A4XX_CP_PROTECT(8), 0x60003300);
/* HLSQ registers */
gpu_write(gpu, REG_A4XX_CP_PROTECT(9), 0x60003800);
/* VPC registers */
gpu_write(gpu, REG_A4XX_CP_PROTECT(10), 0x61003980);
/* SMMU registers */
gpu_write(gpu, REG_A4XX_CP_PROTECT(11), 0x6e010000);
gpu_write(gpu, REG_A4XX_RBBM_INT_0_MASK, A4XX_INT0_MASK);
ret = adreno_hw_init(gpu);
if (ret)
return ret;
/* Load PM4: */
ptr = (uint32_t *)(adreno_gpu->pm4->data);
len = adreno_gpu->pm4->size / 4;
DBG("loading PM4 ucode version: %u", ptr[0]);
gpu_write(gpu, REG_A4XX_CP_ME_RAM_WADDR, 0);
for (i = 1; i < len; i++)
gpu_write(gpu, REG_A4XX_CP_ME_RAM_DATA, ptr[i]);
/* Load PFP: */
ptr = (uint32_t *)(adreno_gpu->pfp->data);
len = adreno_gpu->pfp->size / 4;
DBG("loading PFP ucode version: %u", ptr[0]);
gpu_write(gpu, REG_A4XX_CP_PFP_UCODE_ADDR, 0);
for (i = 1; i < len; i++)
gpu_write(gpu, REG_A4XX_CP_PFP_UCODE_DATA, ptr[i]);
/* clear ME_HALT to start micro engine */
gpu_write(gpu, REG_A4XX_CP_ME_CNTL, 0);
return a4xx_me_init(gpu) ? 0 : -EINVAL;
}
/*
* a4xx_enable_hwcg() - Program the clock control registers
* @device: The adreno device pointer
*/
static void a4xx_enable_hwcg(struct msm_gpu *gpu)
{
struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
unsigned int i;
for (i = 0; i < 4; i++)
gpu_write(gpu, REG_A4XX_RBBM_CLOCK_CTL_TP(i), 0x02222202);
for (i = 0; i < 4; i++)
gpu_write(gpu, REG_A4XX_RBBM_CLOCK_CTL2_TP(i), 0x00002222);
for (i = 0; i < 4; i++)
gpu_write(gpu, REG_A4XX_RBBM_CLOCK_HYST_TP(i), 0x0E739CE7);
for (i = 0; i < 4; i++)
gpu_write(gpu, REG_A4XX_RBBM_CLOCK_DELAY_TP(i), 0x00111111);
for (i = 0; i < 4; i++)
gpu_write(gpu, REG_A4XX_RBBM_CLOCK_CTL_SP(i), 0x22222222);
for (i = 0; i < 4; i++)
gpu_write(gpu, REG_A4XX_RBBM_CLOCK_CTL2_SP(i), 0x00222222);
for (i = 0; i < 4; i++)
gpu_write(gpu, REG_A4XX_RBBM_CLOCK_HYST_SP(i), 0x00000104);
for (i = 0; i < 4; i++)
gpu_write(gpu, REG_A4XX_RBBM_CLOCK_DELAY_SP(i), 0x00000081);
gpu_write(gpu, REG_A4XX_RBBM_CLOCK_CTL_UCHE, 0x22222222);
gpu_write(gpu, REG_A4XX_RBBM_CLOCK_CTL2_UCHE, 0x02222222);
gpu_write(gpu, REG_A4XX_RBBM_CLOCK_CTL3_UCHE, 0x00000000);
gpu_write(gpu, REG_A4XX_RBBM_CLOCK_CTL4_UCHE, 0x00000000);
gpu_write(gpu, REG_A4XX_RBBM_CLOCK_HYST_UCHE, 0x00004444);
gpu_write(gpu, REG_A4XX_RBBM_CLOCK_DELAY_UCHE, 0x00001112);
for (i = 0; i < 4; i++)
gpu_write(gpu, REG_A4XX_RBBM_CLOCK_CTL_RB(i), 0x22222222);
/* Disable L1 clocking in A420 due to CCU issues with it */
for (i = 0; i < 4; i++) {
if (adreno_is_a420(adreno_gpu)) {
gpu_write(gpu, REG_A4XX_RBBM_CLOCK_CTL2_RB(i),
0x00002020);
} else {
gpu_write(gpu, REG_A4XX_RBBM_CLOCK_CTL2_RB(i),
0x00022020);
}
}
for (i = 0; i < 4; i++) {
gpu_write(gpu, REG_A4XX_RBBM_CLOCK_CTL_MARB_CCU(i),
0x00000922);
}
for (i = 0; i < 4; i++) {
gpu_write(gpu, REG_A4XX_RBBM_CLOCK_HYST_RB_MARB_CCU(i),
0x00000000);
}
for (i = 0; i < 4; i++) {
gpu_write(gpu, REG_A4XX_RBBM_CLOCK_DELAY_RB_MARB_CCU_L1(i),
0x00000001);
}
gpu_write(gpu, REG_A4XX_RBBM_CLOCK_MODE_GPC, 0x02222222);
gpu_write(gpu, REG_A4XX_RBBM_CLOCK_HYST_GPC, 0x04100104);
gpu_write(gpu, REG_A4XX_RBBM_CLOCK_DELAY_GPC, 0x00022222);
gpu_write(gpu, REG_A4XX_RBBM_CLOCK_CTL_COM_DCOM, 0x00000022);
gpu_write(gpu, REG_A4XX_RBBM_CLOCK_HYST_COM_DCOM, 0x0000010F);
gpu_write(gpu, REG_A4XX_RBBM_CLOCK_DELAY_COM_DCOM, 0x00000022);
gpu_write(gpu, REG_A4XX_RBBM_CLOCK_CTL_TSE_RAS_RBBM, 0x00222222);
gpu_write(gpu, REG_A4XX_RBBM_CLOCK_HYST_TSE_RAS_RBBM, 0x00004104);
gpu_write(gpu, REG_A4XX_RBBM_CLOCK_DELAY_TSE_RAS_RBBM, 0x00000222);
gpu_write(gpu, REG_A4XX_RBBM_CLOCK_CTL_HLSQ , 0x00000000);
gpu_write(gpu, REG_A4XX_RBBM_CLOCK_HYST_HLSQ, 0x00000000);
gpu_write(gpu, REG_A4XX_RBBM_CLOCK_DELAY_HLSQ, 0x00220000);
/* Early A430's have a timing issue with SP/TP power collapse;
disabling HW clock gating prevents it. */
if (adreno_is_a430(adreno_gpu) && adreno_gpu->rev.patchid < 2)
gpu_write(gpu, REG_A4XX_RBBM_CLOCK_CTL, 0);
else
gpu_write(gpu, REG_A4XX_RBBM_CLOCK_CTL, 0xAAAAAAAA);
gpu_write(gpu, REG_A4XX_RBBM_CLOCK_CTL2, 0);
}
/**
* _ringbuffer_bootstrap_ucode() - Bootstrap GPU Ucode
* @rb: Pointer to adreno ringbuffer
* @load_jt: If non zero only load Jump tables
*
* Bootstrap ucode for GPU
* load_jt == 0, bootstrap full microcode
* load_jt == 1, bootstrap jump tables of microcode
*
* For example a bootstrap packet would like below
* Setup a type3 bootstrap packet
* PFP size to bootstrap
* PFP addr to write the PFP data
* PM4 size to bootstrap
* PM4 addr to write the PM4 data
* PFP dwords from microcode to bootstrap
* PM4 size dwords from microcode to bootstrap
*/
static int _ringbuffer_bootstrap_ucode(struct adreno_ringbuffer *rb,
unsigned int load_jt)
{
unsigned int *cmds, bootstrap_size, rb_size;
int i = 0;
int ret;
struct kgsl_device *device = rb->device;
struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
unsigned int pm4_size, pm4_idx, pm4_addr, pfp_size, pfp_idx, pfp_addr;
/* Only bootstrap jump tables of ucode */
if (load_jt) {
pm4_idx = adreno_dev->gpucore->pm4_jt_idx;
pm4_addr = adreno_dev->gpucore->pm4_jt_addr;
pfp_idx = adreno_dev->gpucore->pfp_jt_idx;
pfp_addr = adreno_dev->gpucore->pfp_jt_addr;
} else {
/* Bootstrap full ucode */
pm4_idx = 1;
pm4_addr = 0;
pfp_idx = 1;
pfp_addr = 0;
}
pm4_size = (adreno_dev->pm4_fw_size - pm4_idx);
pfp_size = (adreno_dev->pfp_fw_size - pfp_idx);
bootstrap_size = (pm4_size + pfp_size + 5);
/*
* Overwrite the first entry in the jump table with the special
* bootstrap opcode
*/
if (adreno_is_a4xx(adreno_dev)) {
adreno_writereg(adreno_dev, ADRENO_REG_CP_PFP_UCODE_ADDR,
0x400);
adreno_writereg(adreno_dev, ADRENO_REG_CP_PFP_UCODE_DATA,
0x6f0009);
/*
* The support packets (the RMW and INTERRUPT) that are sent
* after the bootstrap packet should not be included in the size
* of the bootstrap packet but we do need to reserve enough
* space for those too
*/
rb_size = bootstrap_size + 6;
} else {
adreno_writereg(adreno_dev, ADRENO_REG_CP_PFP_UCODE_ADDR,
0x200);
adreno_writereg(adreno_dev, ADRENO_REG_CP_PFP_UCODE_DATA,
0x6f0005);
rb_size = bootstrap_size;
}
/* clear ME_HALT to start micro engine */
adreno_writereg(adreno_dev, ADRENO_REG_CP_ME_CNTL, 0);
cmds = adreno_ringbuffer_allocspace(rb, rb_size);
if (IS_ERR(cmds))
return PTR_ERR(cmds);
if (cmds == NULL)
return -ENOSPC;
/* Construct the packet that bootsraps the ucode */
*cmds++ = cp_type3_packet(CP_BOOTSTRAP_UCODE, (bootstrap_size - 1));
*cmds++ = pfp_size;
*cmds++ = pfp_addr;
*cmds++ = pm4_size;
*cmds++ = pm4_addr;
/**
* Theory of operation:
*
* In A4x, we cannot have the PFP executing instructions while its instruction
* RAM is loading. We load the PFP's instruction RAM using type-0 writes
* from the ME.
*
* To make sure the PFP is not fetching instructions at the same time,
* we put it in a one-instruction loop:
* mvc (ME), (ringbuffer)
* which executes repeatedly until all of the data has been moved from
* the ring buffer to the ME.
*/
if (adreno_is_a4xx(adreno_dev)) {
for (i = pm4_idx; i < adreno_dev->pm4_fw_size; i++)
*cmds++ = adreno_dev->pm4_fw[i];
for (i = pfp_idx; i < adreno_dev->pfp_fw_size; i++)
*cmds++ = adreno_dev->pfp_fw[i];
*cmds++ = cp_type3_packet(CP_REG_RMW, 3);
*cmds++ = 0x20000000 + A4XX_CP_RB_WPTR;
*cmds++ = 0xffffffff;
*cmds++ = 0x00000002;
*cmds++ = cp_type3_packet(CP_INTERRUPT, 1);
*cmds++ = 0;
rb->wptr = rb->wptr - 2;
adreno_ringbuffer_submit(rb, NULL);
rb->wptr = rb->wptr + 2;
} else {
for (i = pfp_idx; i < adreno_dev->pfp_fw_size; i++)
*cmds++ = adreno_dev->pfp_fw[i];
for (i = pm4_idx; i < adreno_dev->pm4_fw_size; i++)
*cmds++ = adreno_dev->pm4_fw[i];
adreno_ringbuffer_submit(rb, NULL);
}
/* idle device to validate bootstrap */
ret = adreno_spin_idle(device);
if (ret) {
KGSL_DRV_ERR(rb->device,
"microcode bootstrap failed to idle\n");
kgsl_device_snapshot(device, NULL);
}
/* Clear the chicken bit for speed up on A430 and its derivatives */
if (!adreno_is_a420(adreno_dev))
kgsl_regwrite(device, A4XX_CP_DEBUG,
A4XX_CP_DEBUG_DEFAULT & ~(1 << 14));
return ret;
}