int adreno_hw_init(struct msm_gpu *gpu)
{
struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
int ret;
DBG("%s", gpu->name);
ret = msm_gem_get_iova(gpu->rb->bo, gpu->id, &gpu->rb_iova);
if (ret) {
gpu->rb_iova = 0;
dev_err(gpu->dev->dev, "could not map ringbuffer: %d\n", ret);
return ret;
}
/* Setup REG_CP_RB_CNTL: */
adreno_gpu_write(adreno_gpu, REG_ADRENO_CP_RB_CNTL,
/* size is log2(quad-words): */
AXXX_CP_RB_CNTL_BUFSZ(ilog2(gpu->rb->size / 8)) |
AXXX_CP_RB_CNTL_BLKSZ(ilog2(RB_BLKSIZE / 8)) |
(adreno_is_a430(adreno_gpu) ? AXXX_CP_RB_CNTL_NO_UPDATE : 0));
/* Setup ringbuffer address: */
adreno_gpu_write64(adreno_gpu, REG_ADRENO_CP_RB_BASE,
REG_ADRENO_CP_RB_BASE_HI, gpu->rb_iova);
if (!adreno_is_a430(adreno_gpu)) {
adreno_gpu_write64(adreno_gpu, REG_ADRENO_CP_RB_RPTR_ADDR,
REG_ADRENO_CP_RB_RPTR_ADDR_HI,
rbmemptr(adreno_gpu, rptr));
}
return 0;
}
/* Use this helper to read rptr, since a430 doesn't update rptr in memory */
static uint32_t get_rptr(struct adreno_gpu *adreno_gpu)
{
if (adreno_is_a430(adreno_gpu))
return adreno_gpu->memptrs->rptr = adreno_gpu_read(
adreno_gpu, REG_ADRENO_CP_RB_RPTR);
else
return adreno_gpu->memptrs->rptr;
}
/*
* 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);
}
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;
}
void adreno_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit,
struct msm_file_private *ctx)
{
struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
struct msm_drm_private *priv = gpu->dev->dev_private;
struct msm_ringbuffer *ring = gpu->rb;
unsigned i;
for (i = 0; i < submit->nr_cmds; i++) {
switch (submit->cmd[i].type) {
case MSM_SUBMIT_CMD_IB_TARGET_BUF:
/* ignore IB-targets */
break;
case MSM_SUBMIT_CMD_CTX_RESTORE_BUF:
/* ignore if there has not been a ctx switch: */
if (priv->lastctx == ctx)
break;
case MSM_SUBMIT_CMD_BUF:
OUT_PKT3(ring, adreno_is_a430(adreno_gpu) ?
CP_INDIRECT_BUFFER_PFE : CP_INDIRECT_BUFFER_PFD, 2);
OUT_RING(ring, submit->cmd[i].iova);
OUT_RING(ring, submit->cmd[i].size);
OUT_PKT2(ring);
break;
}
}
OUT_PKT0(ring, REG_AXXX_CP_SCRATCH_REG2, 1);
OUT_RING(ring, submit->fence->seqno);
if (adreno_is_a3xx(adreno_gpu) || adreno_is_a4xx(adreno_gpu)) {
/* Flush HLSQ lazy updates to make sure there is nothing
* pending for indirect loads after the timestamp has
* passed:
*/
OUT_PKT3(ring, CP_EVENT_WRITE, 1);
OUT_RING(ring, HLSQ_FLUSH);
OUT_PKT3(ring, CP_WAIT_FOR_IDLE, 1);
OUT_RING(ring, 0x00000000);
}
OUT_PKT3(ring, CP_EVENT_WRITE, 3);
OUT_RING(ring, CACHE_FLUSH_TS);
OUT_RING(ring, rbmemptr(adreno_gpu, fence));
OUT_RING(ring, submit->fence->seqno);
/* we could maybe be clever and only CP_COND_EXEC the interrupt: */
OUT_PKT3(ring, CP_INTERRUPT, 1);
OUT_RING(ring, 0x80000000);
/* Workaround for missing irq issue on 8x16/a306. Unsure if the
* root cause is a platform issue or some a306 quirk, but this
* keeps things humming along:
*/
if (adreno_is_a306(adreno_gpu)) {
OUT_PKT3(ring, CP_WAIT_FOR_IDLE, 1);
OUT_RING(ring, 0x00000000);
OUT_PKT3(ring, CP_INTERRUPT, 1);
OUT_RING(ring, 0x80000000);
}
#if 0
if (adreno_is_a3xx(adreno_gpu)) {
/* Dummy set-constant to trigger context rollover */
OUT_PKT3(ring, CP_SET_CONSTANT, 2);
OUT_RING(ring, CP_REG(REG_A3XX_HLSQ_CL_KERNEL_GROUP_X_REG));
OUT_RING(ring, 0x00000000);
}
#endif
gpu->funcs->flush(gpu);
}
