/**
* adreno_context_restore() - generic context restore handler
* @adreno_dev: the device
* @context: the context
*
* Basic context restore handler that writes the context identifier
* to the ringbuffer and issues pagetable switch commands if necessary.
*/
static int adreno_context_restore(struct adreno_device *adreno_dev,
struct adreno_context *context)
{
struct kgsl_device *device;
unsigned int cmds[8];
if (adreno_dev == NULL || context == NULL)
return -EINVAL;
device = &adreno_dev->dev;
/* write the context identifier to the ringbuffer */
cmds[0] = cp_nop_packet(1);
cmds[1] = KGSL_CONTEXT_TO_MEM_IDENTIFIER;
cmds[2] = cp_type3_packet(CP_MEM_WRITE, 2);
cmds[3] = device->memstore.gpuaddr +
KGSL_MEMSTORE_OFFSET(KGSL_MEMSTORE_GLOBAL, current_context);
cmds[4] = context->base.id;
/* Flush the UCHE for new context */
cmds[5] = cp_type0_packet(
adreno_getreg(adreno_dev, ADRENO_REG_UCHE_INVALIDATE0), 2);
cmds[6] = 0;
if (adreno_is_a4xx(adreno_dev))
cmds[7] = 0x12;
else if (adreno_is_a3xx(adreno_dev))
cmds[7] = 0x90000000;
return adreno_ringbuffer_issuecmds(device, context,
KGSL_CMD_FLAGS_NONE, cmds, 8);
}
static void _snapshot_hlsq_regs(struct kgsl_snapshot_registers *regs,
struct kgsl_snapshot_registers_list *list,
struct adreno_device *adreno_dev)
{
if (!adreno_is_a3xx(adreno_dev)) {
regs[list->count].regs = (unsigned int *) a3xx_hlsq_registers;
regs[list->count].count = a3xx_hlsq_registers_count;
list->count++;
}
}
/**
* adreno_context_restore() - generic context restore handler
* @rb: The RB in which context is to be restored
*
* Basic context restore handler that writes the context identifier
* to the ringbuffer and issues pagetable switch commands if necessary.
*/
static void adreno_context_restore(struct adreno_ringbuffer *rb)
{
struct kgsl_device *device = rb->device;
struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
struct adreno_context *drawctxt = rb->drawctxt_active;
unsigned int cmds[11];
int ret;
if (!drawctxt)
return;
/*
* write the context identifier to the ringbuffer, write to both
* the global index and the index of the RB in which the context
* operates. The global values will always be reliable since we
* could be in middle of RB switch in which case the RB value may
* not be accurate
*/
cmds[0] = cp_nop_packet(1);
cmds[1] = KGSL_CONTEXT_TO_MEM_IDENTIFIER;
cmds[2] = cp_type3_packet(CP_MEM_WRITE, 2);
cmds[3] = device->memstore.gpuaddr +
KGSL_MEMSTORE_RB_OFFSET(rb, current_context);
cmds[4] = drawctxt->base.id;
cmds[5] = cp_type3_packet(CP_MEM_WRITE, 2);
cmds[6] = device->memstore.gpuaddr +
KGSL_MEMSTORE_OFFSET(KGSL_MEMSTORE_GLOBAL,
current_context);
cmds[7] = drawctxt->base.id;
/* Flush the UCHE for new context */
cmds[8] = cp_type0_packet(
adreno_getreg(adreno_dev, ADRENO_REG_UCHE_INVALIDATE0), 2);
cmds[9] = 0;
if (adreno_is_a4xx(adreno_dev))
cmds[10] = 0x12;
else if (adreno_is_a3xx(adreno_dev))
cmds[10] = 0x90000000;
ret = adreno_ringbuffer_issuecmds(rb, KGSL_CMD_FLAGS_NONE, cmds, 11);
if (ret) {
/*
* A failure to submit commands to ringbuffer means RB may
* be full, in this case wait for idle and use CPU
*/
ret = adreno_idle(device);
BUG_ON(ret);
_adreno_context_restore_cpu(rb, drawctxt);
}
}
void adreno_ringbuffer_submit(struct adreno_ringbuffer *rb,
struct adreno_submit_time *time)
{
struct adreno_device *adreno_dev = ADRENO_DEVICE(rb->device);
BUG_ON(rb->wptr == 0);
_cff_write_ringbuffer(rb);
if (time != NULL) {
unsigned long flags;
local_irq_save(flags);
if (!adreno_is_a3xx(adreno_dev)) {
adreno_readreg64(adreno_dev,
ADRENO_REG_RBBM_ALWAYSON_COUNTER_LO,
ADRENO_REG_RBBM_ALWAYSON_COUNTER_HI,
&time->ticks);
if (ADRENO_GPUREV(adreno_dev) >= 400 &&
ADRENO_GPUREV(adreno_dev) <= ADRENO_REV_A530)
time->ticks &= 0xFFFFFFFF;
}
else
time->ticks = 0;
time->ktime = local_clock();
getnstimeofday(&time->utime);
local_irq_restore(flags);
}
mb();
if (adreno_preempt_state(adreno_dev, ADRENO_DISPATCHER_PREEMPT_CLEAR) &&
(adreno_dev->cur_rb == rb)) {
kgsl_pwrscale_busy(rb->device);
adreno_writereg(adreno_dev, ADRENO_REG_CP_RB_WPTR, rb->wptr);
}
}
int adreno_ringbuffer_load_pm4_ucode(struct kgsl_device *device)
{
struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
int i;
if (adreno_dev->pm4_fw == NULL) {
int ret = adreno_ringbuffer_read_pm4_ucode(device);
if (ret)
return ret;
}
KGSL_DRV_INFO(device, "loading pm4 ucode version: %d\n",
adreno_dev->pm4_fw_version);
if (adreno_is_a3xx(adreno_dev))
adreno_regwrite(device, REG_CP_DEBUG, CP_DEBUG_DEFAULT);
else
adreno_regwrite(device, REG_CP_DEBUG, 0x02000000);
adreno_regwrite(device, REG_CP_ME_RAM_WADDR, 0);
for (i = 1; i < adreno_dev->pm4_fw_size; i++)
adreno_regwrite(device, REG_CP_ME_RAM_DATA,
adreno_dev->pm4_fw[i]);
return 0;
}
static uint32_t
adreno_ringbuffer_addcmds(struct adreno_ringbuffer *rb,
struct adreno_context *context,
unsigned int flags, unsigned int *cmds,
int sizedwords)
{
struct adreno_device *adreno_dev = ADRENO_DEVICE(rb->device);
unsigned int *ringcmds;
unsigned int timestamp;
unsigned int total_sizedwords = sizedwords;
unsigned int i;
unsigned int rcmd_gpu;
unsigned int context_id = KGSL_MEMSTORE_GLOBAL;
unsigned int gpuaddr = rb->device->memstore.gpuaddr;
if (context != NULL) {
/*
* if the context was not created with per context timestamp
* support, we must use the global timestamp since issueibcmds
* will be returning that one.
*/
if (context->flags & CTXT_FLAGS_PER_CONTEXT_TS)
context_id = context->id;
}
/* reserve space to temporarily turn off protected mode
* error checking if needed
*/
total_sizedwords += flags & KGSL_CMD_FLAGS_PMODE ? 4 : 0;
total_sizedwords += !(flags & KGSL_CMD_FLAGS_NO_TS_CMP) ? 7 : 0;
total_sizedwords += !(flags & KGSL_CMD_FLAGS_NOT_KERNEL_CMD) ? 2 : 0;
if (adreno_is_a3xx(adreno_dev))
total_sizedwords += 7;
total_sizedwords += 2; /* scratchpad ts for recovery */
if (context) {
total_sizedwords += 3; /* sop timestamp */
total_sizedwords += 4; /* eop timestamp */
total_sizedwords += 3; /* global timestamp without cache
* flush for non-zero context */
} else {
total_sizedwords += 4; /* global timestamp for recovery*/
}
ringcmds = adreno_ringbuffer_allocspace(rb, total_sizedwords);
rcmd_gpu = rb->buffer_desc.gpuaddr
+ sizeof(uint)*(rb->wptr-total_sizedwords);
if (!(flags & KGSL_CMD_FLAGS_NOT_KERNEL_CMD)) {
GSL_RB_WRITE(ringcmds, rcmd_gpu, cp_nop_packet(1));
GSL_RB_WRITE(ringcmds, rcmd_gpu, KGSL_CMD_IDENTIFIER);
}
if (flags & KGSL_CMD_FLAGS_PMODE) {
/* disable protected mode error checking */
GSL_RB_WRITE(ringcmds, rcmd_gpu,
cp_type3_packet(CP_SET_PROTECTED_MODE, 1));
GSL_RB_WRITE(ringcmds, rcmd_gpu, 0);
}
for (i = 0; i < sizedwords; i++) {
GSL_RB_WRITE(ringcmds, rcmd_gpu, *cmds);
cmds++;
}
if (flags & KGSL_CMD_FLAGS_PMODE) {
/* re-enable protected mode error checking */
GSL_RB_WRITE(ringcmds, rcmd_gpu,
cp_type3_packet(CP_SET_PROTECTED_MODE, 1));
GSL_RB_WRITE(ringcmds, rcmd_gpu, 1);
}
/* always increment the global timestamp. once. */
rb->timestamp[KGSL_MEMSTORE_GLOBAL]++;
if (context) {
if (context_id == KGSL_MEMSTORE_GLOBAL)
rb->timestamp[context_id] =
rb->timestamp[KGSL_MEMSTORE_GLOBAL];
else
rb->timestamp[context_id]++;
}
timestamp = rb->timestamp[context_id];
/* scratchpad ts for recovery */
GSL_RB_WRITE(ringcmds, rcmd_gpu, cp_type0_packet(REG_CP_TIMESTAMP, 1));
GSL_RB_WRITE(ringcmds, rcmd_gpu, rb->timestamp[KGSL_MEMSTORE_GLOBAL]);
if (adreno_is_a3xx(adreno_dev)) {
/*
* FLush HLSQ lazy updates to make sure there are no
* rsources pending for indirect loads after the timestamp
*/
GSL_RB_WRITE(ringcmds, rcmd_gpu,
cp_type3_packet(CP_EVENT_WRITE, 1));
GSL_RB_WRITE(ringcmds, rcmd_gpu, 0x07); /* HLSQ_FLUSH */
GSL_RB_WRITE(ringcmds, rcmd_gpu,
cp_type3_packet(CP_WAIT_FOR_IDLE, 1));
GSL_RB_WRITE(ringcmds, rcmd_gpu, 0x00);
}
if (context) {
/* start-of-pipeline timestamp */
GSL_RB_WRITE(ringcmds, rcmd_gpu,
cp_type3_packet(CP_MEM_WRITE, 2));
GSL_RB_WRITE(ringcmds, rcmd_gpu, (gpuaddr +
KGSL_MEMSTORE_OFFSET(context->id, soptimestamp)));
GSL_RB_WRITE(ringcmds, rcmd_gpu, timestamp);
/* end-of-pipeline timestamp */
GSL_RB_WRITE(ringcmds, rcmd_gpu,
cp_type3_packet(CP_EVENT_WRITE, 3));
GSL_RB_WRITE(ringcmds, rcmd_gpu, CACHE_FLUSH_TS);
GSL_RB_WRITE(ringcmds, rcmd_gpu, (gpuaddr +
KGSL_MEMSTORE_OFFSET(context->id, eoptimestamp)));
GSL_RB_WRITE(ringcmds, rcmd_gpu, timestamp);
GSL_RB_WRITE(ringcmds, rcmd_gpu,
cp_type3_packet(CP_MEM_WRITE, 2));
GSL_RB_WRITE(ringcmds, rcmd_gpu, (gpuaddr +
KGSL_MEMSTORE_OFFSET(KGSL_MEMSTORE_GLOBAL,
eoptimestamp)));
GSL_RB_WRITE(ringcmds, rcmd_gpu,
rb->timestamp[KGSL_MEMSTORE_GLOBAL]);
} else {
GSL_RB_WRITE(ringcmds, rcmd_gpu,
cp_type3_packet(CP_EVENT_WRITE, 3));
GSL_RB_WRITE(ringcmds, rcmd_gpu, CACHE_FLUSH_TS);
GSL_RB_WRITE(ringcmds, rcmd_gpu, (gpuaddr +
KGSL_MEMSTORE_OFFSET(KGSL_MEMSTORE_GLOBAL,
eoptimestamp)));
GSL_RB_WRITE(ringcmds, rcmd_gpu,
rb->timestamp[KGSL_MEMSTORE_GLOBAL]);
}
if (!(flags & KGSL_CMD_FLAGS_NO_TS_CMP)) {
/* Conditional execution based on memory values */
GSL_RB_WRITE(ringcmds, rcmd_gpu,
cp_type3_packet(CP_COND_EXEC, 4));
GSL_RB_WRITE(ringcmds, rcmd_gpu, (gpuaddr +
KGSL_MEMSTORE_OFFSET(
context_id, ts_cmp_enable)) >> 2);
GSL_RB_WRITE(ringcmds, rcmd_gpu, (gpuaddr +
KGSL_MEMSTORE_OFFSET(
context_id, ref_wait_ts)) >> 2);
GSL_RB_WRITE(ringcmds, rcmd_gpu, timestamp);
/* # of conditional command DWORDs */
GSL_RB_WRITE(ringcmds, rcmd_gpu, 2);
GSL_RB_WRITE(ringcmds, rcmd_gpu,
cp_type3_packet(CP_INTERRUPT, 1));
GSL_RB_WRITE(ringcmds, rcmd_gpu, CP_INT_CNTL__RB_INT_MASK);
}
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;
}
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);
}
static uint32_t
adreno_ringbuffer_addcmds(struct adreno_ringbuffer *rb,
unsigned int flags, unsigned int *cmds,
int sizedwords)
{
struct adreno_device *adreno_dev = ADRENO_DEVICE(rb->device);
unsigned int *ringcmds;
unsigned int timestamp;
unsigned int total_sizedwords = sizedwords + 6;
unsigned int i;
unsigned int rcmd_gpu;
/* reserve space to temporarily turn off protected mode
* error checking if needed
*/
total_sizedwords += flags & KGSL_CMD_FLAGS_PMODE ? 4 : 0;
total_sizedwords += !(flags & KGSL_CMD_FLAGS_NO_TS_CMP) ? 7 : 0;
total_sizedwords += !(flags & KGSL_CMD_FLAGS_NOT_KERNEL_CMD) ? 2 : 0;
if (adreno_is_a3xx(adreno_dev))
total_sizedwords += 7;
ringcmds = adreno_ringbuffer_allocspace(rb, total_sizedwords);
rcmd_gpu = rb->buffer_desc.gpuaddr
+ sizeof(uint)*(rb->wptr-total_sizedwords);
if (!(flags & KGSL_CMD_FLAGS_NOT_KERNEL_CMD)) {
GSL_RB_WRITE(ringcmds, rcmd_gpu, cp_nop_packet(1));
GSL_RB_WRITE(ringcmds, rcmd_gpu, KGSL_CMD_IDENTIFIER);
}
if (flags & KGSL_CMD_FLAGS_PMODE) {
/* disable protected mode error checking */
GSL_RB_WRITE(ringcmds, rcmd_gpu,
cp_type3_packet(CP_SET_PROTECTED_MODE, 1));
GSL_RB_WRITE(ringcmds, rcmd_gpu, 0);
}
for (i = 0; i < sizedwords; i++) {
GSL_RB_WRITE(ringcmds, rcmd_gpu, *cmds);
cmds++;
}
if (flags & KGSL_CMD_FLAGS_PMODE) {
/* re-enable protected mode error checking */
GSL_RB_WRITE(ringcmds, rcmd_gpu,
cp_type3_packet(CP_SET_PROTECTED_MODE, 1));
GSL_RB_WRITE(ringcmds, rcmd_gpu, 1);
}
rb->timestamp++;
timestamp = rb->timestamp;
/* start-of-pipeline and end-of-pipeline timestamps */
GSL_RB_WRITE(ringcmds, rcmd_gpu, cp_type0_packet(REG_CP_TIMESTAMP, 1));
GSL_RB_WRITE(ringcmds, rcmd_gpu, rb->timestamp);
if (adreno_is_a3xx(adreno_dev)) {
/*
* FLush HLSQ lazy updates to make sure there are no
* rsources pending for indirect loads after the timestamp
*/
GSL_RB_WRITE(ringcmds, rcmd_gpu,
cp_type3_packet(CP_EVENT_WRITE, 1));
GSL_RB_WRITE(ringcmds, rcmd_gpu, 0x07); /* HLSQ_FLUSH */
GSL_RB_WRITE(ringcmds, rcmd_gpu,
cp_type3_packet(CP_WAIT_FOR_IDLE, 1));
GSL_RB_WRITE(ringcmds, rcmd_gpu, 0x00);
}
GSL_RB_WRITE(ringcmds, rcmd_gpu, cp_type3_packet(CP_EVENT_WRITE, 3));
GSL_RB_WRITE(ringcmds, rcmd_gpu, CACHE_FLUSH_TS);
GSL_RB_WRITE(ringcmds, rcmd_gpu,
(rb->device->memstore.gpuaddr +
KGSL_DEVICE_MEMSTORE_OFFSET(eoptimestamp)));
GSL_RB_WRITE(ringcmds, rcmd_gpu, rb->timestamp);
if (!(flags & KGSL_CMD_FLAGS_NO_TS_CMP)) {
/* Conditional execution based on memory values */
GSL_RB_WRITE(ringcmds, rcmd_gpu,
cp_type3_packet(CP_COND_EXEC, 4));
GSL_RB_WRITE(ringcmds, rcmd_gpu, (rb->device->memstore.gpuaddr +
KGSL_DEVICE_MEMSTORE_OFFSET(ts_cmp_enable)) >> 2);
GSL_RB_WRITE(ringcmds, rcmd_gpu, (rb->device->memstore.gpuaddr +
KGSL_DEVICE_MEMSTORE_OFFSET(ref_wait_ts)) >> 2);
GSL_RB_WRITE(ringcmds, rcmd_gpu, rb->timestamp);
/* # of conditional command DWORDs */
GSL_RB_WRITE(ringcmds, rcmd_gpu, 2);
GSL_RB_WRITE(ringcmds, rcmd_gpu,
cp_type3_packet(CP_INTERRUPT, 1));
GSL_RB_WRITE(ringcmds, rcmd_gpu, CP_INT_CNTL__RB_INT_MASK);
}
static uint32_t
adreno_ringbuffer_addcmds(struct adreno_ringbuffer *rb,
struct adreno_context *context,
unsigned int flags, unsigned int *cmds,
int sizedwords)
{
struct adreno_device *adreno_dev = ADRENO_DEVICE(rb->device);
unsigned int *ringcmds;
unsigned int timestamp;
unsigned int total_sizedwords = sizedwords;
unsigned int i;
unsigned int rcmd_gpu;
unsigned int context_id = KGSL_MEMSTORE_GLOBAL;
unsigned int gpuaddr = rb->device->memstore.gpuaddr;
if (context->flags & CTXT_FLAGS_PER_CONTEXT_TS)
context_id = context->id;
total_sizedwords += flags & KGSL_CMD_FLAGS_PMODE ? 4 : 0;
total_sizedwords += !(flags & KGSL_CMD_FLAGS_NO_TS_CMP) ? 7 : 0;
total_sizedwords += 2;
if (adreno_is_a3xx(adreno_dev))
total_sizedwords += 7;
total_sizedwords += 2;
if (context->flags & CTXT_FLAGS_PER_CONTEXT_TS) {
total_sizedwords += 3;
total_sizedwords += 4;
total_sizedwords += 3;
} else {
total_sizedwords += 4;
}
ringcmds = adreno_ringbuffer_allocspace(rb, total_sizedwords);
if (context->flags & CTXT_FLAGS_GPU_HANG) {
KGSL_CTXT_WARN(rb->device,
"Context %p caused a gpu hang. Will not accept commands for context %d\n",
context, context->id);
return rb->timestamp[context_id];
}
rcmd_gpu = rb->buffer_desc.gpuaddr
+ sizeof(uint)*(rb->wptr-total_sizedwords);
GSL_RB_WRITE(ringcmds, rcmd_gpu, cp_nop_packet(1));
GSL_RB_WRITE(ringcmds, rcmd_gpu, KGSL_CMD_IDENTIFIER);
if (flags & KGSL_CMD_FLAGS_PMODE) {
GSL_RB_WRITE(ringcmds, rcmd_gpu,
cp_type3_packet(CP_SET_PROTECTED_MODE, 1));
GSL_RB_WRITE(ringcmds, rcmd_gpu, 0);
}
for (i = 0; i < sizedwords; i++) {
GSL_RB_WRITE(ringcmds, rcmd_gpu, *cmds);
cmds++;
}
if (flags & KGSL_CMD_FLAGS_PMODE) {
GSL_RB_WRITE(ringcmds, rcmd_gpu,
cp_type3_packet(CP_SET_PROTECTED_MODE, 1));
GSL_RB_WRITE(ringcmds, rcmd_gpu, 1);
}
rb->timestamp[KGSL_MEMSTORE_GLOBAL]++;
if (context) {
if (context_id == KGSL_MEMSTORE_GLOBAL)
rb->timestamp[context_id] =
rb->timestamp[KGSL_MEMSTORE_GLOBAL];
else
rb->timestamp[context_id]++;
}
timestamp = rb->timestamp[context_id];
GSL_RB_WRITE(ringcmds, rcmd_gpu, cp_type0_packet(REG_CP_TIMESTAMP, 1));
GSL_RB_WRITE(ringcmds, rcmd_gpu, rb->timestamp[KGSL_MEMSTORE_GLOBAL]);
if (adreno_is_a3xx(adreno_dev)) {
GSL_RB_WRITE(ringcmds, rcmd_gpu,
cp_type3_packet(CP_EVENT_WRITE, 1));
GSL_RB_WRITE(ringcmds, rcmd_gpu, 0x07);
GSL_RB_WRITE(ringcmds, rcmd_gpu,
cp_type3_packet(CP_WAIT_FOR_IDLE, 1));
GSL_RB_WRITE(ringcmds, rcmd_gpu, 0x00);
}
if (context->flags & CTXT_FLAGS_PER_CONTEXT_TS) {
GSL_RB_WRITE(ringcmds, rcmd_gpu,
cp_type3_packet(CP_MEM_WRITE, 2));
GSL_RB_WRITE(ringcmds, rcmd_gpu, (gpuaddr +
KGSL_MEMSTORE_OFFSET(context->id, soptimestamp)));
GSL_RB_WRITE(ringcmds, rcmd_gpu, timestamp);
GSL_RB_WRITE(ringcmds, rcmd_gpu,
cp_type3_packet(CP_EVENT_WRITE, 3));
GSL_RB_WRITE(ringcmds, rcmd_gpu, CACHE_FLUSH_TS);
GSL_RB_WRITE(ringcmds, rcmd_gpu, (gpuaddr +
KGSL_MEMSTORE_OFFSET(context->id, eoptimestamp)));
GSL_RB_WRITE(ringcmds, rcmd_gpu, timestamp);
GSL_RB_WRITE(ringcmds, rcmd_gpu,
cp_type3_packet(CP_MEM_WRITE, 2));
GSL_RB_WRITE(ringcmds, rcmd_gpu, (gpuaddr +
KGSL_MEMSTORE_OFFSET(KGSL_MEMSTORE_GLOBAL,
eoptimestamp)));
GSL_RB_WRITE(ringcmds, rcmd_gpu,
rb->timestamp[KGSL_MEMSTORE_GLOBAL]);
} else {
GSL_RB_WRITE(ringcmds, rcmd_gpu,
cp_type3_packet(CP_EVENT_WRITE, 3));
GSL_RB_WRITE(ringcmds, rcmd_gpu, CACHE_FLUSH_TS);
GSL_RB_WRITE(ringcmds, rcmd_gpu, (gpuaddr +
KGSL_MEMSTORE_OFFSET(KGSL_MEMSTORE_GLOBAL,
eoptimestamp)));
GSL_RB_WRITE(ringcmds, rcmd_gpu,
rb->timestamp[KGSL_MEMSTORE_GLOBAL]);
}
if (!(flags & KGSL_CMD_FLAGS_NO_TS_CMP)) {
GSL_RB_WRITE(ringcmds, rcmd_gpu,
cp_type3_packet(CP_COND_EXEC, 4));
GSL_RB_WRITE(ringcmds, rcmd_gpu, (gpuaddr +
KGSL_MEMSTORE_OFFSET(
context_id, ts_cmp_enable)) >> 2);
GSL_RB_WRITE(ringcmds, rcmd_gpu, (gpuaddr +
KGSL_MEMSTORE_OFFSET(
context_id, ref_wait_ts)) >> 2);
GSL_RB_WRITE(ringcmds, rcmd_gpu, timestamp);
GSL_RB_WRITE(ringcmds, rcmd_gpu, 2);
GSL_RB_WRITE(ringcmds, rcmd_gpu,
cp_type3_packet(CP_INTERRUPT, 1));
GSL_RB_WRITE(ringcmds, rcmd_gpu, CP_INT_CNTL__RB_INT_MASK);
}
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;
}
