static uint32_t
kgsl_ringbuffer_addcmds(struct kgsl_ringbuffer *rb,
unsigned int flags, unsigned int *cmds,
int sizedwords)
{
unsigned int *ringcmds;
unsigned int timestamp;
unsigned int total_sizedwords = sizedwords + 6;
total_sizedwords += flags & KGSL_CMD_FLAGS_PMODE ? 4 : 0;
total_sizedwords += !(flags & KGSL_CMD_FLAGS_NO_TS_CMP) ? 9 : 0;
ringcmds = kgsl_ringbuffer_allocspace(rb, total_sizedwords);
if (flags & KGSL_CMD_FLAGS_PMODE) {
*ringcmds++ = pm4_type3_packet(PM4_SET_PROTECTED_MODE, 1);
*ringcmds++ = 0;
}
memcpy(ringcmds, cmds, (sizedwords << 2));
ringcmds += sizedwords;
if (flags & KGSL_CMD_FLAGS_PMODE) {
*ringcmds++ = pm4_type3_packet(PM4_SET_PROTECTED_MODE, 1);
*ringcmds++ = 1;
}
rb->timestamp++;
timestamp = rb->timestamp;
*ringcmds++ = pm4_type0_packet(REG_CP_TIMESTAMP, 1);
*ringcmds++ = rb->timestamp;
*ringcmds++ = pm4_type3_packet(PM4_EVENT_WRITE, 3);
*ringcmds++ = CACHE_FLUSH_TS;
*ringcmds++ =
(rb->device->memstore.gpuaddr +
KGSL_DEVICE_MEMSTORE_OFFSET(eoptimestamp));
*ringcmds++ = rb->timestamp;
if (!(flags & KGSL_CMD_FLAGS_NO_TS_CMP)) {
*ringcmds++ = pm4_type3_packet(PM4_COND_EXEC, 4);
*ringcmds++ = (rb->device->memstore.gpuaddr +
KGSL_DEVICE_MEMSTORE_OFFSET(ts_cmp_enable)) >> 2;
*ringcmds++ = (rb->device->memstore.gpuaddr +
KGSL_DEVICE_MEMSTORE_OFFSET(ref_wait_ts)) >> 2;
*ringcmds++ = rb->timestamp;
*ringcmds++ = 4;
*ringcmds++ = pm4_type3_packet(PM4_WAIT_FOR_IDLE, 1);
*ringcmds++ = 0x00000000;
*ringcmds++ = pm4_type3_packet(PM4_INTERRUPT, 1);
*ringcmds++ = CP_INT_CNTL__RB_INT_MASK;
}
static uint32_t
kgsl_ringbuffer_addcmds(struct kgsl_ringbuffer *rb,
unsigned int flags, unsigned int *cmds,
int sizedwords)
{
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;
ringcmds = kgsl_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, pm4_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,
pm4_type3_packet(PM4_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,
pm4_type3_packet(PM4_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, pm4_type0_packet(REG_CP_TIMESTAMP, 1));
GSL_RB_WRITE(ringcmds, rcmd_gpu, rb->timestamp);
GSL_RB_WRITE(ringcmds, rcmd_gpu, pm4_type3_packet(PM4_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,
pm4_type3_packet(PM4_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,
pm4_type3_packet(PM4_INTERRUPT, 1));
GSL_RB_WRITE(ringcmds, rcmd_gpu, CP_INT_CNTL__RB_INT_MASK);
}
int kgsl_ringbuffer_start(struct kgsl_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 *cmds, rb_cntl;
struct kgsl_device *device = rb->device;
uint cmds_gpu;
if (rb->flags & KGSL_FLAGS_STARTED)
return 0;
if (init_ram) {
rb->timestamp = 0;
GSL_RB_INIT_TIMESTAMP(rb);
}
kgsl_sharedmem_set(&rb->memptrs_desc, 0, 0,
sizeof(struct kgsl_rbmemptrs));
kgsl_sharedmem_set(&rb->buffer_desc, 0, 0xAA,
(rb->sizedwords << 2));
kgsl_yamato_regwrite(device, REG_CP_RB_WPTR_BASE,
(rb->memptrs_desc.gpuaddr
+ GSL_RB_MEMPTRS_WPTRPOLL_OFFSET));
/* setup WPTR delay */
kgsl_yamato_regwrite(device, REG_CP_RB_WPTR_DELAY, 0 /*0x70000010 */);
/*setup REG_CP_RB_CNTL */
kgsl_yamato_regread(device, REG_CP_RB_CNTL, &rb_cntl);
cp_rb_cntl.val = rb_cntl;
/* size of ringbuffer */
cp_rb_cntl.f.rb_bufsz =
kgsl_ringbuffer_sizelog2quadwords(rb->sizedwords);
/* quadwords to read before updating mem RPTR */
cp_rb_cntl.f.rb_blksz = rb->blksizequadwords;
cp_rb_cntl.f.rb_poll_en = GSL_RB_CNTL_POLL_EN; /* WPTR polling */
/* mem RPTR writebacks */
cp_rb_cntl.f.rb_no_update = GSL_RB_CNTL_NO_UPDATE;
kgsl_yamato_regwrite(device, REG_CP_RB_CNTL, cp_rb_cntl.val);
kgsl_yamato_regwrite(device, REG_CP_RB_BASE, rb->buffer_desc.gpuaddr);
kgsl_yamato_regwrite(device, REG_CP_RB_RPTR_ADDR,
rb->memptrs_desc.gpuaddr +
GSL_RB_MEMPTRS_RPTR_OFFSET);
/* explicitly clear all cp interrupts */
kgsl_yamato_regwrite(device, REG_CP_INT_ACK, 0xFFFFFFFF);
/* setup scratch/timestamp */
kgsl_yamato_regwrite(device, REG_SCRATCH_ADDR,
device->memstore.gpuaddr +
KGSL_DEVICE_MEMSTORE_OFFSET(soptimestamp));
kgsl_yamato_regwrite(device, REG_SCRATCH_UMSK,
GSL_RB_MEMPTRS_SCRATCH_MASK);
/* load the CP ucode */
status = kgsl_ringbuffer_load_pm4_ucode(device);
if (status != 0)
return status;
/* load the prefetch parser ucode */
status = kgsl_ringbuffer_load_pfp_ucode(device);
if (status != 0)
return status;
kgsl_yamato_regwrite(device, REG_CP_QUEUE_THRESHOLDS, 0x000C0804);
rb->rptr = 0;
rb->wptr = 0;
/* clear ME_HALT to start micro engine */
kgsl_yamato_regwrite(device, REG_CP_ME_CNTL, 0);
/* ME_INIT */
cmds = kgsl_ringbuffer_allocspace(rb, 19);
cmds_gpu = rb->buffer_desc.gpuaddr + sizeof(uint)*(rb->wptr-19);
GSL_RB_WRITE(cmds, cmds_gpu, PM4_HDR_ME_INIT);
/* All fields present (bits 9:0) */
GSL_RB_WRITE(cmds, cmds_gpu, 0x000003ff);
/* Disable/Enable Real-Time Stream processing (present but ignored) */
GSL_RB_WRITE(cmds, cmds_gpu, 0x00000000);
/* Enable (2D <-> 3D) implicit synchronization (present but ignored) */
GSL_RB_WRITE(cmds, cmds_gpu, 0x00000000);
GSL_RB_WRITE(cmds, cmds_gpu,
GSL_HAL_SUBBLOCK_OFFSET(REG_RB_SURFACE_INFO));
GSL_RB_WRITE(cmds, cmds_gpu,
GSL_HAL_SUBBLOCK_OFFSET(REG_PA_SC_WINDOW_OFFSET));
GSL_RB_WRITE(cmds, cmds_gpu,
GSL_HAL_SUBBLOCK_OFFSET(REG_VGT_MAX_VTX_INDX));
GSL_RB_WRITE(cmds, cmds_gpu,
GSL_HAL_SUBBLOCK_OFFSET(REG_SQ_PROGRAM_CNTL));
GSL_RB_WRITE(cmds, cmds_gpu,
GSL_HAL_SUBBLOCK_OFFSET(REG_RB_DEPTHCONTROL));
GSL_RB_WRITE(cmds, cmds_gpu,
GSL_HAL_SUBBLOCK_OFFSET(REG_PA_SU_POINT_SIZE));
GSL_RB_WRITE(cmds, cmds_gpu,
GSL_HAL_SUBBLOCK_OFFSET(REG_PA_SC_LINE_CNTL));
GSL_RB_WRITE(cmds, cmds_gpu,
GSL_HAL_SUBBLOCK_OFFSET(REG_PA_SU_POLY_OFFSET_FRONT_SCALE));
/* Vertex and Pixel Shader Start Addresses in instructions
* (3 DWORDS per instruction) */
GSL_RB_WRITE(cmds, cmds_gpu, 0x80000180);
/* Maximum Contexts */
GSL_RB_WRITE(cmds, cmds_gpu, 0x00000001);
/* Write Confirm Interval and The CP will wait the
* wait_interval * 16 clocks between polling */
GSL_RB_WRITE(cmds, cmds_gpu, 0x00000000);
/* NQ and External Memory Swap */
GSL_RB_WRITE(cmds, cmds_gpu, 0x00000000);
/* Protected mode error checking */
GSL_RB_WRITE(cmds, cmds_gpu, GSL_RB_PROTECTED_MODE_CONTROL);
/* Disable header dumping and Header dump address */
GSL_RB_WRITE(cmds, cmds_gpu, 0x00000000);
/* Header dump size */
GSL_RB_WRITE(cmds, cmds_gpu, 0x00000000);
kgsl_ringbuffer_submit(rb);
/* idle device to validate ME INIT */
status = kgsl_yamato_idle(device, KGSL_TIMEOUT_DEFAULT);
kgsl_yamato_regwrite(rb->device, REG_CP_INT_CNTL, GSL_CP_INT_MASK);
if (status == 0)
rb->flags |= KGSL_FLAGS_STARTED;
return status;
}
static int kgsl_ringbuffer_start(struct kgsl_ringbuffer *rb)
{
int status;
union reg_cp_rb_cntl cp_rb_cntl;
unsigned int *cmds, rb_cntl;
struct kgsl_device *device = rb->device;
KGSL_CMD_VDBG("enter (rb=%p)\n", rb);
if (rb->flags & KGSL_FLAGS_STARTED) {
KGSL_CMD_VDBG("return %d\n", 0);
return 0;
}
kgsl_sharedmem_set(&rb->memptrs_desc, 0, 0,
sizeof(struct kgsl_rbmemptrs));
kgsl_sharedmem_set(&rb->buffer_desc, 0, 0xAA,
(rb->sizedwords << 2));
kgsl_yamato_regwrite(device, REG_CP_RB_WPTR_BASE,
(rb->memptrs_desc.gpuaddr
+ GSL_RB_MEMPTRS_WPTRPOLL_OFFSET));
kgsl_yamato_regwrite(device, REG_CP_RB_WPTR_DELAY, 0 );
kgsl_yamato_regread(device, REG_CP_RB_CNTL, &rb_cntl);
cp_rb_cntl.val = rb_cntl;
cp_rb_cntl.f.rb_bufsz =
kgsl_ringbuffer_sizelog2quadwords(rb->sizedwords);
cp_rb_cntl.f.rb_blksz = rb->blksizequadwords;
cp_rb_cntl.f.rb_poll_en = GSL_RB_CNTL_POLL_EN;
cp_rb_cntl.f.rb_no_update = GSL_RB_CNTL_NO_UPDATE;
kgsl_yamato_regwrite(device, REG_CP_RB_CNTL, cp_rb_cntl.val);
kgsl_yamato_regwrite(device, REG_CP_RB_BASE, rb->buffer_desc.gpuaddr);
kgsl_yamato_regwrite(device, REG_CP_RB_RPTR_ADDR,
rb->memptrs_desc.gpuaddr +
GSL_RB_MEMPTRS_RPTR_OFFSET);
kgsl_yamato_regwrite(device, REG_CP_INT_ACK, 0xFFFFFFFF);
kgsl_yamato_regwrite(device, REG_SCRATCH_ADDR,
device->memstore.gpuaddr +
KGSL_DEVICE_MEMSTORE_OFFSET(soptimestamp));
kgsl_yamato_regwrite(device, REG_SCRATCH_UMSK,
GSL_RB_MEMPTRS_SCRATCH_MASK);
status = kgsl_ringbuffer_load_pm4_ucode(device);
if (status != 0) {
KGSL_DRV_ERR("kgsl_ringbuffer_load_pm4_ucode failed %d\n",
status);
return status;
}
status = kgsl_ringbuffer_load_pfp_ucode(device);
if (status != 0) {
KGSL_DRV_ERR("kgsl_ringbuffer_load_pm4_ucode failed %d\n",
status);
return status;
}
kgsl_yamato_regwrite(device, REG_CP_QUEUE_THRESHOLDS, 0x000C0804);
rb->rptr = 0;
rb->wptr = 0;
rb->timestamp = 0;
GSL_RB_INIT_TIMESTAMP(rb);
INIT_LIST_HEAD(&rb->memqueue);
kgsl_yamato_regwrite(device, REG_CP_ME_CNTL, 0);
cmds = kgsl_ringbuffer_allocspace(rb, 19);
GSL_RB_WRITE(cmds, PM4_HDR_ME_INIT);
GSL_RB_WRITE(cmds, 0x000003ff);
GSL_RB_WRITE(cmds, 0x00000000);
GSL_RB_WRITE(cmds, 0x00000000);
GSL_RB_WRITE(cmds, GSL_HAL_SUBBLOCK_OFFSET(REG_RB_SURFACE_INFO));
GSL_RB_WRITE(cmds, GSL_HAL_SUBBLOCK_OFFSET(REG_PA_SC_WINDOW_OFFSET));
GSL_RB_WRITE(cmds, GSL_HAL_SUBBLOCK_OFFSET(REG_VGT_MAX_VTX_INDX));
GSL_RB_WRITE(cmds, GSL_HAL_SUBBLOCK_OFFSET(REG_SQ_PROGRAM_CNTL));
GSL_RB_WRITE(cmds, GSL_HAL_SUBBLOCK_OFFSET(REG_RB_DEPTHCONTROL));
GSL_RB_WRITE(cmds, GSL_HAL_SUBBLOCK_OFFSET(REG_PA_SU_POINT_SIZE));
GSL_RB_WRITE(cmds, GSL_HAL_SUBBLOCK_OFFSET(REG_PA_SC_LINE_CNTL));
GSL_RB_WRITE(cmds,
GSL_HAL_SUBBLOCK_OFFSET(REG_PA_SU_POLY_OFFSET_FRONT_SCALE));
GSL_RB_WRITE(cmds, 0x80000180);
GSL_RB_WRITE(cmds, 0x00000001);
GSL_RB_WRITE(cmds, 0x00000000);
GSL_RB_WRITE(cmds, 0x00000000);
GSL_RB_WRITE(cmds, GSL_RB_PROTECTED_MODE_CONTROL);
GSL_RB_WRITE(cmds, 0x00000000);
GSL_RB_WRITE(cmds, 0x00000000);
kgsl_ringbuffer_submit(rb);
status = kgsl_yamato_idle(device, KGSL_TIMEOUT_DEFAULT);
KGSL_CMD_DBG("enabling CP interrupts: mask %08lx\n", GSL_CP_INT_MASK);
kgsl_yamato_regwrite(rb->device, REG_CP_INT_CNTL, GSL_CP_INT_MASK);
if (status == 0)
rb->flags |= KGSL_FLAGS_STARTED;
KGSL_CMD_VDBG("return %d\n", status);
return status;
}
static uint32_t
kgsl_ringbuffer_addcmds(struct kgsl_ringbuffer *rb,
unsigned int flags, unsigned int *cmds,
int sizedwords)
{
unsigned int *ringcmds;
unsigned int timestamp;
unsigned int total_sizedwords = sizedwords + 6;
/* 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) ? 9 : 0;
ringcmds = kgsl_ringbuffer_allocspace(rb, total_sizedwords);
if (flags & KGSL_CMD_FLAGS_PMODE) {
/* disable protected mode error checking */
*ringcmds++ = pm4_type3_packet(PM4_SET_PROTECTED_MODE, 1);
*ringcmds++ = 0;
}
memcpy(ringcmds, cmds, (sizedwords << 2));
ringcmds += sizedwords;
if (flags & KGSL_CMD_FLAGS_PMODE) {
/* re-enable protected mode error checking */
*ringcmds++ = pm4_type3_packet(PM4_SET_PROTECTED_MODE, 1);
*ringcmds++ = 1;
}
rb->timestamp++;
timestamp = rb->timestamp;
/* start-of-pipeline and end-of-pipeline timestamps */
*ringcmds++ = pm4_type0_packet(REG_CP_TIMESTAMP, 1);
*ringcmds++ = rb->timestamp;
*ringcmds++ = pm4_type3_packet(PM4_EVENT_WRITE, 3);
*ringcmds++ = CACHE_FLUSH_TS;
*ringcmds++ =
(rb->device->memstore.gpuaddr +
KGSL_DEVICE_MEMSTORE_OFFSET(eoptimestamp));
*ringcmds++ = rb->timestamp;
if (!(flags & KGSL_CMD_FLAGS_NO_TS_CMP)) {
/* Conditional execution based on memory values */
*ringcmds++ = pm4_type3_packet(PM4_COND_EXEC, 4);
*ringcmds++ = (rb->device->memstore.gpuaddr +
KGSL_DEVICE_MEMSTORE_OFFSET(ts_cmp_enable)) >> 2;
*ringcmds++ = (rb->device->memstore.gpuaddr +
KGSL_DEVICE_MEMSTORE_OFFSET(ref_wait_ts)) >> 2;
*ringcmds++ = rb->timestamp;
/* # of conditional command DWORDs */
*ringcmds++ = 4;
*ringcmds++ = pm4_type3_packet(PM4_WAIT_FOR_IDLE, 1);
*ringcmds++ = 0x00000000;
*ringcmds++ = pm4_type3_packet(PM4_INTERRUPT, 1);
*ringcmds++ = CP_INT_CNTL__RB_INT_MASK;
}
