static int adreno_ringbuffer_load_pm4_ucode(struct kgsl_device *device)
{
struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
const char *fwfile;
int i, ret = 0;
if (adreno_is_a220(adreno_dev)) {
fwfile = A220_PM4_470_FW;
} else if (adreno_is_a225(adreno_dev)) {
fwfile = A225_PM4_FW;
} else if (adreno_is_a20x(adreno_dev)) {
fwfile = A200_PM4_FW;
} else {
KGSL_DRV_ERR(device, "Could not load PM4 file\n");
return -EINVAL;
}
if (adreno_dev->pm4_fw == NULL) {
int len;
unsigned int *ptr;
ret = _load_firmware(device, fwfile, (void *) &ptr, &len);
if (ret)
goto err;
/* PM4 size is 3 dword aligned plus 1 dword of version */
if (len % ((sizeof(uint32_t) * 3)) != sizeof(uint32_t)) {
KGSL_DRV_ERR(device, "Bad firmware size: %d\n", len);
ret = -EINVAL;
goto err;
}
adreno_dev->pm4_fw_size = len / sizeof(uint32_t);
adreno_dev->pm4_fw = ptr;
}
KGSL_DRV_INFO(device, "loading pm4 ucode version: %d\n",
adreno_dev->pm4_fw[0]);
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]);
err:
return ret;
}
static int adreno_ringbuffer_load_pfp_ucode(struct kgsl_device *device)
{
struct adreno_device *adreno_dev = ADRENO_DEVICE(device);
const char *fwfile;
int i, ret = 0;
if (adreno_is_a220(adreno_dev)) {
fwfile = A220_PFP_470_FW;
} else if (adreno_is_a225(adreno_dev)) {
fwfile = A225_PFP_FW;
} else if (adreno_is_a20x(adreno_dev)) {
fwfile = A200_PFP_FW;
} else {
KGSL_DRV_ERR(device, "Could not load PFP firmware\n");
return -EINVAL;
}
if (adreno_dev->pfp_fw == NULL) {
int len;
unsigned int *ptr;
ret = _load_firmware(device, fwfile, (void *) &ptr, &len);
if (ret)
goto err;
/* PFP size shold be dword aligned */
if (len % sizeof(uint32_t) != 0) {
KGSL_DRV_ERR(device, "Bad firmware size: %d\n", len);
ret = -EINVAL;
goto err;
}
adreno_dev->pfp_fw_size = len / sizeof(uint32_t);
adreno_dev->pfp_fw = ptr;
}
KGSL_DRV_INFO(device, "loading pfp ucode version: %d\n",
adreno_dev->pfp_fw[0]);
adreno_regwrite(device, REG_CP_PFP_UCODE_ADDR, 0);
for (i = 1; i < adreno_dev->pfp_fw_size; i++)
adreno_regwrite(device, REG_CP_PFP_UCODE_DATA,
adreno_dev->pfp_fw[i]);
err:
return ret;
}
static void build_regrestore_cmds(struct adreno_device *adreno_dev,
struct adreno_context *drawctxt)
{
unsigned int *start = tmp_ctx.cmd;
unsigned int *cmd = start;
unsigned int i = 0;
unsigned int reg_array_size = 0;
const unsigned int *ptr_register_ranges;
*cmd++ = cp_type3_packet(CP_WAIT_FOR_IDLE, 1);
*cmd++ = 0;
/* H/W Registers */
/* deferred cp_type3_packet(CP_LOAD_CONSTANT_CONTEXT, ???); */
cmd++;
#ifdef CONFIG_MSM_KGSL_DISABLE_SHADOW_WRITES
/* Force mismatch */
*cmd++ = ((drawctxt->gpustate.gpuaddr + REG_OFFSET) & 0xFFFFE000) | 1;
#else
*cmd++ = (drawctxt->gpustate.gpuaddr + REG_OFFSET) & 0xFFFFE000;
#endif
/* Based on chip id choose the registers ranges*/
if (adreno_is_a220(adreno_dev)) {
ptr_register_ranges = register_ranges_a220;
reg_array_size = ARRAY_SIZE(register_ranges_a220);
} else if (adreno_is_a225(adreno_dev)) {
ptr_register_ranges = register_ranges_a225;
reg_array_size = ARRAY_SIZE(register_ranges_a225);
} else {
ptr_register_ranges = register_ranges_a20x;
reg_array_size = ARRAY_SIZE(register_ranges_a20x);
}
for (i = 0; i < (reg_array_size/2); i++) {
cmd = reg_range(cmd, ptr_register_ranges[i*2],
ptr_register_ranges[i*2+1]);
}
/* Now we know how many register blocks we have, we can compute command
* length
*/
start[2] =
cp_type3_packet(CP_LOAD_CONSTANT_CONTEXT, (cmd - start) - 3);
/* Enable shadowing for the entire register block. */
#ifdef CONFIG_MSM_KGSL_DISABLE_SHADOW_WRITES
start[4] |= (0 << 24) | (4 << 16); /* Disable shadowing. */
#else
start[4] |= (1 << 24) | (4 << 16);
#endif
/* Need to handle some of the registers separately */
*cmd++ = cp_type0_packet(REG_SQ_GPR_MANAGEMENT, 1);
tmp_ctx.reg_values[0] = virt2gpu(cmd, &drawctxt->gpustate);
*cmd++ = 0x00040400;
*cmd++ = cp_type3_packet(CP_WAIT_FOR_IDLE, 1);
*cmd++ = 0;
*cmd++ = cp_type0_packet(REG_TP0_CHICKEN, 1);
tmp_ctx.reg_values[1] = virt2gpu(cmd, &drawctxt->gpustate);
*cmd++ = 0x00000000;
if (adreno_is_a22x(adreno_dev)) {
unsigned int i;
unsigned int j = 2;
for (i = REG_A220_VSC_BIN_SIZE; i <=
REG_A220_VSC_PIPE_DATA_LENGTH_7; i++) {
*cmd++ = cp_type0_packet(i, 1);
tmp_ctx.reg_values[j] = virt2gpu(cmd,
&drawctxt->gpustate);
*cmd++ = 0x00000000;
j++;
}
}
/* ALU Constants */
*cmd++ = cp_type3_packet(CP_LOAD_CONSTANT_CONTEXT, 3);
*cmd++ = drawctxt->gpustate.gpuaddr & 0xFFFFE000;
#ifdef CONFIG_MSM_KGSL_DISABLE_SHADOW_WRITES
*cmd++ = (0 << 24) | (0 << 16) | 0; /* Disable shadowing */
#else
*cmd++ = (1 << 24) | (0 << 16) | 0;
#endif
*cmd++ = ALU_CONSTANTS;
/* Texture Constants */
*cmd++ = cp_type3_packet(CP_LOAD_CONSTANT_CONTEXT, 3);
*cmd++ = (drawctxt->gpustate.gpuaddr + TEX_OFFSET) & 0xFFFFE000;
#ifdef CONFIG_MSM_KGSL_DISABLE_SHADOW_WRITES
/* Disable shadowing */
*cmd++ = (0 << 24) | (1 << 16) | 0;
#else
*cmd++ = (1 << 24) | (1 << 16) | 0;
#endif
*cmd++ = TEX_CONSTANTS;
/* Boolean Constants */
*cmd++ = cp_type3_packet(CP_SET_CONSTANT, 1 + BOOL_CONSTANTS);
*cmd++ = (2 << 16) | 0;
/* the next BOOL_CONSTANT dwords is the shadow area for
* boolean constants.
*/
tmp_ctx.bool_shadow = virt2gpu(cmd, &drawctxt->gpustate);
cmd += BOOL_CONSTANTS;
/* Loop Constants */
*cmd++ = cp_type3_packet(CP_SET_CONSTANT, 1 + LOOP_CONSTANTS);
*cmd++ = (3 << 16) | 0;
/* the next LOOP_CONSTANTS dwords is the shadow area for
* loop constants.
*/
tmp_ctx.loop_shadow = virt2gpu(cmd, &drawctxt->gpustate);
cmd += LOOP_CONSTANTS;
/* create indirect buffer command for above command sequence */
create_ib1(drawctxt, drawctxt->reg_restore, start, cmd);
tmp_ctx.cmd = cmd;
}
/* save h/w regs, alu constants, texture contants, etc. ...
* requires: bool_shadow_gpuaddr, loop_shadow_gpuaddr
*/
static void build_regsave_cmds(struct adreno_device *adreno_dev,
struct adreno_context *drawctxt)
{
unsigned int *start = tmp_ctx.cmd;
unsigned int *cmd = start;
*cmd++ = cp_type3_packet(CP_WAIT_FOR_IDLE, 1);
*cmd++ = 0;
#ifdef CONFIG_MSM_KGSL_DISABLE_SHADOW_WRITES
/* Make sure the HW context has the correct register values
* before reading them. */
*cmd++ = cp_type3_packet(CP_CONTEXT_UPDATE, 1);
*cmd++ = 0;
{
unsigned int i = 0;
unsigned int reg_array_size = 0;
const unsigned int *ptr_register_ranges;
/* Based on chip id choose the register ranges */
if (adreno_is_a220(adreno_dev)) {
ptr_register_ranges = register_ranges_a220;
reg_array_size = ARRAY_SIZE(register_ranges_a220);
} else if (adreno_is_a225(adreno_dev)) {
ptr_register_ranges = register_ranges_a225;
reg_array_size = ARRAY_SIZE(register_ranges_a225);
} else {
ptr_register_ranges = register_ranges_a20x;
reg_array_size = ARRAY_SIZE(register_ranges_a20x);
}
/* Write HW registers into shadow */
for (i = 0; i < (reg_array_size/2) ; i++) {
build_reg_to_mem_range(ptr_register_ranges[i*2],
ptr_register_ranges[i*2+1],
&cmd, drawctxt);
}
}
/* Copy ALU constants */
cmd =
reg_to_mem(cmd, (drawctxt->gpustate.gpuaddr) & 0xFFFFE000,
REG_SQ_CONSTANT_0, ALU_CONSTANTS);
/* Copy Tex constants */
cmd =
reg_to_mem(cmd,
(drawctxt->gpustate.gpuaddr + TEX_OFFSET) & 0xFFFFE000,
REG_SQ_FETCH_0, TEX_CONSTANTS);
#else
/* Insert a wait for idle packet before reading the registers.
* This is to fix a hang/reset seen during stress testing. In this
* hang, CP encountered a timeout reading SQ's boolean constant
* register. There is logic in the HW that blocks reading of this
* register when the SQ block is not idle, which we believe is
* contributing to the hang.*/
*cmd++ = cp_type3_packet(CP_WAIT_FOR_IDLE, 1);
*cmd++ = 0;
/* H/w registers are already shadowed; just need to disable shadowing
* to prevent corruption.
*/
*cmd++ = cp_type3_packet(CP_LOAD_CONSTANT_CONTEXT, 3);
*cmd++ = (drawctxt->gpustate.gpuaddr + REG_OFFSET) & 0xFFFFE000;
*cmd++ = 4 << 16; /* regs, start=0 */
*cmd++ = 0x0; /* count = 0 */
/* ALU constants are already shadowed; just need to disable shadowing
* to prevent corruption.
*/
*cmd++ = cp_type3_packet(CP_LOAD_CONSTANT_CONTEXT, 3);
*cmd++ = drawctxt->gpustate.gpuaddr & 0xFFFFE000;
*cmd++ = 0 << 16; /* ALU, start=0 */
*cmd++ = 0x0; /* count = 0 */
/* Tex constants are already shadowed; just need to disable shadowing
* to prevent corruption.
*/
*cmd++ = cp_type3_packet(CP_LOAD_CONSTANT_CONTEXT, 3);
*cmd++ = (drawctxt->gpustate.gpuaddr + TEX_OFFSET) & 0xFFFFE000;
*cmd++ = 1 << 16; /* Tex, start=0 */
*cmd++ = 0x0; /* count = 0 */
#endif
/* Need to handle some of the registers separately */
*cmd++ = cp_type3_packet(CP_REG_TO_MEM, 2);
*cmd++ = REG_SQ_GPR_MANAGEMENT;
*cmd++ = tmp_ctx.reg_values[0];
*cmd++ = cp_type3_packet(CP_REG_TO_MEM, 2);
*cmd++ = REG_TP0_CHICKEN;
*cmd++ = tmp_ctx.reg_values[1];
if (adreno_is_a22x(adreno_dev)) {
unsigned int i;
unsigned int j = 2;
for (i = REG_A220_VSC_BIN_SIZE; i <=
REG_A220_VSC_PIPE_DATA_LENGTH_7; i++) {
*cmd++ = cp_type3_packet(CP_REG_TO_MEM, 2);
*cmd++ = i;
*cmd++ = tmp_ctx.reg_values[j];
j++;
}
}
/* Copy Boolean constants */
cmd = reg_to_mem(cmd, tmp_ctx.bool_shadow, REG_SQ_CF_BOOLEANS,
BOOL_CONSTANTS);
/* Copy Loop constants */
cmd = reg_to_mem(cmd, tmp_ctx.loop_shadow,
REG_SQ_CF_LOOP, LOOP_CONSTANTS);
/* create indirect buffer command for above command sequence */
create_ib1(drawctxt, drawctxt->reg_save, start, cmd);
tmp_ctx.cmd = cmd;
}
/*copy colour, depth, & stencil buffers from system memory to graphics memory*/
static unsigned int *build_sys2gmem_cmds(struct adreno_device *adreno_dev,
struct adreno_context *drawctxt,
struct tmp_ctx *ctx,
struct gmem_shadow_t *shadow)
{
unsigned int *cmds = shadow->gmem_restore_commands;
unsigned int *start = cmds;
/* Store TP0_CHICKEN register */
*cmds++ = pm4_type3_packet(PM4_REG_TO_MEM, 2);
*cmds++ = REG_TP0_CHICKEN;
if (ctx)
*cmds++ = ctx->chicken_restore;
else
cmds++;
*cmds++ = pm4_type3_packet(PM4_WAIT_FOR_IDLE, 1);
*cmds++ = 0;
/* Set TP0_CHICKEN to zero */
*cmds++ = pm4_type0_packet(REG_TP0_CHICKEN, 1);
*cmds++ = 0x00000000;
/* Set PA_SC_AA_CONFIG to 0 */
*cmds++ = pm4_type0_packet(REG_PA_SC_AA_CONFIG, 1);
*cmds++ = 0x00000000;
/* shader constants */
/* vertex buffer constants */
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 7);
*cmds++ = (0x1 << 16) | (9 * 6);
/* valid(?) vtx constant flag & addr */
*cmds++ = shadow->quad_vertices.gpuaddr | 0x3;
/* limit = 12 dwords */
*cmds++ = 0x00000030;
/* valid(?) vtx constant flag & addr */
*cmds++ = shadow->quad_texcoords.gpuaddr | 0x3;
/* limit = 8 dwords */
*cmds++ = 0x00000020;
*cmds++ = 0;
*cmds++ = 0;
/* Invalidate L2 cache to make sure vertices are updated */
*cmds++ = pm4_type0_packet(REG_TC_CNTL_STATUS, 1);
*cmds++ = 0x1;
cmds = program_shader(cmds, 0, sys2gmem_vtx_pgm, SYS2GMEM_VTX_PGM_LEN);
/* Load the patched fragment shader stream */
cmds =
program_shader(cmds, 1, sys2gmem_frag_pgm, SYS2GMEM_FRAG_PGM_LEN);
/* SQ_PROGRAM_CNTL / SQ_CONTEXT_MISC */
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 3);
*cmds++ = PM4_REG(REG_SQ_PROGRAM_CNTL);
*cmds++ = 0x10030002;
*cmds++ = 0x00000008;
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 2);
*cmds++ = PM4_REG(REG_PA_SC_AA_MASK);
*cmds++ = 0x0000ffff; /* REG_PA_SC_AA_MASK */
if (!adreno_is_a220(adreno_dev)) {
/* PA_SC_VIZ_QUERY */
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 2);
*cmds++ = PM4_REG(REG_PA_SC_VIZ_QUERY);
*cmds++ = 0x0; /*REG_PA_SC_VIZ_QUERY */
}
/* RB_COLORCONTROL */
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 2);
*cmds++ = PM4_REG(REG_RB_COLORCONTROL);
*cmds++ = 0x00000c20;
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 4);
*cmds++ = PM4_REG(REG_VGT_MAX_VTX_INDX);
*cmds++ = 0x00ffffff; /* mmVGT_MAX_VTX_INDX */
*cmds++ = 0x0; /* mmVGT_MIN_VTX_INDX */
*cmds++ = 0x00000000; /* mmVGT_INDX_OFFSET */
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 3);
*cmds++ = PM4_REG(REG_VGT_VERTEX_REUSE_BLOCK_CNTL);
*cmds++ = 0x00000002; /* mmVGT_VERTEX_REUSE_BLOCK_CNTL */
*cmds++ = 0x00000002; /* mmVGT_OUT_DEALLOC_CNTL */
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 2);
*cmds++ = PM4_REG(REG_SQ_INTERPOLATOR_CNTL);
*cmds++ = 0xffffffff; /* mmSQ_INTERPOLATOR_CNTL */
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 2);
*cmds++ = PM4_REG(REG_PA_SC_AA_CONFIG);
*cmds++ = 0x00000000; /* REG_PA_SC_AA_CONFIG */
/* set REG_PA_SU_SC_MODE_CNTL
* Front_ptype = draw triangles
* Back_ptype = draw triangles
* Provoking vertex = last
*/
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 2);
*cmds++ = PM4_REG(REG_PA_SU_SC_MODE_CNTL);
*cmds++ = 0x00080240;
/* texture constants */
*cmds++ =
pm4_type3_packet(PM4_SET_CONSTANT, (SYS2GMEM_TEX_CONST_LEN + 1));
*cmds++ = (0x1 << 16) | (0 * 6);
memcpy(cmds, sys2gmem_tex_const, SYS2GMEM_TEX_CONST_LEN << 2);
cmds[0] |= (shadow->pitch >> 5) << 22;
cmds[1] |=
shadow->gmemshadow.gpuaddr | surface_format_table[shadow->format];
cmds[2] |= (shadow->width - 1) | (shadow->height - 1) << 13;
cmds += SYS2GMEM_TEX_CONST_LEN;
/* program surface info */
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 3);
*cmds++ = PM4_REG(REG_RB_SURFACE_INFO);
*cmds++ = shadow->gmem_pitch; /* pitch, MSAA = 1 */
/* RB_COLOR_INFO Endian=none, Linear, Format=RGBA8888, Swap=0,
* Base=gmem_base
*/
if (ctx)
*cmds++ =
(shadow->
format << RB_COLOR_INFO__COLOR_FORMAT__SHIFT) | ctx->
gmem_base;
else {
unsigned int temp = *cmds;
*cmds++ = (temp & ~RB_COLOR_INFO__COLOR_FORMAT_MASK) |
(shadow->format << RB_COLOR_INFO__COLOR_FORMAT__SHIFT);
}
/* RB_DEPTHCONTROL */
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 2);
*cmds++ = PM4_REG(REG_RB_DEPTHCONTROL);
if (adreno_is_a220(adreno_dev))
*cmds++ = 8; /* disable Z */
else
*cmds++ = 0; /* disable Z */
/* Use maximum scissor values -- quad vertices already
* have the correct bounds */
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 3);
*cmds++ = PM4_REG(REG_PA_SC_SCREEN_SCISSOR_TL);
*cmds++ = (0 << 16) | 0;
*cmds++ = ((0x1fff) << 16) | 0x1fff;
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 3);
*cmds++ = PM4_REG(REG_PA_SC_WINDOW_SCISSOR_TL);
*cmds++ = (unsigned int)((1U << 31) | (0 << 16) | 0);
*cmds++ = ((0x1fff) << 16) | 0x1fff;
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 2);
*cmds++ = PM4_REG(REG_PA_CL_VTE_CNTL);
/* disable X/Y/Z transforms, X/Y/Z are premultiplied by W */
*cmds++ = 0x00000b00;
/*load the viewport so that z scale = clear depth and z offset = 0.0f */
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 3);
*cmds++ = PM4_REG(REG_PA_CL_VPORT_ZSCALE);
*cmds++ = 0xbf800000;
*cmds++ = 0x0;
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 2);
*cmds++ = PM4_REG(REG_RB_COLOR_MASK);
*cmds++ = 0x0000000f; /* R = G = B = 1:enabled */
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 2);
*cmds++ = PM4_REG(REG_RB_COLOR_DEST_MASK);
*cmds++ = 0xffffffff;
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 3);
*cmds++ = PM4_REG(REG_SQ_WRAPPING_0);
*cmds++ = 0x00000000;
*cmds++ = 0x00000000;
/* load the stencil ref value
* $AAM - do this later
*/
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 2);
*cmds++ = PM4_REG(REG_RB_MODECONTROL);
/* draw pixels with color and depth/stencil component */
*cmds++ = 0x4;
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 2);
*cmds++ = PM4_REG(REG_PA_CL_CLIP_CNTL);
*cmds++ = 0x00010000;
if (adreno_is_a220(adreno_dev)) {
*cmds++ = pm4_type3_packet(PM4_SET_DRAW_INIT_FLAGS, 1);
*cmds++ = 0;
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 2);
*cmds++ = PM4_REG(REG_LEIA_RB_LRZ_VSC_CONTROL);
*cmds++ = 0x0000000;
*cmds++ = pm4_type3_packet(PM4_DRAW_INDX, 3);
*cmds++ = 0; /* viz query info. */
/* PrimType=RectList, SrcSel=AutoIndex, VisCullMode=Ignore */
*cmds++ = 0x00004088;
*cmds++ = 3; /* NumIndices=3 */
} else {
/* queue the draw packet */
*cmds++ = pm4_type3_packet(PM4_DRAW_INDX, 2);
*cmds++ = 0; /* viz query info. */
/* PrimType=RectList, NumIndices=3, SrcSel=AutoIndex */
*cmds++ = 0x00030088;
}
/* create indirect buffer command for above command sequence */
create_ib1(drawctxt, shadow->gmem_restore, start, cmds);
return cmds;
}
/*copy colour, depth, & stencil buffers from graphics memory to system memory*/
static unsigned int *build_gmem2sys_cmds(struct adreno_device *adreno_dev,
struct adreno_context *drawctxt,
struct tmp_ctx *ctx,
struct gmem_shadow_t *shadow)
{
unsigned int *cmds = shadow->gmem_save_commands;
unsigned int *start = cmds;
/* Calculate the new offset based on the adjusted base */
unsigned int bytesperpixel = format2bytesperpixel[shadow->format];
unsigned int addr = shadow->gmemshadow.gpuaddr;
unsigned int offset = (addr - (addr & 0xfffff000)) / bytesperpixel;
/* Store TP0_CHICKEN register */
*cmds++ = pm4_type3_packet(PM4_REG_TO_MEM, 2);
*cmds++ = REG_TP0_CHICKEN;
if (ctx)
*cmds++ = ctx->chicken_restore;
else
cmds++;
*cmds++ = pm4_type3_packet(PM4_WAIT_FOR_IDLE, 1);
*cmds++ = 0;
/* Set TP0_CHICKEN to zero */
*cmds++ = pm4_type0_packet(REG_TP0_CHICKEN, 1);
*cmds++ = 0x00000000;
/* Set PA_SC_AA_CONFIG to 0 */
*cmds++ = pm4_type0_packet(REG_PA_SC_AA_CONFIG, 1);
*cmds++ = 0x00000000;
/* program shader */
/* load shader vtx constants ... 5 dwords */
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 4);
*cmds++ = (0x1 << 16) | SHADER_CONST_ADDR;
*cmds++ = 0;
/* valid(?) vtx constant flag & addr */
*cmds++ = shadow->quad_vertices.gpuaddr | 0x3;
/* limit = 12 dwords */
*cmds++ = 0x00000030;
/* Invalidate L2 cache to make sure vertices are updated */
*cmds++ = pm4_type0_packet(REG_TC_CNTL_STATUS, 1);
*cmds++ = 0x1;
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 4);
*cmds++ = PM4_REG(REG_VGT_MAX_VTX_INDX);
*cmds++ = 0x00ffffff; /* REG_VGT_MAX_VTX_INDX */
*cmds++ = 0x0; /* REG_VGT_MIN_VTX_INDX */
*cmds++ = 0x00000000; /* REG_VGT_INDX_OFFSET */
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 2);
*cmds++ = PM4_REG(REG_PA_SC_AA_MASK);
*cmds++ = 0x0000ffff; /* REG_PA_SC_AA_MASK */
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 2);
*cmds++ = PM4_REG(REG_RB_COLORCONTROL);
*cmds++ = 0x00000c20;
/* load the patched vertex shader stream */
cmds = program_shader(cmds, 0, gmem2sys_vtx_pgm, GMEM2SYS_VTX_PGM_LEN);
/* Load the patched fragment shader stream */
cmds =
program_shader(cmds, 1, gmem2sys_frag_pgm, GMEM2SYS_FRAG_PGM_LEN);
/* SQ_PROGRAM_CNTL / SQ_CONTEXT_MISC */
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 3);
*cmds++ = PM4_REG(REG_SQ_PROGRAM_CNTL);
if (adreno_is_a220(adreno_dev))
*cmds++ = 0x10018001;
else
*cmds++ = 0x10010001;
*cmds++ = 0x00000008;
/* resolve */
/* PA_CL_VTE_CNTL */
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 2);
*cmds++ = PM4_REG(REG_PA_CL_VTE_CNTL);
/* disable X/Y/Z transforms, X/Y/Z are premultiplied by W */
*cmds++ = 0x00000b00;
/* program surface info */
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 3);
*cmds++ = PM4_REG(REG_RB_SURFACE_INFO);
*cmds++ = shadow->gmem_pitch; /* pitch, MSAA = 1 */
/* RB_COLOR_INFO Endian=none, Linear, Format=RGBA8888, Swap=0,
* Base=gmem_base
*/
/* gmem base assumed 4K aligned. */
if (ctx) {
BUG_ON(ctx->gmem_base & 0xFFF);
*cmds++ =
(shadow->
format << RB_COLOR_INFO__COLOR_FORMAT__SHIFT) | ctx->
gmem_base;
} else {
unsigned int temp = *cmds;
*cmds++ = (temp & ~RB_COLOR_INFO__COLOR_FORMAT_MASK) |
(shadow->format << RB_COLOR_INFO__COLOR_FORMAT__SHIFT);
}
/* disable Z */
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 2);
*cmds++ = PM4_REG(REG_RB_DEPTHCONTROL);
if (adreno_is_a220(adreno_dev))
*cmds++ = 0x08;
else
*cmds++ = 0;
/* set REG_PA_SU_SC_MODE_CNTL
* Front_ptype = draw triangles
* Back_ptype = draw triangles
* Provoking vertex = last
*/
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 2);
*cmds++ = PM4_REG(REG_PA_SU_SC_MODE_CNTL);
*cmds++ = 0x00080240;
/* Use maximum scissor values -- quad vertices already have the
* correct bounds */
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 3);
*cmds++ = PM4_REG(REG_PA_SC_SCREEN_SCISSOR_TL);
*cmds++ = (0 << 16) | 0;
*cmds++ = (0x1fff << 16) | (0x1fff);
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 3);
*cmds++ = PM4_REG(REG_PA_SC_WINDOW_SCISSOR_TL);
*cmds++ = (unsigned int)((1U << 31) | (0 << 16) | 0);
*cmds++ = (0x1fff << 16) | (0x1fff);
/* load the viewport so that z scale = clear depth and
* z offset = 0.0f
*/
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 3);
*cmds++ = PM4_REG(REG_PA_CL_VPORT_ZSCALE);
*cmds++ = 0xbf800000; /* -1.0f */
*cmds++ = 0x0;
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 2);
*cmds++ = PM4_REG(REG_RB_COLOR_MASK);
*cmds++ = 0x0000000f; /* R = G = B = 1:enabled */
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 2);
*cmds++ = PM4_REG(REG_RB_COLOR_DEST_MASK);
*cmds++ = 0xffffffff;
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 3);
*cmds++ = PM4_REG(REG_SQ_WRAPPING_0);
*cmds++ = 0x00000000;
*cmds++ = 0x00000000;
/* load the stencil ref value
* $AAM - do this later
*/
/* load the COPY state */
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 6);
*cmds++ = PM4_REG(REG_RB_COPY_CONTROL);
*cmds++ = 0; /* RB_COPY_CONTROL */
*cmds++ = addr & 0xfffff000; /* RB_COPY_DEST_BASE */
*cmds++ = shadow->pitch >> 5; /* RB_COPY_DEST_PITCH */
/* Endian=none, Linear, Format=RGBA8888,Swap=0,!Dither,
* MaskWrite:R=G=B=A=1
*/
*cmds++ = 0x0003c008 |
(shadow->format << RB_COPY_DEST_INFO__COPY_DEST_FORMAT__SHIFT);
/* Make sure we stay in offsetx field. */
BUG_ON(offset & 0xfffff000);
*cmds++ = offset;
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 2);
*cmds++ = PM4_REG(REG_RB_MODECONTROL);
*cmds++ = 0x6; /* EDRAM copy */
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 2);
*cmds++ = PM4_REG(REG_PA_CL_CLIP_CNTL);
*cmds++ = 0x00010000;
if (adreno_is_a220(adreno_dev)) {
*cmds++ = pm4_type3_packet(PM4_SET_DRAW_INIT_FLAGS, 1);
*cmds++ = 0;
*cmds++ = pm4_type3_packet(PM4_SET_CONSTANT, 2);
*cmds++ = PM4_REG(REG_LEIA_RB_LRZ_VSC_CONTROL);
*cmds++ = 0x0000000;
*cmds++ = pm4_type3_packet(PM4_DRAW_INDX, 3);
*cmds++ = 0; /* viz query info. */
/* PrimType=RectList, SrcSel=AutoIndex, VisCullMode=Ignore */
*cmds++ = 0x00004088;
*cmds++ = 3; /* NumIndices=3 */
} else {
/* queue the draw packet */
*cmds++ = pm4_type3_packet(PM4_DRAW_INDX, 2);
*cmds++ = 0; /* viz query info. */
/* PrimType=RectList, NumIndices=3, SrcSel=AutoIndex */
*cmds++ = 0x00030088;
}
/* create indirect buffer command for above command sequence */
create_ib1(drawctxt, shadow->gmem_save, start, cmds);
return cmds;
}
/* switch drawing contexts */
void
adreno_drawctxt_switch(struct adreno_device *adreno_dev,
struct adreno_context *drawctxt,
unsigned int flags)
{
struct adreno_context *active_ctxt =
adreno_dev->drawctxt_active;
struct kgsl_device *device = &adreno_dev->dev;
unsigned int cmds[5];
if (drawctxt) {
if (flags & KGSL_CONTEXT_SAVE_GMEM)
/* Set the flag in context so that the save is done
* when this context is switched out. */
drawctxt->flags |= CTXT_FLAGS_GMEM_SAVE;
else
/* Remove GMEM saving flag from the context */
drawctxt->flags &= ~CTXT_FLAGS_GMEM_SAVE;
}
/* already current? */
if (active_ctxt == drawctxt)
return;
KGSL_CTXT_INFO(device, "from %p to %p flags %d\n",
adreno_dev->drawctxt_active, drawctxt, flags);
/* save old context*/
if (active_ctxt && active_ctxt->flags & CTXT_FLAGS_GPU_HANG)
KGSL_CTXT_WARN(device,
"Current active context has caused gpu hang\n");
if (active_ctxt != NULL) {
KGSL_CTXT_INFO(device,
"active_ctxt flags %08x\n", active_ctxt->flags);
/* save registers and constants. */
adreno_ringbuffer_issuecmds(device, 0,
active_ctxt->reg_save, 3);
if (active_ctxt->flags & CTXT_FLAGS_SHADER_SAVE) {
/* save shader partitioning and instructions. */
adreno_ringbuffer_issuecmds(device,
KGSL_CMD_FLAGS_PMODE,
active_ctxt->shader_save, 3);
/* fixup shader partitioning parameter for
* SET_SHADER_BASES.
*/
adreno_ringbuffer_issuecmds(device, 0,
active_ctxt->shader_fixup, 3);
active_ctxt->flags |= CTXT_FLAGS_SHADER_RESTORE;
}
if (active_ctxt->flags & CTXT_FLAGS_GMEM_SAVE
&& active_ctxt->flags & CTXT_FLAGS_GMEM_SHADOW) {
/* save gmem.
* (note: changes shader. shader must already be saved.)
*/
adreno_ringbuffer_issuecmds(device,
KGSL_CMD_FLAGS_PMODE,
active_ctxt->context_gmem_shadow.gmem_save, 3);
/* Restore TP0_CHICKEN */
adreno_ringbuffer_issuecmds(device, 0,
active_ctxt->chicken_restore, 3);
active_ctxt->flags |= CTXT_FLAGS_GMEM_RESTORE;
}
}
adreno_dev->drawctxt_active = drawctxt;
/* restore new context */
if (drawctxt != NULL) {
KGSL_CTXT_INFO(device,
"drawctxt flags %08x\n", drawctxt->flags);
cmds[0] = pm4_nop_packet(1);
cmds[1] = KGSL_CONTEXT_TO_MEM_IDENTIFIER;
cmds[2] = pm4_type3_packet(PM4_MEM_WRITE, 2);
cmds[3] = device->memstore.gpuaddr +
KGSL_DEVICE_MEMSTORE_OFFSET(current_context);
cmds[4] = (unsigned int)adreno_dev->drawctxt_active;
adreno_ringbuffer_issuecmds(device, 0, cmds, 5);
kgsl_mmu_setstate(device, drawctxt->pagetable);
#ifndef CONFIG_MSM_KGSL_CFF_DUMP_NO_CONTEXT_MEM_DUMP
kgsl_cffdump_syncmem(NULL, &drawctxt->gpustate,
drawctxt->gpustate.gpuaddr, LCC_SHADOW_SIZE +
REG_SHADOW_SIZE + CMD_BUFFER_SIZE + TEX_SHADOW_SIZE,
false);
#endif
/* restore gmem.
* (note: changes shader. shader must not already be restored.)
*/
if (drawctxt->flags & CTXT_FLAGS_GMEM_RESTORE) {
adreno_ringbuffer_issuecmds(device,
KGSL_CMD_FLAGS_PMODE,
drawctxt->context_gmem_shadow.gmem_restore, 3);
/* Restore TP0_CHICKEN */
adreno_ringbuffer_issuecmds(device, 0,
drawctxt->chicken_restore, 3);
drawctxt->flags &= ~CTXT_FLAGS_GMEM_RESTORE;
}
/* restore registers and constants. */
adreno_ringbuffer_issuecmds(device, 0,
drawctxt->reg_restore, 3);
/* restore shader instructions & partitioning. */
if (drawctxt->flags & CTXT_FLAGS_SHADER_RESTORE) {
adreno_ringbuffer_issuecmds(device, 0,
drawctxt->shader_restore, 3);
}
cmds[0] = pm4_type3_packet(PM4_SET_BIN_BASE_OFFSET, 1);
cmds[1] = drawctxt->bin_base_offset;
if (!adreno_is_a220(adreno_dev))
adreno_ringbuffer_issuecmds(device, 0, cmds, 2);
} else
kgsl_mmu_setstate(device, device->mmu.defaultpagetable);
}
void *a2xx_snapshot(struct adreno_device *adreno_dev, void *snapshot,
int *remain, int hang)
{
struct kgsl_device *device = &adreno_dev->dev;
struct kgsl_snapshot_registers_list list;
struct kgsl_snapshot_registers regs;
unsigned int pmoverride;
/* Choose the register set to dump */
if (adreno_is_a20x(adreno_dev)) {
regs.regs = (unsigned int *) a200_registers;
regs.count = a200_registers_count;
} else if (adreno_is_a220(adreno_dev)) {
regs.regs = (unsigned int *) a220_registers;
regs.count = a220_registers_count;
} else if (adreno_is_a225(adreno_dev)) {
regs.regs = (unsigned int *) a225_registers;
regs.count = a225_registers_count;
}
list.registers = ®s;
list.count = 1;
/* Master set of (non debug) registers */
snapshot = kgsl_snapshot_add_section(device,
KGSL_SNAPSHOT_SECTION_REGS, snapshot, remain,
kgsl_snapshot_dump_regs, &list);
/* CP_STATE_DEBUG indexed registers */
snapshot = kgsl_snapshot_indexed_registers(device, snapshot,
remain, REG_CP_STATE_DEBUG_INDEX,
REG_CP_STATE_DEBUG_DATA, 0x0, 0x14);
/* CP_ME indexed registers */
snapshot = kgsl_snapshot_indexed_registers(device, snapshot,
remain, REG_CP_ME_CNTL, REG_CP_ME_STATUS,
64, 44);
/*
* Need to temporarily turn off clock gating for the debug bus to
* work
*/
adreno_regread(device, REG_RBBM_PM_OVERRIDE2, &pmoverride);
adreno_regwrite(device, REG_RBBM_PM_OVERRIDE2, 0xFF);
/* SX debug registers */
snapshot = kgsl_snapshot_add_section(device,
KGSL_SNAPSHOT_SECTION_DEBUG, snapshot, remain,
a2xx_snapshot_sxdebug, NULL);
/* SU debug indexed registers (only for < 470) */
if (!adreno_is_a22x(adreno_dev))
snapshot = kgsl_snapshot_indexed_registers(device, snapshot,
remain, REG_PA_SU_DEBUG_CNTL,
REG_PA_SU_DEBUG_DATA,
0, 0x1B);
/* CP debug registers */
snapshot = kgsl_snapshot_add_section(device,
KGSL_SNAPSHOT_SECTION_DEBUG, snapshot, remain,
a2xx_snapshot_cpdebug, NULL);
/* MH debug indexed registers */
snapshot = kgsl_snapshot_indexed_registers(device, snapshot,
remain, MH_DEBUG_CTRL, MH_DEBUG_DATA, 0x0, 0x40);
/* Leia only register sets */
if (adreno_is_a22x(adreno_dev)) {
/* RB DEBUG indexed regisers */
snapshot = kgsl_snapshot_indexed_registers(device, snapshot,
remain, REG_RB_DEBUG_CNTL, REG_RB_DEBUG_DATA, 0, 8);
/* RB DEBUG indexed registers bank 2 */
snapshot = kgsl_snapshot_indexed_registers(device, snapshot,
remain, REG_RB_DEBUG_CNTL, REG_RB_DEBUG_DATA + 0x1000,
0, 8);
/* PC_DEBUG indexed registers */
snapshot = kgsl_snapshot_indexed_registers(device, snapshot,
remain, REG_PC_DEBUG_CNTL, REG_PC_DEBUG_DATA, 0, 8);
/* GRAS_DEBUG indexed registers */
snapshot = kgsl_snapshot_indexed_registers(device, snapshot,
remain, REG_GRAS_DEBUG_CNTL, REG_GRAS_DEBUG_DATA, 0, 4);
/* MIU debug registers */
snapshot = kgsl_snapshot_add_section(device,
KGSL_SNAPSHOT_SECTION_DEBUG, snapshot, remain,
a2xx_snapshot_miudebug, NULL);
/* SQ DEBUG debug registers */
snapshot = kgsl_snapshot_add_section(device,
KGSL_SNAPSHOT_SECTION_DEBUG, snapshot, remain,
a2xx_snapshot_sqdebug, NULL);
/*
* Reading SQ THREAD causes bad things to happen on a running
* system, so only read it if the GPU is already hung
*/
if (hang) {
/* SQ THREAD debug registers */
snapshot = kgsl_snapshot_add_section(device,
KGSL_SNAPSHOT_SECTION_DEBUG, snapshot, remain,
a2xx_snapshot_sqthreaddebug, NULL);
}
}
/* Reset the clock gating */
adreno_regwrite(device, REG_RBBM_PM_OVERRIDE2, pmoverride);
return snapshot;
}
static void build_regrestore_cmds(struct adreno_device *adreno_dev,
struct adreno_context *drawctxt)
{
unsigned int *start = tmp_ctx.cmd;
unsigned int *cmd = start;
unsigned int i = 0;
unsigned int reg_array_size = 0;
const unsigned int *ptr_register_ranges;
*cmd++ = cp_type3_packet(CP_WAIT_FOR_IDLE, 1);
*cmd++ = 0;
cmd++;
#ifdef CONFIG_MSM_KGSL_DISABLE_SHADOW_WRITES
*cmd++ = ((drawctxt->gpustate.gpuaddr + REG_OFFSET) & 0xFFFFE000) | 1;
#else
*cmd++ = (drawctxt->gpustate.gpuaddr + REG_OFFSET) & 0xFFFFE000;
#endif
if (adreno_is_a220(adreno_dev)) {
ptr_register_ranges = register_ranges_a220;
reg_array_size = ARRAY_SIZE(register_ranges_a220);
} else if (adreno_is_a225(adreno_dev)) {
ptr_register_ranges = register_ranges_a225;
reg_array_size = ARRAY_SIZE(register_ranges_a225);
} else {
ptr_register_ranges = register_ranges_a20x;
reg_array_size = ARRAY_SIZE(register_ranges_a20x);
}
for (i = 0; i < (reg_array_size/2); i++) {
cmd = reg_range(cmd, ptr_register_ranges[i*2],
ptr_register_ranges[i*2+1]);
}
start[2] =
cp_type3_packet(CP_LOAD_CONSTANT_CONTEXT, (cmd - start) - 3);
#ifdef CONFIG_MSM_KGSL_DISABLE_SHADOW_WRITES
start[4] |= (0 << 24) | (4 << 16);
#else
start[4] |= (1 << 24) | (4 << 16);
#endif
*cmd++ = cp_type0_packet(REG_SQ_GPR_MANAGEMENT, 1);
tmp_ctx.reg_values[0] = virt2gpu(cmd, &drawctxt->gpustate);
*cmd++ = 0x00040400;
*cmd++ = cp_type3_packet(CP_WAIT_FOR_IDLE, 1);
*cmd++ = 0;
*cmd++ = cp_type0_packet(REG_TP0_CHICKEN, 1);
tmp_ctx.reg_values[1] = virt2gpu(cmd, &drawctxt->gpustate);
*cmd++ = 0x00000000;
if (adreno_is_a20x(adreno_dev)) {
*cmd++ = cp_type0_packet(REG_RB_BC_CONTROL, 1);
tmp_ctx.reg_values[2] = virt2gpu(cmd, &drawctxt->gpustate);
*cmd++ = 0x00000000;
}
if (adreno_is_a22x(adreno_dev)) {
unsigned int i;
unsigned int j = 2;
for (i = REG_A220_VSC_BIN_SIZE; i <=
REG_A220_VSC_PIPE_DATA_LENGTH_7; i++) {
*cmd++ = cp_type0_packet(i, 1);
tmp_ctx.reg_values[j] = virt2gpu(cmd,
&drawctxt->gpustate);
*cmd++ = 0x00000000;
j++;
}
}
*cmd++ = cp_type3_packet(CP_LOAD_CONSTANT_CONTEXT, 3);
*cmd++ = drawctxt->gpustate.gpuaddr & 0xFFFFE000;
#ifdef CONFIG_MSM_KGSL_DISABLE_SHADOW_WRITES
*cmd++ = (0 << 24) | (0 << 16) | 0;
#else
*cmd++ = (1 << 24) | (0 << 16) | 0;
#endif
*cmd++ = ALU_CONSTANTS;
*cmd++ = cp_type3_packet(CP_LOAD_CONSTANT_CONTEXT, 3);
*cmd++ = (drawctxt->gpustate.gpuaddr + TEX_OFFSET) & 0xFFFFE000;
#ifdef CONFIG_MSM_KGSL_DISABLE_SHADOW_WRITES
*cmd++ = (0 << 24) | (1 << 16) | 0;
#else
*cmd++ = (1 << 24) | (1 << 16) | 0;
#endif
*cmd++ = TEX_CONSTANTS;
*cmd++ = cp_type3_packet(CP_SET_CONSTANT, 1 + BOOL_CONSTANTS);
*cmd++ = (2 << 16) | 0;
tmp_ctx.bool_shadow = virt2gpu(cmd, &drawctxt->gpustate);
cmd += BOOL_CONSTANTS;
*cmd++ = cp_type3_packet(CP_SET_CONSTANT, 1 + LOOP_CONSTANTS);
*cmd++ = (3 << 16) | 0;
tmp_ctx.loop_shadow = virt2gpu(cmd, &drawctxt->gpustate);
cmd += LOOP_CONSTANTS;
create_ib1(drawctxt, drawctxt->reg_restore, start, cmd);
tmp_ctx.cmd = cmd;
}
static void build_regsave_cmds(struct adreno_device *adreno_dev,
struct adreno_context *drawctxt)
{
unsigned int *start = tmp_ctx.cmd;
unsigned int *cmd = start;
*cmd++ = cp_type3_packet(CP_WAIT_FOR_IDLE, 1);
*cmd++ = 0;
#ifdef CONFIG_MSM_KGSL_DISABLE_SHADOW_WRITES
*cmd++ = cp_type3_packet(CP_CONTEXT_UPDATE, 1);
*cmd++ = 0;
{
unsigned int i = 0;
unsigned int reg_array_size = 0;
const unsigned int *ptr_register_ranges;
if (adreno_is_a220(adreno_dev)) {
ptr_register_ranges = register_ranges_a220;
reg_array_size = ARRAY_SIZE(register_ranges_a220);
} else if (adreno_is_a225(adreno_dev)) {
ptr_register_ranges = register_ranges_a225;
reg_array_size = ARRAY_SIZE(register_ranges_a225);
} else {
ptr_register_ranges = register_ranges_a20x;
reg_array_size = ARRAY_SIZE(register_ranges_a20x);
}
for (i = 0; i < (reg_array_size/2) ; i++) {
build_reg_to_mem_range(ptr_register_ranges[i*2],
ptr_register_ranges[i*2+1],
&cmd, drawctxt);
}
}
cmd =
reg_to_mem(cmd, (drawctxt->gpustate.gpuaddr) & 0xFFFFE000,
REG_SQ_CONSTANT_0, ALU_CONSTANTS);
cmd =
reg_to_mem(cmd,
(drawctxt->gpustate.gpuaddr + TEX_OFFSET) & 0xFFFFE000,
REG_SQ_FETCH_0, TEX_CONSTANTS);
#else
*cmd++ = cp_type3_packet(CP_WAIT_FOR_IDLE, 1);
*cmd++ = 0;
*cmd++ = cp_type3_packet(CP_LOAD_CONSTANT_CONTEXT, 3);
*cmd++ = (drawctxt->gpustate.gpuaddr + REG_OFFSET) & 0xFFFFE000;
*cmd++ = 4 << 16;
*cmd++ = 0x0;
*cmd++ = cp_type3_packet(CP_LOAD_CONSTANT_CONTEXT, 3);
*cmd++ = drawctxt->gpustate.gpuaddr & 0xFFFFE000;
*cmd++ = 0 << 16;
*cmd++ = 0x0;
*cmd++ = cp_type3_packet(CP_LOAD_CONSTANT_CONTEXT, 3);
*cmd++ = (drawctxt->gpustate.gpuaddr + TEX_OFFSET) & 0xFFFFE000;
*cmd++ = 1 << 16;
*cmd++ = 0x0;
#endif
*cmd++ = cp_type3_packet(CP_REG_TO_MEM, 2);
*cmd++ = REG_SQ_GPR_MANAGEMENT;
*cmd++ = tmp_ctx.reg_values[0];
*cmd++ = cp_type3_packet(CP_REG_TO_MEM, 2);
*cmd++ = REG_TP0_CHICKEN;
*cmd++ = tmp_ctx.reg_values[1];
if (adreno_is_a20x(adreno_dev)) {
*cmd++ = cp_type3_packet(CP_REG_TO_MEM, 2);
*cmd++ = REG_RB_BC_CONTROL;
*cmd++ = tmp_ctx.reg_values[2];
}
if (adreno_is_a22x(adreno_dev)) {
unsigned int i;
unsigned int j = 2;
for (i = REG_A220_VSC_BIN_SIZE; i <=
REG_A220_VSC_PIPE_DATA_LENGTH_7; i++) {
*cmd++ = cp_type3_packet(CP_REG_TO_MEM, 2);
*cmd++ = i;
*cmd++ = tmp_ctx.reg_values[j];
j++;
}
}
cmd = reg_to_mem(cmd, tmp_ctx.bool_shadow, REG_SQ_CF_BOOLEANS,
BOOL_CONSTANTS);
cmd = reg_to_mem(cmd, tmp_ctx.loop_shadow,
REG_SQ_CF_LOOP, LOOP_CONSTANTS);
create_ib1(drawctxt, drawctxt->reg_save, start, cmd);
tmp_ctx.cmd = cmd;
}
