static void gfx_v9_4_3_set_kiq_pm4_funcs(struct amdgpu_device *adev)
{
- adev->gfx.kiq[0].pmf = &gfx_v9_4_3_kiq_pm4_funcs;
+ int i;
+ for (i = 0; i < adev->gfx.num_xcd; i++)
+ adev->gfx.kiq[i].pmf = &gfx_v9_4_3_kiq_pm4_funcs;
}
static void gfx_v9_4_3_init_golden_registers(struct amdgpu_device *adev)
const struct gfx_firmware_header_v1_0 *mec_hdr;
- bitmap_zero(adev->gfx.mec_bitmap[0].queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
+ for (i = 0; i < adev->gfx.num_xcd; i++)
+ bitmap_zero(adev->gfx.mec_bitmap[i].queue_bitmap,
+ AMDGPU_MAX_COMPUTE_QUEUES);
/* take ownership of the relevant compute queues */
amdgpu_gfx_compute_queue_acquire(adev);
}
static int gfx_v9_4_3_compute_ring_init(struct amdgpu_device *adev, int ring_id,
- int mec, int pipe, int queue)
+ int xcc_id, int mec, int pipe, int queue)
{
unsigned irq_type;
struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id];
ring = &adev->gfx.compute_ring[ring_id];
/* mec0 is me1 */
+ ring->xcc_id = xcc_id;
ring->me = mec + 1;
ring->pipe = pipe;
ring->queue = queue;
ring->eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr
+ (ring_id * GFX9_MEC_HPD_SIZE);
ring->vm_hub = AMDGPU_GFXHUB_0;
- sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue);
+ sprintf(ring->name, "comp_%d.%d.%d.%d",
+ ring->xcc_id, ring->me, ring->pipe, ring->queue);
irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP
+ ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec)
static int gfx_v9_4_3_sw_init(void *handle)
{
- int i, j, k, r, ring_id;
+ int i, j, k, r, ring_id, xcc_id;
struct amdgpu_kiq *kiq;
struct amdgpu_device *adev = (struct amdgpu_device *)handle;
/* set up the compute queues - allocate horizontally across pipes */
ring_id = 0;
- for (i = 0; i < adev->gfx.mec.num_mec; ++i) {
- for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) {
- for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) {
- if (!amdgpu_gfx_is_mec_queue_enabled(adev, 0, i,
- k, j))
- continue;
-
- r = gfx_v9_4_3_compute_ring_init(adev,
- ring_id,
- i, k, j);
- if (r)
- return r;
-
- ring_id++;
+ for (xcc_id = 0; xcc_id < adev->gfx.num_xcd; xcc_id++) {
+
+ for (i = 0; i < adev->gfx.mec.num_mec; ++i) {
+ for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) {
+ for (k = 0; k < adev->gfx.mec.num_pipe_per_mec;
+ k++) {
+ if (!amdgpu_gfx_is_mec_queue_enabled(
+ adev, xcc_id, i, k, j))
+ continue;
+
+ r = gfx_v9_4_3_compute_ring_init(adev,
+ ring_id,
+ xcc_id,
+ i, k, j);
+ if (r)
+ return r;
+
+ ring_id++;
+ }
}
}
- }
- r = amdgpu_gfx_kiq_init(adev, GFX9_MEC_HPD_SIZE, 0);
- if (r) {
- DRM_ERROR("Failed to init KIQ BOs!\n");
- return r;
- }
+ r = amdgpu_gfx_kiq_init(adev, GFX9_MEC_HPD_SIZE, xcc_id);
+ if (r) {
+ DRM_ERROR("Failed to init KIQ BOs!\n");
+ return r;
+ }
- kiq = &adev->gfx.kiq[0];
- r = amdgpu_gfx_kiq_init_ring(adev, &kiq->ring, &kiq->irq, 0);
- if (r)
- return r;
+ kiq = &adev->gfx.kiq[xcc_id];
+ r = amdgpu_gfx_kiq_init_ring(adev, &kiq->ring, &kiq->irq, xcc_id);
+ if (r)
+ return r;
- /* create MQD for all compute queues as wel as KIQ for SRIOV case */
- r = amdgpu_gfx_mqd_sw_init(adev, sizeof(struct v9_mqd_allocation), 0);
- if (r)
- return r;
+ /* create MQD for all compute queues as wel as KIQ for SRIOV case */
+ r = amdgpu_gfx_mqd_sw_init(adev,
+ sizeof(struct v9_mqd_allocation), xcc_id);
+ if (r)
+ return r;
+ }
r = gfx_v9_4_3_gpu_early_init(adev);
if (r)
int i;
struct amdgpu_device *adev = (struct amdgpu_device *)handle;
- for (i = 0; i < adev->gfx.num_compute_rings; i++)
+ for (i = 0; i < adev->gfx.num_compute_rings *
+ adev->gfx.num_xcd; i++)
amdgpu_ring_fini(&adev->gfx.compute_ring[i]);
- amdgpu_gfx_mqd_sw_fini(adev, 0);
- amdgpu_gfx_kiq_free_ring(&adev->gfx.kiq[0].ring);
- amdgpu_gfx_kiq_fini(adev, 0);
+ for (i = 0; i < adev->gfx.num_xcd; i++) {
+ amdgpu_gfx_mqd_sw_fini(adev, i);
+ amdgpu_gfx_kiq_free_ring(&adev->gfx.kiq[i].ring);
+ amdgpu_gfx_kiq_fini(adev, i);
+ }
gfx_v9_4_3_mec_fini(adev);
amdgpu_bo_unref(&adev->gfx.rlc.clear_state_obj);
return (~data) & mask;
}
-static void gfx_v9_4_3_setup_rb(struct amdgpu_device *adev)
+static void gfx_v9_4_3_setup_rb(struct amdgpu_device *adev, int xcc_id)
{
int i, j;
u32 data;
}
#define DEFAULT_SH_MEM_BASES (0x6000)
-static void gfx_v9_4_3_init_compute_vmid(struct amdgpu_device *adev)
+static void gfx_v9_4_3_init_compute_vmid(struct amdgpu_device *adev, int xcc_id)
{
int i;
uint32_t sh_mem_config;
mutex_lock(&adev->srbm_mutex);
for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) {
- soc15_grbm_select(adev, 0, 0, 0, i, 0);
+ soc15_grbm_select(adev, 0, 0, 0, i, xcc_id);
/* CP and shaders */
- WREG32_SOC15_RLC(GC, 0, regSH_MEM_CONFIG, sh_mem_config);
- WREG32_SOC15_RLC(GC, 0, regSH_MEM_BASES, sh_mem_bases);
+ WREG32_SOC15_RLC(GC, xcc_id, regSH_MEM_CONFIG, sh_mem_config);
+ WREG32_SOC15_RLC(GC, xcc_id, regSH_MEM_BASES, sh_mem_bases);
}
- soc15_grbm_select(adev, 0, 0, 0, 0, 0);
+ soc15_grbm_select(adev, 0, 0, 0, 0, xcc_id);
mutex_unlock(&adev->srbm_mutex);
/* Initialize all compute VMIDs to have no GDS, GWS, or OA
acccess. These should be enabled by FW for target VMIDs. */
for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) {
- WREG32_SOC15_OFFSET(GC, 0, regGDS_VMID0_BASE, 2 * i, 0);
- WREG32_SOC15_OFFSET(GC, 0, regGDS_VMID0_SIZE, 2 * i, 0);
- WREG32_SOC15_OFFSET(GC, 0, regGDS_GWS_VMID0, i, 0);
- WREG32_SOC15_OFFSET(GC, 0, regGDS_OA_VMID0, i, 0);
+ WREG32_SOC15_OFFSET(GC, xcc_id, regGDS_VMID0_BASE, 2 * i, 0);
+ WREG32_SOC15_OFFSET(GC, xcc_id, regGDS_VMID0_SIZE, 2 * i, 0);
+ WREG32_SOC15_OFFSET(GC, xcc_id, regGDS_GWS_VMID0, i, 0);
+ WREG32_SOC15_OFFSET(GC, xcc_id, regGDS_OA_VMID0, i, 0);
}
}
-static void gfx_v9_4_3_init_gds_vmid(struct amdgpu_device *adev)
+static void gfx_v9_4_3_init_gds_vmid(struct amdgpu_device *adev, int xcc_id)
{
int vmid;
* access so that HWS firmware can save/restore entries.
*/
for (vmid = 1; vmid < AMDGPU_NUM_VMID; vmid++) {
- WREG32_SOC15_OFFSET(GC, 0, regGDS_VMID0_BASE, 2 * vmid, 0);
- WREG32_SOC15_OFFSET(GC, 0, regGDS_VMID0_SIZE, 2 * vmid, 0);
- WREG32_SOC15_OFFSET(GC, 0, regGDS_GWS_VMID0, vmid, 0);
- WREG32_SOC15_OFFSET(GC, 0, regGDS_OA_VMID0, vmid, 0);
+ WREG32_SOC15_OFFSET(GC, xcc_id, regGDS_VMID0_BASE, 2 * vmid, 0);
+ WREG32_SOC15_OFFSET(GC, xcc_id, regGDS_VMID0_SIZE, 2 * vmid, 0);
+ WREG32_SOC15_OFFSET(GC, xcc_id, regGDS_GWS_VMID0, vmid, 0);
+ WREG32_SOC15_OFFSET(GC, xcc_id, regGDS_OA_VMID0, vmid, 0);
}
}
static void gfx_v9_4_3_constants_init(struct amdgpu_device *adev)
{
u32 tmp;
- int i;
+ int i, j;
- WREG32_FIELD15_PREREG(GC, 0, GRBM_CNTL, READ_TIMEOUT, 0xff);
+ for (i = 0; i < adev->gfx.num_xcd; i++) {
+ WREG32_FIELD15_PREREG(GC, i, GRBM_CNTL, READ_TIMEOUT, 0xff);
+ gfx_v9_4_3_setup_rb(adev, i);
+ }
- gfx_v9_4_3_setup_rb(adev);
gfx_v9_4_3_get_cu_info(adev, &adev->gfx.cu_info);
adev->gfx.config.db_debug2 = RREG32_SOC15(GC, 0, regDB_DEBUG2);
/* where to put LDS, scratch, GPUVM in FSA64 space */
mutex_lock(&adev->srbm_mutex);
for (i = 0; i < adev->vm_manager.id_mgr[AMDGPU_GFXHUB_0].num_ids; i++) {
- soc15_grbm_select(adev, 0, 0, 0, i, 0);
- /* CP and shaders */
- if (i == 0) {
- tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE,
- SH_MEM_ALIGNMENT_MODE_UNALIGNED);
- tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, RETRY_DISABLE,
- !!adev->gmc.noretry);
- WREG32_SOC15_RLC(GC, 0, regSH_MEM_CONFIG, tmp);
- WREG32_SOC15_RLC(GC, 0, regSH_MEM_BASES, 0);
- } else {
- tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE,
- SH_MEM_ALIGNMENT_MODE_UNALIGNED);
- tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, RETRY_DISABLE,
- !!adev->gmc.noretry);
- WREG32_SOC15_RLC(GC, 0, regSH_MEM_CONFIG, tmp);
- tmp = REG_SET_FIELD(0, SH_MEM_BASES, PRIVATE_BASE,
- (adev->gmc.private_aperture_start >> 48));
- tmp = REG_SET_FIELD(tmp, SH_MEM_BASES, SHARED_BASE,
- (adev->gmc.shared_aperture_start >> 48));
- WREG32_SOC15_RLC(GC, 0, regSH_MEM_BASES, tmp);
+ for (j = 0; j < adev->gfx.num_xcd; j++) {
+ soc15_grbm_select(adev, 0, 0, 0, i, j);
+ /* CP and shaders */
+ if (i == 0) {
+ tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE,
+ SH_MEM_ALIGNMENT_MODE_UNALIGNED);
+ tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, RETRY_DISABLE,
+ !!adev->gmc.noretry);
+ WREG32_SOC15_RLC(GC, j, regSH_MEM_CONFIG, tmp);
+ WREG32_SOC15_RLC(GC, j, regSH_MEM_BASES, 0);
+ } else {
+ tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE,
+ SH_MEM_ALIGNMENT_MODE_UNALIGNED);
+ tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, RETRY_DISABLE,
+ !!adev->gmc.noretry);
+ WREG32_SOC15_RLC(GC, j, regSH_MEM_CONFIG, tmp);
+ tmp = REG_SET_FIELD(0, SH_MEM_BASES, PRIVATE_BASE,
+ (adev->gmc.private_aperture_start >> 48));
+ tmp = REG_SET_FIELD(tmp, SH_MEM_BASES, SHARED_BASE,
+ (adev->gmc.shared_aperture_start >> 48));
+ WREG32_SOC15_RLC(GC, j, regSH_MEM_BASES, tmp);
+ }
}
}
soc15_grbm_select(adev, 0, 0, 0, 0, 0);
mutex_unlock(&adev->srbm_mutex);
- gfx_v9_4_3_init_compute_vmid(adev);
- gfx_v9_4_3_init_gds_vmid(adev);
+ for (i = 0; i < adev->gfx.num_xcd; i++) {
+ gfx_v9_4_3_init_compute_vmid(adev, i);
+ gfx_v9_4_3_init_gds_vmid(adev, i);
+ }
}
-static void gfx_v9_4_3_enable_save_restore_machine(struct amdgpu_device *adev)
+static void gfx_v9_4_3_enable_save_restore_machine(struct amdgpu_device *adev,
+ int xcc_id)
{
- WREG32_FIELD15_PREREG(GC, 0, RLC_SRM_CNTL, SRM_ENABLE, 1);
+ WREG32_FIELD15_PREREG(GC, xcc_id, RLC_SRM_CNTL, SRM_ENABLE, 1);
}
-static void gfx_v9_4_3_init_csb(struct amdgpu_device *adev)
+static void gfx_v9_4_3_init_csb(struct amdgpu_device *adev, int xcc_id)
{
adev->gfx.rlc.funcs->get_csb_buffer(adev, adev->gfx.rlc.cs_ptr);
/* csib */
- WREG32_RLC(SOC15_REG_OFFSET(GC, 0, regRLC_CSIB_ADDR_HI),
+ WREG32_RLC(SOC15_REG_OFFSET(GC, xcc_id, regRLC_CSIB_ADDR_HI),
adev->gfx.rlc.clear_state_gpu_addr >> 32);
- WREG32_RLC(SOC15_REG_OFFSET(GC, 0, regRLC_CSIB_ADDR_LO),
+ WREG32_RLC(SOC15_REG_OFFSET(GC, xcc_id, regRLC_CSIB_ADDR_LO),
adev->gfx.rlc.clear_state_gpu_addr & 0xfffffffc);
- WREG32_RLC(SOC15_REG_OFFSET(GC, 0, regRLC_CSIB_LENGTH),
+ WREG32_RLC(SOC15_REG_OFFSET(GC, xcc_id, regRLC_CSIB_LENGTH),
adev->gfx.rlc.clear_state_size);
}
-static void gfx_v9_4_3_init_pg(struct amdgpu_device *adev)
+static void gfx_v9_4_3_init_pg(struct amdgpu_device *adev, int xcc_id)
{
- gfx_v9_4_3_init_csb(adev);
+ gfx_v9_4_3_init_csb(adev, xcc_id);
/*
* Rlc save restore list is workable since v2_1.
* And it's needed by gfxoff feature.
*/
if (adev->gfx.rlc.is_rlc_v2_1)
- gfx_v9_4_3_enable_save_restore_machine(adev);
+ gfx_v9_4_3_enable_save_restore_machine(adev, xcc_id);
if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
AMD_PG_SUPPORT_GFX_SMG |
return 0;
}
-static void gfx_v9_4_3_wait_for_rlc_serdes(struct amdgpu_device *adev)
+static void gfx_v9_4_3_wait_for_rlc_serdes(struct amdgpu_device *adev,
+ int xcc_id)
{
u32 i, j, k;
u32 mask;
static void gfx_v9_4_3_rlc_stop(struct amdgpu_device *adev)
{
- WREG32_FIELD15_PREREG(GC, 0, RLC_CNTL, RLC_ENABLE_F32, 0);
- gfx_v9_4_3_enable_gui_idle_interrupt(adev, false);
- gfx_v9_4_3_wait_for_rlc_serdes(adev);
+ int i;
+
+ for (i = 0; i < adev->gfx.num_xcd; i++) {
+ WREG32_FIELD15_PREREG(GC, i, RLC_CNTL, RLC_ENABLE_F32, 0);
+ gfx_v9_4_3_enable_gui_idle_interrupt(adev, false);
+ gfx_v9_4_3_wait_for_rlc_serdes(adev, i);
+ }
}
static void gfx_v9_4_3_rlc_reset(struct amdgpu_device *adev)
{
- WREG32_FIELD15_PREREG(GC, 0, GRBM_SOFT_RESET, SOFT_RESET_RLC, 1);
- udelay(50);
- WREG32_FIELD15_PREREG(GC, 0, GRBM_SOFT_RESET, SOFT_RESET_RLC, 0);
- udelay(50);
+ int i;
+
+ for (i = 0; i < adev->gfx.num_xcd; i++) {
+ WREG32_FIELD15_PREREG(GC, i, GRBM_SOFT_RESET, SOFT_RESET_RLC, 1);
+ udelay(50);
+ WREG32_FIELD15_PREREG(GC, i, GRBM_SOFT_RESET, SOFT_RESET_RLC, 0);
+ udelay(50);
+ }
}
static void gfx_v9_4_3_rlc_start(struct amdgpu_device *adev)
#ifdef AMDGPU_RLC_DEBUG_RETRY
u32 rlc_ucode_ver;
#endif
+ int i;
- WREG32_FIELD15_PREREG(GC, 0, RLC_CNTL, RLC_ENABLE_F32, 1);
- udelay(50);
-
- /* carrizo do enable cp interrupt after cp inited */
- if (!(adev->flags & AMD_IS_APU)) {
- gfx_v9_4_3_enable_gui_idle_interrupt(adev, true);
+ for (i = 0; i < adev->gfx.num_xcd; i++) {
+ WREG32_FIELD15_PREREG(GC, i, RLC_CNTL, RLC_ENABLE_F32, 1);
udelay(50);
- }
+
+ /* carrizo do enable cp interrupt after cp inited */
+ if (!(adev->flags & AMD_IS_APU)) {
+ gfx_v9_4_3_enable_gui_idle_interrupt(adev, true);
+ udelay(50);
+ }
#ifdef AMDGPU_RLC_DEBUG_RETRY
- /* RLC_GPM_GENERAL_6 : RLC Ucode version */
- rlc_ucode_ver = RREG32_SOC15(GC, 0, regRLC_GPM_GENERAL_6);
- if (rlc_ucode_ver == 0x108) {
- dev_info(adev->dev,
- "Using rlc debug ucode. regRLC_GPM_GENERAL_6 ==0x08%x / fw_ver == %i \n",
- rlc_ucode_ver, adev->gfx.rlc_fw_version);
- /* RLC_GPM_TIMER_INT_3 : Timer interval in RefCLK cycles,
- * default is 0x9C4 to create a 100us interval */
- WREG32_SOC15(GC, 0, regRLC_GPM_TIMER_INT_3, 0x9C4);
- /* RLC_GPM_GENERAL_12 : Minimum gap between wptr and rptr
- * to disable the page fault retry interrupts, default is
- * 0x100 (256) */
- WREG32_SOC15(GC, 0, regRLC_GPM_GENERAL_12, 0x100);
- }
+ /* RLC_GPM_GENERAL_6 : RLC Ucode version */
+ rlc_ucode_ver = RREG32_SOC15(GC, i, regRLC_GPM_GENERAL_6);
+ if (rlc_ucode_ver == 0x108) {
+ dev_info(adev->dev,
+ "Using rlc debug ucode. regRLC_GPM_GENERAL_6 ==0x08%x / fw_ver == %i \n",
+ rlc_ucode_ver, adev->gfx.rlc_fw_version);
+ /* RLC_GPM_TIMER_INT_3 : Timer interval in RefCLK cycles,
+ * default is 0x9C4 to create a 100us interval */
+ WREG32_SOC15(GC, i, regRLC_GPM_TIMER_INT_3, 0x9C4);
+ /* RLC_GPM_GENERAL_12 : Minimum gap between wptr and rptr
+ * to disable the page fault retry interrupts, default is
+ * 0x100 (256) */
+ WREG32_SOC15(GC, i, regRLC_GPM_GENERAL_12, 0x100);
+ }
#endif
+ }
}
-static int gfx_v9_4_3_rlc_load_microcode(struct amdgpu_device *adev)
+static int gfx_v9_4_3_rlc_load_microcode(struct amdgpu_device *adev, int xcc_id)
{
const struct rlc_firmware_header_v2_0 *hdr;
const __le32 *fw_data;
le32_to_cpu(hdr->header.ucode_array_offset_bytes));
fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
- WREG32_SOC15(GC, 0, regRLC_GPM_UCODE_ADDR,
+ WREG32_SOC15(GC, xcc_id, regRLC_GPM_UCODE_ADDR,
RLCG_UCODE_LOADING_START_ADDRESS);
for (i = 0; i < fw_size; i++) {
if (amdgpu_emu_mode == 1 && i % 100 == 0) {
dev_info(adev->dev, "Write RLC ucode data %u DWs\n", i);
msleep(1);
}
- WREG32_SOC15(GC, 0, regRLC_GPM_UCODE_DATA, le32_to_cpup(fw_data++));
+ WREG32_SOC15(GC, xcc_id, regRLC_GPM_UCODE_DATA, le32_to_cpup(fw_data++));
}
- WREG32_SOC15(GC, 0, regRLC_GPM_UCODE_ADDR, adev->gfx.rlc_fw_version);
+ WREG32_SOC15(GC, xcc_id, regRLC_GPM_UCODE_ADDR, adev->gfx.rlc_fw_version);
return 0;
}
static int gfx_v9_4_3_rlc_resume(struct amdgpu_device *adev)
{
- int r;
+ int r, i;
adev->gfx.rlc.funcs->stop(adev);
- /* disable CG */
- WREG32_SOC15(GC, 0, regRLC_CGCG_CGLS_CTRL, 0);
+ for (i = 0; i < adev->gfx.num_xcd; i++) {
+ /* disable CG */
+ WREG32_SOC15(GC, i, regRLC_CGCG_CGLS_CTRL, 0);
- gfx_v9_4_3_init_pg(adev);
+ gfx_v9_4_3_init_pg(adev, i);
- if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
- /* legacy rlc firmware loading */
- r = gfx_v9_4_3_rlc_load_microcode(adev);
- if (r)
- return r;
+ if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
+ /* legacy rlc firmware loading */
+ r = gfx_v9_4_3_rlc_load_microcode(adev, i);
+ if (r)
+ return r;
+ }
}
adev->gfx.rlc.funcs->start(adev);
return 0;
}
-static void gfx_v9_4_3_update_spm_vmid(struct amdgpu_device *adev, unsigned vmid)
+static void gfx_v9_4_3_update_spm_vmid(struct amdgpu_device *adev,
+ unsigned vmid)
{
u32 reg, data;
ARRAY_SIZE(rlcg_access_gc_9_4_3));
}
-static void gfx_v9_4_3_cp_compute_enable(struct amdgpu_device *adev, bool enable)
+static void gfx_v9_4_3_cp_compute_enable(struct amdgpu_device *adev,
+ bool enable, int xcc_id)
{
if (enable) {
- WREG32_SOC15_RLC(GC, 0, regCP_MEC_CNTL, 0);
+ WREG32_SOC15_RLC(GC, xcc_id, regCP_MEC_CNTL, 0);
} else {
- WREG32_SOC15_RLC(GC, 0, regCP_MEC_CNTL,
+ WREG32_SOC15_RLC(GC, xcc_id, regCP_MEC_CNTL,
(CP_MEC_CNTL__MEC_ME1_HALT_MASK | CP_MEC_CNTL__MEC_ME2_HALT_MASK));
- adev->gfx.kiq[0].ring.sched.ready = false;
+ adev->gfx.kiq[xcc_id].ring.sched.ready = false;
}
udelay(50);
}
-static int gfx_v9_4_3_cp_compute_load_microcode(struct amdgpu_device *adev)
+static int gfx_v9_4_3_cp_compute_load_microcode(struct amdgpu_device *adev,
+ int xcc_id)
{
const struct gfx_firmware_header_v1_0 *mec_hdr;
const __le32 *fw_data;
if (!adev->gfx.mec_fw)
return -EINVAL;
- gfx_v9_4_3_cp_compute_enable(adev, false);
+ gfx_v9_4_3_cp_compute_enable(adev, false, xcc_id);
mec_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
amdgpu_ucode_print_gfx_hdr(&mec_hdr->header);
tmp = 0;
tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, VMID, 0);
tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, CACHE_POLICY, 0);
- WREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_CNTL, tmp);
+ WREG32_SOC15(GC, xcc_id, regCP_CPC_IC_BASE_CNTL, tmp);
- WREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_LO,
+ WREG32_SOC15(GC, xcc_id, regCP_CPC_IC_BASE_LO,
adev->gfx.mec.mec_fw_gpu_addr & 0xFFFFF000);
- WREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_HI,
+ WREG32_SOC15(GC, xcc_id, regCP_CPC_IC_BASE_HI,
upper_32_bits(adev->gfx.mec.mec_fw_gpu_addr));
mec_ucode_addr_offset =
- SOC15_REG_OFFSET(GC, 0, regCP_MEC_ME1_UCODE_ADDR);
+ SOC15_REG_OFFSET(GC, xcc_id, regCP_MEC_ME1_UCODE_ADDR);
mec_ucode_data_offset =
- SOC15_REG_OFFSET(GC, 0, regCP_MEC_ME1_UCODE_DATA);
+ SOC15_REG_OFFSET(GC, xcc_id, regCP_MEC_ME1_UCODE_DATA);
/* MEC1 */
WREG32(mec_ucode_addr_offset, mec_hdr->jt_offset);
}
/* KIQ functions */
-static void gfx_v9_4_3_kiq_setting(struct amdgpu_ring *ring)
+static void gfx_v9_4_3_kiq_setting(struct amdgpu_ring *ring, int xcc_id)
{
uint32_t tmp;
struct amdgpu_device *adev = ring->adev;
/* tell RLC which is KIQ queue */
- tmp = RREG32_SOC15(GC, 0, regRLC_CP_SCHEDULERS);
+ tmp = RREG32_SOC15(GC, xcc_id, regRLC_CP_SCHEDULERS);
tmp &= 0xffffff00;
tmp |= (ring->me << 5) | (ring->pipe << 3) | (ring->queue);
- WREG32_SOC15_RLC(GC, 0, regRLC_CP_SCHEDULERS, tmp);
+ WREG32_SOC15_RLC(GC, xcc_id, regRLC_CP_SCHEDULERS, tmp);
tmp |= 0x80;
- WREG32_SOC15_RLC(GC, 0, regRLC_CP_SCHEDULERS, tmp);
+ WREG32_SOC15_RLC(GC, xcc_id, regRLC_CP_SCHEDULERS, tmp);
}
static void gfx_v9_4_3_mqd_set_priority(struct amdgpu_ring *ring, struct v9_mqd *mqd)
return 0;
}
-static int gfx_v9_4_3_kiq_init_register(struct amdgpu_ring *ring)
+static int gfx_v9_4_3_kiq_init_register(struct amdgpu_ring *ring, int xcc_id)
{
struct amdgpu_device *adev = ring->adev;
struct v9_mqd *mqd = ring->mqd_ptr;
int j;
/* disable wptr polling */
- WREG32_FIELD15_PREREG(GC, 0, CP_PQ_WPTR_POLL_CNTL, EN, 0);
+ WREG32_FIELD15_PREREG(GC, xcc_id, CP_PQ_WPTR_POLL_CNTL, EN, 0);
- WREG32_SOC15_RLC(GC, 0, regCP_HQD_EOP_BASE_ADDR,
+ WREG32_SOC15_RLC(GC, xcc_id, regCP_HQD_EOP_BASE_ADDR,
mqd->cp_hqd_eop_base_addr_lo);
- WREG32_SOC15_RLC(GC, 0, regCP_HQD_EOP_BASE_ADDR_HI,
+ WREG32_SOC15_RLC(GC, xcc_id, regCP_HQD_EOP_BASE_ADDR_HI,
mqd->cp_hqd_eop_base_addr_hi);
/* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
- WREG32_SOC15_RLC(GC, 0, regCP_HQD_EOP_CONTROL,
+ WREG32_SOC15_RLC(GC, xcc_id, regCP_HQD_EOP_CONTROL,
mqd->cp_hqd_eop_control);
/* enable doorbell? */
- WREG32_SOC15_RLC(GC, 0, regCP_HQD_PQ_DOORBELL_CONTROL,
+ WREG32_SOC15_RLC(GC, xcc_id, regCP_HQD_PQ_DOORBELL_CONTROL,
mqd->cp_hqd_pq_doorbell_control);
/* disable the queue if it's active */
- if (RREG32_SOC15(GC, 0, regCP_HQD_ACTIVE) & 1) {
- WREG32_SOC15_RLC(GC, 0, regCP_HQD_DEQUEUE_REQUEST, 1);
+ if (RREG32_SOC15(GC, xcc_id, regCP_HQD_ACTIVE) & 1) {
+ WREG32_SOC15_RLC(GC, xcc_id, regCP_HQD_DEQUEUE_REQUEST, 1);
for (j = 0; j < adev->usec_timeout; j++) {
- if (!(RREG32_SOC15(GC, 0, regCP_HQD_ACTIVE) & 1))
+ if (!(RREG32_SOC15(GC, xcc_id, regCP_HQD_ACTIVE) & 1))
break;
udelay(1);
}
- WREG32_SOC15_RLC(GC, 0, regCP_HQD_DEQUEUE_REQUEST,
+ WREG32_SOC15_RLC(GC, xcc_id, regCP_HQD_DEQUEUE_REQUEST,
mqd->cp_hqd_dequeue_request);
- WREG32_SOC15_RLC(GC, 0, regCP_HQD_PQ_RPTR,
+ WREG32_SOC15_RLC(GC, xcc_id, regCP_HQD_PQ_RPTR,
mqd->cp_hqd_pq_rptr);
- WREG32_SOC15_RLC(GC, 0, regCP_HQD_PQ_WPTR_LO,
+ WREG32_SOC15_RLC(GC, xcc_id, regCP_HQD_PQ_WPTR_LO,
mqd->cp_hqd_pq_wptr_lo);
- WREG32_SOC15_RLC(GC, 0, regCP_HQD_PQ_WPTR_HI,
+ WREG32_SOC15_RLC(GC, xcc_id, regCP_HQD_PQ_WPTR_HI,
mqd->cp_hqd_pq_wptr_hi);
}
/* set the pointer to the MQD */
- WREG32_SOC15_RLC(GC, 0, regCP_MQD_BASE_ADDR,
+ WREG32_SOC15_RLC(GC, xcc_id, regCP_MQD_BASE_ADDR,
mqd->cp_mqd_base_addr_lo);
- WREG32_SOC15_RLC(GC, 0, regCP_MQD_BASE_ADDR_HI,
+ WREG32_SOC15_RLC(GC, xcc_id, regCP_MQD_BASE_ADDR_HI,
mqd->cp_mqd_base_addr_hi);
/* set MQD vmid to 0 */
- WREG32_SOC15_RLC(GC, 0, regCP_MQD_CONTROL,
+ WREG32_SOC15_RLC(GC, xcc_id, regCP_MQD_CONTROL,
mqd->cp_mqd_control);
/* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
- WREG32_SOC15_RLC(GC, 0, regCP_HQD_PQ_BASE,
+ WREG32_SOC15_RLC(GC, xcc_id, regCP_HQD_PQ_BASE,
mqd->cp_hqd_pq_base_lo);
- WREG32_SOC15_RLC(GC, 0, regCP_HQD_PQ_BASE_HI,
+ WREG32_SOC15_RLC(GC, xcc_id, regCP_HQD_PQ_BASE_HI,
mqd->cp_hqd_pq_base_hi);
/* set up the HQD, this is similar to CP_RB0_CNTL */
- WREG32_SOC15_RLC(GC, 0, regCP_HQD_PQ_CONTROL,
+ WREG32_SOC15_RLC(GC, xcc_id, regCP_HQD_PQ_CONTROL,
mqd->cp_hqd_pq_control);
/* set the wb address whether it's enabled or not */
- WREG32_SOC15_RLC(GC, 0, regCP_HQD_PQ_RPTR_REPORT_ADDR,
+ WREG32_SOC15_RLC(GC, xcc_id, regCP_HQD_PQ_RPTR_REPORT_ADDR,
mqd->cp_hqd_pq_rptr_report_addr_lo);
- WREG32_SOC15_RLC(GC, 0, regCP_HQD_PQ_RPTR_REPORT_ADDR_HI,
+ WREG32_SOC15_RLC(GC, xcc_id, regCP_HQD_PQ_RPTR_REPORT_ADDR_HI,
mqd->cp_hqd_pq_rptr_report_addr_hi);
/* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
- WREG32_SOC15_RLC(GC, 0, regCP_HQD_PQ_WPTR_POLL_ADDR,
+ WREG32_SOC15_RLC(GC, xcc_id, regCP_HQD_PQ_WPTR_POLL_ADDR,
mqd->cp_hqd_pq_wptr_poll_addr_lo);
- WREG32_SOC15_RLC(GC, 0, regCP_HQD_PQ_WPTR_POLL_ADDR_HI,
+ WREG32_SOC15_RLC(GC, xcc_id, regCP_HQD_PQ_WPTR_POLL_ADDR_HI,
mqd->cp_hqd_pq_wptr_poll_addr_hi);
/* enable the doorbell if requested */
if (ring->use_doorbell) {
- WREG32_SOC15(GC, 0, regCP_MEC_DOORBELL_RANGE_LOWER,
+ WREG32_SOC15(GC, xcc_id, regCP_MEC_DOORBELL_RANGE_LOWER,
(adev->doorbell_index.kiq * 2) << 2);
- WREG32_SOC15(GC, 0, regCP_MEC_DOORBELL_RANGE_UPPER,
+ WREG32_SOC15(GC, xcc_id, regCP_MEC_DOORBELL_RANGE_UPPER,
(adev->doorbell_index.userqueue_end * 2) << 2);
}
- WREG32_SOC15_RLC(GC, 0, regCP_HQD_PQ_DOORBELL_CONTROL,
+ WREG32_SOC15_RLC(GC, xcc_id, regCP_HQD_PQ_DOORBELL_CONTROL,
mqd->cp_hqd_pq_doorbell_control);
/* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */
- WREG32_SOC15_RLC(GC, 0, regCP_HQD_PQ_WPTR_LO,
+ WREG32_SOC15_RLC(GC, xcc_id, regCP_HQD_PQ_WPTR_LO,
mqd->cp_hqd_pq_wptr_lo);
- WREG32_SOC15_RLC(GC, 0, regCP_HQD_PQ_WPTR_HI,
+ WREG32_SOC15_RLC(GC, xcc_id, regCP_HQD_PQ_WPTR_HI,
mqd->cp_hqd_pq_wptr_hi);
/* set the vmid for the queue */
- WREG32_SOC15_RLC(GC, 0, regCP_HQD_VMID, mqd->cp_hqd_vmid);
+ WREG32_SOC15_RLC(GC, xcc_id, regCP_HQD_VMID, mqd->cp_hqd_vmid);
- WREG32_SOC15_RLC(GC, 0, regCP_HQD_PERSISTENT_STATE,
+ WREG32_SOC15_RLC(GC, xcc_id, regCP_HQD_PERSISTENT_STATE,
mqd->cp_hqd_persistent_state);
/* activate the queue */
- WREG32_SOC15_RLC(GC, 0, regCP_HQD_ACTIVE,
+ WREG32_SOC15_RLC(GC, xcc_id, regCP_HQD_ACTIVE,
mqd->cp_hqd_active);
if (ring->use_doorbell)
- WREG32_FIELD15_PREREG(GC, 0, CP_PQ_STATUS, DOORBELL_ENABLE, 1);
+ WREG32_FIELD15_PREREG(GC, xcc_id, CP_PQ_STATUS, DOORBELL_ENABLE, 1);
return 0;
}
-static int gfx_v9_4_3_kiq_fini_register(struct amdgpu_ring *ring)
+static int gfx_v9_4_3_kiq_fini_register(struct amdgpu_ring *ring, int xcc_id)
{
struct amdgpu_device *adev = ring->adev;
int j;
/* disable the queue if it's active */
- if (RREG32_SOC15(GC, 0, regCP_HQD_ACTIVE) & 1) {
+ if (RREG32_SOC15(GC, xcc_id, regCP_HQD_ACTIVE) & 1) {
- WREG32_SOC15_RLC(GC, 0, regCP_HQD_DEQUEUE_REQUEST, 1);
+ WREG32_SOC15_RLC(GC, xcc_id, regCP_HQD_DEQUEUE_REQUEST, 1);
for (j = 0; j < adev->usec_timeout; j++) {
- if (!(RREG32_SOC15(GC, 0, regCP_HQD_ACTIVE) & 1))
+ if (!(RREG32_SOC15(GC, xcc_id, regCP_HQD_ACTIVE) & 1))
break;
udelay(1);
}
DRM_DEBUG("KIQ dequeue request failed.\n");
/* Manual disable if dequeue request times out */
- WREG32_SOC15_RLC(GC, 0, regCP_HQD_ACTIVE, 0);
+ WREG32_SOC15_RLC(GC, xcc_id, regCP_HQD_ACTIVE, 0);
}
- WREG32_SOC15_RLC(GC, 0, regCP_HQD_DEQUEUE_REQUEST,
+ WREG32_SOC15_RLC(GC, xcc_id, regCP_HQD_DEQUEUE_REQUEST,
0);
}
- WREG32_SOC15_RLC(GC, 0, regCP_HQD_IQ_TIMER, 0);
- WREG32_SOC15_RLC(GC, 0, regCP_HQD_IB_CONTROL, 0);
- WREG32_SOC15_RLC(GC, 0, regCP_HQD_PERSISTENT_STATE, 0);
- WREG32_SOC15_RLC(GC, 0, regCP_HQD_PQ_DOORBELL_CONTROL, 0x40000000);
- WREG32_SOC15_RLC(GC, 0, regCP_HQD_PQ_DOORBELL_CONTROL, 0);
- WREG32_SOC15_RLC(GC, 0, regCP_HQD_PQ_RPTR, 0);
- WREG32_SOC15_RLC(GC, 0, regCP_HQD_PQ_WPTR_HI, 0);
- WREG32_SOC15_RLC(GC, 0, regCP_HQD_PQ_WPTR_LO, 0);
+ WREG32_SOC15_RLC(GC, xcc_id, regCP_HQD_IQ_TIMER, 0);
+ WREG32_SOC15_RLC(GC, xcc_id, regCP_HQD_IB_CONTROL, 0);
+ WREG32_SOC15_RLC(GC, xcc_id, regCP_HQD_PERSISTENT_STATE, 0);
+ WREG32_SOC15_RLC(GC, xcc_id, regCP_HQD_PQ_DOORBELL_CONTROL, 0x40000000);
+ WREG32_SOC15_RLC(GC, xcc_id, regCP_HQD_PQ_DOORBELL_CONTROL, 0);
+ WREG32_SOC15_RLC(GC, xcc_id, regCP_HQD_PQ_RPTR, 0);
+ WREG32_SOC15_RLC(GC, xcc_id, regCP_HQD_PQ_WPTR_HI, 0);
+ WREG32_SOC15_RLC(GC, xcc_id, regCP_HQD_PQ_WPTR_LO, 0);
return 0;
}
-static int gfx_v9_4_3_kiq_init_queue(struct amdgpu_ring *ring)
+static int gfx_v9_4_3_kiq_init_queue(struct amdgpu_ring *ring, int xcc_id)
{
struct amdgpu_device *adev = ring->adev;
struct v9_mqd *mqd = ring->mqd_ptr;
struct v9_mqd *tmp_mqd;
- gfx_v9_4_3_kiq_setting(ring);
+ gfx_v9_4_3_kiq_setting(ring, xcc_id);
/* GPU could be in bad state during probe, driver trigger the reset
* after load the SMU, in this case , the mqd is not be initialized.
* driver need to re-init the mqd.
* check mqd->cp_hqd_pq_control since this value should not be 0
*/
- tmp_mqd = (struct v9_mqd *)adev->gfx.kiq[0].mqd_backup;
+ tmp_mqd = (struct v9_mqd *)adev->gfx.kiq[xcc_id].mqd_backup;
if (amdgpu_in_reset(adev) && tmp_mqd->cp_hqd_pq_control) {
/* for GPU_RESET case , reset MQD to a clean status */
- if (adev->gfx.kiq[0].mqd_backup)
- memcpy(mqd, adev->gfx.kiq[0].mqd_backup, sizeof(struct v9_mqd_allocation));
+ if (adev->gfx.kiq[xcc_id].mqd_backup)
+ memcpy(mqd, adev->gfx.kiq[xcc_id].mqd_backup, sizeof(struct v9_mqd_allocation));
/* reset ring buffer */
ring->wptr = 0;
amdgpu_ring_clear_ring(ring);
-
mutex_lock(&adev->srbm_mutex);
- soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0, 0);
- gfx_v9_4_3_kiq_init_register(ring);
- soc15_grbm_select(adev, 0, 0, 0, 0, 0);
+ soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0, xcc_id);
+ gfx_v9_4_3_kiq_init_register(ring, xcc_id);
+ soc15_grbm_select(adev, 0, 0, 0, 0, xcc_id);
mutex_unlock(&adev->srbm_mutex);
} else {
memset((void *)mqd, 0, sizeof(struct v9_mqd_allocation));
((struct v9_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF;
((struct v9_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF;
mutex_lock(&adev->srbm_mutex);
- soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0, 0);
+ soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0, xcc_id);
gfx_v9_4_3_mqd_init(ring);
- gfx_v9_4_3_kiq_init_register(ring);
- soc15_grbm_select(adev, 0, 0, 0, 0, 0);
+ gfx_v9_4_3_kiq_init_register(ring, xcc_id);
+ soc15_grbm_select(adev, 0, 0, 0, 0, xcc_id);
mutex_unlock(&adev->srbm_mutex);
- if (adev->gfx.kiq[0].mqd_backup)
- memcpy(adev->gfx.kiq[0].mqd_backup, mqd, sizeof(struct v9_mqd_allocation));
+ if (adev->gfx.kiq[xcc_id].mqd_backup)
+ memcpy(adev->gfx.kiq[xcc_id].mqd_backup, mqd, sizeof(struct v9_mqd_allocation));
}
return 0;
}
-static int gfx_v9_4_3_kcq_init_queue(struct amdgpu_ring *ring)
+static int gfx_v9_4_3_kcq_init_queue(struct amdgpu_ring *ring, int xcc_id)
{
struct amdgpu_device *adev = ring->adev;
struct v9_mqd *mqd = ring->mqd_ptr;
((struct v9_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF;
((struct v9_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF;
mutex_lock(&adev->srbm_mutex);
- soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0, 0);
+ soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0, xcc_id);
gfx_v9_4_3_mqd_init(ring);
- soc15_grbm_select(adev, 0, 0, 0, 0, 0);
+ soc15_grbm_select(adev, 0, 0, 0, 0, xcc_id);
mutex_unlock(&adev->srbm_mutex);
if (adev->gfx.mec.mqd_backup[mqd_idx])
return 0;
}
-static int gfx_v9_4_3_kiq_resume(struct amdgpu_device *adev)
+static int gfx_v9_4_3_kiq_resume(struct amdgpu_device *adev, int xcc_id)
{
struct amdgpu_ring *ring;
int r;
- ring = &adev->gfx.kiq[0].ring;
+ ring = &adev->gfx.kiq[xcc_id].ring;
r = amdgpu_bo_reserve(ring->mqd_obj, false);
if (unlikely(r != 0))
if (unlikely(r != 0))
return r;
- gfx_v9_4_3_kiq_init_queue(ring);
+ gfx_v9_4_3_kiq_init_queue(ring, xcc_id);
amdgpu_bo_kunmap(ring->mqd_obj);
ring->mqd_ptr = NULL;
amdgpu_bo_unreserve(ring->mqd_obj);
return 0;
}
-static int gfx_v9_4_3_kcq_resume(struct amdgpu_device *adev)
+static int gfx_v9_4_3_kcq_resume(struct amdgpu_device *adev, int xcc_id)
{
struct amdgpu_ring *ring = NULL;
int r = 0, i;
- gfx_v9_4_3_cp_compute_enable(adev, true);
+ gfx_v9_4_3_cp_compute_enable(adev, true, xcc_id);
for (i = 0; i < adev->gfx.num_compute_rings; i++) {
- ring = &adev->gfx.compute_ring[i];
+ ring = &adev->gfx.compute_ring[i + xcc_id * adev->gfx.num_compute_rings];
r = amdgpu_bo_reserve(ring->mqd_obj, false);
if (unlikely(r != 0))
goto done;
r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr);
if (!r) {
- r = gfx_v9_4_3_kcq_init_queue(ring);
+ r = gfx_v9_4_3_kcq_init_queue(ring, xcc_id);
amdgpu_bo_kunmap(ring->mqd_obj);
ring->mqd_ptr = NULL;
}
goto done;
}
- r = amdgpu_gfx_enable_kcq(adev, 0);
+ r = amdgpu_gfx_enable_kcq(adev, xcc_id);
done:
return r;
}
static int gfx_v9_4_3_cp_resume(struct amdgpu_device *adev)
{
- int r, i;
+ int r, i, j;
struct amdgpu_ring *ring;
- gfx_v9_4_3_enable_gui_idle_interrupt(adev, false);
+ for (i = 0; i < adev->gfx.num_xcd; i++) {
+ gfx_v9_4_3_enable_gui_idle_interrupt(adev, false);
+
+ if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
+ gfx_v9_4_3_disable_gpa_mode(adev);
- if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
- gfx_v9_4_3_disable_gpa_mode(adev);
+ r = gfx_v9_4_3_cp_compute_load_microcode(adev, i);
+ if (r)
+ return r;
+ }
- r = gfx_v9_4_3_cp_compute_load_microcode(adev);
+ r = gfx_v9_4_3_kiq_resume(adev, i);
if (r)
return r;
- }
- r = gfx_v9_4_3_kiq_resume(adev);
- if (r)
- return r;
+ r = gfx_v9_4_3_kcq_resume(adev, i);
+ if (r)
+ return r;
- r = gfx_v9_4_3_kcq_resume(adev);
- if (r)
- return r;
+ for (j = 0; j < adev->gfx.num_compute_rings; j++) {
+ ring = &adev->gfx.compute_ring[j + i * adev->gfx.num_compute_rings];
+ amdgpu_ring_test_helper(ring);
+ }
- for (i = 0; i < adev->gfx.num_compute_rings; i++) {
- ring = &adev->gfx.compute_ring[i];
- amdgpu_ring_test_helper(ring);
+ gfx_v9_4_3_enable_gui_idle_interrupt(adev, true);
}
- gfx_v9_4_3_enable_gui_idle_interrupt(adev, true);
-
return 0;
}
-static void gfx_v9_4_3_cp_enable(struct amdgpu_device *adev, bool enable)
+static void gfx_v9_4_3_cp_enable(struct amdgpu_device *adev, bool enable,
+ int xcc_id)
{
- gfx_v9_4_3_cp_compute_enable(adev, enable);
+ gfx_v9_4_3_cp_compute_enable(adev, enable, xcc_id);
}
static int gfx_v9_4_3_hw_init(void *handle)
static int gfx_v9_4_3_hw_fini(void *handle)
{
struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+ int i;
amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0);
amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0);
- if (amdgpu_gfx_disable_kcq(adev, 0))
- DRM_ERROR("KCQ disable failed\n");
+ for (i = 0; i < adev->gfx.num_xcd; i++) {
+ if (amdgpu_gfx_disable_kcq(adev, i))
+ DRM_ERROR("XCD %d KCQ disable failed\n", i);
- /* Use deinitialize sequence from CAIL when unbinding device from driver,
- * otherwise KIQ is hanging when binding back
- */
- if (!amdgpu_in_reset(adev) && !adev->in_suspend) {
- mutex_lock(&adev->srbm_mutex);
- soc15_grbm_select(adev, adev->gfx.kiq[0].ring.me,
- adev->gfx.kiq[0].ring.pipe,
- adev->gfx.kiq[0].ring.queue, 0, 0);
- gfx_v9_4_3_kiq_fini_register(&adev->gfx.kiq[0].ring);
- soc15_grbm_select(adev, 0, 0, 0, 0, 0);
- mutex_unlock(&adev->srbm_mutex);
- }
+ /* Use deinitialize sequence from CAIL when unbinding device
+ * from driver, otherwise KIQ is hanging when binding back
+ */
+ if (!amdgpu_in_reset(adev) && !adev->in_suspend) {
+ mutex_lock(&adev->srbm_mutex);
+ soc15_grbm_select(adev, adev->gfx.kiq[i].ring.me,
+ adev->gfx.kiq[i].ring.pipe,
+ adev->gfx.kiq[i].ring.queue, 0, i);
+ gfx_v9_4_3_kiq_fini_register(&adev->gfx.kiq[i].ring, i);
+ soc15_grbm_select(adev, 0, 0, 0, 0, i);
+ mutex_unlock(&adev->srbm_mutex);
+ }
- gfx_v9_4_3_cp_enable(adev, false);
+ gfx_v9_4_3_cp_enable(adev, false, i);
+ }
/* Skip suspend with A+A reset */
if (adev->gmc.xgmi.connected_to_cpu && amdgpu_in_reset(adev)) {
static bool gfx_v9_4_3_is_idle(void *handle)
{
struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+ int i;
- if (REG_GET_FIELD(RREG32_SOC15(GC, 0, regGRBM_STATUS),
- GRBM_STATUS, GUI_ACTIVE))
- return false;
- else
- return true;
+ for (i = 0; i < adev->gfx.num_xcd; i++) {
+ if (REG_GET_FIELD(RREG32_SOC15(GC, i, regGRBM_STATUS),
+ GRBM_STATUS, GUI_ACTIVE))
+ return false;
+ }
+ return true;
}
static int gfx_v9_4_3_wait_for_idle(void *handle)
adev->gfx.rlc.funcs->stop(adev);
/* Disable MEC parsing/prefetching */
- gfx_v9_4_3_cp_compute_enable(adev, false);
+ gfx_v9_4_3_cp_compute_enable(adev, false, 0);
if (grbm_soft_reset) {
tmp = RREG32_SOC15(GC, 0, regGRBM_SOFT_RESET);
{
struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+ /* hardcode in emulation phase */
+ adev->gfx.num_xcd = 1;
+ adev->gfx.num_xcc_per_xcp = 1;
+ adev->gfx.partition_mode = AMDGPU_SPX_PARTITION_MODE;
+
adev->gfx.num_compute_rings = min(amdgpu_gfx_get_num_kcq(adev),
AMDGPU_MAX_COMPUTE_RINGS);
gfx_v9_4_3_set_kiq_pm4_funcs(adev);
}
static void gfx_v9_4_3_update_medium_grain_clock_gating(struct amdgpu_device *adev,
- bool enable)
+ bool enable, int xcc_id)
{
uint32_t data, def;
/* It is disabled by HW by default */
if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG)) {
/* 1 - RLC_CGTT_MGCG_OVERRIDE */
- def = data = RREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE);
+ def = data = RREG32_SOC15(GC, xcc_id, regRLC_CGTT_MGCG_OVERRIDE);
data &= ~(RLC_CGTT_MGCG_OVERRIDE__GRBM_CGTT_SCLK_OVERRIDE_MASK |
RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK |
data |= RLC_CGTT_MGCG_OVERRIDE__RLC_CGTT_SCLK_OVERRIDE_MASK;
if (def != data)
- WREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE, data);
+ WREG32_SOC15(GC, xcc_id, regRLC_CGTT_MGCG_OVERRIDE, data);
/* MGLS is a global flag to control all MGLS in GFX */
if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
/* 2 - RLC memory Light sleep */
if (adev->cg_flags & AMD_CG_SUPPORT_GFX_RLC_LS) {
- def = data = RREG32_SOC15(GC, 0, regRLC_MEM_SLP_CNTL);
+ def = data = RREG32_SOC15(GC, xcc_id, regRLC_MEM_SLP_CNTL);
data |= RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK;
if (def != data)
- WREG32_SOC15(GC, 0, regRLC_MEM_SLP_CNTL, data);
+ WREG32_SOC15(GC, xcc_id, regRLC_MEM_SLP_CNTL, data);
}
/* 3 - CP memory Light sleep */
if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS) {
- def = data = RREG32_SOC15(GC, 0, regCP_MEM_SLP_CNTL);
+ def = data = RREG32_SOC15(GC, xcc_id, regCP_MEM_SLP_CNTL);
data |= CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
if (def != data)
- WREG32_SOC15(GC, 0, regCP_MEM_SLP_CNTL, data);
+ WREG32_SOC15(GC, xcc_id, regCP_MEM_SLP_CNTL, data);
}
}
} else {
/* 1 - MGCG_OVERRIDE */
- def = data = RREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE);
+ def = data = RREG32_SOC15(GC, xcc_id, regRLC_CGTT_MGCG_OVERRIDE);
data |= (RLC_CGTT_MGCG_OVERRIDE__RLC_CGTT_SCLK_OVERRIDE_MASK |
RLC_CGTT_MGCG_OVERRIDE__GRBM_CGTT_SCLK_OVERRIDE_MASK |
RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGLS_OVERRIDE_MASK);
if (def != data)
- WREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE, data);
+ WREG32_SOC15(GC, xcc_id, regRLC_CGTT_MGCG_OVERRIDE, data);
/* 2 - disable MGLS in RLC */
- data = RREG32_SOC15(GC, 0, regRLC_MEM_SLP_CNTL);
+ data = RREG32_SOC15(GC, xcc_id, regRLC_MEM_SLP_CNTL);
if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK) {
data &= ~RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK;
- WREG32_SOC15(GC, 0, regRLC_MEM_SLP_CNTL, data);
+ WREG32_SOC15(GC, xcc_id, regRLC_MEM_SLP_CNTL, data);
}
/* 3 - disable MGLS in CP */
- data = RREG32_SOC15(GC, 0, regCP_MEM_SLP_CNTL);
+ data = RREG32_SOC15(GC, xcc_id, regCP_MEM_SLP_CNTL);
if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK) {
data &= ~CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
- WREG32_SOC15(GC, 0, regCP_MEM_SLP_CNTL, data);
+ WREG32_SOC15(GC, xcc_id, regCP_MEM_SLP_CNTL, data);
}
}
}
static void gfx_v9_4_3_update_coarse_grain_clock_gating(struct amdgpu_device *adev,
- bool enable)
+ bool enable, int xcc_id)
{
uint32_t def, data;
amdgpu_gfx_rlc_enter_safe_mode(adev);
if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)) {
- def = data = RREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE);
+ def = data = RREG32_SOC15(GC, xcc_id, regRLC_CGTT_MGCG_OVERRIDE);
/* unset CGCG override */
data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGCG_OVERRIDE_MASK;
if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS)
data |= RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGLS_OVERRIDE_MASK;
/* update CGCG and CGLS override bits */
if (def != data)
- WREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE, data);
+ WREG32_SOC15(GC, xcc_id, regRLC_CGTT_MGCG_OVERRIDE, data);
/* enable cgcg FSM(0x0000363F) */
- def = RREG32_SOC15(GC, 0, regRLC_CGCG_CGLS_CTRL);
+ def = RREG32_SOC15(GC, xcc_id, regRLC_CGCG_CGLS_CTRL);
if (adev->asic_type == CHIP_ARCTURUS)
data = (0x2000 << RLC_CGCG_CGLS_CTRL__CGCG_GFX_IDLE_THRESHOLD__SHIFT) |
data |= (0x000F << RLC_CGCG_CGLS_CTRL__CGLS_REP_COMPANSAT_DELAY__SHIFT) |
RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK;
if (def != data)
- WREG32_SOC15(GC, 0, regRLC_CGCG_CGLS_CTRL, data);
+ WREG32_SOC15(GC, xcc_id, regRLC_CGCG_CGLS_CTRL, data);
/* set IDLE_POLL_COUNT(0x00900100) */
- def = RREG32_SOC15(GC, 0, regCP_RB_WPTR_POLL_CNTL);
+ def = RREG32_SOC15(GC, xcc_id, regCP_RB_WPTR_POLL_CNTL);
data = (0x0100 << CP_RB_WPTR_POLL_CNTL__POLL_FREQUENCY__SHIFT) |
(0x0090 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT);
if (def != data)
- WREG32_SOC15(GC, 0, regCP_RB_WPTR_POLL_CNTL, data);
+ WREG32_SOC15(GC, xcc_id, regCP_RB_WPTR_POLL_CNTL, data);
} else {
- def = data = RREG32_SOC15(GC, 0, regRLC_CGCG_CGLS_CTRL);
+ def = data = RREG32_SOC15(GC, xcc_id, regRLC_CGCG_CGLS_CTRL);
/* reset CGCG/CGLS bits */
data &= ~(RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK | RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK);
/* disable cgcg and cgls in FSM */
if (def != data)
- WREG32_SOC15(GC, 0, regRLC_CGCG_CGLS_CTRL, data);
+ WREG32_SOC15(GC, xcc_id, regRLC_CGCG_CGLS_CTRL, data);
}
amdgpu_gfx_rlc_exit_safe_mode(adev);
}
static int gfx_v9_4_3_update_gfx_clock_gating(struct amdgpu_device *adev,
- bool enable)
+ bool enable, int xcc_id)
{
if (enable) {
/* CGCG/CGLS should be enabled after MGCG/MGLS
* === MGCG + MGLS ===
*/
- gfx_v9_4_3_update_medium_grain_clock_gating(adev, enable);
+ gfx_v9_4_3_update_medium_grain_clock_gating(adev, enable, xcc_id);
/* === CGCG + CGLS === */
- gfx_v9_4_3_update_coarse_grain_clock_gating(adev, enable);
+ gfx_v9_4_3_update_coarse_grain_clock_gating(adev, enable, xcc_id);
} else {
/* CGCG/CGLS should be disabled before MGCG/MGLS
* === CGCG + CGLS ===
*/
- gfx_v9_4_3_update_coarse_grain_clock_gating(adev, enable);
+ gfx_v9_4_3_update_coarse_grain_clock_gating(adev, enable, xcc_id);
/* === MGCG + MGLS === */
- gfx_v9_4_3_update_medium_grain_clock_gating(adev, enable);
+ gfx_v9_4_3_update_medium_grain_clock_gating(adev, enable, xcc_id);
}
return 0;
}
enum amd_clockgating_state state)
{
struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+ int i;
if (amdgpu_sriov_vf(adev))
return 0;
switch (adev->ip_versions[GC_HWIP][0]) {
case IP_VERSION(9, 4, 3):
- gfx_v9_4_3_update_gfx_clock_gating(adev,
- state == AMD_CG_STATE_GATE);
+ for (i = 0; i < adev->gfx.num_xcd; i++)
+ gfx_v9_4_3_update_gfx_clock_gating(adev,
+ state == AMD_CG_STATE_GATE, i);
break;
default:
break;
}
static void gfx_v9_4_3_set_compute_eop_interrupt_state(struct amdgpu_device *adev,
- int me, int pipe,
- enum amdgpu_interrupt_state state)
+ int me, int pipe,
+ enum amdgpu_interrupt_state state,
+ int xcc_id)
{
u32 mec_int_cntl, mec_int_cntl_reg;
if (me == 1) {
switch (pipe) {
case 0:
- mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, regCP_ME1_PIPE0_INT_CNTL);
+ mec_int_cntl_reg = SOC15_REG_OFFSET(GC, xcc_id, regCP_ME1_PIPE0_INT_CNTL);
break;
case 1:
- mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, regCP_ME1_PIPE1_INT_CNTL);
+ mec_int_cntl_reg = SOC15_REG_OFFSET(GC, xcc_id, regCP_ME1_PIPE1_INT_CNTL);
break;
case 2:
- mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, regCP_ME1_PIPE2_INT_CNTL);
+ mec_int_cntl_reg = SOC15_REG_OFFSET(GC, xcc_id, regCP_ME1_PIPE2_INT_CNTL);
break;
case 3:
- mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, regCP_ME1_PIPE3_INT_CNTL);
+ mec_int_cntl_reg = SOC15_REG_OFFSET(GC, xcc_id, regCP_ME1_PIPE3_INT_CNTL);
break;
default:
DRM_DEBUG("invalid pipe %d\n", pipe);
unsigned type,
enum amdgpu_interrupt_state state)
{
+ int i;
+
switch (state) {
case AMDGPU_IRQ_STATE_DISABLE:
case AMDGPU_IRQ_STATE_ENABLE:
- WREG32_FIELD15_PREREG(GC, 0, CP_INT_CNTL_RING0,
- PRIV_REG_INT_ENABLE,
- state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
+ for (i = 0; i < adev->gfx.num_xcd; i++)
+ WREG32_FIELD15_PREREG(GC, i, CP_INT_CNTL_RING0,
+ PRIV_REG_INT_ENABLE,
+ state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
break;
default:
break;
unsigned type,
enum amdgpu_interrupt_state state)
{
+ int i;
+
switch (state) {
case AMDGPU_IRQ_STATE_DISABLE:
case AMDGPU_IRQ_STATE_ENABLE:
- WREG32_FIELD15_PREREG(GC, 0, CP_INT_CNTL_RING0,
- PRIV_INSTR_INT_ENABLE,
- state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
+ for (i = 0; i < adev->gfx.num_xcd; i++)
+ WREG32_FIELD15_PREREG(GC, i, CP_INT_CNTL_RING0,
+ PRIV_INSTR_INT_ENABLE,
+ state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
break;
default:
break;
unsigned type,
enum amdgpu_interrupt_state state)
{
- switch (type) {
- case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP:
- gfx_v9_4_3_set_compute_eop_interrupt_state(adev, 1, 0, state);
- break;
- case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE1_EOP:
- gfx_v9_4_3_set_compute_eop_interrupt_state(adev, 1, 1, state);
- break;
- case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE2_EOP:
- gfx_v9_4_3_set_compute_eop_interrupt_state(adev, 1, 2, state);
- break;
- case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE3_EOP:
- gfx_v9_4_3_set_compute_eop_interrupt_state(adev, 1, 3, state);
- break;
- case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE0_EOP:
- gfx_v9_4_3_set_compute_eop_interrupt_state(adev, 2, 0, state);
- break;
- case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE1_EOP:
- gfx_v9_4_3_set_compute_eop_interrupt_state(adev, 2, 1, state);
- break;
- case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE2_EOP:
- gfx_v9_4_3_set_compute_eop_interrupt_state(adev, 2, 2, state);
- break;
- case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE3_EOP:
- gfx_v9_4_3_set_compute_eop_interrupt_state(adev, 2, 3, state);
- break;
- default:
- break;
+ int i;
+ for (i = 0; i < adev->gfx.num_xcd; i++) {
+ switch (type) {
+ case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP:
+ gfx_v9_4_3_set_compute_eop_interrupt_state(adev, 1, 0, state, i);
+ break;
+ case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE1_EOP:
+ gfx_v9_4_3_set_compute_eop_interrupt_state(adev, 1, 1, state, i);
+ break;
+ case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE2_EOP:
+ gfx_v9_4_3_set_compute_eop_interrupt_state(adev, 1, 2, state, i);
+ break;
+ case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE3_EOP:
+ gfx_v9_4_3_set_compute_eop_interrupt_state(adev, 1, 3, state, i);
+ break;
+ case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE0_EOP:
+ gfx_v9_4_3_set_compute_eop_interrupt_state(adev, 2, 0, state, i);
+ break;
+ case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE1_EOP:
+ gfx_v9_4_3_set_compute_eop_interrupt_state(adev, 2, 1, state, i);
+ break;
+ case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE2_EOP:
+ gfx_v9_4_3_set_compute_eop_interrupt_state(adev, 2, 2, state, i);
+ break;
+ case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE3_EOP:
+ gfx_v9_4_3_set_compute_eop_interrupt_state(adev, 2, 3, state, i);
+ break;
+ default:
+ break;
+ }
}
+
return 0;
}
static void gfx_v9_4_3_set_ring_funcs(struct amdgpu_device *adev)
{
- int i;
+ int i, j;
- adev->gfx.kiq[0].ring.funcs = &gfx_v9_4_3_ring_funcs_kiq;
+ for (i = 0; i < adev->gfx.num_xcd; i++) {
+ adev->gfx.kiq[i].ring.funcs = &gfx_v9_4_3_ring_funcs_kiq;
- for (i = 0; i < adev->gfx.num_compute_rings; i++)
- adev->gfx.compute_ring[i].funcs = &gfx_v9_4_3_ring_funcs_compute;
+ for (j = 0; j < adev->gfx.num_compute_rings; j++)
+ adev->gfx.compute_ring[j + i * adev->gfx.num_compute_rings].funcs
+ = &gfx_v9_4_3_ring_funcs_compute;
+ }
}
static const struct amdgpu_irq_src_funcs gfx_v9_4_3_eop_irq_funcs = {
projects / platform / kernel / linux-rpi.git / commitdiff