#include "amdgpu.h"
#include "amdgpu_amdkfd.h"
#include "amdgpu_amdkfd_gfx_v9.h"
#include "amdgpu_amdkfd_aldebaran.h"
#include "gc/gc_9_4_3_offset.h"
#include "gc/gc_9_4_3_sh_mask.h"
#include "athub/athub_1_8_0_offset.h"
#include "athub/athub_1_8_0_sh_mask.h"
#include "oss/osssys_4_4_2_offset.h"
#include "oss/osssys_4_4_2_sh_mask.h"
#include "v9_structs.h"
#include "soc15.h"
#include "sdma/sdma_4_4_2_offset.h"
#include "sdma/sdma_4_4_2_sh_mask.h"
#include <uapi/linux/kfd_ioctl.h>
static inline struct v9_sdma_mqd *get_sdma_mqd(void *mqd)
{
return (struct v9_sdma_mqd *)mqd;
}
static uint32_t get_sdma_rlc_reg_offset(struct amdgpu_device *adev,
unsigned int engine_id,
unsigned int queue_id)
{
uint32_t sdma_engine_reg_base =
SOC15_REG_OFFSET(SDMA0, GET_INST(SDMA0, engine_id),
regSDMA_RLC0_RB_CNTL) -
regSDMA_RLC0_RB_CNTL;
uint32_t retval = sdma_engine_reg_base +
queue_id * (regSDMA_RLC1_RB_CNTL - regSDMA_RLC0_RB_CNTL);
pr_debug("RLC register offset for SDMA%d RLC%d: 0x%x\n", engine_id,
queue_id, retval);
return retval;
}
static int kgd_gfx_v9_4_3_hqd_sdma_load(struct amdgpu_device *adev, void *mqd,
uint32_t __user *wptr, struct mm_struct *mm)
{
struct v9_sdma_mqd *m;
uint32_t sdma_rlc_reg_offset;
unsigned long end_jiffies;
uint32_t data;
uint64_t data64;
uint64_t __user *wptr64 = (uint64_t __user *)wptr;
m = get_sdma_mqd(mqd);
sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev, m->sdma_engine_id,
m->sdma_queue_id);
WREG32(sdma_rlc_reg_offset + regSDMA_RLC0_RB_CNTL,
m->sdmax_rlcx_rb_cntl & (~SDMA_RLC0_RB_CNTL__RB_ENABLE_MASK));
end_jiffies = msecs_to_jiffies(2000) + jiffies;
while (true) {
data = RREG32(sdma_rlc_reg_offset + regSDMA_RLC0_CONTEXT_STATUS);
if (data & SDMA_RLC0_CONTEXT_STATUS__IDLE_MASK)
break;
if (time_after(jiffies, end_jiffies)) {
pr_err("SDMA RLC not idle in %s\n", __func__);
return -ETIME;
}
usleep_range(500, 1000);
}
WREG32(sdma_rlc_reg_offset + regSDMA_RLC0_DOORBELL_OFFSET,
m->sdmax_rlcx_doorbell_offset);
data = REG_SET_FIELD(m->sdmax_rlcx_doorbell, SDMA_RLC0_DOORBELL,
ENABLE, 1);
WREG32(sdma_rlc_reg_offset + regSDMA_RLC0_DOORBELL, data);
WREG32(sdma_rlc_reg_offset + regSDMA_RLC0_RB_RPTR,
m->sdmax_rlcx_rb_rptr);
WREG32(sdma_rlc_reg_offset + regSDMA_RLC0_RB_RPTR_HI,
m->sdmax_rlcx_rb_rptr_hi);
WREG32(sdma_rlc_reg_offset + regSDMA_RLC0_MINOR_PTR_UPDATE, 1);
if (read_user_wptr(mm, wptr64, data64)) {
WREG32(sdma_rlc_reg_offset + regSDMA_RLC0_RB_WPTR,
lower_32_bits(data64));
WREG32(sdma_rlc_reg_offset + regSDMA_RLC0_RB_WPTR_HI,
upper_32_bits(data64));
} else {
WREG32(sdma_rlc_reg_offset + regSDMA_RLC0_RB_WPTR,
m->sdmax_rlcx_rb_rptr);
WREG32(sdma_rlc_reg_offset + regSDMA_RLC0_RB_WPTR_HI,
m->sdmax_rlcx_rb_rptr_hi);
}
WREG32(sdma_rlc_reg_offset + regSDMA_RLC0_MINOR_PTR_UPDATE, 0);
WREG32(sdma_rlc_reg_offset + regSDMA_RLC0_RB_BASE, m->sdmax_rlcx_rb_base);
WREG32(sdma_rlc_reg_offset + regSDMA_RLC0_RB_BASE_HI,
m->sdmax_rlcx_rb_base_hi);
WREG32(sdma_rlc_reg_offset + regSDMA_RLC0_RB_RPTR_ADDR_LO,
m->sdmax_rlcx_rb_rptr_addr_lo);
WREG32(sdma_rlc_reg_offset + regSDMA_RLC0_RB_RPTR_ADDR_HI,
m->sdmax_rlcx_rb_rptr_addr_hi);
data = REG_SET_FIELD(m->sdmax_rlcx_rb_cntl, SDMA_RLC0_RB_CNTL,
RB_ENABLE, 1);
WREG32(sdma_rlc_reg_offset + regSDMA_RLC0_RB_CNTL, data);
return 0;
}
static int kgd_gfx_v9_4_3_hqd_sdma_dump(struct amdgpu_device *adev,
uint32_t engine_id, uint32_t queue_id,
uint32_t (**dump)[2], uint32_t *n_regs)
{
uint32_t sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev,
engine_id, queue_id);
uint32_t i = 0, reg;
#undef HQD_N_REGS
#define HQD_N_REGS (19+6+7+12)
#define DUMP_REG(addr) do { \
if (WARN_ON_ONCE(i >= HQD_N_REGS)) \
break; \
(*dump)[i][0] = (addr) << 2; \
(*dump)[i++][1] = RREG32(addr); \
} while (0)
*dump = kmalloc_array(HQD_N_REGS * 2, sizeof(uint32_t), GFP_KERNEL);
if (*dump == NULL)
return -ENOMEM;
for (reg = regSDMA_RLC0_RB_CNTL; reg <= regSDMA_RLC0_DOORBELL; reg++)
DUMP_REG(sdma_rlc_reg_offset + reg);
for (reg = regSDMA_RLC0_STATUS; reg <= regSDMA_RLC0_CSA_ADDR_HI; reg++)
DUMP_REG(sdma_rlc_reg_offset + reg);
for (reg = regSDMA_RLC0_IB_SUB_REMAIN;
reg <= regSDMA_RLC0_MINOR_PTR_UPDATE; reg++)
DUMP_REG(sdma_rlc_reg_offset + reg);
for (reg = regSDMA_RLC0_MIDCMD_DATA0;
reg <= regSDMA_RLC0_MIDCMD_CNTL; reg++)
DUMP_REG(sdma_rlc_reg_offset + reg);
WARN_ON_ONCE(i != HQD_N_REGS);
*n_regs = i;
return 0;
}
static bool kgd_gfx_v9_4_3_hqd_sdma_is_occupied(struct amdgpu_device *adev, void *mqd)
{
struct v9_sdma_mqd *m;
uint32_t sdma_rlc_reg_offset;
uint32_t sdma_rlc_rb_cntl;
m = get_sdma_mqd(mqd);
sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev, m->sdma_engine_id,
m->sdma_queue_id);
sdma_rlc_rb_cntl = RREG32(sdma_rlc_reg_offset + regSDMA_RLC0_RB_CNTL);
if (sdma_rlc_rb_cntl & SDMA_RLC0_RB_CNTL__RB_ENABLE_MASK)
return true;
return false;
}
static int kgd_gfx_v9_4_3_hqd_sdma_destroy(struct amdgpu_device *adev, void *mqd,
unsigned int utimeout)
{
struct v9_sdma_mqd *m;
uint32_t sdma_rlc_reg_offset;
uint32_t temp;
unsigned long end_jiffies = (utimeout * HZ / 1000) + jiffies;
m = get_sdma_mqd(mqd);
sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev, m->sdma_engine_id,
m->sdma_queue_id);
temp = RREG32(sdma_rlc_reg_offset + regSDMA_RLC0_RB_CNTL);
temp = temp & ~SDMA_RLC0_RB_CNTL__RB_ENABLE_MASK;
WREG32(sdma_rlc_reg_offset + regSDMA_RLC0_RB_CNTL, temp);
while (true) {
temp = RREG32(sdma_rlc_reg_offset + regSDMA_RLC0_CONTEXT_STATUS);
if (temp & SDMA_RLC0_CONTEXT_STATUS__IDLE_MASK)
break;
if (time_after(jiffies, end_jiffies)) {
pr_err("SDMA RLC not idle in %s\n", __func__);
return -ETIME;
}
usleep_range(500, 1000);
}
WREG32(sdma_rlc_reg_offset + regSDMA_RLC0_DOORBELL, 0);
WREG32(sdma_rlc_reg_offset + regSDMA_RLC0_RB_CNTL,
RREG32(sdma_rlc_reg_offset + regSDMA_RLC0_RB_CNTL) |
SDMA_RLC0_RB_CNTL__RB_ENABLE_MASK);
m->sdmax_rlcx_rb_rptr =
RREG32(sdma_rlc_reg_offset + regSDMA_RLC0_RB_RPTR);
m->sdmax_rlcx_rb_rptr_hi =
RREG32(sdma_rlc_reg_offset + regSDMA_RLC0_RB_RPTR_HI);
return 0;
}
static int kgd_gfx_v9_4_3_set_pasid_vmid_mapping(struct amdgpu_device *adev,
u32 pasid, unsigned int vmid, uint32_t xcc_inst)
{
unsigned long timeout;
unsigned int reg;
unsigned int phy_inst = GET_INST(GC, xcc_inst);
unsigned int aid = phy_inst / 2;
uint32_t pasid_mapping = (pasid == 0) ? 0 : (uint32_t)pasid |
ATC_VMID0_PASID_MAPPING__VALID_MASK;
WREG32(SOC15_REG_OFFSET(ATHUB, 0,
regATC_VMID0_PASID_MAPPING) + vmid, pasid_mapping);
timeout = jiffies + msecs_to_jiffies(10);
while (!(RREG32(SOC15_REG_OFFSET(ATHUB, 0,
regATC_VMID_PASID_MAPPING_UPDATE_STATUS)) &
(1U << vmid))) {
if (time_after(jiffies, timeout)) {
pr_err("Fail to program VMID-PASID mapping\n");
return -ETIME;
}
cpu_relax();
}
WREG32(SOC15_REG_OFFSET(ATHUB, 0,
regATC_VMID_PASID_MAPPING_UPDATE_STATUS),
1U << vmid);
reg = RREG32(SOC15_REG_OFFSET(OSSSYS, 0, regIH_VMID_LUT_INDEX));
WREG32(SOC15_REG_OFFSET(OSSSYS, 0, regIH_VMID_LUT_INDEX),
aid * 4 + (phy_inst % 2) + 1);
WREG32(SOC15_REG_OFFSET(OSSSYS, 0, regIH_VMID_0_LUT) + vmid,
pasid_mapping);
WREG32(SOC15_REG_OFFSET(OSSSYS, 0, regIH_VMID_LUT_INDEX),
aid * 4);
WREG32(SOC15_REG_OFFSET(OSSSYS, 0, regIH_VMID_0_LUT_MM) + vmid,
pasid_mapping);
WREG32(SOC15_REG_OFFSET(OSSSYS, 0, regIH_VMID_LUT_INDEX), reg);
return 0;
}
static inline struct v9_mqd *get_mqd(void *mqd)
{
return (struct v9_mqd *)mqd;
}
static int kgd_gfx_v9_4_3_hqd_load(struct amdgpu_device *adev, void *mqd,
uint32_t pipe_id, uint32_t queue_id,
uint32_t __user *wptr, uint32_t wptr_shift,
uint32_t wptr_mask, struct mm_struct *mm, uint32_t inst)
{
struct v9_mqd *m;
uint32_t *mqd_hqd;
uint32_t reg, hqd_base, hqd_end, data;
m = get_mqd(mqd);
kgd_gfx_v9_acquire_queue(adev, pipe_id, queue_id, inst);
mqd_hqd = &m->cp_mqd_base_addr_lo;
hqd_base = SOC15_REG_OFFSET(GC, GET_INST(GC, inst), regCP_MQD_BASE_ADDR);
hqd_end = SOC15_REG_OFFSET(GC, GET_INST(GC, inst), regCP_HQD_AQL_DISPATCH_ID_HI);
for (reg = hqd_base; reg <= hqd_end; reg++)
WREG32_RLC(reg, mqd_hqd[reg - hqd_base]);
data = REG_SET_FIELD(m->cp_hqd_pq_doorbell_control,
CP_HQD_PQ_DOORBELL_CONTROL, DOORBELL_EN, 1);
WREG32_RLC(SOC15_REG_OFFSET(GC, GET_INST(GC, inst), regCP_HQD_PQ_DOORBELL_CONTROL),
data);
if (wptr) {
uint32_t queue_size =
2 << REG_GET_FIELD(m->cp_hqd_pq_control,
CP_HQD_PQ_CONTROL, QUEUE_SIZE);
uint64_t guessed_wptr = m->cp_hqd_pq_rptr & (queue_size - 1);
if ((m->cp_hqd_pq_wptr_lo & (queue_size - 1)) < guessed_wptr)
guessed_wptr += queue_size;
guessed_wptr += m->cp_hqd_pq_wptr_lo & ~(queue_size - 1);
guessed_wptr += (uint64_t)m->cp_hqd_pq_wptr_hi << 32;
WREG32_RLC(SOC15_REG_OFFSET(GC, GET_INST(GC, inst), regCP_HQD_PQ_WPTR_LO),
lower_32_bits(guessed_wptr));
WREG32_RLC(SOC15_REG_OFFSET(GC, GET_INST(GC, inst), regCP_HQD_PQ_WPTR_HI),
upper_32_bits(guessed_wptr));
WREG32_RLC(SOC15_REG_OFFSET(GC, GET_INST(GC, inst), regCP_HQD_PQ_WPTR_POLL_ADDR),
lower_32_bits((uintptr_t)wptr));
WREG32_RLC(SOC15_REG_OFFSET(GC, GET_INST(GC, inst),
regCP_HQD_PQ_WPTR_POLL_ADDR_HI),
upper_32_bits((uintptr_t)wptr));
WREG32(SOC15_REG_OFFSET(GC, GET_INST(GC, inst), regCP_PQ_WPTR_POLL_CNTL1),
(uint32_t)kgd_gfx_v9_get_queue_mask(adev, pipe_id,
queue_id));
}
WREG32_RLC(SOC15_REG_OFFSET(GC, GET_INST(GC, inst), regCP_HQD_EOP_RPTR),
REG_SET_FIELD(m->cp_hqd_eop_rptr,
CP_HQD_EOP_RPTR, INIT_FETCHER, 1));
data = REG_SET_FIELD(m->cp_hqd_active, CP_HQD_ACTIVE, ACTIVE, 1);
WREG32_RLC(SOC15_REG_OFFSET(GC, GET_INST(GC, inst), regCP_HQD_ACTIVE), data);
kgd_gfx_v9_release_queue(adev, inst);
return 0;
}
static uint32_t kgd_gfx_v9_4_3_disable_debug_trap(struct amdgpu_device *adev,
bool keep_trap_enabled,
uint32_t vmid)
{
uint32_t data = 0;
data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, TRAP_EN, 1);
data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, EXCP_EN, 0);
data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, EXCP_REPLACE, 0);
return data;
}
static int kgd_gfx_v9_4_3_validate_trap_override_request(
struct amdgpu_device *adev,
uint32_t trap_override,
uint32_t *trap_mask_supported)
{
*trap_mask_supported &= KFD_DBG_TRAP_MASK_FP_INVALID |
KFD_DBG_TRAP_MASK_FP_INPUT_DENORMAL |
KFD_DBG_TRAP_MASK_FP_DIVIDE_BY_ZERO |
KFD_DBG_TRAP_MASK_FP_OVERFLOW |
KFD_DBG_TRAP_MASK_FP_UNDERFLOW |
KFD_DBG_TRAP_MASK_FP_INEXACT |
KFD_DBG_TRAP_MASK_INT_DIVIDE_BY_ZERO |
KFD_DBG_TRAP_MASK_DBG_ADDRESS_WATCH |
KFD_DBG_TRAP_MASK_DBG_MEMORY_VIOLATION |
KFD_DBG_TRAP_MASK_TRAP_ON_WAVE_START |
KFD_DBG_TRAP_MASK_TRAP_ON_WAVE_END;
if (trap_override != KFD_DBG_TRAP_OVERRIDE_OR &&
trap_override != KFD_DBG_TRAP_OVERRIDE_REPLACE)
return -EPERM;
return 0;
}
static uint32_t trap_mask_map_sw_to_hw(uint32_t mask)
{
uint32_t trap_on_start = (mask & KFD_DBG_TRAP_MASK_TRAP_ON_WAVE_START) ? 1 : 0;
uint32_t trap_on_end = (mask & KFD_DBG_TRAP_MASK_TRAP_ON_WAVE_END) ? 1 : 0;
uint32_t excp_en = mask & (KFD_DBG_TRAP_MASK_FP_INVALID |
KFD_DBG_TRAP_MASK_FP_INPUT_DENORMAL |
KFD_DBG_TRAP_MASK_FP_DIVIDE_BY_ZERO |
KFD_DBG_TRAP_MASK_FP_OVERFLOW |
KFD_DBG_TRAP_MASK_FP_UNDERFLOW |
KFD_DBG_TRAP_MASK_FP_INEXACT |
KFD_DBG_TRAP_MASK_INT_DIVIDE_BY_ZERO |
KFD_DBG_TRAP_MASK_DBG_ADDRESS_WATCH |
KFD_DBG_TRAP_MASK_DBG_MEMORY_VIOLATION);
uint32_t ret;
ret = REG_SET_FIELD(0, SPI_GDBG_PER_VMID_CNTL, EXCP_EN, excp_en);
ret = REG_SET_FIELD(ret, SPI_GDBG_PER_VMID_CNTL, TRAP_ON_START, trap_on_start);
ret = REG_SET_FIELD(ret, SPI_GDBG_PER_VMID_CNTL, TRAP_ON_END, trap_on_end);
return ret;
}
static uint32_t trap_mask_map_hw_to_sw(uint32_t mask)
{
uint32_t ret = REG_GET_FIELD(mask, SPI_GDBG_PER_VMID_CNTL, EXCP_EN);
if (REG_GET_FIELD(mask, SPI_GDBG_PER_VMID_CNTL, TRAP_ON_START))
ret |= KFD_DBG_TRAP_MASK_TRAP_ON_WAVE_START;
if (REG_GET_FIELD(mask, SPI_GDBG_PER_VMID_CNTL, TRAP_ON_END))
ret |= KFD_DBG_TRAP_MASK_TRAP_ON_WAVE_END;
return ret;
}
static uint32_t kgd_gfx_v9_4_3_set_wave_launch_trap_override(
struct amdgpu_device *adev,
uint32_t vmid,
uint32_t trap_override,
uint32_t trap_mask_bits,
uint32_t trap_mask_request,
uint32_t *trap_mask_prev,
uint32_t kfd_dbg_trap_cntl_prev)
{
uint32_t data = 0;
*trap_mask_prev = trap_mask_map_hw_to_sw(kfd_dbg_trap_cntl_prev);
data = (trap_mask_bits & trap_mask_request) |
(*trap_mask_prev & ~trap_mask_request);
data = trap_mask_map_sw_to_hw(data);
data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, TRAP_EN, 1);
data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, EXCP_REPLACE, trap_override);
return data;
}
#define TCP_WATCH_STRIDE (regTCP_WATCH1_ADDR_H - regTCP_WATCH0_ADDR_H)
static uint32_t kgd_gfx_v9_4_3_set_address_watch(
struct amdgpu_device *adev,
uint64_t watch_address,
uint32_t watch_address_mask,
uint32_t watch_id,
uint32_t watch_mode,
uint32_t debug_vmid,
uint32_t inst)
{
uint32_t watch_address_high;
uint32_t watch_address_low;
uint32_t watch_address_cntl;
watch_address_cntl = 0;
watch_address_low = lower_32_bits(watch_address);
watch_address_high = upper_32_bits(watch_address) & 0xffff;
watch_address_cntl = REG_SET_FIELD(watch_address_cntl,
TCP_WATCH0_CNTL,
MODE,
watch_mode);
watch_address_cntl = REG_SET_FIELD(watch_address_cntl,
TCP_WATCH0_CNTL,
MASK,
watch_address_mask >> 7);
watch_address_cntl = REG_SET_FIELD(watch_address_cntl,
TCP_WATCH0_CNTL,
VALID,
1);
WREG32_RLC((SOC15_REG_OFFSET(GC, GET_INST(GC, inst),
regTCP_WATCH0_ADDR_H) +
(watch_id * TCP_WATCH_STRIDE)),
watch_address_high);
WREG32_RLC((SOC15_REG_OFFSET(GC, GET_INST(GC, inst),
regTCP_WATCH0_ADDR_L) +
(watch_id * TCP_WATCH_STRIDE)),
watch_address_low);
return watch_address_cntl;
}
static uint32_t kgd_gfx_v9_4_3_clear_address_watch(struct amdgpu_device *adev,
uint32_t watch_id)
{
return 0;
}
const struct kfd2kgd_calls gc_9_4_3_kfd2kgd = {
.program_sh_mem_settings = kgd_gfx_v9_program_sh_mem_settings,
.set_pasid_vmid_mapping = kgd_gfx_v9_4_3_set_pasid_vmid_mapping,
.init_interrupts = kgd_gfx_v9_init_interrupts,
.hqd_load = kgd_gfx_v9_4_3_hqd_load,
.hiq_mqd_load = kgd_gfx_v9_hiq_mqd_load,
.hqd_sdma_load = kgd_gfx_v9_4_3_hqd_sdma_load,
.hqd_dump = kgd_gfx_v9_hqd_dump,
.hqd_sdma_dump = kgd_gfx_v9_4_3_hqd_sdma_dump,
.hqd_is_occupied = kgd_gfx_v9_hqd_is_occupied,
.hqd_sdma_is_occupied = kgd_gfx_v9_4_3_hqd_sdma_is_occupied,
.hqd_destroy = kgd_gfx_v9_hqd_destroy,
.hqd_sdma_destroy = kgd_gfx_v9_4_3_hqd_sdma_destroy,
.wave_control_execute = kgd_gfx_v9_wave_control_execute,
.get_atc_vmid_pasid_mapping_info =
kgd_gfx_v9_get_atc_vmid_pasid_mapping_info,
.set_vm_context_page_table_base =
kgd_gfx_v9_set_vm_context_page_table_base,
.get_cu_occupancy = kgd_gfx_v9_get_cu_occupancy,
.program_trap_handler_settings =
kgd_gfx_v9_program_trap_handler_settings,
.build_grace_period_packet_info =
kgd_gfx_v9_build_grace_period_packet_info,
.get_iq_wait_times = kgd_gfx_v9_get_iq_wait_times,
.enable_debug_trap = kgd_aldebaran_enable_debug_trap,
.disable_debug_trap = kgd_gfx_v9_4_3_disable_debug_trap,
.validate_trap_override_request =
kgd_gfx_v9_4_3_validate_trap_override_request,
.set_wave_launch_trap_override =
kgd_gfx_v9_4_3_set_wave_launch_trap_override,
.set_wave_launch_mode = kgd_aldebaran_set_wave_launch_mode,
.set_address_watch = kgd_gfx_v9_4_3_set_address_watch,
.clear_address_watch = kgd_gfx_v9_4_3_clear_address_watch
}