#include "amdgpu.h"
#include "amdgpu_discovery.h"
#include "soc15.h"
#include "soc15_common.h"
#include "soc_v1_0.h"
#include "amdgpu_ip.h"
#include "amdgpu_imu.h"
#include "gfxhub_v12_1.h"
#include "sdma_v7_1.h"
#include "gfx_v12_1.h"
#include "amdgpu_video_codecs.h"
#include "amdgpu_reset.h"
#include "gc/gc_12_1_0_offset.h"
#include "gc/gc_12_1_0_sh_mask.h"
#include "mp/mp_15_0_8_offset.h"
#define XCC_REG_RANGE_0_LOW 0x1260 /* XCC gfxdec0 lower Bound */
#define XCC_REG_RANGE_0_HIGH 0x3C00 /* XCC gfxdec0 upper Bound */
#define XCC_REG_RANGE_1_LOW 0xA000 /* XCC gfxdec1 lower Bound */
#define XCC_REG_RANGE_1_HIGH 0x10000 /* XCC gfxdec1 upper Bound */
#define NORMALIZE_XCC_REG_OFFSET(offset) \
(offset & 0xFFFF)
#define MID1_REG_RANGE_0_LOW 0x40000
#define MID1_REG_RANGE_0_HIGH 0x80000
#define NORMALIZE_MID_REG_OFFSET(offset) \
(offset & 0x3FFFF)
static const struct amdgpu_video_codecs vcn_5_0_2_video_codecs_encode_vcn0 = {
.codec_count = 0,
.codec_array = NULL,
};
static const struct amdgpu_video_codec_info vcn_5_0_2_video_codecs_decode_array_vcn0[] = {
{codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4_AVC, 4096, 4096, 52)},
{codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_HEVC, 8192, 4352, 186)},
{codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_JPEG, 16384, 16384, 0)},
{codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_VP9, 8192, 4352, 0)},
{codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_AV1, 8192, 4352, 0)},
};
static const struct amdgpu_video_codecs vcn_5_0_2_video_codecs_decode_vcn0 = {
.codec_count = ARRAY_SIZE(vcn_5_0_2_video_codecs_decode_array_vcn0),
.codec_array = vcn_5_0_2_video_codecs_decode_array_vcn0,
};
static int soc_v1_0_query_video_codecs(struct amdgpu_device *adev, bool encode,
const struct amdgpu_video_codecs **codecs)
{
switch (amdgpu_ip_version(adev, UVD_HWIP, 0)) {
case IP_VERSION(5, 0, 2):
if (encode)
*codecs = &vcn_5_0_2_video_codecs_encode_vcn0;
else
*codecs = &vcn_5_0_2_video_codecs_decode_vcn0;
return 0;
default:
return -EINVAL;
}
}
static void soc_v1_0_doorbell_index_init(struct amdgpu_device *adev)
{
int i;
adev->doorbell_index.kiq = AMDGPU_SOC_V1_0_DOORBELL_KIQ_START;
adev->doorbell_index.mec_ring0 = AMDGPU_SOC_V1_0_DOORBELL_MEC_RING_START;
adev->doorbell_index.mes_ring0 = AMDGPU_SOC_V1_0_DOORBELL_MES_RING0;
adev->doorbell_index.mes_ring1 = AMDGPU_SOC_V1_0_DOORBELL_MES_RING1;
adev->doorbell_index.userqueue_start = AMDGPU_SOC_V1_0_DOORBELL_USERQUEUE_START;
adev->doorbell_index.userqueue_end = AMDGPU_SOC_V1_0_DOORBELL_USERQUEUE_END;
adev->doorbell_index.xcc_doorbell_range = AMDGPU_SOC_V1_0_DOORBELL_XCC_RANGE;
adev->doorbell_index.sdma_doorbell_range = 14;
for (i = 0; i < adev->sdma.num_instances; i++)
adev->doorbell_index.sdma_engine[i] =
AMDGPU_SOC_V1_0_DOORBELL_sDMA_ENGINE_START +
i * (adev->doorbell_index.sdma_doorbell_range >> 1);
adev->doorbell_index.ih = AMDGPU_SOC_V1_0_DOORBELL_IH;
adev->doorbell_index.vcn.vcn_ring0_1 = AMDGPU_SOC_V1_0_DOORBELL_VCN_START;
adev->doorbell_index.first_non_cp = AMDGPU_SOC_V1_0_DOORBELL_FIRST_NON_CP;
adev->doorbell_index.last_non_cp = AMDGPU_SOC_V1_0_DOORBELL_LAST_NON_CP;
adev->doorbell_index.max_assignment = AMDGPU_SOC_V1_0_DOORBELL_MAX_ASSIGNMENT << 1;
}
u64 soc_v1_0_encode_ext_smn_addressing(int ext_id)
{
u64 ext_offset;
int socket_id, die_id;
if (ext_id == 0)
return 0;
die_id = ext_id & 0x3;
socket_id = (ext_id >> 6) & 0xff;
if (socket_id == 0)
ext_offset = ((u64)die_id << 34) | (1ULL << 32);
else if (socket_id != 0 && die_id != 0)
ext_offset = ((u64)socket_id << 40) | ((u64)die_id << 34) |
(3ULL << 32);
else
ext_offset = ((u64)socket_id << 40) | (1ULL << 33);
return ext_offset;
}
static u32 soc_v1_0_get_config_memsize(struct amdgpu_device *adev)
{
return adev->nbio.funcs->get_memsize(adev);
}
static u32 soc_v1_0_get_xclk(struct amdgpu_device *adev)
{
if (!adev->bios)
return 10000;
else
return adev->clock.spll.reference_freq;
}
void soc_v1_0_grbm_select(struct amdgpu_device *adev,
u32 me, u32 pipe,
u32 queue, u32 vmid,
int xcc_id)
{
u32 grbm_gfx_cntl = 0;
grbm_gfx_cntl = REG_SET_FIELD(grbm_gfx_cntl, GRBM_GFX_CNTL, PIPEID, pipe);
grbm_gfx_cntl = REG_SET_FIELD(grbm_gfx_cntl, GRBM_GFX_CNTL, MEID, me);
grbm_gfx_cntl = REG_SET_FIELD(grbm_gfx_cntl, GRBM_GFX_CNTL, VMID, vmid);
grbm_gfx_cntl = REG_SET_FIELD(grbm_gfx_cntl, GRBM_GFX_CNTL, QUEUEID, queue);
WREG32_SOC15_RLC_SHADOW(GC, xcc_id, regGRBM_GFX_CNTL, grbm_gfx_cntl);
}
static struct soc15_allowed_register_entry soc_v1_0_allowed_read_registers[] = {
{ SOC15_REG_ENTRY(GC, 0, regGRBM_STATUS) },
{ SOC15_REG_ENTRY(GC, 0, regGRBM_STATUS2) },
{ SOC15_REG_ENTRY(GC, 0, regGRBM_STATUS3) },
{ SOC15_REG_ENTRY(GC, 0, regGRBM_STATUS_SE0) },
{ SOC15_REG_ENTRY(GC, 0, regGRBM_STATUS_SE1) },
{ SOC15_REG_ENTRY(GC, 0, regCP_STAT) },
{ SOC15_REG_ENTRY(GC, 0, regCP_STALLED_STAT1) },
{ SOC15_REG_ENTRY(GC, 0, regCP_STALLED_STAT2) },
{ SOC15_REG_ENTRY(GC, 0, regCP_STALLED_STAT3) },
{ SOC15_REG_ENTRY(GC, 0, regCP_CPF_BUSY_STAT) },
{ SOC15_REG_ENTRY(GC, 0, regCP_CPF_STALLED_STAT1) },
{ SOC15_REG_ENTRY(GC, 0, regCP_CPF_STATUS) },
{ SOC15_REG_ENTRY(GC, 0, regCP_CPC_BUSY_STAT) },
{ SOC15_REG_ENTRY(GC, 0, regCP_CPC_STALLED_STAT1) },
{ SOC15_REG_ENTRY(GC, 0, regCP_CPC_STATUS) },
{ SOC15_REG_ENTRY(GC, 0, regGB_ADDR_CONFIG_1) },
};
static uint32_t soc_v1_0_get_register_value(struct amdgpu_device *adev,
bool indexed, u32 se_num,
u32 sh_num, u32 reg_offset)
{
if (indexed) {
return amdgpu_read_indexed_register(adev, se_num, sh_num, reg_offset);
} else {
if (reg_offset == SOC15_REG_OFFSET(GC, 0, regGB_ADDR_CONFIG_1) &&
adev->gfx.config.gb_addr_config)
return adev->gfx.config.gb_addr_config;
return RREG32(reg_offset);
}
}
static int soc_v1_0_read_register(struct amdgpu_device *adev,
u32 se_num, u32 sh_num,
u32 reg_offset, u32 *value)
{
uint32_t i;
struct soc15_allowed_register_entry *en;
*value = 0;
for (i = 0; i < ARRAY_SIZE(soc_v1_0_allowed_read_registers); i++) {
en = &soc_v1_0_allowed_read_registers[i];
if (!adev->reg_offset[en->hwip][en->inst])
continue;
else if (reg_offset != (adev->reg_offset[en->hwip][en->inst][en->seg]
+ en->reg_offset))
continue;
*value = soc_v1_0_get_register_value(adev,
soc_v1_0_allowed_read_registers[i].grbm_indexed,
se_num, sh_num, reg_offset);
return 0;
}
return -EINVAL;
}
static bool soc_v1_0_need_reset_on_init(struct amdgpu_device *adev)
{
return false;
}
static enum amd_reset_method
soc_v1_0_asic_reset_method(struct amdgpu_device *adev)
{
if ((adev->gmc.xgmi.supported && adev->gmc.xgmi.connected_to_cpu) ||
(amdgpu_ip_version(adev, MP1_HWIP, 0) == IP_VERSION(15, 0, 8))) {
if (amdgpu_reset_method != -1)
dev_warn_once(adev->dev, "Reset override isn't supported, using Mode2 instead.\n");
return AMD_RESET_METHOD_MODE2;
}
return amdgpu_reset_method;
}
static int soc_v1_0_asic_reset(struct amdgpu_device *adev)
{
switch (soc_v1_0_asic_reset_method(adev)) {
case AMD_RESET_METHOD_MODE2:
dev_info(adev->dev, "MODE2 reset\n");
return amdgpu_dpm_mode2_reset(adev);
default:
dev_info(adev->dev, "Invalid reset method Not supported\n");
return -EOPNOTSUPP;
}
return 0;
}
static void soc_v1_0_get_fw_reserved_info(struct amdgpu_device *adev,
u64 *reserve_size,
u64 *offset)
{
struct mem_reserved_info umf_info, vram_info;
u64 vram_base;
if (amdgpu_discovery_get_mem_reserved_region_by_id(adev,
MASTER_DIE_UMF_REGION_ID, &umf_info)) {
dev_warn(adev->dev,
"MASTER_DIE_UMF region not found in discovery\n");
return;
}
if (!amdgpu_discovery_get_mem_reserved_region_by_id(adev,
SPECIFIC_PURPOSE_REGION_ID, &vram_info) &&
vram_info.reserved_region_size)
vram_base = vram_info.reserved_region_start;
else
vram_base = adev->gmc.vram_start;
dev_dbg(adev->dev, "%s: vram_base=0x%llx\n", __func__, vram_base);
if (umf_info.reserved_region_size == 0 ||
umf_info.reserved_region_start < vram_base ||
umf_info.reserved_region_start + umf_info.reserved_region_size >
vram_base + adev->gmc.real_vram_size) {
dev_warn(adev->dev,
"MASTER_DIE_UMF region out of VRAM: start=0x%llx size=0x%llx vram=[0x%llx,+0x%llx)\n",
umf_info.reserved_region_start, umf_info.reserved_region_size,
vram_base, adev->gmc.real_vram_size);
return;
}
*reserve_size = umf_info.reserved_region_size;
*offset = umf_info.reserved_region_start - vram_base;
}
static const struct amdgpu_asic_funcs soc_v1_0_asic_funcs = {
.read_bios_from_rom = &amdgpu_soc15_read_bios_from_rom,
.read_register = &soc_v1_0_read_register,
.get_config_memsize = &soc_v1_0_get_config_memsize,
.get_xclk = &soc_v1_0_get_xclk,
.init_doorbell_index = &soc_v1_0_doorbell_index_init,
.need_reset_on_init = &soc_v1_0_need_reset_on_init,
.encode_ext_smn_addressing = &soc_v1_0_encode_ext_smn_addressing,
.reset = soc_v1_0_asic_reset,
.reset_method = &soc_v1_0_asic_reset_method,
.query_video_codecs = &soc_v1_0_query_video_codecs,
.get_fw_reserved_info = &soc_v1_0_get_fw_reserved_info,
};
enum soc_v1_0_external_rev_id {
SOC_V1_0_MID_A0_AID_A0_XCD_A0 = 0x1,
SOC_V1_0_MID_A0_AID_A0_XCD_B0 = 0x2,
SOC_V1_0_MID_A0_AID_A1_XCD_A0 = 0x3,
SOC_V1_0_MID_A0_AID_A1_XCD_B0 = 0x4,
};
static int soc_v1_0_set_rev_id(struct amdgpu_device *adev)
{
u16 die_rev_id = 0, xcd_rev_id;
int r;
xcd_rev_id = amdgpu_device_get_rev_id(adev);
r = amdgpu_discovery_get_die_rev_id(adev, &die_rev_id);
if (r) {
die_rev_id = 0;
dev_warn(adev->dev,
"missing die rev id from ip discovery, assuming 0\n");
}
die_rev_id |= xcd_rev_id << SOC_V1_0_DIE_REV_XCD__SHIFT;
adev->rev_id = die_rev_id;
switch (die_rev_id) {
case SOC_V1_0_DIE_REV(0, 0, 0):
adev->external_rev_id = SOC_V1_0_MID_A0_AID_A0_XCD_A0;
break;
case SOC_V1_0_DIE_REV(0, 0, 1):
adev->external_rev_id = SOC_V1_0_MID_A0_AID_A0_XCD_B0;
break;
case SOC_V1_0_DIE_REV(0, 1, 0):
adev->external_rev_id = SOC_V1_0_MID_A0_AID_A1_XCD_A0;
break;
case SOC_V1_0_DIE_REV(0, 1, 1):
adev->external_rev_id = SOC_V1_0_MID_A0_AID_A1_XCD_B0;
break;
default:
dev_warn(adev->dev, "unknown die rev id 0x%x\n", die_rev_id);
break;
}
return 0;
}
static int soc_v1_0_common_early_init(struct amdgpu_ip_block *ip_block)
{
struct amdgpu_device *adev = ip_block->adev;
int r;
adev->reg.pcie.rreg = &amdgpu_device_indirect_rreg;
adev->reg.pcie.wreg = &amdgpu_device_indirect_wreg;
adev->reg.pcie.rreg_ext = &amdgpu_device_indirect_rreg_ext;
adev->reg.pcie.wreg_ext = &amdgpu_device_indirect_wreg_ext;
adev->reg.pcie.rreg64 = &amdgpu_device_indirect_rreg64;
adev->reg.pcie.wreg64 = &amdgpu_device_indirect_wreg64;
adev->reg.pcie.port_rreg = &amdgpu_device_pcie_port_rreg;
adev->reg.pcie.port_wreg = &amdgpu_device_pcie_port_wreg;
adev->reg.pcie.rreg64_ext = &amdgpu_device_indirect_rreg64_ext;
adev->reg.pcie.wreg64_ext = &amdgpu_device_indirect_wreg64_ext;
adev->asic_funcs = &soc_v1_0_asic_funcs;
r = soc_v1_0_set_rev_id(adev);
if (r)
return r;
switch (amdgpu_ip_version(adev, GC_HWIP, 0)) {
case IP_VERSION(12, 1, 0):
adev->cg_flags = 0;
adev->pg_flags = AMD_PG_SUPPORT_VCN_DPG;
break;
default:
return -EINVAL;
}
adev->nbio.funcs->init_registers(adev);
return 0;
}
static int soc_v1_0_common_late_init(struct amdgpu_ip_block *ip_block)
{
struct amdgpu_device *adev = ip_block->adev;
adev->nbio.funcs->enable_doorbell_selfring_aperture(adev, true);
return 0;
}
static int soc_v1_0_common_hw_init(struct amdgpu_ip_block *ip_block)
{
struct amdgpu_device *adev = ip_block->adev;
adev->nbio.funcs->enable_doorbell_aperture(adev, true);
return 0;
}
static int soc_v1_0_common_hw_fini(struct amdgpu_ip_block *ip_block)
{
struct amdgpu_device *adev = ip_block->adev;
adev->nbio.funcs->enable_doorbell_aperture(adev, false);
adev->nbio.funcs->enable_doorbell_selfring_aperture(adev, false);
return 0;
}
static int soc_v1_0_common_suspend(struct amdgpu_ip_block *ip_block)
{
return soc_v1_0_common_hw_fini(ip_block);
}
static int soc_v1_0_common_resume(struct amdgpu_ip_block *ip_block)
{
return soc_v1_0_common_hw_init(ip_block);
}
static int soc_v1_0_common_set_clockgating_state(struct amdgpu_ip_block *ip_block,
enum amd_clockgating_state state)
{
return 0;
}
static int soc_v1_0_common_set_powergating_state(struct amdgpu_ip_block *ip_block,
enum amd_powergating_state state)
{
return 0;
}
static void soc_v1_0_common_get_clockgating_state(struct amdgpu_ip_block *ip_block,
u64 *flags)
{
return;
}
static const struct amd_ip_funcs soc_v1_0_common_ip_funcs = {
.name = "soc_v1_0_common",
.early_init = soc_v1_0_common_early_init,
.late_init = soc_v1_0_common_late_init,
.hw_init = soc_v1_0_common_hw_init,
.hw_fini = soc_v1_0_common_hw_fini,
.suspend = soc_v1_0_common_suspend,
.resume = soc_v1_0_common_resume,
.set_clockgating_state = soc_v1_0_common_set_clockgating_state,
.set_powergating_state = soc_v1_0_common_set_powergating_state,
.get_clockgating_state = soc_v1_0_common_get_clockgating_state,
};
const struct amdgpu_ip_block_version soc_v1_0_common_ip_block = {
.type = AMD_IP_BLOCK_TYPE_COMMON,
.major = 1,
.minor = 0,
.rev = 0,
.funcs = &soc_v1_0_common_ip_funcs,
};
static struct amdgpu_reset_handler *
soc_v1_0_get_reset_handler(struct amdgpu_reset_control *reset_ctl,
struct amdgpu_reset_context *reset_context)
{
struct amdgpu_device *adev = (struct amdgpu_device *)reset_ctl->handle;
struct amdgpu_reset_handler *handler;
enum amd_reset_method method;
int i;
method = (reset_context->method == AMD_RESET_METHOD_NONE) ?
amdgpu_asic_reset_method(adev) : reset_context->method;
for_each_handler(i, handler, reset_ctl) {
if (handler->reset_method == method)
return handler;
}
return NULL;
}
static inline u32 soc_v1_0_get_ip_block_mask(struct amdgpu_device *adev)
{
u32 ip_block_mask = BIT(AMD_IP_BLOCK_TYPE_GFX) |
BIT(AMD_IP_BLOCK_TYPE_MES) |
BIT(AMD_IP_BLOCK_TYPE_SDMA) |
BIT(AMD_IP_BLOCK_TYPE_IH);
return ip_block_mask;
}
static int soc_v1_0_mode2_suspend_ip(struct amdgpu_device *adev)
{
u32 ip_block_mask = soc_v1_0_get_ip_block_mask(adev);
u32 ip_block;
int r, i;
amdgpu_device_set_cg_state(adev, AMD_CG_STATE_UNGATE);
ip_block_mask &= ~BIT(AMD_IP_BLOCK_TYPE_SDMA);
for (i = adev->num_ip_blocks - 1; i >= 0; i--) {
if (!adev->ip_blocks[i].status.valid)
continue;
ip_block = BIT(adev->ip_blocks[i].version->type);
if (!(ip_block_mask & ip_block))
continue;
r = amdgpu_ip_block_suspend(&adev->ip_blocks[i]);
if (r)
return r;
}
return 0;
}
static int
soc_v1_0_mode2_prepare_hwcontext(struct amdgpu_reset_control *reset_ctl,
struct amdgpu_reset_context *reset_context)
{
struct amdgpu_device *adev = (struct amdgpu_device *)reset_ctl->handle;
return soc_v1_0_mode2_suspend_ip(adev);
}
static int soc_v1_0_mode2_reset(struct amdgpu_device *adev)
{
adev->asic_reset_res = amdgpu_dpm_mode2_reset(adev);
return adev->asic_reset_res;
}
static void soc_v1_0_async_reset(struct work_struct *work)
{
struct amdgpu_reset_handler *handler;
struct amdgpu_reset_control *reset_ctl =
container_of(work, struct amdgpu_reset_control, reset_work);
struct amdgpu_device *adev = (struct amdgpu_device *)reset_ctl->handle;
int i;
for_each_handler(i, handler, reset_ctl) {
if (handler->reset_method == reset_ctl->active_reset) {
dev_dbg(adev->dev, "Resetting device\n");
handler->do_reset(adev);
break;
}
}
}
static int
soc_v1_0_mode2_perform_reset(struct amdgpu_reset_control *reset_ctl,
struct amdgpu_reset_context *reset_context)
{
struct amdgpu_device *adev = (struct amdgpu_device *)reset_ctl->handle;
struct list_head *reset_device_list = reset_context->reset_device_list;
struct amdgpu_device *tmp_adev = NULL;
int r = 0;
dev_dbg(adev->dev, "soc_v1_0 perform hw reset\n");
if (!reset_device_list)
return -EINVAL;
list_for_each_entry(tmp_adev, reset_device_list, reset_list) {
mutex_lock(&tmp_adev->reset_cntl->reset_lock);
tmp_adev->reset_cntl->active_reset = AMD_RESET_METHOD_MODE2;
}
list_for_each_entry(tmp_adev, reset_device_list, reset_list) {
r = soc_v1_0_mode2_reset(tmp_adev);
if (r) {
dev_err(tmp_adev->dev,
"ASIC reset failed with error, %d for drm dev, %s",
r, adev_to_drm(tmp_adev)->unique);
break;
}
}
list_for_each_entry(tmp_adev, reset_device_list, reset_list) {
mutex_unlock(&tmp_adev->reset_cntl->reset_lock);
tmp_adev->reset_cntl->active_reset = AMD_RESET_METHOD_NONE;
}
return r;
}
static int soc_v1_0_mode2_restore_ip(struct amdgpu_device *adev)
{
u32 ip_block_mask = soc_v1_0_get_ip_block_mask(adev);
struct amdgpu_ip_block *ih_block;
u32 ip_block;
int i, r;
if (ip_block_mask & BIT(AMD_IP_BLOCK_TYPE_IH)) {
ih_block = amdgpu_device_ip_get_ip_block(adev,
AMD_IP_BLOCK_TYPE_IH);
if (unlikely(!ih_block)) {
dev_err(adev->dev, "Failed to get IH handle\n");
return -EINVAL;
}
r = amdgpu_ip_block_resume(ih_block);
if (r)
return r;
}
ip_block_mask &= ~BIT(AMD_IP_BLOCK_TYPE_IH);
adev->gfxhub.funcs->init(adev);
r = adev->gfxhub.funcs->gart_enable(adev);
if (r) {
dev_err(adev->dev, "GFXHUB gart reenable failed after reset\n");
return r;
}
for (i = 0; i < adev->num_ip_blocks; i++) {
if (!adev->ip_blocks[i].status.valid)
continue;
ip_block = BIT(adev->ip_blocks[i].version->type);
if (!(ip_block_mask & ip_block))
continue;
r = amdgpu_ip_block_resume(&adev->ip_blocks[i]);
if (r)
return r;
}
for (i = 0; i < adev->num_ip_blocks; i++) {
if (!adev->ip_blocks[i].status.valid)
continue;
ip_block = BIT(adev->ip_blocks[i].version->type);
if (!(ip_block_mask & ip_block))
continue;
if (adev->ip_blocks[i].version->funcs->late_init) {
r = adev->ip_blocks[i].version->funcs->late_init(&adev->ip_blocks[i]);
if (r) {
dev_err(adev->dev,
"late_init of IP block <%s> failed %d after reset\n",
adev->ip_blocks[i].version->funcs->name,
r);
return r;
}
}
adev->ip_blocks[i].status.late_initialized = true;
}
amdgpu_device_set_cg_state(adev, AMD_CG_STATE_GATE);
return r;
}
static int
soc_v1_0_mode2_restore_hwcontext(struct amdgpu_reset_control *reset_ctl,
struct amdgpu_reset_context *reset_context)
{
struct list_head *reset_device_list = reset_context->reset_device_list;
struct amdgpu_device *tmp_adev = NULL;
int r;
if (!reset_device_list)
return -EINVAL;
list_for_each_entry(tmp_adev, reset_device_list, reset_list) {
amdgpu_set_init_level(tmp_adev,
AMDGPU_INIT_LEVEL_RESET_RECOVERY);
dev_info(tmp_adev->dev,
"GPU reset succeeded, trying to resume\n");
amdgpu_ras_clear_err_state(tmp_adev);
r = soc_v1_0_mode2_restore_ip(tmp_adev);
if (r)
goto end;
amdgpu_register_gpu_instance(tmp_adev);
amdgpu_ras_resume(tmp_adev);
if (!r) {
amdgpu_set_init_level(tmp_adev,
AMDGPU_INIT_LEVEL_DEFAULT);
amdgpu_irq_gpu_reset_resume_helper(tmp_adev);
r = amdgpu_ib_ring_tests(tmp_adev);
if (r) {
dev_err(tmp_adev->dev,
"ib ring test failed (%d).\n", r);
r = -EAGAIN;
tmp_adev->asic_reset_res = r;
goto end;
}
}
}
end:
return r;
}
static struct amdgpu_reset_handler soc_v1_0_mode2_handler = {
.reset_method = AMD_RESET_METHOD_MODE2,
.prepare_env = NULL,
.prepare_hwcontext = soc_v1_0_mode2_prepare_hwcontext,
.perform_reset = soc_v1_0_mode2_perform_reset,
.restore_hwcontext = soc_v1_0_mode2_restore_hwcontext,
.restore_env = NULL,
.do_reset = soc_v1_0_mode2_reset,
};
static struct amdgpu_reset_handler
*soc_v1_0_rst_handlers[AMDGPU_RESET_MAX_HANDLERS] = {
&soc_v1_0_mode2_handler,
};
int soc_v1_0_reset_init(struct amdgpu_device *adev)
{
struct amdgpu_reset_control *reset_ctl;
reset_ctl = kzalloc_obj(*reset_ctl);
if (!reset_ctl)
return -ENOMEM;
reset_ctl->handle = adev;
reset_ctl->async_reset = soc_v1_0_async_reset;
reset_ctl->active_reset = AMD_RESET_METHOD_NONE;
reset_ctl->get_reset_handler = soc_v1_0_get_reset_handler;
INIT_WORK(&reset_ctl->reset_work, reset_ctl->async_reset);
reset_ctl->reset_handlers = &soc_v1_0_rst_handlers;
adev->reset_cntl = reset_ctl;
return 0;
}
int soc_v1_0_reset_fini(struct amdgpu_device *adev)
{
kfree(adev->reset_cntl);
adev->reset_cntl = NULL;
return 0;
}
static enum amdgpu_gfx_partition __soc_v1_0_calc_xcp_mode(struct amdgpu_xcp_mgr *xcp_mgr)
{
struct amdgpu_device *adev = xcp_mgr->adev;
int num_xcc, num_xcc_per_xcp = 0, mode = 0;
num_xcc = NUM_XCC(xcp_mgr->adev->gfx.xcc_mask);
if (adev->gfx.funcs &&
adev->gfx.funcs->get_xccs_per_xcp)
num_xcc_per_xcp = adev->gfx.funcs->get_xccs_per_xcp(adev);
if ((num_xcc_per_xcp) && (num_xcc % num_xcc_per_xcp == 0))
mode = num_xcc / num_xcc_per_xcp;
if (num_xcc_per_xcp == 1)
return AMDGPU_CPX_PARTITION_MODE;
switch (mode) {
case 1:
return AMDGPU_SPX_PARTITION_MODE;
case 2:
return AMDGPU_DPX_PARTITION_MODE;
case 3:
return AMDGPU_TPX_PARTITION_MODE;
case 4:
return AMDGPU_QPX_PARTITION_MODE;
default:
return AMDGPU_UNKNOWN_COMPUTE_PARTITION_MODE;
}
return AMDGPU_UNKNOWN_COMPUTE_PARTITION_MODE;
}
static int soc_v1_0_query_partition_mode(struct amdgpu_xcp_mgr *xcp_mgr)
{
enum amdgpu_gfx_partition derv_mode, mode;
struct amdgpu_device *adev = xcp_mgr->adev;
mode = AMDGPU_UNKNOWN_COMPUTE_PARTITION_MODE;
derv_mode = __soc_v1_0_calc_xcp_mode(xcp_mgr);
if (amdgpu_sriov_vf(adev) || !adev->psp.funcs)
return derv_mode;
if (adev->nbio.funcs &&
adev->nbio.funcs->get_compute_partition_mode) {
mode = adev->nbio.funcs->get_compute_partition_mode(adev);
if (mode != derv_mode)
dev_warn(adev->dev,
"Mismatch in compute partition mode - reported : %d derived : %d",
mode, derv_mode);
}
return mode;
}
static int __soc_v1_0_get_xcc_per_xcp(struct amdgpu_xcp_mgr *xcp_mgr, int mode)
{
int num_xcc, num_xcc_per_xcp = 0;
num_xcc = NUM_XCC(xcp_mgr->adev->gfx.xcc_mask);
switch (mode) {
case AMDGPU_SPX_PARTITION_MODE:
num_xcc_per_xcp = num_xcc;
break;
case AMDGPU_DPX_PARTITION_MODE:
num_xcc_per_xcp = num_xcc / 2;
break;
case AMDGPU_TPX_PARTITION_MODE:
num_xcc_per_xcp = num_xcc / 3;
break;
case AMDGPU_QPX_PARTITION_MODE:
num_xcc_per_xcp = num_xcc / 4;
break;
case AMDGPU_CPX_PARTITION_MODE:
num_xcc_per_xcp = 1;
break;
}
return num_xcc_per_xcp;
}
static int __soc_v1_0_get_xcp_ip_info(struct amdgpu_xcp_mgr *xcp_mgr, int xcp_id,
enum AMDGPU_XCP_IP_BLOCK ip_id,
struct amdgpu_xcp_ip *ip)
{
struct amdgpu_device *adev = xcp_mgr->adev;
int num_sdma, num_vcn, num_shared_vcn, num_xcp;
int num_xcc_xcp, num_sdma_xcp, num_vcn_xcp;
num_sdma = adev->sdma.num_instances;
num_vcn = adev->vcn.num_vcn_inst;
num_shared_vcn = 1;
num_xcc_xcp = adev->gfx.num_xcc_per_xcp;
num_xcp = NUM_XCC(adev->gfx.xcc_mask) / num_xcc_xcp;
switch (xcp_mgr->mode) {
case AMDGPU_SPX_PARTITION_MODE:
case AMDGPU_DPX_PARTITION_MODE:
case AMDGPU_TPX_PARTITION_MODE:
case AMDGPU_QPX_PARTITION_MODE:
case AMDGPU_CPX_PARTITION_MODE:
num_sdma_xcp = DIV_ROUND_UP(num_sdma, num_xcp);
num_vcn_xcp = DIV_ROUND_UP(num_vcn, num_xcp);
break;
default:
return -EINVAL;
}
if (num_vcn && num_xcp > num_vcn)
num_shared_vcn = num_xcp / num_vcn;
switch (ip_id) {
case AMDGPU_XCP_GFXHUB:
ip->inst_mask = XCP_INST_MASK(num_xcc_xcp, xcp_id);
ip->ip_funcs = &gfxhub_v12_1_xcp_funcs;
break;
case AMDGPU_XCP_GFX:
ip->inst_mask = XCP_INST_MASK(num_xcc_xcp, xcp_id);
ip->ip_funcs = &gfx_v12_1_xcp_funcs;
break;
case AMDGPU_XCP_SDMA:
ip->inst_mask = XCP_INST_MASK(num_sdma_xcp, xcp_id);
ip->ip_funcs = &sdma_v7_1_xcp_funcs;
break;
case AMDGPU_XCP_VCN:
ip->inst_mask =
XCP_INST_MASK(num_vcn_xcp, xcp_id / num_shared_vcn);
break;
default:
return -EINVAL;
}
ip->ip_id = ip_id;
return 0;
}
static int soc_v1_0_get_xcp_res_info(struct amdgpu_xcp_mgr *xcp_mgr,
int mode,
struct amdgpu_xcp_cfg *xcp_cfg)
{
struct amdgpu_device *adev = xcp_mgr->adev;
int max_res[AMDGPU_XCP_RES_MAX] = {};
bool res_lt_xcp;
int num_xcp, i;
u16 nps_modes;
if (!(xcp_mgr->supp_xcp_modes & BIT(mode)))
return -EINVAL;
max_res[AMDGPU_XCP_RES_XCC] = NUM_XCC(adev->gfx.xcc_mask);
max_res[AMDGPU_XCP_RES_DMA] = adev->sdma.num_instances;
max_res[AMDGPU_XCP_RES_DEC] = adev->vcn.num_vcn_inst;
max_res[AMDGPU_XCP_RES_JPEG] = adev->jpeg.num_jpeg_inst;
switch (mode) {
case AMDGPU_SPX_PARTITION_MODE:
num_xcp = 1;
nps_modes = BIT(AMDGPU_NPS1_PARTITION_MODE);
break;
case AMDGPU_DPX_PARTITION_MODE:
num_xcp = 2;
nps_modes = BIT(AMDGPU_NPS1_PARTITION_MODE) |
BIT(AMDGPU_NPS2_PARTITION_MODE);
break;
case AMDGPU_TPX_PARTITION_MODE:
num_xcp = 3;
nps_modes = BIT(AMDGPU_NPS1_PARTITION_MODE) |
BIT(AMDGPU_NPS4_PARTITION_MODE);
break;
case AMDGPU_QPX_PARTITION_MODE:
num_xcp = 4;
nps_modes = BIT(AMDGPU_NPS1_PARTITION_MODE) |
BIT(AMDGPU_NPS4_PARTITION_MODE);
break;
case AMDGPU_CPX_PARTITION_MODE:
num_xcp = NUM_XCC(adev->gfx.xcc_mask);
nps_modes = BIT(AMDGPU_NPS1_PARTITION_MODE) |
BIT(AMDGPU_NPS4_PARTITION_MODE);
break;
default:
return -EINVAL;
}
xcp_cfg->compatible_nps_modes =
(adev->gmc.supported_nps_modes & nps_modes);
xcp_cfg->num_res = ARRAY_SIZE(max_res);
for (i = 0; i < xcp_cfg->num_res; i++) {
xcp_cfg->xcp_res[i].id = i;
if (!max_res[i])
continue;
res_lt_xcp = max_res[i] < num_xcp;
xcp_cfg->xcp_res[i].num_inst =
res_lt_xcp ? 1 : max_res[i] / num_xcp;
xcp_cfg->xcp_res[i].num_inst =
i == AMDGPU_XCP_RES_JPEG ?
xcp_cfg->xcp_res[i].num_inst *
adev->jpeg.num_jpeg_rings : xcp_cfg->xcp_res[i].num_inst;
xcp_cfg->xcp_res[i].num_shared =
res_lt_xcp ? num_xcp / max_res[i] : 1;
}
return 0;
}
static enum amdgpu_gfx_partition __soc_v1_0_get_auto_mode(struct amdgpu_xcp_mgr *xcp_mgr)
{
struct amdgpu_device *adev = xcp_mgr->adev;
int num_xcc;
num_xcc = NUM_XCC(xcp_mgr->adev->gfx.xcc_mask);
if (adev->gmc.num_mem_partitions == 1)
return AMDGPU_SPX_PARTITION_MODE;
if (adev->gmc.num_mem_partitions == num_xcc)
return AMDGPU_CPX_PARTITION_MODE;
if (adev->gmc.num_mem_partitions == 2)
return AMDGPU_DPX_PARTITION_MODE;
return AMDGPU_UNKNOWN_COMPUTE_PARTITION_MODE;
}
static bool __soc_v1_0_is_valid_mode(struct amdgpu_xcp_mgr *xcp_mgr,
enum amdgpu_gfx_partition mode)
{
struct amdgpu_device *adev = xcp_mgr->adev;
int num_xcc, num_xccs_per_xcp;
num_xcc = NUM_XCC(adev->gfx.xcc_mask);
switch (mode) {
case AMDGPU_SPX_PARTITION_MODE:
return adev->gmc.num_mem_partitions == 1 && num_xcc > 0;
case AMDGPU_DPX_PARTITION_MODE:
return adev->gmc.num_mem_partitions <= 2 && (num_xcc % 2) == 0;
case AMDGPU_TPX_PARTITION_MODE:
return (adev->gmc.num_mem_partitions == 1 ||
adev->gmc.num_mem_partitions == 3) &&
((num_xcc % 3) == 0);
case AMDGPU_QPX_PARTITION_MODE:
num_xccs_per_xcp = num_xcc / 4;
return (adev->gmc.num_mem_partitions == 1 ||
adev->gmc.num_mem_partitions == 4) &&
(num_xccs_per_xcp >= 2);
case AMDGPU_CPX_PARTITION_MODE:
return (adev->gmc.num_mem_partitions == 1 ||
adev->gmc.num_mem_partitions == 4) &&
(num_xcc > 1) &&
(num_xcc % adev->gmc.num_mem_partitions) == 0;
default:
return false;
}
return false;
}
static void __soc_v1_0_update_available_partition_mode(struct amdgpu_xcp_mgr *xcp_mgr)
{
int mode;
xcp_mgr->avail_xcp_modes = 0;
for_each_inst(mode, xcp_mgr->supp_xcp_modes) {
if (__soc_v1_0_is_valid_mode(xcp_mgr, mode))
xcp_mgr->avail_xcp_modes |= BIT(mode);
}
}
static int soc_v1_0_switch_partition_mode(struct amdgpu_xcp_mgr *xcp_mgr,
int mode, int *num_xcps)
{
int num_xcc_per_xcp, num_xcc, ret;
struct amdgpu_device *adev;
u32 flags = 0;
adev = xcp_mgr->adev;
num_xcc = NUM_XCC(adev->gfx.xcc_mask);
if (mode == AMDGPU_AUTO_COMPUTE_PARTITION_MODE) {
mode = __soc_v1_0_get_auto_mode(xcp_mgr);
if (mode == AMDGPU_UNKNOWN_COMPUTE_PARTITION_MODE) {
dev_err(adev->dev,
"Invalid config, no compatible compute partition mode found, available memory partitions: %d",
adev->gmc.num_mem_partitions);
return -EINVAL;
}
} else if (!__soc_v1_0_is_valid_mode(xcp_mgr, mode)) {
dev_err(adev->dev,
"Invalid compute partition mode requested, requested: %s, available memory partitions: %d",
amdgpu_gfx_compute_mode_desc(mode), adev->gmc.num_mem_partitions);
return -EINVAL;
}
if (adev->kfd.init_complete && !amdgpu_in_reset(adev))
flags |= AMDGPU_XCP_OPS_KFD;
ret = amdgpu_xcp_pre_partition_switch(xcp_mgr, flags);
if (ret)
goto out;
num_xcc_per_xcp = __soc_v1_0_get_xcc_per_xcp(xcp_mgr, mode);
if (adev->gfx.imu.funcs &&
adev->gfx.imu.funcs->switch_compute_partition) {
ret = adev->gfx.imu.funcs->switch_compute_partition(xcp_mgr->adev, num_xcc_per_xcp, mode);
if (ret)
goto out;
}
if (adev->gfx.imu.funcs &&
adev->gfx.imu.funcs->init_mcm_addr_lut &&
amdgpu_emu_mode)
adev->gfx.imu.funcs->init_mcm_addr_lut(adev);
*num_xcps = num_xcc / num_xcc_per_xcp;
amdgpu_xcp_init(xcp_mgr, *num_xcps, mode);
ret = amdgpu_xcp_post_partition_switch(xcp_mgr, flags);
if (!ret)
__soc_v1_0_update_available_partition_mode(xcp_mgr);
out:
return ret;
}
#ifdef HAVE_ACPI_DEV_GET_FIRST_MATCH_DEV
static int __soc_v1_0_get_xcp_mem_id(struct amdgpu_device *adev,
int xcc_id, uint8_t *mem_id)
{
*mem_id = xcc_id / adev->gfx.num_xcc_per_xcp;
*mem_id /= adev->xcp_mgr->num_xcp_per_mem_partition;
return 0;
}
static int soc_v1_0_get_xcp_mem_id(struct amdgpu_xcp_mgr *xcp_mgr,
struct amdgpu_xcp *xcp, uint8_t *mem_id)
{
struct amdgpu_numa_info numa_info;
struct amdgpu_device *adev;
uint32_t xcc_mask;
int r, i, xcc_id;
adev = xcp_mgr->adev;
if (adev->gmc.num_mem_partitions == 1) {
*mem_id = 0;
return 0;
}
r = amdgpu_xcp_get_inst_details(xcp, AMDGPU_XCP_GFX, &xcc_mask);
if (r || !xcc_mask)
return -EINVAL;
xcc_id = ffs(xcc_mask) - 1;
if (!adev->gmc.is_app_apu)
return __soc_v1_0_get_xcp_mem_id(adev, xcc_id, mem_id);
r = amdgpu_acpi_get_mem_info(adev, xcc_id, &numa_info);
if (r)
return r;
r = -EINVAL;
for (i = 0; i < adev->gmc.num_mem_partitions; ++i) {
if (adev->gmc.mem_partitions[i].numa.node == numa_info.nid) {
*mem_id = i;
r = 0;
break;
}
}
return r;
}
#endif
static int soc_v1_0_get_xcp_ip_details(struct amdgpu_xcp_mgr *xcp_mgr, int xcp_id,
enum AMDGPU_XCP_IP_BLOCK ip_id,
struct amdgpu_xcp_ip *ip)
{
if (!ip)
return -EINVAL;
return __soc_v1_0_get_xcp_ip_info(xcp_mgr, xcp_id, ip_id, ip);
}
struct amdgpu_xcp_mgr_funcs soc_v1_0_xcp_funcs = {
.switch_partition_mode = &soc_v1_0_switch_partition_mode,
.query_partition_mode = &soc_v1_0_query_partition_mode,
.get_ip_details = &soc_v1_0_get_xcp_ip_details,
.get_xcp_res_info = &soc_v1_0_get_xcp_res_info,
#ifdef HAVE_ACPI_DEV_GET_FIRST_MATCH_DEV
.get_xcp_mem_id = &soc_v1_0_get_xcp_mem_id,
#endif
};
static int soc_v1_0_xcp_mgr_init(struct amdgpu_device *adev)
{
int ret;
if (amdgpu_sriov_vf(adev))
soc_v1_0_xcp_funcs.switch_partition_mode = NULL;
ret = amdgpu_xcp_mgr_init(adev, AMDGPU_UNKNOWN_COMPUTE_PARTITION_MODE,
1, &soc_v1_0_xcp_funcs);
if (ret)
return ret;
amdgpu_xcp_update_supported_modes(adev->xcp_mgr);
return ret;
}
uint32_t soc_v1_0_get_aid_mask(uint16_t xcc_mask)
{
int xcc_inst_per_aid = 4;
uint32_t aid_mask = 0;
int i;
for (i = 0; xcc_mask; xcc_mask >>= xcc_inst_per_aid, i++) {
if (xcc_mask & GENMASK(xcc_inst_per_aid - 1, 0))
aid_mask |= BIT(i);
}
return aid_mask;
}
int soc_v1_0_init_soc_config(struct amdgpu_device *adev)
{
int ret, i;
uint16_t sdma_mask = 0;
adev->aid_mask = soc_v1_0_get_aid_mask(adev->gfx.xcc_mask);
adev->sdma.num_inst_per_xcc = 2;
for_each_inst(i, adev->gfx.xcc_mask)
sdma_mask |=
GENMASK(adev->sdma.num_inst_per_xcc - 1, 0) <<
(i * adev->sdma.num_inst_per_xcc);
adev->sdma.sdma_mask = sdma_mask;
adev->sdma.num_instances = NUM_XCC(adev->sdma.sdma_mask);
adev->vcn.harvest_config = 0;
adev->vcn.num_inst_per_aid = 2;
adev->vcn.num_vcn_inst = hweight32(adev->vcn.inst_mask);
adev->jpeg.harvest_config = 0;
adev->jpeg.num_inst_per_aid = 2;
adev->jpeg.num_jpeg_inst = hweight32(adev->jpeg.inst_mask);
ret = soc_v1_0_xcp_mgr_init(adev);
if (ret)
return ret;
amdgpu_ip_map_init(adev);
return 0;
}
bool soc_v1_0_normalize_xcc_reg_range(uint32_t reg)
{
if (((reg >= XCC_REG_RANGE_0_LOW) && (reg < XCC_REG_RANGE_0_HIGH)) ||
((reg >= XCC_REG_RANGE_1_LOW) && (reg < XCC_REG_RANGE_1_HIGH)))
return true;
else
return false;
}
uint32_t soc_v1_0_normalize_xcc_reg_offset(uint32_t reg)
{
uint32_t normalized_reg = NORMALIZE_XCC_REG_OFFSET(reg);
if (soc_v1_0_normalize_xcc_reg_range(normalized_reg))
return normalized_reg;
else
return reg;
}
bool soc_v1_0_mid1_reg_range(uint32_t reg)
{
uint32_t normalized_reg = soc_v1_0_normalize_xcc_reg_offset(reg);
if (soc_v1_0_normalize_xcc_reg_range(normalized_reg))
return false;
if ((reg >= MID1_REG_RANGE_0_LOW) && (reg < MID1_REG_RANGE_0_HIGH))
return true;
else
return false;
}
uint32_t soc_v1_0_normalize_reg_offset(uint32_t reg)
{
uint32_t normalized_reg = soc_v1_0_normalize_xcc_reg_offset(reg);
if (soc_v1_0_normalize_xcc_reg_range(normalized_reg))
return soc_v1_0_normalize_xcc_reg_offset(reg);
if (soc_v1_0_mid1_reg_range(reg))
return NORMALIZE_MID_REG_OFFSET(reg);
return reg;
}