linuxmsm/laptops-kernel
0
0 Fork 0
mirror of https://github.com/linux-msm/laptops-kernel.git synced 2026-08-13 14:19:53 -07:00
Code Issues Packages Projects Releases Wiki Activity

Merge tag 'amd-drm-next-6.5-2023-06-09' of https://gitlab.freedesktop.org/agd5f/linux into drm-next

Browse Source 
amd-drm-next-6.5-2023-06-02:

amdgpu:
- SR-IOV fixes
- Warning fixes
- Misc code cleanups and spelling fixes
- DCN 3.2 updates
- Improved DC FAMS support for better power management
- Improved DC SubVP support for better power management
- DCN 3.1.x fixes
- Max IB size query
- DC GPU reset fixes
- RAS updates
- DCN 3.0.x fixes
- S/G display fixes
- CP shadow buffer support
- Implement connector force callback
- Z8 power improvements
- PSP 13.0.10 vbflash support
- Mode2 reset fixes
- Store MQDs in VRAM to improve queue switch latency
- VCN 3.x fixes
- JPEG 3.x fixes
- Enable DC_FP on LoongArch
- GFXOFF fixes
- GC 9.4.3 partition support
- SDMA 4.4.2 partition support
- VCN/JPEG 4.0.3 partition support
- VCN 4.0.3 updates
- NBIO 7.9 updates
- GC 9.4.3 updates
- Take NUMA into account when allocating memory
- Handle NUMA for partitions
- SMU 13.0.6 updates
- GC 9.4.3 RAS updates
- Stop including unused swiotlb.h
- SMU 13.0.7 fixes
- Fix clock output ordering on some APUs
- Clean up DC FPGA code
- GFX9 preemption fixes
- Misc irq fixes
- S0ix fixes
- Add new DRM_AMDGPU_WERROR config parameter to help with CI
- PCIe fix for RDNA2
- kdoc fixes
- Documentation updates

amdkfd:
- Query TTM mem limit rather than hardcoding it
- GC 9.4.3 partition support
- Handle NUMA for partitions

radeon:
- Fix possible double free
- Stop including unused swiotlb.h
- Fix possible division by zero

ttm:
- Add query for TTM mem limit
- Add NUMA awareness to pools
- Export ttm_pool_fini()

UAPI:
- Add new ctx query flag to better handle GPU resets
  Mesa MR: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22290
- Add new interface to query and set shadow buffer for RDNA3
  Mesa MR: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21986
- Add new INFO query for max IB size
  Proposed userspace: https://gitlab.freedesktop.org/bnieuwenhuizen/mesa/-/commits/ib-rejection-v3

amd-drm-next-6.5-2023-06-09:

amdgpu:
- S0ix fixes
- Initial SMU13 Overdrive support
- kdoc fixes
- Misc clode cleanups
- Flexible array fixes
- Display OTG fixes
- SMU 13.0.6 updates
- Revert some broken clock counter updates
- Misc display fixes
- GFX9 preemption fixes
- Add support for newer EEPROM bad page table format
- Add missing radeon secondary id
- Add support for new colorspace KMS API
- CSA fix
- Stable pstate fixes for APUs
- make vbl interface admin only
- Handle PCI accelerator class

amdkfd:
- Add debugger support for gdb

radeon:
- Fix possible UAF

drm:
- Add Colorspace functionality

UAPI:
- Add debugger interface for enabling gdb
  Proposed userspace: https://github.com/ROCm-Developer-Tools/ROCdbgapi/tree/wip-dbgapi
- Add KMS colorspace API
  Discussion: https://lists.freedesktop.org/archives/dri-devel/2023-June/408128.html

From: Alex Deucher <alexander.deucher@amd.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230609174817.7764-1-alexander.deucher@amd.com
This commit is contained in:
Dave Airlie committed 2023-06-15 14:11:22 +10:00
2 parents ba57b9b11f 3b718dcaf1
commit 901bdf5ea1
498 files changed
+45772 -8021
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

Documentation/gpu/amdgpu/apu-asic-info-table.csv
+3 -1
View File
@@ -5,6 +5,8 @@ Ryzen 4000 series, RENOIR, DCN 2.1, 9.3, VCN 2.2, 4.1.2, 11.0.3
Ryzen 3000 series / AMD Ryzen Embedded V1*/R1* with Radeon Vega Gfx, RAVEN2, DCN 1.0, 9.2.2, VCN 1.0.1, 4.1.1, 10.0.1
SteamDeck, VANGOGH, DCN 3.0.1, 10.3.1, VCN 3.1.0, 5.2.1, 11.5.0
Ryzen 5000 series / Ryzen 7x30 series, GREEN SARDINE / Cezanne / Barcelo / Barcelo-R, DCN 2.1, 9.3, VCN 2.2, 4.1.1, 12.0.1
Ryzen 6000 series / Ryzen 7x35 series, YELLOW CARP / Rembrandt / Rembrandt+, 3.1.2, 10.3.3, VCN 3.1.1, 5.2.3, 13.0.3
Ryzen 6000 series / Ryzen 7x35 series / Ryzen 7x36 series, YELLOW CARP / Rembrandt / Rembrandt-R, 3.1.2, 10.3.3, VCN 3.1.1, 5.2.3, 13.0.3
Ryzen 7000 series (AM5), Raphael, 3.1.5, 10.3.6, 3.1.2, 5.2.6, 13.0.5
Ryzen 7x45 series (FL1), / Dragon Range, 3.1.5, 10.3.6, 3.1.2, 5.2.6, 13.0.5
Ryzen 7x20 series, Mendocino, 3.1.6, 10.3.7, 3.1.1, 5.2.7, 13.0.8
Ryzen 7x40 series, Phoenix, 3.1.4, 11.0.1 / 11.0.4, 4.0.2, 6.0.1, 13.0.4 / 13.0.11
1 Product Name Code Reference DCN/DCE version GC version VCE/UVD/VCN version SDMA version MP0 version
5 Ryzen 3000 series / AMD Ryzen Embedded V1*/R1* with Radeon Vega Gfx RAVEN2 DCN 1.0 9.2.2 VCN 1.0.1 4.1.1 10.0.1
6 SteamDeck VANGOGH DCN 3.0.1 10.3.1 VCN 3.1.0 5.2.1 11.5.0
7 Ryzen 5000 series / Ryzen 7x30 series GREEN SARDINE / Cezanne / Barcelo / Barcelo-R DCN 2.1 9.3 VCN 2.2 4.1.1 12.0.1
8 Ryzen 6000 series / Ryzen 7x35 series Ryzen 6000 series / Ryzen 7x35 series / Ryzen 7x36 series YELLOW CARP / Rembrandt / Rembrandt+ YELLOW CARP / Rembrandt / Rembrandt-R 3.1.2 10.3.3 VCN 3.1.1 5.2.3 13.0.3
9 Ryzen 7000 series (AM5) Raphael 3.1.5 10.3.6 3.1.2 5.2.6 13.0.5
10 Ryzen 7x45 series (FL1) / Dragon Range 3.1.5 10.3.6 3.1.2 5.2.6 13.0.5
11 Ryzen 7x20 series Mendocino 3.1.6 10.3.7 3.1.1 5.2.7 13.0.8
12 Ryzen 7x40 series Phoenix 3.1.4 11.0.1 / 11.0.4 4.0.2 6.0.1 13.0.4 / 13.0.11
drivers/gpu/drm/Makefile
+1
View File
@@ -140,6 +140,7 @@ obj-$(CONFIG_DRM_TTM) += ttm/
obj-$(CONFIG_DRM_SCHED) += scheduler/
obj-$(CONFIG_DRM_RADEON)+= radeon/
obj-$(CONFIG_DRM_AMDGPU)+= amd/amdgpu/
obj-$(CONFIG_DRM_AMDGPU)+= amd/amdxcp/
obj-$(CONFIG_DRM_I915) += i915/
obj-$(CONFIG_DRM_KMB_DISPLAY) += kmb/
obj-$(CONFIG_DRM_MGAG200) += mgag200/
drivers/gpu/drm/amd/amdgpu/Kconfig
+10
View File
@@ -69,6 +69,16 @@ config DRM_AMDGPU_USERPTR
This option selects CONFIG_HMM and CONFIG_HMM_MIRROR if it
isn't already selected to enabled full userptr support.
config DRM_AMDGPU_WERROR
bool "Force the compiler to throw an error instead of a warning when compiling"
depends on DRM_AMDGPU
depends on EXPERT
depends on !COMPILE_TEST
default n
help
Add -Werror to the build flags for amdgpu.ko.
Only enable this if you are warning code for amdgpu.ko.
source "drivers/gpu/drm/amd/acp/Kconfig"
source "drivers/gpu/drm/amd/display/Kconfig"
source "drivers/gpu/drm/amd/amdkfd/Kconfig"
drivers/gpu/drm/amd/amdgpu/Makefile
+16 -3
View File
@@ -39,6 +39,15 @@ ccflags-y := -I$(FULL_AMD_PATH)/include/asic_reg \
-I$(FULL_AMD_DISPLAY_PATH)/amdgpu_dm \
-I$(FULL_AMD_PATH)/amdkfd
subdir-ccflags-y := -Wextra
subdir-ccflags-y += $(call cc-option, -Wunused-but-set-variable)
subdir-ccflags-y += -Wno-unused-parameter
subdir-ccflags-y += -Wno-type-limits
subdir-ccflags-y += -Wno-sign-compare
subdir-ccflags-y += -Wno-missing-field-initializers
subdir-ccflags-y += -Wno-override-init
subdir-ccflags-$(CONFIG_DRM_AMDGPU_WERROR) += -Werror
amdgpu-y := amdgpu_drv.o
# add KMS driver
@@ -60,7 +69,7 @@ amdgpu-y += amdgpu_device.o amdgpu_kms.o \
amdgpu_umc.o smu_v11_0_i2c.o amdgpu_fru_eeprom.o amdgpu_rap.o \
amdgpu_fw_attestation.o amdgpu_securedisplay.o \
amdgpu_eeprom.o amdgpu_mca.o amdgpu_psp_ta.o amdgpu_lsdma.o \
amdgpu_ring_mux.o
amdgpu_ring_mux.o amdgpu_xcp.o
amdgpu-$(CONFIG_PROC_FS) += amdgpu_fdinfo.o
@@ -78,7 +87,7 @@ amdgpu-y += \
vega20_reg_init.o nbio_v7_4.o nbio_v2_3.o nv.o arct_reg_init.o mxgpu_nv.o \
nbio_v7_2.o hdp_v4_0.o hdp_v5_0.o aldebaran_reg_init.o aldebaran.o soc21.o \
sienna_cichlid.o smu_v13_0_10.o nbio_v4_3.o hdp_v6_0.o nbio_v7_7.o hdp_v5_2.o lsdma_v6_0.o \
nbio_v7_9.o
nbio_v7_9.o aqua_vanjaram_reg_init.o
# add DF block
amdgpu-y += \
@@ -183,12 +192,14 @@ amdgpu-y += \
vcn_v2_5.o \
vcn_v3_0.o \
vcn_v4_0.o \
vcn_v4_0_3.o \
amdgpu_jpeg.o \
jpeg_v1_0.o \
jpeg_v2_0.o \
jpeg_v2_5.o \
jpeg_v3_0.o \
jpeg_v4_0.o
jpeg_v4_0.o \
jpeg_v4_0_3.o
# add ATHUB block
amdgpu-y += \
@@ -203,6 +214,7 @@ amdgpu-y += \
smuio_v11_0.o \
smuio_v11_0_6.o \
smuio_v13_0.o \
smuio_v13_0_3.o \
smuio_v13_0_6.o
# add reset block
@@ -228,6 +240,7 @@ amdgpu-y += \
amdgpu_amdkfd_gfx_v9.o \
amdgpu_amdkfd_arcturus.o \
amdgpu_amdkfd_aldebaran.o \
amdgpu_amdkfd_gc_9_4_3.o \
amdgpu_amdkfd_gfx_v10.o \
amdgpu_amdkfd_gfx_v10_3.o \
amdgpu_amdkfd_gfx_v11.o
drivers/gpu/drm/amd/amdgpu/amdgpu.h
+74 -5
View File
@@ -107,8 +107,9 @@
#include "amdgpu_fdinfo.h"
#include "amdgpu_mca.h"
#include "amdgpu_ras.h"
#include "amdgpu_xcp.h"
#define MAX_GPU_INSTANCE 16
#define MAX_GPU_INSTANCE 64
struct amdgpu_gpu_instance
{
@@ -212,6 +213,8 @@ extern int amdgpu_noretry;
extern int amdgpu_force_asic_type;
extern int amdgpu_smartshift_bias;
extern int amdgpu_use_xgmi_p2p;
extern int amdgpu_mtype_local;
extern bool enforce_isolation;
#ifdef CONFIG_HSA_AMD
extern int sched_policy;
extern bool debug_evictions;
@@ -242,9 +245,10 @@ extern int amdgpu_num_kcq;
extern int amdgpu_vcnfw_log;
extern int amdgpu_sg_display;
extern int amdgpu_user_partt_mode;
#define AMDGPU_VM_MAX_NUM_CTX 4096
#define AMDGPU_SG_THRESHOLD (256*1024*1024)
#define AMDGPU_DEFAULT_GTT_SIZE_MB 3072ULL /* 3GB by default */
#define AMDGPU_WAIT_IDLE_TIMEOUT_IN_MS 3000
#define AMDGPU_MAX_USEC_TIMEOUT 100000 /* 100 ms */
#define AMDGPU_FENCE_JIFFIES_TIMEOUT (HZ / 2)
@@ -282,6 +286,7 @@ extern int amdgpu_sg_display;
#define AMDGPU_SMARTSHIFT_MAX_BIAS (100)
#define AMDGPU_SMARTSHIFT_MIN_BIAS (-100)
struct amdgpu_xcp_mgr;
struct amdgpu_device;
struct amdgpu_irq_src;
struct amdgpu_fpriv;
@@ -463,6 +468,8 @@ struct amdgpu_fpriv {
struct mutex bo_list_lock;
struct idr bo_list_handles;
struct amdgpu_ctx_mgr ctx_mgr;
/** GPU partition selection */
uint32_t xcp_id;
};
int amdgpu_file_to_fpriv(struct file *filp, struct amdgpu_fpriv **fpriv);
@@ -573,6 +580,8 @@ struct amdgpu_asic_funcs {
/* query video codecs */
int (*query_video_codecs)(struct amdgpu_device *adev, bool encode,
const struct amdgpu_video_codecs **codecs);
/* encode "> 32bits" smn addressing */
u64 (*encode_ext_smn_addressing)(int ext_id);
};
/*
@@ -607,6 +616,9 @@ void amdgpu_cgs_destroy_device(struct cgs_device *cgs_device);
typedef uint32_t (*amdgpu_rreg_t)(struct amdgpu_device*, uint32_t);
typedef void (*amdgpu_wreg_t)(struct amdgpu_device*, uint32_t, uint32_t);
typedef uint32_t (*amdgpu_rreg_ext_t)(struct amdgpu_device*, uint64_t);
typedef void (*amdgpu_wreg_ext_t)(struct amdgpu_device*, uint64_t, uint32_t);
typedef uint64_t (*amdgpu_rreg64_t)(struct amdgpu_device*, uint32_t);
typedef void (*amdgpu_wreg64_t)(struct amdgpu_device*, uint32_t, uint64_t);
@@ -657,7 +669,7 @@ enum amd_hw_ip_block_type {
MAX_HWIP
};
#define HWIP_MAX_INSTANCE 28
#define HWIP_MAX_INSTANCE 44
#define HW_ID_MAX 300
#define IP_VERSION(mj, mn, rv) (((mj) << 16) | ((mn) << 8) | (rv))
@@ -665,6 +677,17 @@ enum amd_hw_ip_block_type {
#define IP_VERSION_MIN(ver) (((ver) >> 8) & 0xFF)
#define IP_VERSION_REV(ver) ((ver) & 0xFF)
struct amdgpu_ip_map_info {
/* Map of logical to actual dev instances/mask */
uint32_t dev_inst[MAX_HWIP][HWIP_MAX_INSTANCE];
int8_t (*logical_to_dev_inst)(struct amdgpu_device *adev,
enum amd_hw_ip_block_type block,
int8_t inst);
uint32_t (*logical_to_dev_mask)(struct amdgpu_device *adev,
enum amd_hw_ip_block_type block,
uint32_t mask);
};
struct amd_powerplay {
void *pp_handle;
const struct amd_pm_funcs *pp_funcs;
@@ -750,6 +773,7 @@ struct amdgpu_device {
struct amdgpu_acp acp;
#endif
struct amdgpu_hive_info *hive;
struct amdgpu_xcp_mgr *xcp_mgr;
/* ASIC */
enum amd_asic_type asic_type;
uint32_t family;
@@ -797,6 +821,8 @@ struct amdgpu_device {
amdgpu_wreg_t pcie_wreg;
amdgpu_rreg_t pciep_rreg;
amdgpu_wreg_t pciep_wreg;
amdgpu_rreg_ext_t pcie_rreg_ext;
amdgpu_wreg_ext_t pcie_wreg_ext;
amdgpu_rreg64_t pcie_rreg64;
amdgpu_wreg64_t pcie_wreg64;
/* protects concurrent UVD register access */
@@ -830,7 +856,7 @@ struct amdgpu_device {
dma_addr_t dummy_page_addr;
struct amdgpu_vm_manager vm_manager;
struct amdgpu_vmhub vmhub[AMDGPU_MAX_VMHUBS];
unsigned num_vmhubs;
DECLARE_BITMAP(vmhubs_mask, AMDGPU_MAX_VMHUBS);
/* memory management */
struct amdgpu_mman mman;
@@ -962,6 +988,7 @@ struct amdgpu_device {
/* soc15 register offset based on ip, instance and segment */
uint32_t *reg_offset[MAX_HWIP][HWIP_MAX_INSTANCE];
struct amdgpu_ip_map_info ip_map;
/* delayed work_func for deferring clockgating during resume */
struct delayed_work delayed_init_work;
@@ -1020,6 +1047,9 @@ struct amdgpu_device {
struct pci_saved_state *pci_state;
pci_channel_state_t pci_channel_state;
/* Track auto wait count on s_barrier settings */
bool barrier_has_auto_waitcnt;
struct amdgpu_reset_control *reset_cntl;
uint32_t ip_versions[MAX_HWIP][HWIP_MAX_INSTANCE];
@@ -1050,6 +1080,8 @@ struct amdgpu_device {
bool job_hang;
bool dc_enabled;
/* Mask of active clusters */
uint32_t aid_mask;
};
static inline struct amdgpu_device *drm_to_adev(struct drm_device *ddev)
@@ -1081,11 +1113,18 @@ size_t amdgpu_device_aper_access(struct amdgpu_device *adev, loff_t pos,
void amdgpu_device_vram_access(struct amdgpu_device *adev, loff_t pos,
void *buf, size_t size, bool write);
uint32_t amdgpu_device_wait_on_rreg(struct amdgpu_device *adev,
uint32_t inst, uint32_t reg_addr, char reg_name[],
uint32_t expected_value, uint32_t mask);
uint32_t amdgpu_device_rreg(struct amdgpu_device *adev,
uint32_t reg, uint32_t acc_flags);
u32 amdgpu_device_indirect_rreg_ext(struct amdgpu_device *adev,
u64 reg_addr);
void amdgpu_device_wreg(struct amdgpu_device *adev,
uint32_t reg, uint32_t v,
uint32_t acc_flags);
void amdgpu_device_indirect_wreg_ext(struct amdgpu_device *adev,
u64 reg_addr, u32 reg_data);
void amdgpu_mm_wreg_mmio_rlc(struct amdgpu_device *adev,
uint32_t reg, uint32_t v);
void amdgpu_mm_wreg8(struct amdgpu_device *adev, uint32_t offset, uint8_t value);
@@ -1137,6 +1176,8 @@ int emu_soc_asic_init(struct amdgpu_device *adev);
#define WREG32_PCIE(reg, v) adev->pcie_wreg(adev, (reg), (v))
#define RREG32_PCIE_PORT(reg) adev->pciep_rreg(adev, (reg))
#define WREG32_PCIE_PORT(reg, v) adev->pciep_wreg(adev, (reg), (v))
#define RREG32_PCIE_EXT(reg) adev->pcie_rreg_ext(adev, (reg))
#define WREG32_PCIE_EXT(reg, v) adev->pcie_wreg_ext(adev, (reg), (v))
#define RREG64_PCIE(reg) adev->pcie_rreg64(adev, (reg))
#define WREG64_PCIE(reg, v) adev->pcie_wreg64(adev, (reg), (v))
#define RREG32_SMC(reg) adev->smc_rreg(adev, (reg))
@@ -1204,7 +1245,8 @@ int emu_soc_asic_init(struct amdgpu_device *adev);
/*
* ASICs macro.
*/
#define amdgpu_asic_set_vga_state(adev, state) (adev)->asic_funcs->set_vga_state((adev), (state))
#define amdgpu_asic_set_vga_state(adev, state) \
((adev)->asic_funcs->set_vga_state ? (adev)->asic_funcs->set_vga_state((adev), (state)) : 0)
#define amdgpu_asic_reset(adev) (adev)->asic_funcs->reset((adev))
#define amdgpu_asic_reset_method(adev) (adev)->asic_funcs->reset_method((adev))
#define amdgpu_asic_get_xclk(adev) (adev)->asic_funcs->get_xclk((adev))
@@ -1235,6 +1277,10 @@ int emu_soc_asic_init(struct amdgpu_device *adev);
#define amdgpu_inc_vram_lost(adev) atomic_inc(&((adev)->vram_lost_counter));
#define for_each_inst(i, inst_mask) \
for (i = ffs(inst_mask) - 1; inst_mask; \
inst_mask &= ~(1U << i), i = ffs(inst_mask) - 1)
#define MIN(X, Y) ((X) < (Y) ? (X) : (Y))
/* Common functions */
@@ -1348,6 +1394,12 @@ struct amdgpu_afmt_acr amdgpu_afmt_acr(uint32_t clock);
/* amdgpu_acpi.c */
struct amdgpu_numa_info {
uint64_t size;
int pxm;
int nid;
};
/* ATCS Device/Driver State */
#define AMDGPU_ATCS_PSC_DEV_STATE_D0 0
#define AMDGPU_ATCS_PSC_DEV_STATE_D3_HOT 3
@@ -1365,15 +1417,32 @@ int amdgpu_acpi_power_shift_control(struct amdgpu_device *adev,
u8 dev_state, bool drv_state);
int amdgpu_acpi_smart_shift_update(struct drm_device *dev, enum amdgpu_ss ss_state);
int amdgpu_acpi_pcie_notify_device_ready(struct amdgpu_device *adev);
int amdgpu_acpi_get_tmr_info(struct amdgpu_device *adev, u64 *tmr_offset,
u64 *tmr_size);
int amdgpu_acpi_get_mem_info(struct amdgpu_device *adev, int xcc_id,
struct amdgpu_numa_info *numa_info);
void amdgpu_acpi_get_backlight_caps(struct amdgpu_dm_backlight_caps *caps);
bool amdgpu_acpi_should_gpu_reset(struct amdgpu_device *adev);
void amdgpu_acpi_detect(void);
void amdgpu_acpi_release(void);
#else
static inline int amdgpu_acpi_init(struct amdgpu_device *adev) { return 0; }
static inline int amdgpu_acpi_get_tmr_info(struct amdgpu_device *adev,
u64 *tmr_offset, u64 *tmr_size)
{
return -EINVAL;
}
static inline int amdgpu_acpi_get_mem_info(struct amdgpu_device *adev,
int xcc_id,
struct amdgpu_numa_info *numa_info)
{
return -EINVAL;
}
static inline void amdgpu_acpi_fini(struct amdgpu_device *adev) { }
static inline bool amdgpu_acpi_should_gpu_reset(struct amdgpu_device *adev) { return false; }
static inline void amdgpu_acpi_detect(void) { }
static inline void amdgpu_acpi_release(void) { }
static inline bool amdgpu_acpi_is_power_shift_control_supported(void) { return false; }
static inline int amdgpu_acpi_power_shift_control(struct amdgpu_device *adev,
u8 dev_state, bool drv_state) { return 0; }
drivers/gpu/drm/amd/amdgpu/amdgpu_acpi.c
+415 -4
View File
@@ -26,6 +26,7 @@
#include <linux/acpi.h>
#include <linux/backlight.h>
#include <linux/slab.h>
#include <linux/xarray.h>
#include <linux/power_supply.h>
#include <linux/pm_runtime.h>
#include <linux/suspend.h>
@@ -38,6 +39,45 @@
#include "amd_acpi.h"
#include "atom.h"
/* Declare GUID for AMD _DSM method for XCCs */
static const guid_t amd_xcc_dsm_guid = GUID_INIT(0x8267f5d5, 0xa556, 0x44f2,
0xb8, 0xb4, 0x45, 0x56, 0x2e,
0x8c, 0x5b, 0xec);
#define AMD_XCC_HID_START 3000
#define AMD_XCC_DSM_GET_NUM_FUNCS 0
#define AMD_XCC_DSM_GET_SUPP_MODE 1
#define AMD_XCC_DSM_GET_XCP_MODE 2
#define AMD_XCC_DSM_GET_VF_XCC_MAPPING 4
#define AMD_XCC_DSM_GET_TMR_INFO 5
#define AMD_XCC_DSM_NUM_FUNCS 5
#define AMD_XCC_MAX_HID 24
struct xarray numa_info_xa;
/* Encapsulates the XCD acpi object information */
struct amdgpu_acpi_xcc_info {
struct list_head list;
struct amdgpu_numa_info *numa_info;
uint8_t xcp_node;
uint8_t phy_id;
acpi_handle handle;
};
struct amdgpu_acpi_dev_info {
struct list_head list;
struct list_head xcc_list;
uint16_t bdf;
uint16_t supp_xcp_mode;
uint16_t xcp_mode;
uint16_t mem_mode;
uint64_t tmr_base;
uint64_t tmr_size;
};
struct list_head amdgpu_acpi_dev_list;
struct amdgpu_atif_notification_cfg {
bool enabled;
int command_code;
@@ -801,6 +841,343 @@ int amdgpu_acpi_smart_shift_update(struct drm_device *dev, enum amdgpu_ss ss_sta
return r;
}
#ifdef CONFIG_ACPI_NUMA
static inline uint64_t amdgpu_acpi_get_numa_size(int nid)
{
/* This is directly using si_meminfo_node implementation as the
* function is not exported.
*/
int zone_type;
uint64_t managed_pages = 0;
pg_data_t *pgdat = NODE_DATA(nid);
for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
managed_pages +=
zone_managed_pages(&pgdat->node_zones[zone_type]);
return managed_pages * PAGE_SIZE;
}
static struct amdgpu_numa_info *amdgpu_acpi_get_numa_info(uint32_t pxm)
{
struct amdgpu_numa_info *numa_info;
int nid;
numa_info = xa_load(&numa_info_xa, pxm);
if (!numa_info) {
struct sysinfo info;
numa_info = kzalloc(sizeof *numa_info, GFP_KERNEL);
if (!numa_info)
return NULL;
nid = pxm_to_node(pxm);
numa_info->pxm = pxm;
numa_info->nid = nid;
if (numa_info->nid == NUMA_NO_NODE) {
si_meminfo(&info);
numa_info->size = info.totalram * info.mem_unit;
} else {
numa_info->size = amdgpu_acpi_get_numa_size(nid);
}
xa_store(&numa_info_xa, numa_info->pxm, numa_info, GFP_KERNEL);
}
return numa_info;
}
#endif
/**
* amdgpu_acpi_get_node_id - obtain the NUMA node id for corresponding amdgpu
* acpi device handle
*
* @handle: acpi handle
* @numa_info: amdgpu_numa_info structure holding numa information
*
* Queries the ACPI interface to fetch the corresponding NUMA Node ID for a
* given amdgpu acpi device.
*
* Returns ACPI STATUS OK with Node ID on success or the corresponding failure reason
*/
static acpi_status amdgpu_acpi_get_node_id(acpi_handle handle,
struct amdgpu_numa_info **numa_info)
{
#ifdef CONFIG_ACPI_NUMA
u64 pxm;
acpi_status status;
if (!numa_info)
return_ACPI_STATUS(AE_ERROR);
status = acpi_evaluate_integer(handle, "_PXM", NULL, &pxm);
if (ACPI_FAILURE(status))
return status;
*numa_info = amdgpu_acpi_get_numa_info(pxm);
if (!*numa_info)
return_ACPI_STATUS(AE_ERROR);
return_ACPI_STATUS(AE_OK);
#else
return_ACPI_STATUS(AE_NOT_EXIST);
#endif
}
static struct amdgpu_acpi_dev_info *amdgpu_acpi_get_dev(u16 bdf)
{
struct amdgpu_acpi_dev_info *acpi_dev;
if (list_empty(&amdgpu_acpi_dev_list))
return NULL;
list_for_each_entry(acpi_dev, &amdgpu_acpi_dev_list, list)
if (acpi_dev->bdf == bdf)
return acpi_dev;
return NULL;
}
static int amdgpu_acpi_dev_init(struct amdgpu_acpi_dev_info **dev_info,
struct amdgpu_acpi_xcc_info *xcc_info, u16 bdf)
{
struct amdgpu_acpi_dev_info *tmp;
union acpi_object *obj;
int ret = -ENOENT;
*dev_info = NULL;
tmp = kzalloc(sizeof(struct amdgpu_acpi_dev_info), GFP_KERNEL);
if (!tmp)
return -ENOMEM;
INIT_LIST_HEAD(&tmp->xcc_list);
INIT_LIST_HEAD(&tmp->list);
tmp->bdf = bdf;
obj = acpi_evaluate_dsm_typed(xcc_info->handle, &amd_xcc_dsm_guid, 0,
AMD_XCC_DSM_GET_SUPP_MODE, NULL,
ACPI_TYPE_INTEGER);
if (!obj) {
acpi_handle_debug(xcc_info->handle,
"_DSM function %d evaluation failed",
AMD_XCC_DSM_GET_SUPP_MODE);
ret = -ENOENT;
goto out;
}
tmp->supp_xcp_mode = obj->integer.value & 0xFFFF;
ACPI_FREE(obj);
obj = acpi_evaluate_dsm_typed(xcc_info->handle, &amd_xcc_dsm_guid, 0,
AMD_XCC_DSM_GET_XCP_MODE, NULL,
ACPI_TYPE_INTEGER);
if (!obj) {
acpi_handle_debug(xcc_info->handle,
"_DSM function %d evaluation failed",
AMD_XCC_DSM_GET_XCP_MODE);
ret = -ENOENT;
goto out;
}
tmp->xcp_mode = obj->integer.value & 0xFFFF;
tmp->mem_mode = (obj->integer.value >> 32) & 0xFFFF;
ACPI_FREE(obj);
/* Evaluate DSMs and fill XCC information */
obj = acpi_evaluate_dsm_typed(xcc_info->handle, &amd_xcc_dsm_guid, 0,
AMD_XCC_DSM_GET_TMR_INFO, NULL,
ACPI_TYPE_PACKAGE);
if (!obj || obj->package.count < 2) {
acpi_handle_debug(xcc_info->handle,
"_DSM function %d evaluation failed",
AMD_XCC_DSM_GET_TMR_INFO);
ret = -ENOENT;
goto out;
}
tmp->tmr_base = obj->package.elements[0].integer.value;
tmp->tmr_size = obj->package.elements[1].integer.value;
ACPI_FREE(obj);
DRM_DEBUG_DRIVER(
"New dev(%x): Supported xcp mode: %x curr xcp_mode : %x mem mode : %x, tmr base: %llx tmr size: %llx ",
tmp->bdf, tmp->supp_xcp_mode, tmp->xcp_mode, tmp->mem_mode,
tmp->tmr_base, tmp->tmr_size);
list_add_tail(&tmp->list, &amdgpu_acpi_dev_list);
*dev_info = tmp;
return 0;
out:
if (obj)
ACPI_FREE(obj);
kfree(tmp);
return ret;
}
static int amdgpu_acpi_get_xcc_info(struct amdgpu_acpi_xcc_info *xcc_info,
u16 *bdf)
{
union acpi_object *obj;
acpi_status status;
int ret = -ENOENT;
obj = acpi_evaluate_dsm_typed(xcc_info->handle, &amd_xcc_dsm_guid, 0,
AMD_XCC_DSM_GET_NUM_FUNCS, NULL,
ACPI_TYPE_INTEGER);
if (!obj || obj->integer.value != AMD_XCC_DSM_NUM_FUNCS)
goto out;
ACPI_FREE(obj);
/* Evaluate DSMs and fill XCC information */
obj = acpi_evaluate_dsm_typed(xcc_info->handle, &amd_xcc_dsm_guid, 0,
AMD_XCC_DSM_GET_VF_XCC_MAPPING, NULL,
ACPI_TYPE_INTEGER);
if (!obj) {
acpi_handle_debug(xcc_info->handle,
"_DSM function %d evaluation failed",
AMD_XCC_DSM_GET_VF_XCC_MAPPING);
ret = -EINVAL;
goto out;
}
/* PF xcc id [39:32] */
xcc_info->phy_id = (obj->integer.value >> 32) & 0xFF;
/* xcp node of this xcc [47:40] */
xcc_info->xcp_node = (obj->integer.value >> 40) & 0xFF;
/* PF bus/dev/fn of this xcc [63:48] */
*bdf = (obj->integer.value >> 48) & 0xFFFF;
ACPI_FREE(obj);
obj = NULL;
status =
amdgpu_acpi_get_node_id(xcc_info->handle, &xcc_info->numa_info);
/* TODO: check if this check is required */
if (ACPI_SUCCESS(status))
ret = 0;
out:
if (obj)
ACPI_FREE(obj);
return ret;
}
static int amdgpu_acpi_enumerate_xcc(void)
{
struct amdgpu_acpi_dev_info *dev_info = NULL;
struct amdgpu_acpi_xcc_info *xcc_info;
struct acpi_device *acpi_dev;
char hid[ACPI_ID_LEN];
int ret, id;
u16 bdf;
INIT_LIST_HEAD(&amdgpu_acpi_dev_list);
xa_init(&numa_info_xa);
for (id = 0; id < AMD_XCC_MAX_HID; id++) {
sprintf(hid, "%s%d", "AMD", AMD_XCC_HID_START + id);
acpi_dev = acpi_dev_get_first_match_dev(hid, NULL, -1);
/* These ACPI objects are expected to be in sequential order. If
* one is not found, no need to check the rest.
*/
if (!acpi_dev) {
DRM_DEBUG_DRIVER("No matching acpi device found for %s",
hid);
break;
}
xcc_info = kzalloc(sizeof(struct amdgpu_acpi_xcc_info),
GFP_KERNEL);
if (!xcc_info) {
DRM_ERROR("Failed to allocate memory for xcc info\n");
return -ENOMEM;
}
INIT_LIST_HEAD(&xcc_info->list);
xcc_info->handle = acpi_device_handle(acpi_dev);
acpi_dev_put(acpi_dev);
ret = amdgpu_acpi_get_xcc_info(xcc_info, &bdf);
if (ret) {
kfree(xcc_info);
continue;
}
dev_info = amdgpu_acpi_get_dev(bdf);
if (!dev_info)
ret = amdgpu_acpi_dev_init(&dev_info, xcc_info, bdf);
if (ret == -ENOMEM)
return ret;
if (!dev_info) {
kfree(xcc_info);
continue;
}
list_add_tail(&xcc_info->list, &dev_info->xcc_list);
}
return 0;
}
int amdgpu_acpi_get_tmr_info(struct amdgpu_device *adev, u64 *tmr_offset,
u64 *tmr_size)
{
struct amdgpu_acpi_dev_info *dev_info;
u16 bdf;
if (!tmr_offset || !tmr_size)
return -EINVAL;
bdf = (adev->pdev->bus->number << 8) | adev->pdev->devfn;
dev_info = amdgpu_acpi_get_dev(bdf);
if (!dev_info)
return -ENOENT;
*tmr_offset = dev_info->tmr_base;
*tmr_size = dev_info->tmr_size;
return 0;
}
int amdgpu_acpi_get_mem_info(struct amdgpu_device *adev, int xcc_id,
struct amdgpu_numa_info *numa_info)
{
struct amdgpu_acpi_dev_info *dev_info;
struct amdgpu_acpi_xcc_info *xcc_info;
u16 bdf;
if (!numa_info)
return -EINVAL;
bdf = (adev->pdev->bus->number << 8) | adev->pdev->devfn;
dev_info = amdgpu_acpi_get_dev(bdf);
if (!dev_info)
return -ENOENT;
list_for_each_entry(xcc_info, &dev_info->xcc_list, list) {
if (xcc_info->phy_id == xcc_id) {
memcpy(numa_info, xcc_info->numa_info,
sizeof(*numa_info));
return 0;
}
}
return -ENOENT;
}
/**
* amdgpu_acpi_event - handle notify events
*
@@ -1054,6 +1431,36 @@ void amdgpu_acpi_detect(void)
} else {
atif->backlight_caps.caps_valid = false;
}
amdgpu_acpi_enumerate_xcc();
}
void amdgpu_acpi_release(void)
{
struct amdgpu_acpi_dev_info *dev_info, *dev_tmp;
struct amdgpu_acpi_xcc_info *xcc_info, *xcc_tmp;
struct amdgpu_numa_info *numa_info;
unsigned long index;
xa_for_each(&numa_info_xa, index, numa_info) {
kfree(numa_info);
xa_erase(&numa_info_xa, index);
}
if (list_empty(&amdgpu_acpi_dev_list))
return;
list_for_each_entry_safe(dev_info, dev_tmp, &amdgpu_acpi_dev_list,
list) {
list_for_each_entry_safe(xcc_info, xcc_tmp, &dev_info->xcc_list,
list) {
list_del(&xcc_info->list);
kfree(xcc_info);
}
list_del(&dev_info->list);
kfree(dev_info);
}
}
#if IS_ENABLED(CONFIG_SUSPEND)
@@ -1092,16 +1499,20 @@ bool amdgpu_acpi_is_s0ix_active(struct amdgpu_device *adev)
* S0ix even though the system is suspending to idle, so return false
* in that case.
*/
if (!(acpi_gbl_FADT.flags & ACPI_FADT_LOW_POWER_S0))
dev_warn_once(adev->dev,
if (!(acpi_gbl_FADT.flags & ACPI_FADT_LOW_POWER_S0)) {
dev_err_once(adev->dev,
"Power consumption will be higher as BIOS has not been configured for suspend-to-idle.\n"
"To use suspend-to-idle change the sleep mode in BIOS setup.\n");
return false;
}
#if !IS_ENABLED(CONFIG_AMD_PMC)
dev_warn_once(adev->dev,
dev_err_once(adev->dev,
"Power consumption will be higher as the kernel has not been compiled with CONFIG_AMD_PMC.\n");
#endif /* CONFIG_AMD_PMC */
return false;
#else
return true;
#endif /* CONFIG_AMD_PMC */
}
#endif /* CONFIG_SUSPEND */
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
+71 -11
View File
@@ -53,7 +53,6 @@ int amdgpu_amdkfd_init(void)
amdgpu_amdkfd_total_mem_size *= si.mem_unit;
ret = kgd2kfd_init();
amdgpu_amdkfd_gpuvm_init_mem_limits();
kfd_initialized = !ret;
return ret;
@@ -143,6 +142,8 @@ void amdgpu_amdkfd_device_init(struct amdgpu_device *adev)
int i;
int last_valid_bit;
amdgpu_amdkfd_gpuvm_init_mem_limits();
if (adev->kfd.dev) {
struct kgd2kfd_shared_resources gpu_resources = {
.compute_vmid_bitmap =
@@ -162,7 +163,7 @@ void amdgpu_amdkfd_device_init(struct amdgpu_device *adev)
* clear
*/
bitmap_complement(gpu_resources.cp_queue_bitmap,
adev->gfx.mec.queue_bitmap,
adev->gfx.mec_bitmap[0].queue_bitmap,
KGD_MAX_QUEUES);
/* According to linux/bitmap.h we shouldn't use bitmap_clear if
@@ -427,14 +428,23 @@ uint32_t amdgpu_amdkfd_get_fw_version(struct amdgpu_device *adev,
}
void amdgpu_amdkfd_get_local_mem_info(struct amdgpu_device *adev,
struct kfd_local_mem_info *mem_info)
struct kfd_local_mem_info *mem_info,
struct amdgpu_xcp *xcp)
{
memset(mem_info, 0, sizeof(*mem_info));
mem_info->local_mem_size_public = adev->gmc.visible_vram_size;
mem_info->local_mem_size_private = adev->gmc.real_vram_size -
if (xcp) {
if (adev->gmc.real_vram_size == adev->gmc.visible_vram_size)
mem_info->local_mem_size_public =
KFD_XCP_MEMORY_SIZE(adev, xcp->id);
else
mem_info->local_mem_size_private =
KFD_XCP_MEMORY_SIZE(adev, xcp->id);
} else {
mem_info->local_mem_size_public = adev->gmc.visible_vram_size;
mem_info->local_mem_size_private = adev->gmc.real_vram_size -
adev->gmc.visible_vram_size;
}
mem_info->vram_width = adev->gmc.vram_width;
pr_debug("Address base: %pap public 0x%llx private 0x%llx\n",
@@ -497,7 +507,7 @@ int amdgpu_amdkfd_get_dmabuf_info(struct amdgpu_device *adev, int dma_buf_fd,
struct amdgpu_device **dmabuf_adev,
uint64_t *bo_size, void *metadata_buffer,
size_t buffer_size, uint32_t *metadata_size,
uint32_t *flags)
uint32_t *flags, int8_t *xcp_id)
{
struct dma_buf *dma_buf;
struct drm_gem_object *obj;
@@ -541,6 +551,8 @@ int amdgpu_amdkfd_get_dmabuf_info(struct amdgpu_device *adev, int dma_buf_fd,
if (bo->flags & AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED)
*flags |= KFD_IOC_ALLOC_MEM_FLAGS_PUBLIC;
}
if (xcp_id)
*xcp_id = bo->xcp_id;
out_put:
dma_buf_put(dma_buf);
@@ -732,17 +744,19 @@ int amdgpu_amdkfd_flush_gpu_tlb_vmid(struct amdgpu_device *adev,
if (adev->family == AMDGPU_FAMILY_AI) {
int i;
for (i = 0; i < adev->num_vmhubs; i++)
for_each_set_bit(i, adev->vmhubs_mask, AMDGPU_MAX_VMHUBS)
amdgpu_gmc_flush_gpu_tlb(adev, vmid, i, 0);
} else {
amdgpu_gmc_flush_gpu_tlb(adev, vmid, AMDGPU_GFXHUB_0, 0);
amdgpu_gmc_flush_gpu_tlb(adev, vmid, AMDGPU_GFXHUB(0), 0);
}
return 0;
}
int amdgpu_amdkfd_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
uint16_t pasid, enum TLB_FLUSH_TYPE flush_type)
uint16_t pasid,
enum TLB_FLUSH_TYPE flush_type,
uint32_t inst)
{
bool all_hub = false;
@@ -750,7 +764,7 @@ int amdgpu_amdkfd_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
adev->family == AMDGPU_FAMILY_RV)
all_hub = true;
return amdgpu_gmc_flush_gpu_tlb_pasid(adev, pasid, flush_type, all_hub);
return amdgpu_gmc_flush_gpu_tlb_pasid(adev, pasid, flush_type, all_hub, inst);
}
bool amdgpu_amdkfd_have_atomics_support(struct amdgpu_device *adev)
@@ -758,11 +772,32 @@ bool amdgpu_amdkfd_have_atomics_support(struct amdgpu_device *adev)
return adev->have_atomics_support;
}
void amdgpu_amdkfd_debug_mem_fence(struct amdgpu_device *adev)
{
amdgpu_device_flush_hdp(adev, NULL);
}
void amdgpu_amdkfd_ras_poison_consumption_handler(struct amdgpu_device *adev, bool reset)
{
amdgpu_umc_poison_handler(adev, reset);
}
int amdgpu_amdkfd_send_close_event_drain_irq(struct amdgpu_device *adev,
uint32_t *payload)
{
int ret;
/* Device or IH ring is not ready so bail. */
ret = amdgpu_ih_wait_on_checkpoint_process_ts(adev, &adev->irq.ih);
if (ret)
return ret;
/* Send payload to fence KFD interrupts */
amdgpu_amdkfd_interrupt(adev, payload);
return 0;
}
bool amdgpu_amdkfd_ras_query_utcl2_poison_status(struct amdgpu_device *adev)
{
if (adev->gfx.ras && adev->gfx.ras->query_utcl2_poison_status)
@@ -770,3 +805,28 @@ bool amdgpu_amdkfd_ras_query_utcl2_poison_status(struct amdgpu_device *adev)
else
return false;
}
int amdgpu_amdkfd_check_and_lock_kfd(struct amdgpu_device *adev)
{
return kgd2kfd_check_and_lock_kfd();
}
void amdgpu_amdkfd_unlock_kfd(struct amdgpu_device *adev)
{
kgd2kfd_unlock_kfd();
}
u64 amdgpu_amdkfd_xcp_memory_size(struct amdgpu_device *adev, int xcp_id)
{
u64 tmp;
s8 mem_id = KFD_XCP_MEM_ID(adev, xcp_id);
if (adev->gmc.num_mem_partitions && xcp_id >= 0 && mem_id >= 0) {
tmp = adev->gmc.mem_partitions[mem_id].size;
do_div(tmp, adev->xcp_mgr->num_xcp_per_mem_partition);
return ALIGN_DOWN(tmp, PAGE_SIZE);
} else {
return adev->gmc.real_vram_size;
}
}
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
+52 -8
View File
@@ -30,10 +30,12 @@
#include <linux/kthread.h>
#include <linux/workqueue.h>
#include <linux/mmu_notifier.h>
#include <linux/memremap.h>
#include <kgd_kfd_interface.h>
#include <drm/ttm/ttm_execbuf_util.h>
#include "amdgpu_sync.h"
#include "amdgpu_vm.h"
#include "amdgpu_xcp.h"
extern uint64_t amdgpu_amdkfd_total_mem_size;
@@ -97,10 +99,13 @@ struct amdgpu_amdkfd_fence {
struct amdgpu_kfd_dev {
struct kfd_dev *dev;
int64_t vram_used;
uint64_t vram_used_aligned;
int64_t vram_used[MAX_XCP];
uint64_t vram_used_aligned[MAX_XCP];
bool init_complete;
struct work_struct reset_work;
/* HMM page migration MEMORY_DEVICE_PRIVATE mapping */
struct dev_pagemap pgmap;
};
enum kgd_engine_type {
@@ -151,6 +156,8 @@ void amdgpu_amdkfd_interrupt(struct amdgpu_device *adev,
void amdgpu_amdkfd_device_probe(struct amdgpu_device *adev);
void amdgpu_amdkfd_device_init(struct amdgpu_device *adev);
void amdgpu_amdkfd_device_fini_sw(struct amdgpu_device *adev);
int amdgpu_amdkfd_check_and_lock_kfd(struct amdgpu_device *adev);
void amdgpu_amdkfd_unlock_kfd(struct amdgpu_device *adev);
int amdgpu_amdkfd_submit_ib(struct amdgpu_device *adev,
enum kgd_engine_type engine,
uint32_t vmid, uint64_t gpu_addr,
@@ -160,7 +167,8 @@ bool amdgpu_amdkfd_have_atomics_support(struct amdgpu_device *adev);
int amdgpu_amdkfd_flush_gpu_tlb_vmid(struct amdgpu_device *adev,
uint16_t vmid);
int amdgpu_amdkfd_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
uint16_t pasid, enum TLB_FLUSH_TYPE flush_type);
uint16_t pasid, enum TLB_FLUSH_TYPE flush_type,
uint32_t inst);
bool amdgpu_amdkfd_is_kfd_vmid(struct amdgpu_device *adev, u32 vmid);
@@ -224,7 +232,8 @@ int amdgpu_amdkfd_remove_gws_from_process(void *info, void *mem);
uint32_t amdgpu_amdkfd_get_fw_version(struct amdgpu_device *adev,
enum kgd_engine_type type);
void amdgpu_amdkfd_get_local_mem_info(struct amdgpu_device *adev,
struct kfd_local_mem_info *mem_info);
struct kfd_local_mem_info *mem_info,
struct amdgpu_xcp *xcp);
uint64_t amdgpu_amdkfd_get_gpu_clock_counter(struct amdgpu_device *adev);
uint32_t amdgpu_amdkfd_get_max_engine_clock_in_mhz(struct amdgpu_device *adev);
@@ -234,13 +243,15 @@ int amdgpu_amdkfd_get_dmabuf_info(struct amdgpu_device *adev, int dma_buf_fd,
struct amdgpu_device **dmabuf_adev,
uint64_t *bo_size, void *metadata_buffer,
size_t buffer_size, uint32_t *metadata_size,
uint32_t *flags);
uint32_t *flags, int8_t *xcp_id);
uint8_t amdgpu_amdkfd_get_xgmi_hops_count(struct amdgpu_device *dst,
struct amdgpu_device *src);
int amdgpu_amdkfd_get_xgmi_bandwidth_mbytes(struct amdgpu_device *dst,
struct amdgpu_device *src,
bool is_min);
int amdgpu_amdkfd_get_pcie_bandwidth_mbytes(struct amdgpu_device *adev, bool is_min);
int amdgpu_amdkfd_send_close_event_drain_irq(struct amdgpu_device *adev,
uint32_t *payload);
/* Read user wptr from a specified user address space with page fault
* disabled. The memory must be pinned and mapped to the hardware when
@@ -279,7 +290,8 @@ int amdgpu_amdkfd_gpuvm_acquire_process_vm(struct amdgpu_device *adev,
void amdgpu_amdkfd_gpuvm_release_process_vm(struct amdgpu_device *adev,
void *drm_priv);
uint64_t amdgpu_amdkfd_gpuvm_get_process_page_dir(void *drm_priv);
size_t amdgpu_amdkfd_get_available_memory(struct amdgpu_device *adev);
size_t amdgpu_amdkfd_get_available_memory(struct amdgpu_device *adev,
uint8_t xcp_id);
int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
struct amdgpu_device *adev, uint64_t va, uint64_t size,
void *drm_priv, struct kgd_mem **mem,
@@ -310,6 +322,7 @@ int amdgpu_amdkfd_gpuvm_import_dmabuf(struct amdgpu_device *adev,
uint64_t *mmap_offset);
int amdgpu_amdkfd_gpuvm_export_dmabuf(struct kgd_mem *mem,
struct dma_buf **dmabuf);
void amdgpu_amdkfd_debug_mem_fence(struct amdgpu_device *adev);
int amdgpu_amdkfd_get_tile_config(struct amdgpu_device *adev,
struct tile_config *config);
void amdgpu_amdkfd_ras_poison_consumption_handler(struct amdgpu_device *adev,
@@ -319,9 +332,18 @@ void amdgpu_amdkfd_block_mmu_notifications(void *p);
int amdgpu_amdkfd_criu_resume(void *p);
bool amdgpu_amdkfd_ras_query_utcl2_poison_status(struct amdgpu_device *adev);
int amdgpu_amdkfd_reserve_mem_limit(struct amdgpu_device *adev,
uint64_t size, u32 alloc_flag);
uint64_t size, u32 alloc_flag, int8_t xcp_id);
void amdgpu_amdkfd_unreserve_mem_limit(struct amdgpu_device *adev,
uint64_t size, u32 alloc_flag);
uint64_t size, u32 alloc_flag, int8_t xcp_id);
u64 amdgpu_amdkfd_xcp_memory_size(struct amdgpu_device *adev, int xcp_id);
#define KFD_XCP_MEM_ID(adev, xcp_id) \
((adev)->xcp_mgr && (xcp_id) >= 0 ?\
(adev)->xcp_mgr->xcp[(xcp_id)].mem_id : -1)
#define KFD_XCP_MEMORY_SIZE(adev, xcp_id) amdgpu_amdkfd_xcp_memory_size((adev), (xcp_id))
#if IS_ENABLED(CONFIG_HSA_AMD)
void amdgpu_amdkfd_gpuvm_init_mem_limits(void);
@@ -352,6 +374,17 @@ void amdgpu_amdkfd_release_notify(struct amdgpu_bo *bo)
{
}
#endif
#if IS_ENABLED(CONFIG_HSA_AMD_SVM)
int kgd2kfd_init_zone_device(struct amdgpu_device *adev);
#else
static inline
int kgd2kfd_init_zone_device(struct amdgpu_device *adev)
{
return 0;
}
#endif
/* KGD2KFD callbacks */
int kgd2kfd_quiesce_mm(struct mm_struct *mm, uint32_t trigger);
int kgd2kfd_resume_mm(struct mm_struct *mm);
@@ -372,6 +405,8 @@ int kgd2kfd_post_reset(struct kfd_dev *kfd);
void kgd2kfd_interrupt(struct kfd_dev *kfd, const void *ih_ring_entry);
void kgd2kfd_set_sram_ecc_flag(struct kfd_dev *kfd);
void kgd2kfd_smi_event_throttle(struct kfd_dev *kfd, uint64_t throttle_bitmask);
int kgd2kfd_check_and_lock_kfd(void);
void kgd2kfd_unlock_kfd(void);
#else
static inline int kgd2kfd_init(void)
{
@@ -437,5 +472,14 @@ static inline
void kgd2kfd_smi_event_throttle(struct kfd_dev *kfd, uint64_t throttle_bitmask)
{
}
static inline int kgd2kfd_check_and_lock_kfd(void)
{
return 0;
}
static inline void kgd2kfd_unlock_kfd(void)
{
}
#endif
#endif /* AMDGPU_AMDKFD_H_INCLUDED */
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c
+153 -1
View File
@@ -23,6 +23,149 @@
#include "amdgpu_amdkfd.h"
#include "amdgpu_amdkfd_arcturus.h"
#include "amdgpu_amdkfd_gfx_v9.h"
#include "gc/gc_9_4_2_offset.h"
#include "gc/gc_9_4_2_sh_mask.h"
#include <uapi/linux/kfd_ioctl.h>
/*
* Returns TRAP_EN, EXCP_EN and EXCP_REPLACE.
*
* restore_dbg_registers is ignored here but is a general interface requirement
* for devices that support GFXOFF and where the RLC save/restore list
* does not support hw registers for debugging i.e. the driver has to manually
* initialize the debug mode registers after it has disabled GFX off during the
* debug session.
*/
static uint32_t kgd_aldebaran_enable_debug_trap(struct amdgpu_device *adev,
bool restore_dbg_registers,
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;
}
/* returns TRAP_EN, EXCP_EN and EXCP_REPLACE. */
static uint32_t kgd_aldebaran_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, keep_trap_enabled);
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_aldebaran_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;
if (trap_override != KFD_DBG_TRAP_OVERRIDE_OR &&
trap_override != KFD_DBG_TRAP_OVERRIDE_REPLACE)
return -EPERM;
return 0;
}
/* returns TRAP_EN, EXCP_EN and EXCP_RPLACE. */
static uint32_t kgd_aldebaran_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 = REG_GET_FIELD(kfd_dbg_trap_cntl_prev, SPI_GDBG_PER_VMID_CNTL, EXCP_EN);
trap_mask_bits = (trap_mask_bits & trap_mask_request) |
(*trap_mask_prev & ~trap_mask_request);
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, trap_mask_bits);
data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, EXCP_REPLACE, trap_override);
return data;
}
static uint32_t kgd_aldebaran_set_wave_launch_mode(struct amdgpu_device *adev,
uint8_t wave_launch_mode,
uint32_t vmid)
{
uint32_t data = 0;
data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, LAUNCH_MODE, wave_launch_mode);
return data;
}
#define TCP_WATCH_STRIDE (regTCP_WATCH1_ADDR_H - regTCP_WATCH0_ADDR_H)
static uint32_t kgd_gfx_aldebaran_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 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 >> 6);
watch_address_cntl = REG_SET_FIELD(watch_address_cntl,
TCP_WATCH0_CNTL,
VALID,
1);
WREG32_RLC((SOC15_REG_OFFSET(GC, 0, regTCP_WATCH0_ADDR_H) +
(watch_id * TCP_WATCH_STRIDE)),
watch_address_high);
WREG32_RLC((SOC15_REG_OFFSET(GC, 0, regTCP_WATCH0_ADDR_L) +
(watch_id * TCP_WATCH_STRIDE)),
watch_address_low);
return watch_address_cntl;
}
static uint32_t kgd_gfx_aldebaran_clear_address_watch(struct amdgpu_device *adev,
uint32_t watch_id)
{
return 0;
}
const struct kfd2kgd_calls aldebaran_kfd2kgd = {
.program_sh_mem_settings = kgd_gfx_v9_program_sh_mem_settings,
@@ -42,5 +185,14 @@ const struct kfd2kgd_calls aldebaran_kfd2kgd = {
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
.enable_debug_trap = kgd_aldebaran_enable_debug_trap,
.disable_debug_trap = kgd_aldebaran_disable_debug_trap,
.validate_trap_override_request = kgd_aldebaran_validate_trap_override_request,
.set_wave_launch_trap_override = kgd_aldebaran_set_wave_launch_trap_override,
.set_wave_launch_mode = kgd_aldebaran_set_wave_launch_mode,
.set_address_watch = kgd_gfx_aldebaran_set_address_watch,
.clear_address_watch = kgd_gfx_aldebaran_clear_address_watch,
.get_iq_wait_times = kgd_gfx_v9_get_iq_wait_times,
.build_grace_period_packet_info = kgd_gfx_v9_build_grace_period_packet_info,
.program_trap_handler_settings = kgd_gfx_v9_program_trap_handler_settings,
};
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c
+123
View File
@@ -26,6 +26,7 @@
#include "amdgpu.h"
#include "amdgpu_amdkfd.h"
#include "amdgpu_amdkfd_arcturus.h"
#include "amdgpu_reset.h"
#include "sdma0/sdma0_4_2_2_offset.h"
#include "sdma0/sdma0_4_2_2_sh_mask.h"
#include "sdma1/sdma1_4_2_2_offset.h"
@@ -48,6 +49,8 @@
#include "amdgpu_amdkfd_gfx_v9.h"
#include "gfxhub_v1_0.h"
#include "mmhub_v9_4.h"
#include "gc/gc_9_0_offset.h"
#include "gc/gc_9_0_sh_mask.h"
#define HQD_N_REGS 56
#define DUMP_REG(addr) do { \
@@ -276,6 +279,117 @@ int kgd_arcturus_hqd_sdma_destroy(struct amdgpu_device *adev, void *mqd,
return 0;
}
/*
* Helper used to suspend/resume gfx pipe for image post process work to set
* barrier behaviour.
*/
static int suspend_resume_compute_scheduler(struct amdgpu_device *adev, bool suspend)
{
int i, r = 0;
for (i = 0; i < adev->gfx.num_compute_rings; i++) {
struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
if (!(ring && ring->sched.thread))
continue;
/* stop secheduler and drain ring. */
if (suspend) {
drm_sched_stop(&ring->sched, NULL);
r = amdgpu_fence_wait_empty(ring);
if (r)
goto out;
} else {
drm_sched_start(&ring->sched, false);
}
}
out:
/* return on resume or failure to drain rings. */
if (!suspend || r)
return r;
return amdgpu_device_ip_wait_for_idle(adev, AMD_IP_BLOCK_TYPE_GFX);
}
static void set_barrier_auto_waitcnt(struct amdgpu_device *adev, bool enable_waitcnt)
{
uint32_t data;
WRITE_ONCE(adev->barrier_has_auto_waitcnt, enable_waitcnt);
if (!down_read_trylock(&adev->reset_domain->sem))
return;
amdgpu_amdkfd_suspend(adev, false);
if (suspend_resume_compute_scheduler(adev, true))
goto out;
data = RREG32(SOC15_REG_OFFSET(GC, 0, mmSQ_CONFIG));
data = REG_SET_FIELD(data, SQ_CONFIG, DISABLE_BARRIER_WAITCNT,
!enable_waitcnt);
WREG32(SOC15_REG_OFFSET(GC, 0, mmSQ_CONFIG), data);
out:
suspend_resume_compute_scheduler(adev, false);
amdgpu_amdkfd_resume(adev, false);
up_read(&adev->reset_domain->sem);
}
/*
* restore_dbg_registers is ignored here but is a general interface requirement
* for devices that support GFXOFF and where the RLC save/restore list
* does not support hw registers for debugging i.e. the driver has to manually
* initialize the debug mode registers after it has disabled GFX off during the
* debug session.
*/
static uint32_t kgd_arcturus_enable_debug_trap(struct amdgpu_device *adev,
bool restore_dbg_registers,
uint32_t vmid)
{
mutex_lock(&adev->grbm_idx_mutex);
kgd_gfx_v9_set_wave_launch_stall(adev, vmid, true);
set_barrier_auto_waitcnt(adev, true);
WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_MASK), 0);
kgd_gfx_v9_set_wave_launch_stall(adev, vmid, false);
mutex_unlock(&adev->grbm_idx_mutex);
return 0;
}
/*
* keep_trap_enabled is ignored here but is a general interface requirement
* for devices that support multi-process debugging where the performance
* overhead from trap temporary setup needs to be bypassed when the debug
* session has ended.
*/
static uint32_t kgd_arcturus_disable_debug_trap(struct amdgpu_device *adev,
bool keep_trap_enabled,
uint32_t vmid)
{
mutex_lock(&adev->grbm_idx_mutex);
kgd_gfx_v9_set_wave_launch_stall(adev, vmid, true);
set_barrier_auto_waitcnt(adev, false);
WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_MASK), 0);
kgd_gfx_v9_set_wave_launch_stall(adev, vmid, false);
mutex_unlock(&adev->grbm_idx_mutex);
return 0;
}
const struct kfd2kgd_calls arcturus_kfd2kgd = {
.program_sh_mem_settings = kgd_gfx_v9_program_sh_mem_settings,
.set_pasid_vmid_mapping = kgd_gfx_v9_set_pasid_vmid_mapping,
@@ -294,6 +408,15 @@ const struct kfd2kgd_calls arcturus_kfd2kgd = {
kgd_gfx_v9_get_atc_vmid_pasid_mapping_info,
.set_vm_context_page_table_base =
kgd_gfx_v9_set_vm_context_page_table_base,
.enable_debug_trap = kgd_arcturus_enable_debug_trap,
.disable_debug_trap = kgd_arcturus_disable_debug_trap,
.validate_trap_override_request = kgd_gfx_v9_validate_trap_override_request,
.set_wave_launch_trap_override = kgd_gfx_v9_set_wave_launch_trap_override,
.set_wave_launch_mode = kgd_gfx_v9_set_wave_launch_mode,
.set_address_watch = kgd_gfx_v9_set_address_watch,
.clear_address_watch = kgd_gfx_v9_clear_address_watch,
.get_iq_wait_times = kgd_gfx_v9_get_iq_wait_times,
.build_grace_period_packet_info = kgd_gfx_v9_build_grace_period_packet_info,
.get_cu_occupancy = kgd_gfx_v9_get_cu_occupancy,
.program_trap_handler_settings = kgd_gfx_v9_program_trap_handler_settings
};
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gc_9_4_3.c
+384
View File
@@ -0,0 +1,384 @@
/*
* Copyright 2021 Advanced Micro Devices, Inc.
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
* OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
* ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
* OTHER DEALINGS IN THE SOFTWARE.
*/
#include "amdgpu.h"
#include "amdgpu_amdkfd.h"
#include "amdgpu_amdkfd_gfx_v9.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"
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);
/* Every two XCCs share one AID */
unsigned int aid = phy_inst / 2;
/*
* We have to assume that there is no outstanding mapping.
* The ATC_VMID_PASID_MAPPING_UPDATE_STATUS bit could be 0 because
* a mapping is in progress or because a mapping finished
* and the SW cleared it.
* So the protocol is to always wait & clear.
*/
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));
/* Every 4 numbers is a cycle. 1st is AID, 2nd and 3rd are XCDs,
* and the 4th is reserved. Therefore "aid * 4 + (xcc_inst % 2) + 1"
* programs _LUT for XCC and "aid * 4" for AID where the XCC connects
* to.
*/
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);
/* HQD registers extend to CP_HQD_AQL_DISPATCH_ID_HI */
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]);
/* Activate doorbell logic before triggering WPTR poll. */
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) {
/* Don't read wptr with get_user because the user
* context may not be accessible (if this function
* runs in a work queue). Instead trigger a one-shot
* polling read from memory in the CP. This assumes
* that wptr is GPU-accessible in the queue's VMID via
* ATC or SVM. WPTR==RPTR before starting the poll so
* the CP starts fetching new commands from the right
* place.
*
* Guessing a 64-bit WPTR from a 32-bit RPTR is a bit
* tricky. Assume that the queue didn't overflow. The
* number of valid bits in the 32-bit RPTR depends on
* the queue size. The remaining bits are taken from
* the saved 64-bit WPTR. If the WPTR wrapped, add the
* queue size.
*/
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));
}
/* Start the EOP fetcher */
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;
}
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,
.program_trap_handler_settings =
kgd_gfx_v9_program_trap_handler_settings
};
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c
+312 -13
View File
@@ -21,6 +21,7 @@
*/
#include "amdgpu.h"
#include "amdgpu_amdkfd.h"
#include "amdgpu_amdkfd_gfx_v10.h"
#include "gc/gc_10_1_0_offset.h"
#include "gc/gc_10_1_0_sh_mask.h"
#include "athub/athub_2_0_0_offset.h"
@@ -31,6 +32,7 @@
#include "v10_structs.h"
#include "nv.h"
#include "nvd.h"
#include <uapi/linux/kfd_ioctl.h>
enum hqd_dequeue_request_type {
NO_ACTION = 0,
@@ -79,7 +81,7 @@ static void kgd_program_sh_mem_settings(struct amdgpu_device *adev, uint32_t vmi
uint32_t sh_mem_config,
uint32_t sh_mem_ape1_base,
uint32_t sh_mem_ape1_limit,
uint32_t sh_mem_bases)
uint32_t sh_mem_bases, uint32_t inst)
{
lock_srbm(adev, 0, 0, 0, vmid);
@@ -91,7 +93,7 @@ static void kgd_program_sh_mem_settings(struct amdgpu_device *adev, uint32_t vmi
}
static int kgd_set_pasid_vmid_mapping(struct amdgpu_device *adev, u32 pasid,
unsigned int vmid)
unsigned int vmid, uint32_t inst)
{
/*
* We have to assume that there is no outstanding mapping.
@@ -135,7 +137,8 @@ static int kgd_set_pasid_vmid_mapping(struct amdgpu_device *adev, u32 pasid,
* but still works
*/
static int kgd_init_interrupts(struct amdgpu_device *adev, uint32_t pipe_id)
static int kgd_init_interrupts(struct amdgpu_device *adev, uint32_t pipe_id,
uint32_t inst)
{
uint32_t mec;
uint32_t pipe;
@@ -205,7 +208,7 @@ static inline struct v10_sdma_mqd *get_sdma_mqd(void *mqd)
static int kgd_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 wptr_mask, struct mm_struct *mm, uint32_t inst)
{
struct v10_compute_mqd *m;
uint32_t *mqd_hqd;
@@ -286,9 +289,9 @@ static int kgd_hqd_load(struct amdgpu_device *adev, void *mqd,
static int kgd_hiq_mqd_load(struct amdgpu_device *adev, void *mqd,
uint32_t pipe_id, uint32_t queue_id,
uint32_t doorbell_off)
uint32_t doorbell_off, uint32_t inst)
{
struct amdgpu_ring *kiq_ring = &adev->gfx.kiq.ring;
struct amdgpu_ring *kiq_ring = &adev->gfx.kiq[0].ring;
struct v10_compute_mqd *m;
uint32_t mec, pipe;
int r;
@@ -303,7 +306,7 @@ static int kgd_hiq_mqd_load(struct amdgpu_device *adev, void *mqd,
pr_debug("kfd: set HIQ, mec:%d, pipe:%d, queue:%d.\n",
mec, pipe, queue_id);
spin_lock(&adev->gfx.kiq.ring_lock);
spin_lock(&adev->gfx.kiq[0].ring_lock);
r = amdgpu_ring_alloc(kiq_ring, 7);
if (r) {
pr_err("Failed to alloc KIQ (%d).\n", r);
@@ -330,7 +333,7 @@ static int kgd_hiq_mqd_load(struct amdgpu_device *adev, void *mqd,
amdgpu_ring_commit(kiq_ring);
out_unlock:
spin_unlock(&adev->gfx.kiq.ring_lock);
spin_unlock(&adev->gfx.kiq[0].ring_lock);
release_queue(adev);
return r;
@@ -338,7 +341,7 @@ out_unlock:
static int kgd_hqd_dump(struct amdgpu_device *adev,
uint32_t pipe_id, uint32_t queue_id,
uint32_t (**dump)[2], uint32_t *n_regs)
uint32_t (**dump)[2], uint32_t *n_regs, uint32_t inst)
{
uint32_t i = 0, reg;
#define HQD_N_REGS 56
@@ -469,7 +472,7 @@ static int kgd_hqd_sdma_dump(struct amdgpu_device *adev,
static bool kgd_hqd_is_occupied(struct amdgpu_device *adev,
uint64_t queue_address, uint32_t pipe_id,
uint32_t queue_id)
uint32_t queue_id, uint32_t inst)
{
uint32_t act;
bool retval = false;
@@ -510,7 +513,7 @@ static bool kgd_hqd_sdma_is_occupied(struct amdgpu_device *adev, void *mqd)
static int kgd_hqd_destroy(struct amdgpu_device *adev, void *mqd,
enum kfd_preempt_type reset_type,
unsigned int utimeout, uint32_t pipe_id,
uint32_t queue_id)
uint32_t queue_id, uint32_t inst)
{
enum hqd_dequeue_request_type type;
unsigned long end_jiffies;
@@ -673,7 +676,7 @@ static bool get_atc_vmid_pasid_mapping_info(struct amdgpu_device *adev,
static int kgd_wave_control_execute(struct amdgpu_device *adev,
uint32_t gfx_index_val,
uint32_t sq_cmd)
uint32_t sq_cmd, uint32_t inst)
{
uint32_t data = 0;
@@ -708,8 +711,295 @@ static void set_vm_context_page_table_base(struct amdgpu_device *adev,
adev->gfxhub.funcs->setup_vm_pt_regs(adev, vmid, page_table_base);
}
/*
* GFX10 helper for wave launch stall requirements on debug trap setting.
*
* vmid:
* Target VMID to stall/unstall.
*
* stall:
* 0-unstall wave launch (enable), 1-stall wave launch (disable).
* After wavefront launch has been stalled, allocated waves must drain from
* SPI in order for debug trap settings to take effect on those waves.
* This is roughly a ~3500 clock cycle wait on SPI where a read on
* SPI_GDBG_WAVE_CNTL translates to ~32 clock cycles.
* KGD_GFX_V10_WAVE_LAUNCH_SPI_DRAIN_LATENCY indicates the number of reads required.
*
* NOTE: We can afford to clear the entire STALL_VMID field on unstall
* because current GFX10 chips cannot support multi-process debugging due to
* trap configuration and masking being limited to global scope. Always
* assume single process conditions.
*
*/
#define KGD_GFX_V10_WAVE_LAUNCH_SPI_DRAIN_LATENCY 110
static void kgd_gfx_v10_set_wave_launch_stall(struct amdgpu_device *adev, uint32_t vmid, bool stall)
{
uint32_t data = RREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL));
int i;
data = REG_SET_FIELD(data, SPI_GDBG_WAVE_CNTL, STALL_VMID,
stall ? 1 << vmid : 0);
WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL), data);
if (!stall)
return;
for (i = 0; i < KGD_GFX_V10_WAVE_LAUNCH_SPI_DRAIN_LATENCY; i++)
RREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL));
}
uint32_t kgd_gfx_v10_enable_debug_trap(struct amdgpu_device *adev,
bool restore_dbg_registers,
uint32_t vmid)
{
mutex_lock(&adev->grbm_idx_mutex);
kgd_gfx_v10_set_wave_launch_stall(adev, vmid, true);
/* assume gfx off is disabled for the debug session if rlc restore not supported. */
if (restore_dbg_registers) {
uint32_t data = 0;
data = REG_SET_FIELD(data, SPI_GDBG_TRAP_CONFIG,
VMID_SEL, 1 << vmid);
data = REG_SET_FIELD(data, SPI_GDBG_TRAP_CONFIG,
TRAP_EN, 1);
WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_CONFIG), data);
WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_DATA0), 0);
WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_DATA1), 0);
kgd_gfx_v10_set_wave_launch_stall(adev, vmid, false);
mutex_unlock(&adev->grbm_idx_mutex);
return 0;
}
WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_MASK), 0);
kgd_gfx_v10_set_wave_launch_stall(adev, vmid, false);
mutex_unlock(&adev->grbm_idx_mutex);
return 0;
}
uint32_t kgd_gfx_v10_disable_debug_trap(struct amdgpu_device *adev,
bool keep_trap_enabled,
uint32_t vmid)
{
mutex_lock(&adev->grbm_idx_mutex);
kgd_gfx_v10_set_wave_launch_stall(adev, vmid, true);
WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_MASK), 0);
kgd_gfx_v10_set_wave_launch_stall(adev, vmid, false);
mutex_unlock(&adev->grbm_idx_mutex);
return 0;
}
int kgd_gfx_v10_validate_trap_override_request(struct amdgpu_device *adev,
uint32_t trap_override,
uint32_t *trap_mask_supported)
{
*trap_mask_supported &= KFD_DBG_TRAP_MASK_DBG_ADDRESS_WATCH;
/* The SPI_GDBG_TRAP_MASK register is global and affects all
* processes. Only allow OR-ing the address-watch bit, since
* this only affects processes under the debugger. Other bits
* should stay 0 to avoid the debugger interfering with other
* processes.
*/
if (trap_override != KFD_DBG_TRAP_OVERRIDE_OR)
return -EINVAL;
return 0;
}
uint32_t kgd_gfx_v10_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, wave_cntl_prev;
mutex_lock(&adev->grbm_idx_mutex);
wave_cntl_prev = RREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL));
kgd_gfx_v10_set_wave_launch_stall(adev, vmid, true);
data = RREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_MASK));
*trap_mask_prev = REG_GET_FIELD(data, SPI_GDBG_TRAP_MASK, EXCP_EN);
trap_mask_bits = (trap_mask_bits & trap_mask_request) |
(*trap_mask_prev & ~trap_mask_request);
data = REG_SET_FIELD(data, SPI_GDBG_TRAP_MASK, EXCP_EN, trap_mask_bits);
data = REG_SET_FIELD(data, SPI_GDBG_TRAP_MASK, REPLACE, trap_override);
WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_MASK), data);
/* We need to preserve wave launch mode stall settings. */
WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL), wave_cntl_prev);
mutex_unlock(&adev->grbm_idx_mutex);
return 0;
}
uint32_t kgd_gfx_v10_set_wave_launch_mode(struct amdgpu_device *adev,
uint8_t wave_launch_mode,
uint32_t vmid)
{
uint32_t data = 0;
bool is_mode_set = !!wave_launch_mode;
mutex_lock(&adev->grbm_idx_mutex);
kgd_gfx_v10_set_wave_launch_stall(adev, vmid, true);
data = REG_SET_FIELD(data, SPI_GDBG_WAVE_CNTL2,
VMID_MASK, is_mode_set ? 1 << vmid : 0);
data = REG_SET_FIELD(data, SPI_GDBG_WAVE_CNTL2,
MODE, is_mode_set ? wave_launch_mode : 0);
WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL2), data);
kgd_gfx_v10_set_wave_launch_stall(adev, vmid, false);
mutex_unlock(&adev->grbm_idx_mutex);
return 0;
}
#define TCP_WATCH_STRIDE (mmTCP_WATCH1_ADDR_H - mmTCP_WATCH0_ADDR_H)
uint32_t kgd_gfx_v10_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 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,
VMID,
debug_vmid);
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);
/* Turning off this watch point until we set all the registers */
watch_address_cntl = REG_SET_FIELD(watch_address_cntl,
TCP_WATCH0_CNTL,
VALID,
0);
WREG32((SOC15_REG_OFFSET(GC, 0, mmTCP_WATCH0_CNTL) +
(watch_id * TCP_WATCH_STRIDE)),
watch_address_cntl);
WREG32((SOC15_REG_OFFSET(GC, 0, mmTCP_WATCH0_ADDR_H) +
(watch_id * TCP_WATCH_STRIDE)),
watch_address_high);
WREG32((SOC15_REG_OFFSET(GC, 0, mmTCP_WATCH0_ADDR_L) +
(watch_id * TCP_WATCH_STRIDE)),
watch_address_low);
/* Enable the watch point */
watch_address_cntl = REG_SET_FIELD(watch_address_cntl,
TCP_WATCH0_CNTL,
VALID,
1);
WREG32((SOC15_REG_OFFSET(GC, 0, mmTCP_WATCH0_CNTL) +
(watch_id * TCP_WATCH_STRIDE)),
watch_address_cntl);
return 0;
}
uint32_t kgd_gfx_v10_clear_address_watch(struct amdgpu_device *adev,
uint32_t watch_id)
{
uint32_t watch_address_cntl;
watch_address_cntl = 0;
WREG32((SOC15_REG_OFFSET(GC, 0, mmTCP_WATCH0_CNTL) +
(watch_id * TCP_WATCH_STRIDE)),
watch_address_cntl);
return 0;
}
/* kgd_gfx_v10_get_iq_wait_times: Returns the mmCP_IQ_WAIT_TIME1/2 values
* The values read are:
* ib_offload_wait_time -- Wait Count for Indirect Buffer Offloads.
* atomic_offload_wait_time -- Wait Count for L2 and GDS Atomics Offloads.
* wrm_offload_wait_time -- Wait Count for WAIT_REG_MEM Offloads.
* gws_wait_time -- Wait Count for Global Wave Syncs.
* que_sleep_wait_time -- Wait Count for Dequeue Retry.
* sch_wave_wait_time -- Wait Count for Scheduling Wave Message.
* sem_rearm_wait_time -- Wait Count for Semaphore re-arm.
* deq_retry_wait_time -- Wait Count for Global Wave Syncs.
*/
void kgd_gfx_v10_get_iq_wait_times(struct amdgpu_device *adev,
uint32_t *wait_times)
{
*wait_times = RREG32(SOC15_REG_OFFSET(GC, 0, mmCP_IQ_WAIT_TIME2));
}
void kgd_gfx_v10_build_grace_period_packet_info(struct amdgpu_device *adev,
uint32_t wait_times,
uint32_t grace_period,
uint32_t *reg_offset,
uint32_t *reg_data)
{
*reg_data = wait_times;
/*
* The CP cannont handle a 0 grace period input and will result in
* an infinite grace period being set so set to 1 to prevent this.
*/
if (grace_period == 0)
grace_period = 1;
*reg_data = REG_SET_FIELD(*reg_data,
CP_IQ_WAIT_TIME2,
SCH_WAVE,
grace_period);
*reg_offset = SOC15_REG_OFFSET(GC, 0, mmCP_IQ_WAIT_TIME2);
}
static void program_trap_handler_settings(struct amdgpu_device *adev,
uint32_t vmid, uint64_t tba_addr, uint64_t tma_addr)
uint32_t vmid, uint64_t tba_addr, uint64_t tma_addr,
uint32_t inst)
{
lock_srbm(adev, 0, 0, 0, vmid);
@@ -750,5 +1040,14 @@ const struct kfd2kgd_calls gfx_v10_kfd2kgd = {
.get_atc_vmid_pasid_mapping_info =
get_atc_vmid_pasid_mapping_info,
.set_vm_context_page_table_base = set_vm_context_page_table_base,
.enable_debug_trap = kgd_gfx_v10_enable_debug_trap,
.disable_debug_trap = kgd_gfx_v10_disable_debug_trap,
.validate_trap_override_request = kgd_gfx_v10_validate_trap_override_request,
.set_wave_launch_trap_override = kgd_gfx_v10_set_wave_launch_trap_override,
.set_wave_launch_mode = kgd_gfx_v10_set_wave_launch_mode,
.set_address_watch = kgd_gfx_v10_set_address_watch,
.clear_address_watch = kgd_gfx_v10_clear_address_watch,
.get_iq_wait_times = kgd_gfx_v10_get_iq_wait_times,
.build_grace_period_packet_info = kgd_gfx_v10_build_grace_period_packet_info,
.program_trap_handler_settings = program_trap_handler_settings,
};
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.h
+55
View File
@@ -0,0 +1,55 @@
/*
* Copyright 2023 Advanced Micro Devices, Inc.
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
* OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
* ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
* OTHER DEALINGS IN THE SOFTWARE.
*/
uint32_t kgd_gfx_v10_enable_debug_trap(struct amdgpu_device *adev,
bool restore_dbg_registers,
uint32_t vmid);
uint32_t kgd_gfx_v10_disable_debug_trap(struct amdgpu_device *adev,
bool keep_trap_enabled,
uint32_t vmid);
int kgd_gfx_v10_validate_trap_override_request(struct amdgpu_device *adev,
uint32_t trap_override,
uint32_t *trap_mask_supported);
uint32_t kgd_gfx_v10_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 kgd_gfx_v10_set_wave_launch_mode(struct amdgpu_device *adev,
uint8_t wave_launch_mode,
uint32_t vmid);
uint32_t kgd_gfx_v10_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 kgd_gfx_v10_clear_address_watch(struct amdgpu_device *adev,
uint32_t watch_id);
void kgd_gfx_v10_get_iq_wait_times(struct amdgpu_device *adev, uint32_t *wait_times);
void kgd_gfx_v10_build_grace_period_packet_info(struct amdgpu_device *adev,
uint32_t wait_times,
uint32_t grace_period,
uint32_t *reg_offset,
uint32_t *reg_data);
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10_3.c
+25 -157
View File
@@ -22,6 +22,7 @@
#include <linux/mmu_context.h>
#include "amdgpu.h"
#include "amdgpu_amdkfd.h"
#include "amdgpu_amdkfd_gfx_v10.h"
#include "gc/gc_10_3_0_offset.h"
#include "gc/gc_10_3_0_sh_mask.h"
#include "oss/osssys_5_0_0_offset.h"
@@ -80,7 +81,7 @@ static void program_sh_mem_settings_v10_3(struct amdgpu_device *adev, uint32_t v
uint32_t sh_mem_config,
uint32_t sh_mem_ape1_base,
uint32_t sh_mem_ape1_limit,
uint32_t sh_mem_bases)
uint32_t sh_mem_bases, uint32_t inst)
{
lock_srbm(adev, 0, 0, 0, vmid);
@@ -93,7 +94,7 @@ static void program_sh_mem_settings_v10_3(struct amdgpu_device *adev, uint32_t v
/* ATC is defeatured on Sienna_Cichlid */
static int set_pasid_vmid_mapping_v10_3(struct amdgpu_device *adev, unsigned int pasid,
unsigned int vmid)
unsigned int vmid, uint32_t inst)
{
uint32_t value = pasid << IH_VMID_0_LUT__PASID__SHIFT;
@@ -105,7 +106,8 @@ static int set_pasid_vmid_mapping_v10_3(struct amdgpu_device *adev, unsigned int
return 0;
}
static int init_interrupts_v10_3(struct amdgpu_device *adev, uint32_t pipe_id)
static int init_interrupts_v10_3(struct amdgpu_device *adev, uint32_t pipe_id,
uint32_t inst)
{
uint32_t mec;
uint32_t pipe;
@@ -177,7 +179,7 @@ static inline struct v10_sdma_mqd *get_sdma_mqd(void *mqd)
static int hqd_load_v10_3(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 wptr_mask, struct mm_struct *mm, uint32_t inst)
{
struct v10_compute_mqd *m;
uint32_t *mqd_hqd;
@@ -273,9 +275,9 @@ static int hqd_load_v10_3(struct amdgpu_device *adev, void *mqd,
static int hiq_mqd_load_v10_3(struct amdgpu_device *adev, void *mqd,
uint32_t pipe_id, uint32_t queue_id,
uint32_t doorbell_off)
uint32_t doorbell_off, uint32_t inst)
{
struct amdgpu_ring *kiq_ring = &adev->gfx.kiq.ring;
struct amdgpu_ring *kiq_ring = &adev->gfx.kiq[0].ring;
struct v10_compute_mqd *m;
uint32_t mec, pipe;
int r;
@@ -290,7 +292,7 @@ static int hiq_mqd_load_v10_3(struct amdgpu_device *adev, void *mqd,
pr_debug("kfd: set HIQ, mec:%d, pipe:%d, queue:%d.\n",
mec, pipe, queue_id);
spin_lock(&adev->gfx.kiq.ring_lock);
spin_lock(&adev->gfx.kiq[0].ring_lock);
r = amdgpu_ring_alloc(kiq_ring, 7);
if (r) {
pr_err("Failed to alloc KIQ (%d).\n", r);
@@ -317,7 +319,7 @@ static int hiq_mqd_load_v10_3(struct amdgpu_device *adev, void *mqd,
amdgpu_ring_commit(kiq_ring);
out_unlock:
spin_unlock(&adev->gfx.kiq.ring_lock);
spin_unlock(&adev->gfx.kiq[0].ring_lock);
release_queue(adev);
return r;
@@ -325,7 +327,7 @@ out_unlock:
static int hqd_dump_v10_3(struct amdgpu_device *adev,
uint32_t pipe_id, uint32_t queue_id,
uint32_t (**dump)[2], uint32_t *n_regs)
uint32_t (**dump)[2], uint32_t *n_regs, uint32_t inst)
{
uint32_t i = 0, reg;
#define HQD_N_REGS 56
@@ -456,7 +458,7 @@ static int hqd_sdma_dump_v10_3(struct amdgpu_device *adev,
static bool hqd_is_occupied_v10_3(struct amdgpu_device *adev,
uint64_t queue_address, uint32_t pipe_id,
uint32_t queue_id)
uint32_t queue_id, uint32_t inst)
{
uint32_t act;
bool retval = false;
@@ -498,7 +500,7 @@ static bool hqd_sdma_is_occupied_v10_3(struct amdgpu_device *adev,
static int hqd_destroy_v10_3(struct amdgpu_device *adev, void *mqd,
enum kfd_preempt_type reset_type,
unsigned int utimeout, uint32_t pipe_id,
uint32_t queue_id)
uint32_t queue_id, uint32_t inst)
{
enum hqd_dequeue_request_type type;
unsigned long end_jiffies;
@@ -586,7 +588,7 @@ static int hqd_sdma_destroy_v10_3(struct amdgpu_device *adev, void *mqd,
static int wave_control_execute_v10_3(struct amdgpu_device *adev,
uint32_t gfx_index_val,
uint32_t sq_cmd)
uint32_t sq_cmd, uint32_t inst)
{
uint32_t data = 0;
@@ -628,7 +630,8 @@ static void set_vm_context_page_table_base_v10_3(struct amdgpu_device *adev,
}
static void program_trap_handler_settings_v10_3(struct amdgpu_device *adev,
uint32_t vmid, uint64_t tba_addr, uint64_t tma_addr)
uint32_t vmid, uint64_t tba_addr, uint64_t tma_addr,
uint32_t inst)
{
lock_srbm(adev, 0, 0, 0, vmid);
@@ -652,142 +655,6 @@ static void program_trap_handler_settings_v10_3(struct amdgpu_device *adev,
unlock_srbm(adev);
}
#if 0
uint32_t enable_debug_trap_v10_3(struct amdgpu_device *adev,
uint32_t trap_debug_wave_launch_mode,
uint32_t vmid)
{
uint32_t data = 0;
uint32_t orig_wave_cntl_value;
uint32_t orig_stall_vmid;
mutex_lock(&adev->grbm_idx_mutex);
orig_wave_cntl_value = RREG32(SOC15_REG_OFFSET(GC,
0,
mmSPI_GDBG_WAVE_CNTL));
orig_stall_vmid = REG_GET_FIELD(orig_wave_cntl_value,
SPI_GDBG_WAVE_CNTL,
STALL_VMID);
data = REG_SET_FIELD(data, SPI_GDBG_WAVE_CNTL, STALL_RA, 1);
WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL), data);
data = 0;
WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_MASK), data);
WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL), orig_stall_vmid);
mutex_unlock(&adev->grbm_idx_mutex);
return 0;
}
uint32_t disable_debug_trap_v10_3(struct amdgpu_device *adev)
{
mutex_lock(&adev->grbm_idx_mutex);
WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_MASK), 0);
mutex_unlock(&adev->grbm_idx_mutex);
return 0;
}
uint32_t set_wave_launch_trap_override_v10_3(struct amdgpu_device *adev,
uint32_t trap_override,
uint32_t trap_mask)
{
uint32_t data = 0;
mutex_lock(&adev->grbm_idx_mutex);
data = RREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL));
data = REG_SET_FIELD(data, SPI_GDBG_WAVE_CNTL, STALL_RA, 1);
WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL), data);
data = 0;
data = REG_SET_FIELD(data, SPI_GDBG_TRAP_MASK,
EXCP_EN, trap_mask);
data = REG_SET_FIELD(data, SPI_GDBG_TRAP_MASK,
REPLACE, trap_override);
WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_MASK), data);
data = RREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL));
data = REG_SET_FIELD(data, SPI_GDBG_WAVE_CNTL, STALL_RA, 0);
WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL), data);
mutex_unlock(&adev->grbm_idx_mutex);
return 0;
}
uint32_t set_wave_launch_mode_v10_3(struct amdgpu_device *adev,
uint8_t wave_launch_mode,
uint32_t vmid)
{
uint32_t data = 0;
bool is_stall_mode;
bool is_mode_set;
is_stall_mode = (wave_launch_mode == 4);
is_mode_set = (wave_launch_mode != 0 && wave_launch_mode != 4);
mutex_lock(&adev->grbm_idx_mutex);
data = REG_SET_FIELD(data, SPI_GDBG_WAVE_CNTL2,
VMID_MASK, is_mode_set ? 1 << vmid : 0);
data = REG_SET_FIELD(data, SPI_GDBG_WAVE_CNTL2,
MODE, is_mode_set ? wave_launch_mode : 0);
WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL2), data);
data = RREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL));
data = REG_SET_FIELD(data, SPI_GDBG_WAVE_CNTL,
STALL_VMID, is_stall_mode ? 1 << vmid : 0);
data = REG_SET_FIELD(data, SPI_GDBG_WAVE_CNTL,
STALL_RA, is_stall_mode ? 1 : 0);
WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL), data);
mutex_unlock(&adev->grbm_idx_mutex);
return 0;
}
/* kgd_get_iq_wait_times: Returns the mmCP_IQ_WAIT_TIME1/2 values
* The values read are:
* ib_offload_wait_time -- Wait Count for Indirect Buffer Offloads.
* atomic_offload_wait_time -- Wait Count for L2 and GDS Atomics Offloads.
* wrm_offload_wait_time -- Wait Count for WAIT_REG_MEM Offloads.
* gws_wait_time -- Wait Count for Global Wave Syncs.
* que_sleep_wait_time -- Wait Count for Dequeue Retry.
* sch_wave_wait_time -- Wait Count for Scheduling Wave Message.
* sem_rearm_wait_time -- Wait Count for Semaphore re-arm.
* deq_retry_wait_time -- Wait Count for Global Wave Syncs.
*/
void get_iq_wait_times_v10_3(struct amdgpu_device *adev,
uint32_t *wait_times)
{
*wait_times = RREG32(SOC15_REG_OFFSET(GC, 0, mmCP_IQ_WAIT_TIME2));
}
void build_grace_period_packet_info_v10_3(struct amdgpu_device *adev,
uint32_t wait_times,
uint32_t grace_period,
uint32_t *reg_offset,
uint32_t *reg_data)
{
*reg_data = wait_times;
*reg_data = REG_SET_FIELD(*reg_data,
CP_IQ_WAIT_TIME2,
SCH_WAVE,
grace_period);
*reg_offset = mmCP_IQ_WAIT_TIME2;
}
#endif
const struct kfd2kgd_calls gfx_v10_3_kfd2kgd = {
.program_sh_mem_settings = program_sh_mem_settings_v10_3,
.set_pasid_vmid_mapping = set_pasid_vmid_mapping_v10_3,
@@ -805,12 +672,13 @@ const struct kfd2kgd_calls gfx_v10_3_kfd2kgd = {
.get_atc_vmid_pasid_mapping_info = get_atc_vmid_pasid_mapping_info_v10_3,
.set_vm_context_page_table_base = set_vm_context_page_table_base_v10_3,
.program_trap_handler_settings = program_trap_handler_settings_v10_3,
#if 0
.enable_debug_trap = enable_debug_trap_v10_3,
.disable_debug_trap = disable_debug_trap_v10_3,
.set_wave_launch_trap_override = set_wave_launch_trap_override_v10_3,
.set_wave_launch_mode = set_wave_launch_mode_v10_3,
.get_iq_wait_times = get_iq_wait_times_v10_3,
.build_grace_period_packet_info = build_grace_period_packet_info_v10_3,
#endif
.get_iq_wait_times = kgd_gfx_v10_get_iq_wait_times,
.build_grace_period_packet_info = kgd_gfx_v10_build_grace_period_packet_info,
.enable_debug_trap = kgd_gfx_v10_enable_debug_trap,
.disable_debug_trap = kgd_gfx_v10_disable_debug_trap,
.validate_trap_override_request = kgd_gfx_v10_validate_trap_override_request,
.set_wave_launch_trap_override = kgd_gfx_v10_set_wave_launch_trap_override,
.set_wave_launch_mode = kgd_gfx_v10_set_wave_launch_mode,
.set_address_watch = kgd_gfx_v10_set_address_watch,
.clear_address_watch = kgd_gfx_v10_clear_address_watch
};
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v11.c
+198 -12
View File
@@ -30,6 +30,7 @@
#include "soc15d.h"
#include "v11_structs.h"
#include "soc21.h"
#include <uapi/linux/kfd_ioctl.h>
enum hqd_dequeue_request_type {
NO_ACTION = 0,
@@ -78,7 +79,7 @@ static void program_sh_mem_settings_v11(struct amdgpu_device *adev, uint32_t vmi
uint32_t sh_mem_config,
uint32_t sh_mem_ape1_base,
uint32_t sh_mem_ape1_limit,
uint32_t sh_mem_bases)
uint32_t sh_mem_bases, uint32_t inst)
{
lock_srbm(adev, 0, 0, 0, vmid);
@@ -89,7 +90,7 @@ static void program_sh_mem_settings_v11(struct amdgpu_device *adev, uint32_t vmi
}
static int set_pasid_vmid_mapping_v11(struct amdgpu_device *adev, unsigned int pasid,
unsigned int vmid)
unsigned int vmid, uint32_t inst)
{
uint32_t value = pasid << IH_VMID_0_LUT__PASID__SHIFT;
@@ -101,7 +102,8 @@ static int set_pasid_vmid_mapping_v11(struct amdgpu_device *adev, unsigned int p
return 0;
}
static int init_interrupts_v11(struct amdgpu_device *adev, uint32_t pipe_id)
static int init_interrupts_v11(struct amdgpu_device *adev, uint32_t pipe_id,
uint32_t inst)
{
uint32_t mec;
uint32_t pipe;
@@ -162,7 +164,7 @@ static inline struct v11_sdma_mqd *get_sdma_mqd(void *mqd)
static int hqd_load_v11(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)
struct mm_struct *mm, uint32_t inst)
{
struct v11_compute_mqd *m;
uint32_t *mqd_hqd;
@@ -258,9 +260,9 @@ static int hqd_load_v11(struct amdgpu_device *adev, void *mqd, uint32_t pipe_id,
static int hiq_mqd_load_v11(struct amdgpu_device *adev, void *mqd,
uint32_t pipe_id, uint32_t queue_id,
uint32_t doorbell_off)
uint32_t doorbell_off, uint32_t inst)
{
struct amdgpu_ring *kiq_ring = &adev->gfx.kiq.ring;
struct amdgpu_ring *kiq_ring = &adev->gfx.kiq[0].ring;
struct v11_compute_mqd *m;
uint32_t mec, pipe;
int r;
@@ -275,7 +277,7 @@ static int hiq_mqd_load_v11(struct amdgpu_device *adev, void *mqd,
pr_debug("kfd: set HIQ, mec:%d, pipe:%d, queue:%d.\n",
mec, pipe, queue_id);
spin_lock(&adev->gfx.kiq.ring_lock);
spin_lock(&adev->gfx.kiq[0].ring_lock);
r = amdgpu_ring_alloc(kiq_ring, 7);
if (r) {
pr_err("Failed to alloc KIQ (%d).\n", r);
@@ -302,7 +304,7 @@ static int hiq_mqd_load_v11(struct amdgpu_device *adev, void *mqd,
amdgpu_ring_commit(kiq_ring);
out_unlock:
spin_unlock(&adev->gfx.kiq.ring_lock);
spin_unlock(&adev->gfx.kiq[0].ring_lock);
release_queue(adev);
return r;
@@ -310,7 +312,7 @@ out_unlock:
static int hqd_dump_v11(struct amdgpu_device *adev,
uint32_t pipe_id, uint32_t queue_id,
uint32_t (**dump)[2], uint32_t *n_regs)
uint32_t (**dump)[2], uint32_t *n_regs, uint32_t inst)
{
uint32_t i = 0, reg;
#define HQD_N_REGS 56
@@ -445,7 +447,7 @@ static int hqd_sdma_dump_v11(struct amdgpu_device *adev,
}
static bool hqd_is_occupied_v11(struct amdgpu_device *adev, uint64_t queue_address,
uint32_t pipe_id, uint32_t queue_id)
uint32_t pipe_id, uint32_t queue_id, uint32_t inst)
{
uint32_t act;
bool retval = false;
@@ -486,7 +488,7 @@ static bool hqd_sdma_is_occupied_v11(struct amdgpu_device *adev, void *mqd)
static int hqd_destroy_v11(struct amdgpu_device *adev, void *mqd,
enum kfd_preempt_type reset_type,
unsigned int utimeout, uint32_t pipe_id,
uint32_t queue_id)
uint32_t queue_id, uint32_t inst)
{
enum hqd_dequeue_request_type type;
unsigned long end_jiffies;
@@ -571,7 +573,7 @@ static int hqd_sdma_destroy_v11(struct amdgpu_device *adev, void *mqd,
static int wave_control_execute_v11(struct amdgpu_device *adev,
uint32_t gfx_index_val,
uint32_t sq_cmd)
uint32_t sq_cmd, uint32_t inst)
{
uint32_t data = 0;
@@ -606,6 +608,183 @@ static void set_vm_context_page_table_base_v11(struct amdgpu_device *adev,
adev->gfxhub.funcs->setup_vm_pt_regs(adev, vmid, page_table_base);
}
/*
* Returns TRAP_EN, EXCP_EN and EXCP_REPLACE.
*
* restore_dbg_registers is ignored here but is a general interface requirement
* for devices that support GFXOFF and where the RLC save/restore list
* does not support hw registers for debugging i.e. the driver has to manually
* initialize the debug mode registers after it has disabled GFX off during the
* debug session.
*/
static uint32_t kgd_gfx_v11_enable_debug_trap(struct amdgpu_device *adev,
bool restore_dbg_registers,
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;
}
/* Returns TRAP_EN, EXCP_EN and EXCP_REPLACE. */
static uint32_t kgd_gfx_v11_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, keep_trap_enabled);
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_v11_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;
if (adev->ip_versions[GC_HWIP][0] >= IP_VERSION(11, 0, 4))
*trap_mask_supported |= 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;
}
/* Returns TRAP_EN, EXCP_EN and EXCP_REPLACE. */
static uint32_t kgd_gfx_v11_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;
}
static uint32_t kgd_gfx_v11_set_wave_launch_mode(struct amdgpu_device *adev,
uint8_t wave_launch_mode,
uint32_t vmid)
{
uint32_t data = 0;
data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, LAUNCH_MODE, wave_launch_mode);
return data;
}
#define TCP_WATCH_STRIDE (regTCP_WATCH1_ADDR_H - regTCP_WATCH0_ADDR_H)
static uint32_t kgd_gfx_v11_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 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, 0, regTCP_WATCH0_ADDR_H) +
(watch_id * TCP_WATCH_STRIDE)),
watch_address_high);
WREG32_RLC((SOC15_REG_OFFSET(GC, 0, regTCP_WATCH0_ADDR_L) +
(watch_id * TCP_WATCH_STRIDE)),
watch_address_low);
return watch_address_cntl;
}
static uint32_t kgd_gfx_v11_clear_address_watch(struct amdgpu_device *adev,
uint32_t watch_id)
{
return 0;
}
const struct kfd2kgd_calls gfx_v11_kfd2kgd = {
.program_sh_mem_settings = program_sh_mem_settings_v11,
.set_pasid_vmid_mapping = set_pasid_vmid_mapping_v11,
@@ -622,4 +801,11 @@ const struct kfd2kgd_calls gfx_v11_kfd2kgd = {
.wave_control_execute = wave_control_execute_v11,
.get_atc_vmid_pasid_mapping_info = NULL,
.set_vm_context_page_table_base = set_vm_context_page_table_base_v11,
.enable_debug_trap = kgd_gfx_v11_enable_debug_trap,
.disable_debug_trap = kgd_gfx_v11_disable_debug_trap,
.validate_trap_override_request = kgd_gfx_v11_validate_trap_override_request,
.set_wave_launch_trap_override = kgd_gfx_v11_set_wave_launch_trap_override,
.set_wave_launch_mode = kgd_gfx_v11_set_wave_launch_mode,
.set_address_watch = kgd_gfx_v11_set_address_watch,
.clear_address_watch = kgd_gfx_v11_clear_address_watch
};
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c
+9 -8
View File
@@ -78,7 +78,7 @@ static void kgd_program_sh_mem_settings(struct amdgpu_device *adev, uint32_t vmi
uint32_t sh_mem_config,
uint32_t sh_mem_ape1_base,
uint32_t sh_mem_ape1_limit,
uint32_t sh_mem_bases)
uint32_t sh_mem_bases, uint32_t inst)
{
lock_srbm(adev, 0, 0, 0, vmid);
@@ -91,7 +91,7 @@ static void kgd_program_sh_mem_settings(struct amdgpu_device *adev, uint32_t vmi
}
static int kgd_set_pasid_vmid_mapping(struct amdgpu_device *adev, u32 pasid,
unsigned int vmid)
unsigned int vmid, uint32_t inst)
{
/*
* We have to assume that there is no outstanding mapping.
@@ -114,7 +114,8 @@ static int kgd_set_pasid_vmid_mapping(struct amdgpu_device *adev, u32 pasid,
return 0;
}
static int kgd_init_interrupts(struct amdgpu_device *adev, uint32_t pipe_id)
static int kgd_init_interrupts(struct amdgpu_device *adev, uint32_t pipe_id,
uint32_t inst)
{
uint32_t mec;
uint32_t pipe;
@@ -158,7 +159,7 @@ static inline struct cik_sdma_rlc_registers *get_sdma_mqd(void *mqd)
static int kgd_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 wptr_mask, struct mm_struct *mm, uint32_t inst)
{
struct cik_mqd *m;
uint32_t *mqd_hqd;
@@ -202,7 +203,7 @@ static int kgd_hqd_load(struct amdgpu_device *adev, void *mqd,
static int kgd_hqd_dump(struct amdgpu_device *adev,
uint32_t pipe_id, uint32_t queue_id,
uint32_t (**dump)[2], uint32_t *n_regs)
uint32_t (**dump)[2], uint32_t *n_regs, uint32_t inst)
{
uint32_t i = 0, reg;
#define HQD_N_REGS (35+4)
@@ -318,7 +319,7 @@ static int kgd_hqd_sdma_dump(struct amdgpu_device *adev,
static bool kgd_hqd_is_occupied(struct amdgpu_device *adev,
uint64_t queue_address, uint32_t pipe_id,
uint32_t queue_id)
uint32_t queue_id, uint32_t inst)
{
uint32_t act;
bool retval = false;
@@ -358,7 +359,7 @@ static bool kgd_hqd_sdma_is_occupied(struct amdgpu_device *adev, void *mqd)
static int kgd_hqd_destroy(struct amdgpu_device *adev, void *mqd,
enum kfd_preempt_type reset_type,
unsigned int utimeout, uint32_t pipe_id,
uint32_t queue_id)
uint32_t queue_id, uint32_t inst)
{
uint32_t temp;
enum hqd_dequeue_request_type type;
@@ -494,7 +495,7 @@ static int kgd_hqd_sdma_destroy(struct amdgpu_device *adev, void *mqd,
static int kgd_wave_control_execute(struct amdgpu_device *adev,
uint32_t gfx_index_val,
uint32_t sq_cmd)
uint32_t sq_cmd, uint32_t inst)
{
uint32_t data;
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c
+9 -8
View File
@@ -72,7 +72,7 @@ static void kgd_program_sh_mem_settings(struct amdgpu_device *adev, uint32_t vmi
uint32_t sh_mem_config,
uint32_t sh_mem_ape1_base,
uint32_t sh_mem_ape1_limit,
uint32_t sh_mem_bases)
uint32_t sh_mem_bases, uint32_t inst)
{
lock_srbm(adev, 0, 0, 0, vmid);
@@ -85,7 +85,7 @@ static void kgd_program_sh_mem_settings(struct amdgpu_device *adev, uint32_t vmi
}
static int kgd_set_pasid_vmid_mapping(struct amdgpu_device *adev, u32 pasid,
unsigned int vmid)
unsigned int vmid, uint32_t inst)
{
/*
* We have to assume that there is no outstanding mapping.
@@ -109,7 +109,8 @@ static int kgd_set_pasid_vmid_mapping(struct amdgpu_device *adev, u32 pasid,
return 0;
}
static int kgd_init_interrupts(struct amdgpu_device *adev, uint32_t pipe_id)
static int kgd_init_interrupts(struct amdgpu_device *adev, uint32_t pipe_id,
uint32_t inst)
{
uint32_t mec;
uint32_t pipe;
@@ -153,7 +154,7 @@ static inline struct vi_sdma_mqd *get_sdma_mqd(void *mqd)
static int kgd_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 wptr_mask, struct mm_struct *mm, uint32_t inst)
{
struct vi_mqd *m;
uint32_t *mqd_hqd;
@@ -226,7 +227,7 @@ static int kgd_hqd_load(struct amdgpu_device *adev, void *mqd,
static int kgd_hqd_dump(struct amdgpu_device *adev,
uint32_t pipe_id, uint32_t queue_id,
uint32_t (**dump)[2], uint32_t *n_regs)
uint32_t (**dump)[2], uint32_t *n_regs, uint32_t inst)
{
uint32_t i = 0, reg;
#define HQD_N_REGS (54+4)
@@ -350,7 +351,7 @@ static int kgd_hqd_sdma_dump(struct amdgpu_device *adev,
static bool kgd_hqd_is_occupied(struct amdgpu_device *adev,
uint64_t queue_address, uint32_t pipe_id,
uint32_t queue_id)
uint32_t queue_id, uint32_t inst)
{
uint32_t act;
bool retval = false;
@@ -390,7 +391,7 @@ static bool kgd_hqd_sdma_is_occupied(struct amdgpu_device *adev, void *mqd)
static int kgd_hqd_destroy(struct amdgpu_device *adev, void *mqd,
enum kfd_preempt_type reset_type,
unsigned int utimeout, uint32_t pipe_id,
uint32_t queue_id)
uint32_t queue_id, uint32_t inst)
{
uint32_t temp;
enum hqd_dequeue_request_type type;
@@ -540,7 +541,7 @@ static bool get_atc_vmid_pasid_mapping_info(struct amdgpu_device *adev,
static int kgd_wave_control_execute(struct amdgpu_device *adev,
uint32_t gfx_index_val,
uint32_t sq_cmd)
uint32_t sq_cmd, uint32_t inst)
{
uint32_t data = 0;
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c
+385 -87
View File
File diff suppressed because it is too large. Load diff
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.h
+54 -14
View File
@@ -20,41 +20,81 @@
* OTHER DEALINGS IN THE SOFTWARE.
*/
void kgd_gfx_v9_program_sh_mem_settings(struct amdgpu_device *adev, uint32_t vmid,
uint32_t sh_mem_config,
uint32_t sh_mem_ape1_base, uint32_t sh_mem_ape1_limit,
uint32_t sh_mem_bases);
uint32_t sh_mem_bases, uint32_t inst);
int kgd_gfx_v9_set_pasid_vmid_mapping(struct amdgpu_device *adev, u32 pasid,
unsigned int vmid);
int kgd_gfx_v9_init_interrupts(struct amdgpu_device *adev, uint32_t pipe_id);
unsigned int vmid, uint32_t inst);
int kgd_gfx_v9_init_interrupts(struct amdgpu_device *adev, uint32_t pipe_id,
uint32_t inst);
int kgd_gfx_v9_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);
struct mm_struct *mm, uint32_t inst);
int kgd_gfx_v9_hiq_mqd_load(struct amdgpu_device *adev, void *mqd,
uint32_t pipe_id, uint32_t queue_id,
uint32_t doorbell_off);
uint32_t doorbell_off, uint32_t inst);
int kgd_gfx_v9_hqd_dump(struct amdgpu_device *adev,
uint32_t pipe_id, uint32_t queue_id,
uint32_t (**dump)[2], uint32_t *n_regs);
uint32_t (**dump)[2], uint32_t *n_regs, uint32_t inst);
bool kgd_gfx_v9_hqd_is_occupied(struct amdgpu_device *adev,
uint64_t queue_address, uint32_t pipe_id,
uint32_t queue_id);
uint32_t queue_id, uint32_t inst);
int kgd_gfx_v9_hqd_destroy(struct amdgpu_device *adev, void *mqd,
enum kfd_preempt_type reset_type,
unsigned int utimeout, uint32_t pipe_id,
uint32_t queue_id);
uint32_t queue_id, uint32_t inst);
int kgd_gfx_v9_wave_control_execute(struct amdgpu_device *adev,
uint32_t gfx_index_val,
uint32_t sq_cmd);
uint32_t sq_cmd, uint32_t inst);
bool kgd_gfx_v9_get_atc_vmid_pasid_mapping_info(struct amdgpu_device *adev,
uint8_t vmid, uint16_t *p_pasid);
void kgd_gfx_v9_set_vm_context_page_table_base(struct amdgpu_device *adev,
uint32_t vmid, uint64_t page_table_base);
void kgd_gfx_v9_get_cu_occupancy(struct amdgpu_device *adev, int pasid,
int *pasid_wave_cnt, int *max_waves_per_cu);
int *pasid_wave_cnt, int *max_waves_per_cu, uint32_t inst);
void kgd_gfx_v9_program_trap_handler_settings(struct amdgpu_device *adev,
uint32_t vmid, uint64_t tba_addr, uint64_t tma_addr);
uint32_t vmid, uint64_t tba_addr, uint64_t tma_addr,
uint32_t inst);
void kgd_gfx_v9_acquire_queue(struct amdgpu_device *adev, uint32_t pipe_id,
uint32_t queue_id, uint32_t inst);
uint64_t kgd_gfx_v9_get_queue_mask(struct amdgpu_device *adev,
uint32_t pipe_id, uint32_t queue_id);
void kgd_gfx_v9_release_queue(struct amdgpu_device *adev, uint32_t inst);
void kgd_gfx_v9_set_wave_launch_stall(struct amdgpu_device *adev,
uint32_t vmid,
bool stall);
uint32_t kgd_gfx_v9_enable_debug_trap(struct amdgpu_device *adev,
bool restore_dbg_registers,
uint32_t vmid);
uint32_t kgd_gfx_v9_disable_debug_trap(struct amdgpu_device *adev,
bool keep_trap_enabled,
uint32_t vmid);
int kgd_gfx_v9_validate_trap_override_request(struct amdgpu_device *adev,
uint32_t trap_override,
uint32_t *trap_mask_supported);
uint32_t kgd_gfx_v9_set_wave_launch_mode(struct amdgpu_device *adev,
uint8_t wave_launch_mode,
uint32_t vmid);
uint32_t kgd_gfx_v9_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 kgd_gfx_v9_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 kgd_gfx_v9_clear_address_watch(struct amdgpu_device *adev,
uint32_t watch_id);
void kgd_gfx_v9_get_iq_wait_times(struct amdgpu_device *adev, uint32_t *wait_times);
void kgd_gfx_v9_build_grace_period_packet_info(struct amdgpu_device *adev,
uint32_t wait_times,
uint32_t grace_period,
uint32_t *reg_offset,
uint32_t *reg_data);
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
+130 -41
View File
@@ -35,7 +35,9 @@
#include "amdgpu_dma_buf.h"
#include <uapi/linux/kfd_ioctl.h>
#include "amdgpu_xgmi.h"
#include "kfd_priv.h"
#include "kfd_smi_events.h"
#include <drm/ttm/ttm_tt.h>
/* Userptr restore delay, just long enough to allow consecutive VM
* changes to accumulate
@@ -110,13 +112,16 @@ void amdgpu_amdkfd_gpuvm_init_mem_limits(void)
struct sysinfo si;
uint64_t mem;
if (kfd_mem_limit.max_system_mem_limit)
return;
si_meminfo(&si);
mem = si.freeram - si.freehigh;
mem *= si.mem_unit;
spin_lock_init(&kfd_mem_limit.mem_limit_lock);
kfd_mem_limit.max_system_mem_limit = mem - (mem >> 4);
kfd_mem_limit.max_ttm_mem_limit = (mem >> 1) - (mem >> 3);
kfd_mem_limit.max_ttm_mem_limit = ttm_tt_pages_limit() << PAGE_SHIFT;
pr_debug("Kernel memory limit %lluM, TTM limit %lluM\n",
(kfd_mem_limit.max_system_mem_limit >> 20),
(kfd_mem_limit.max_ttm_mem_limit >> 20));
@@ -148,16 +153,20 @@ void amdgpu_amdkfd_reserve_system_mem(uint64_t size)
* @size: Size of buffer, in bytes, encapsulated by B0. This should be
* equivalent to amdgpu_bo_size(BO)
* @alloc_flag: Flag used in allocating a BO as noted above
* @xcp_id: xcp_id is used to get xcp from xcp manager, one xcp is
* managed as one compute node in driver for app
*
* Return: returns -ENOMEM in case of error, ZERO otherwise
* Return:
* returns -ENOMEM in case of error, ZERO otherwise
*/
int amdgpu_amdkfd_reserve_mem_limit(struct amdgpu_device *adev,
uint64_t size, u32 alloc_flag)
uint64_t size, u32 alloc_flag, int8_t xcp_id)
{
uint64_t reserved_for_pt =
ESTIMATE_PT_SIZE(amdgpu_amdkfd_total_mem_size);
size_t system_mem_needed, ttm_mem_needed, vram_needed;
int ret = 0;
uint64_t vram_size = 0;
system_mem_needed = 0;
ttm_mem_needed = 0;
@@ -172,6 +181,17 @@ int amdgpu_amdkfd_reserve_mem_limit(struct amdgpu_device *adev,
* 2M BO chunk.
*/
vram_needed = size;
/*
* For GFX 9.4.3, get the VRAM size from XCP structs
*/
if (WARN_ONCE(xcp_id < 0, "invalid XCP ID %d", xcp_id))
return -EINVAL;
vram_size = KFD_XCP_MEMORY_SIZE(adev, xcp_id);
if (adev->gmc.is_app_apu) {
system_mem_needed = size;
ttm_mem_needed = size;
}
} else if (alloc_flag & KFD_IOC_ALLOC_MEM_FLAGS_USERPTR) {
system_mem_needed = size;
} else if (!(alloc_flag &
@@ -191,8 +211,8 @@ int amdgpu_amdkfd_reserve_mem_limit(struct amdgpu_device *adev,
kfd_mem_limit.max_system_mem_limit && !no_system_mem_limit) ||
(kfd_mem_limit.ttm_mem_used + ttm_mem_needed >
kfd_mem_limit.max_ttm_mem_limit) ||
(adev && adev->kfd.vram_used + vram_needed >
adev->gmc.real_vram_size - reserved_for_pt)) {
(adev && xcp_id >= 0 && adev->kfd.vram_used[xcp_id] + vram_needed >
vram_size - reserved_for_pt)) {
ret = -ENOMEM;
goto release;
}
@@ -202,9 +222,11 @@ int amdgpu_amdkfd_reserve_mem_limit(struct amdgpu_device *adev,
*/
WARN_ONCE(vram_needed && !adev,
"adev reference can't be null when vram is used");
if (adev) {
adev->kfd.vram_used += vram_needed;
adev->kfd.vram_used_aligned += ALIGN(vram_needed, VRAM_AVAILABLITY_ALIGN);
if (adev && xcp_id >= 0) {
adev->kfd.vram_used[xcp_id] += vram_needed;
adev->kfd.vram_used_aligned[xcp_id] += adev->gmc.is_app_apu ?
vram_needed :
ALIGN(vram_needed, VRAM_AVAILABLITY_ALIGN);
}
kfd_mem_limit.system_mem_used += system_mem_needed;
kfd_mem_limit.ttm_mem_used += ttm_mem_needed;
@@ -215,7 +237,7 @@ release:
}
void amdgpu_amdkfd_unreserve_mem_limit(struct amdgpu_device *adev,
uint64_t size, u32 alloc_flag)
uint64_t size, u32 alloc_flag, int8_t xcp_id)
{
spin_lock(&kfd_mem_limit.mem_limit_lock);
@@ -225,9 +247,19 @@ void amdgpu_amdkfd_unreserve_mem_limit(struct amdgpu_device *adev,
} else if (alloc_flag & KFD_IOC_ALLOC_MEM_FLAGS_VRAM) {
WARN_ONCE(!adev,
"adev reference can't be null when alloc mem flags vram is set");
if (WARN_ONCE(xcp_id < 0, "invalid XCP ID %d", xcp_id))
goto release;
if (adev) {
adev->kfd.vram_used -= size;
adev->kfd.vram_used_aligned -= ALIGN(size, VRAM_AVAILABLITY_ALIGN);
adev->kfd.vram_used[xcp_id] -= size;
if (adev->gmc.is_app_apu) {
adev->kfd.vram_used_aligned[xcp_id] -= size;
kfd_mem_limit.system_mem_used -= size;
kfd_mem_limit.ttm_mem_used -= size;
} else {
adev->kfd.vram_used_aligned[xcp_id] -=
ALIGN(size, VRAM_AVAILABLITY_ALIGN);
}
}
} else if (alloc_flag & KFD_IOC_ALLOC_MEM_FLAGS_USERPTR) {
kfd_mem_limit.system_mem_used -= size;
@@ -237,8 +269,8 @@ void amdgpu_amdkfd_unreserve_mem_limit(struct amdgpu_device *adev,
pr_err("%s: Invalid BO type %#x\n", __func__, alloc_flag);
goto release;
}
WARN_ONCE(adev && adev->kfd.vram_used < 0,
"KFD VRAM memory accounting unbalanced");
WARN_ONCE(adev && xcp_id >= 0 && adev->kfd.vram_used[xcp_id] < 0,
"KFD VRAM memory accounting unbalanced for xcp: %d", xcp_id);
WARN_ONCE(kfd_mem_limit.ttm_mem_used < 0,
"KFD TTM memory accounting unbalanced");
WARN_ONCE(kfd_mem_limit.system_mem_used < 0,
@@ -254,14 +286,16 @@ void amdgpu_amdkfd_release_notify(struct amdgpu_bo *bo)
u32 alloc_flags = bo->kfd_bo->alloc_flags;
u64 size = amdgpu_bo_size(bo);
amdgpu_amdkfd_unreserve_mem_limit(adev, size, alloc_flags);
amdgpu_amdkfd_unreserve_mem_limit(adev, size, alloc_flags,
bo->xcp_id);
kfree(bo->kfd_bo);
}
/**
* @create_dmamap_sg_bo: Creates a amdgpu_bo object to reflect information
* create_dmamap_sg_bo() - Creates a amdgpu_bo object to reflect information
* about USERPTR or DOOREBELL or MMIO BO.
*
* @adev: Device for which dmamap BO is being created
* @mem: BO of peer device that is being DMA mapped. Provides parameters
* in building the dmamap BO
@@ -285,7 +319,7 @@ create_dmamap_sg_bo(struct amdgpu_device *adev,
ret = amdgpu_gem_object_create(adev, mem->bo->tbo.base.size, 1,
AMDGPU_GEM_DOMAIN_CPU, AMDGPU_GEM_CREATE_PREEMPTIBLE | flags,
ttm_bo_type_sg, mem->bo->tbo.base.resv, &gem_obj);
ttm_bo_type_sg, mem->bo->tbo.base.resv, &gem_obj, 0);
amdgpu_bo_unreserve(mem->bo);
@@ -527,6 +561,12 @@ kfd_mem_dmamap_dmabuf(struct kfd_mem_attachment *attachment)
{
struct ttm_operation_ctx ctx = {.interruptible = true};
struct amdgpu_bo *bo = attachment->bo_va->base.bo;
int ret;
amdgpu_bo_placement_from_domain(bo, AMDGPU_GEM_DOMAIN_CPU);
ret = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
if (ret)
return ret;
amdgpu_bo_placement_from_domain(bo, AMDGPU_GEM_DOMAIN_GTT);
return ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
@@ -659,11 +699,10 @@ kfd_mem_dmaunmap_userptr(struct kgd_mem *mem,
static void
kfd_mem_dmaunmap_dmabuf(struct kfd_mem_attachment *attachment)
{
struct ttm_operation_ctx ctx = {.interruptible = true};
struct amdgpu_bo *bo = attachment->bo_va->base.bo;
amdgpu_bo_placement_from_domain(bo, AMDGPU_GEM_DOMAIN_CPU);
ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
/* This is a no-op. We don't want to trigger eviction fences when
* unmapping DMABufs. Therefore the invalidation (moving to system
* domain) is done in kfd_mem_dmamap_dmabuf.
*/
}
/**
@@ -804,7 +843,7 @@ static int kfd_mem_attach(struct amdgpu_device *adev, struct kgd_mem *mem,
* if peer device has large BAR. In contrast, access over xGMI is
* allowed for both small and large BAR configurations of peer device
*/
if ((adev != bo_adev) &&
if ((adev != bo_adev && !adev->gmc.is_app_apu) &&
((mem->domain == AMDGPU_GEM_DOMAIN_VRAM) ||
(mem->alloc_flags & KFD_IOC_ALLOC_MEM_FLAGS_DOORBELL) ||
(mem->alloc_flags & KFD_IOC_ALLOC_MEM_FLAGS_MMIO_REMAP))) {
@@ -1599,23 +1638,42 @@ out_unlock:
return ret;
}
size_t amdgpu_amdkfd_get_available_memory(struct amdgpu_device *adev)
size_t amdgpu_amdkfd_get_available_memory(struct amdgpu_device *adev,
uint8_t xcp_id)
{
uint64_t reserved_for_pt =
ESTIMATE_PT_SIZE(amdgpu_amdkfd_total_mem_size);
ssize_t available;
uint64_t vram_available, system_mem_available, ttm_mem_available;
spin_lock(&kfd_mem_limit.mem_limit_lock);
available = adev->gmc.real_vram_size
- adev->kfd.vram_used_aligned
vram_available = KFD_XCP_MEMORY_SIZE(adev, xcp_id)
- adev->kfd.vram_used_aligned[xcp_id]
- atomic64_read(&adev->vram_pin_size)
- reserved_for_pt;
if (adev->gmc.is_app_apu) {
system_mem_available = no_system_mem_limit ?
kfd_mem_limit.max_system_mem_limit :
kfd_mem_limit.max_system_mem_limit -
kfd_mem_limit.system_mem_used;
ttm_mem_available = kfd_mem_limit.max_ttm_mem_limit -
kfd_mem_limit.ttm_mem_used;
available = min3(system_mem_available, ttm_mem_available,
vram_available);
available = ALIGN_DOWN(available, PAGE_SIZE);
} else {
available = ALIGN_DOWN(vram_available, VRAM_AVAILABLITY_ALIGN);
}
spin_unlock(&kfd_mem_limit.mem_limit_lock);
if (available < 0)
available = 0;
return ALIGN_DOWN(available, VRAM_AVAILABLITY_ALIGN);
return available;
}
int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
@@ -1624,6 +1682,7 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
uint64_t *offset, uint32_t flags, bool criu_resume)
{
struct amdgpu_vm *avm = drm_priv_to_vm(drm_priv);
struct amdgpu_fpriv *fpriv = container_of(avm, struct amdgpu_fpriv, vm);
enum ttm_bo_type bo_type = ttm_bo_type_device;
struct sg_table *sg = NULL;
uint64_t user_addr = 0;
@@ -1631,6 +1690,7 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
struct drm_gem_object *gobj = NULL;
u32 domain, alloc_domain;
uint64_t aligned_size;
int8_t xcp_id = -1;
u64 alloc_flags;
int ret;
@@ -1639,9 +1699,17 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
*/
if (flags & KFD_IOC_ALLOC_MEM_FLAGS_VRAM) {
domain = alloc_domain = AMDGPU_GEM_DOMAIN_VRAM;
alloc_flags = AMDGPU_GEM_CREATE_VRAM_WIPE_ON_RELEASE;
alloc_flags |= (flags & KFD_IOC_ALLOC_MEM_FLAGS_PUBLIC) ?
if (adev->gmc.is_app_apu) {
domain = AMDGPU_GEM_DOMAIN_GTT;
alloc_domain = AMDGPU_GEM_DOMAIN_GTT;
alloc_flags = 0;
} else {
alloc_flags = AMDGPU_GEM_CREATE_VRAM_WIPE_ON_RELEASE;
alloc_flags |= (flags & KFD_IOC_ALLOC_MEM_FLAGS_PUBLIC) ?
AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED : 0;
}
xcp_id = fpriv->xcp_id == ~0 ? 0 : fpriv->xcp_id;
} else if (flags & KFD_IOC_ALLOC_MEM_FLAGS_GTT) {
domain = alloc_domain = AMDGPU_GEM_DOMAIN_GTT;
alloc_flags = 0;
@@ -1693,17 +1761,19 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
amdgpu_sync_create(&(*mem)->sync);
ret = amdgpu_amdkfd_reserve_mem_limit(adev, aligned_size, flags);
ret = amdgpu_amdkfd_reserve_mem_limit(adev, aligned_size, flags,
xcp_id);
if (ret) {
pr_debug("Insufficient memory\n");
goto err_reserve_limit;
}
pr_debug("\tcreate BO VA 0x%llx size 0x%llx domain %s\n",
va, (*mem)->aql_queue ? size << 1 : size, domain_string(alloc_domain));
pr_debug("\tcreate BO VA 0x%llx size 0x%llx domain %s xcp_id %d\n",
va, (*mem)->aql_queue ? size << 1 : size,
domain_string(alloc_domain), xcp_id);
ret = amdgpu_gem_object_create(adev, aligned_size, 1, alloc_domain, alloc_flags,
bo_type, NULL, &gobj);
bo_type, NULL, &gobj, xcp_id + 1);
if (ret) {
pr_debug("Failed to create BO on domain %s. ret %d\n",
domain_string(alloc_domain), ret);
@@ -1728,6 +1798,7 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
(*mem)->domain = domain;
(*mem)->mapped_to_gpu_memory = 0;
(*mem)->process_info = avm->process_info;
add_kgd_mem_to_kfd_bo_list(*mem, avm->process_info, user_addr);
if (user_addr) {
@@ -1759,7 +1830,7 @@ err_node_allow:
/* Don't unreserve system mem limit twice */
goto err_reserve_limit;
err_bo_create:
amdgpu_amdkfd_unreserve_mem_limit(adev, aligned_size, flags);
amdgpu_amdkfd_unreserve_mem_limit(adev, aligned_size, flags, xcp_id);
err_reserve_limit:
mutex_destroy(&(*mem)->lock);
if (gobj)
@@ -1855,11 +1926,14 @@ int amdgpu_amdkfd_gpuvm_free_memory_of_gpu(
}
/* Update the size of the BO being freed if it was allocated from
* VRAM and is not imported.
* VRAM and is not imported. For APP APU VRAM allocations are done
* in GTT domain
*/
if (size) {
if ((mem->bo->preferred_domains == AMDGPU_GEM_DOMAIN_VRAM) &&
(!is_imported))
if (!is_imported &&
(mem->bo->preferred_domains == AMDGPU_GEM_DOMAIN_VRAM ||
(adev->gmc.is_app_apu &&
mem->bo->preferred_domains == AMDGPU_GEM_DOMAIN_GTT)))
*size = bo_size;
else
*size = 0;
@@ -2282,8 +2356,9 @@ int amdgpu_amdkfd_gpuvm_import_dmabuf(struct amdgpu_device *adev,
(*mem)->dmabuf = dma_buf;
(*mem)->bo = bo;
(*mem)->va = va;
(*mem)->domain = (bo->preferred_domains & AMDGPU_GEM_DOMAIN_VRAM) ?
(*mem)->domain = (bo->preferred_domains & AMDGPU_GEM_DOMAIN_VRAM) && !adev->gmc.is_app_apu ?
AMDGPU_GEM_DOMAIN_VRAM : AMDGPU_GEM_DOMAIN_GTT;
(*mem)->mapped_to_gpu_memory = 0;
(*mem)->process_info = avm->process_info;
add_kgd_mem_to_kfd_bo_list(*mem, avm->process_info, false);
@@ -2445,7 +2520,9 @@ static int update_invalid_user_pages(struct amdkfd_process_info *process_info,
ret = -EAGAIN;
goto unlock_out;
}
mem->invalid = 0;
/* set mem valid if mem has hmm range associated */
if (mem->range)
mem->invalid = 0;
}
unlock_out:
@@ -2577,8 +2654,15 @@ static int confirm_valid_user_pages_locked(struct amdkfd_process_info *process_i
list_for_each_entry_safe(mem, tmp_mem,
&process_info->userptr_inval_list,
validate_list.head) {
bool valid = amdgpu_ttm_tt_get_user_pages_done(
mem->bo->tbo.ttm, mem->range);
bool valid;
/* keep mem without hmm range at userptr_inval_list */
if (!mem->range)
continue;
/* Only check mem with hmm range associated */
valid = amdgpu_ttm_tt_get_user_pages_done(
mem->bo->tbo.ttm, mem->range);
mem->range = NULL;
if (!valid) {
@@ -2586,7 +2670,12 @@ static int confirm_valid_user_pages_locked(struct amdkfd_process_info *process_i
ret = -EAGAIN;
continue;
}
WARN(mem->invalid, "Valid BO is marked invalid");
if (mem->invalid) {
WARN(1, "Valid BO is marked invalid");
ret = -EAGAIN;
continue;
}
list_move_tail(&mem->validate_list.head,
&process_info->userptr_valid_list);
Loaded 20 of 498 files, more files were not shown because too many files have changed in this diff. Show more