| 1 | /* SPDX-License-Identifier: GPL-2.0-only */ |
| 2 | /* |
| 3 | * Copyright (C) 2020-2025 Intel Corporation |
| 4 | */ |
| 5 | |
| 6 | #ifndef __IVPU_DRV_H__ |
| 7 | #define __IVPU_DRV_H__ |
| 8 | |
| 9 | #include <drm/drm_device.h> |
| 10 | #include <drm/drm_drv.h> |
| 11 | #include <drm/drm_managed.h> |
| 12 | #include <drm/drm_mm.h> |
| 13 | #include <drm/drm_print.h> |
| 14 | |
| 15 | #include <linux/pci.h> |
| 16 | #include <linux/xarray.h> |
| 17 | #include <uapi/drm/ivpu_accel.h> |
| 18 | |
| 19 | #include "ivpu_mmu_context.h" |
| 20 | #include "ivpu_ipc.h" |
| 21 | |
| 22 | #define DRIVER_NAME "intel_vpu" |
| 23 | #define DRIVER_DESC "Driver for Intel NPU (Neural Processing Unit)" |
| 24 | |
| 25 | #define PCI_DEVICE_ID_MTL 0x7d1d |
| 26 | #define PCI_DEVICE_ID_ARL 0xad1d |
| 27 | #define PCI_DEVICE_ID_LNL 0x643e |
| 28 | #define PCI_DEVICE_ID_PTL_P 0xb03e |
| 29 | #define PCI_DEVICE_ID_WCL 0xfd3e |
| 30 | #define PCI_DEVICE_ID_NVL 0xd71d |
| 31 | |
| 32 | #define IVPU_HW_IP_37XX 37 |
| 33 | #define IVPU_HW_IP_40XX 40 |
| 34 | #define IVPU_HW_IP_50XX 50 |
| 35 | #define IVPU_HW_IP_60XX 60 |
| 36 | |
| 37 | #define IVPU_HW_IP_REV_LNL_B0 4 |
| 38 | |
| 39 | #define IVPU_HW_BTRS_MTL 1 |
| 40 | #define IVPU_HW_BTRS_LNL 2 |
| 41 | |
| 42 | #define IVPU_GLOBAL_CONTEXT_MMU_SSID 0 |
| 43 | /* SSID 1 is used by the VPU to represent reserved context */ |
| 44 | #define IVPU_RESERVED_CONTEXT_MMU_SSID 1 |
| 45 | #define IVPU_USER_CONTEXT_MIN_SSID 2 |
| 46 | #define IVPU_USER_CONTEXT_MAX_SSID (IVPU_USER_CONTEXT_MIN_SSID + 63) |
| 47 | |
| 48 | #define IVPU_MIN_DB 1 |
| 49 | #define IVPU_MAX_DB 255 |
| 50 | |
| 51 | #define IVPU_JOB_ID_JOB_MASK GENMASK(7, 0) |
| 52 | #define IVPU_JOB_ID_CONTEXT_MASK GENMASK(31, 8) |
| 53 | |
| 54 | #define IVPU_NUM_PRIORITIES 4 |
| 55 | #define IVPU_NUM_CMDQS_PER_CTX (IVPU_NUM_PRIORITIES) |
| 56 | |
| 57 | #define IVPU_CMDQ_MIN_ID 1 |
| 58 | #define IVPU_CMDQ_MAX_ID 255 |
| 59 | |
| 60 | #define IVPU_PLATFORM_SILICON 0 |
| 61 | #define IVPU_PLATFORM_SIMICS 2 |
| 62 | #define IVPU_PLATFORM_FPGA 3 |
| 63 | #define IVPU_PLATFORM_HSLE 4 |
| 64 | #define IVPU_PLATFORM_INVALID 8 |
| 65 | |
| 66 | #define IVPU_SCHED_MODE_AUTO -1 |
| 67 | |
| 68 | #define IVPU_DBG_REG BIT(0) |
| 69 | #define IVPU_DBG_IRQ BIT(1) |
| 70 | #define IVPU_DBG_MMU BIT(2) |
| 71 | #define IVPU_DBG_FILE BIT(3) |
| 72 | #define IVPU_DBG_MISC BIT(4) |
| 73 | #define IVPU_DBG_FW_BOOT BIT(5) |
| 74 | #define IVPU_DBG_PM BIT(6) |
| 75 | #define IVPU_DBG_IPC BIT(7) |
| 76 | #define IVPU_DBG_BO BIT(8) |
| 77 | #define IVPU_DBG_JOB BIT(9) |
| 78 | #define IVPU_DBG_JSM BIT(10) |
| 79 | #define IVPU_DBG_KREF BIT(11) |
| 80 | #define IVPU_DBG_RPM BIT(12) |
| 81 | #define IVPU_DBG_MMU_MAP BIT(13) |
| 82 | #define IVPU_DBG_IOCTL BIT(14) |
| 83 | |
| 84 | #define ivpu_err(vdev, fmt, ...) \ |
| 85 | drm_err(&(vdev)->drm, "%s(): " fmt, __func__, ##__VA_ARGS__) |
| 86 | |
| 87 | #define ivpu_err_ratelimited(vdev, fmt, ...) \ |
| 88 | drm_err_ratelimited(&(vdev)->drm, "%s(): " fmt, __func__, ##__VA_ARGS__) |
| 89 | |
| 90 | #define ivpu_warn(vdev, fmt, ...) \ |
| 91 | drm_warn(&(vdev)->drm, "%s(): " fmt, __func__, ##__VA_ARGS__) |
| 92 | |
| 93 | #define ivpu_warn_ratelimited(vdev, fmt, ...) \ |
| 94 | drm_err_ratelimited(&(vdev)->drm, "%s(): " fmt, __func__, ##__VA_ARGS__) |
| 95 | |
| 96 | #define ivpu_info(vdev, fmt, ...) drm_info(&(vdev)->drm, fmt, ##__VA_ARGS__) |
| 97 | |
| 98 | #define ivpu_dbg(vdev, type, fmt, args...) do { \ |
| 99 | if (unlikely(IVPU_DBG_##type & ivpu_dbg_mask)) \ |
| 100 | dev_dbg((vdev)->drm.dev, "[%s] " fmt, #type, ##args); \ |
| 101 | } while (0) |
| 102 | |
| 103 | #define IVPU_WA(wa_name) (vdev->wa.wa_name) |
| 104 | |
| 105 | #define IVPU_PRINT_WA(wa_name) do { \ |
| 106 | if (IVPU_WA(wa_name)) \ |
| 107 | ivpu_dbg(vdev, MISC, "Using WA: " #wa_name "\n"); \ |
| 108 | } while (0) |
| 109 | |
| 110 | struct ivpu_wa_table { |
| 111 | bool punit_disabled; |
| 112 | bool clear_runtime_mem; |
| 113 | bool interrupt_clear_with_0; |
| 114 | bool disable_clock_relinquish; |
| 115 | bool disable_d0i3_msg; |
| 116 | bool wp0_during_power_up; |
| 117 | bool disable_d0i2; |
| 118 | }; |
| 119 | |
| 120 | struct ivpu_hw_info; |
| 121 | struct ivpu_mmu_info; |
| 122 | struct ivpu_fw_info; |
| 123 | struct ivpu_ipc_info; |
| 124 | struct ivpu_pm_info; |
| 125 | |
| 126 | struct ivpu_device { |
| 127 | struct drm_device drm; |
| 128 | void __iomem *regb; |
| 129 | void __iomem *regv; |
| 130 | u32 platform; |
| 131 | u32 irq; |
| 132 | |
| 133 | struct ivpu_wa_table wa; |
| 134 | struct ivpu_hw_info *hw; |
| 135 | struct ivpu_mmu_info *mmu; |
| 136 | struct ivpu_fw_info *fw; |
| 137 | struct ivpu_ipc_info *ipc; |
| 138 | struct ivpu_pm_info *pm; |
| 139 | |
| 140 | struct ivpu_mmu_context gctx; |
| 141 | struct ivpu_mmu_context rctx; |
| 142 | struct mutex context_list_lock; /* Protects user context addition/removal */ |
| 143 | struct xarray context_xa; |
| 144 | struct xa_limit context_xa_limit; |
| 145 | |
| 146 | struct xarray db_xa; |
| 147 | struct xa_limit db_limit; |
| 148 | u32 db_next; |
| 149 | |
| 150 | struct work_struct irq_ipc_work; |
| 151 | struct work_struct irq_dct_work; |
| 152 | struct work_struct context_abort_work; |
| 153 | |
| 154 | struct mutex bo_list_lock; /* Protects bo_list */ |
| 155 | struct list_head bo_list; |
| 156 | |
| 157 | struct mutex submitted_jobs_lock; /* Protects submitted_jobs */ |
| 158 | struct xarray submitted_jobs_xa; |
| 159 | struct ivpu_ipc_consumer job_done_consumer; |
| 160 | atomic_t job_timeout_counter; |
| 161 | |
| 162 | atomic64_t unique_id_counter; |
| 163 | |
| 164 | ktime_t busy_start_ts; |
| 165 | ktime_t busy_time; |
| 166 | |
| 167 | struct { |
| 168 | int boot; |
| 169 | int jsm; |
| 170 | int tdr; |
| 171 | int inference; |
| 172 | int autosuspend; |
| 173 | int d0i3_entry_msg; |
| 174 | int state_dump_msg; |
| 175 | } timeout; |
| 176 | }; |
| 177 | |
| 178 | /* |
| 179 | * file_priv has its own refcount (ref) that allows user space to close the fd |
| 180 | * without blocking even if VPU is still processing some jobs. |
| 181 | */ |
| 182 | struct ivpu_file_priv { |
| 183 | struct kref ref; |
| 184 | struct ivpu_device *vdev; |
| 185 | struct mutex lock; /* Protects cmdq */ |
| 186 | struct xarray cmdq_xa; |
| 187 | struct ivpu_mmu_context ctx; |
| 188 | struct mutex ms_lock; /* Protects ms_instance_list, ms_info_bo */ |
| 189 | struct list_head ms_instance_list; |
| 190 | struct ivpu_bo *ms_info_bo; |
| 191 | struct xa_limit job_limit; |
| 192 | u32 job_id_next; |
| 193 | struct xa_limit cmdq_limit; |
| 194 | u32 cmdq_id_next; |
| 195 | bool has_mmu_faults; |
| 196 | bool bound; |
| 197 | bool aborted; |
| 198 | }; |
| 199 | |
| 200 | extern int ivpu_dbg_mask; |
| 201 | extern u8 ivpu_pll_min_ratio; |
| 202 | extern u8 ivpu_pll_max_ratio; |
| 203 | extern int ivpu_sched_mode; |
| 204 | extern bool ivpu_disable_mmu_cont_pages; |
| 205 | extern bool ivpu_force_snoop; |
| 206 | |
| 207 | #define IVPU_TEST_MODE_FW_TEST BIT(0) |
| 208 | #define IVPU_TEST_MODE_NULL_HW BIT(1) |
| 209 | #define IVPU_TEST_MODE_NULL_SUBMISSION BIT(2) |
| 210 | #define IVPU_TEST_MODE_D0I3_MSG_DISABLE BIT(4) |
| 211 | #define IVPU_TEST_MODE_D0I3_MSG_ENABLE BIT(5) |
| 212 | #define IVPU_TEST_MODE_MIP_DISABLE BIT(6) |
| 213 | #define IVPU_TEST_MODE_DISABLE_TIMEOUTS BIT(8) |
| 214 | #define IVPU_TEST_MODE_TURBO_ENABLE BIT(9) |
| 215 | #define IVPU_TEST_MODE_TURBO_DISABLE BIT(10) |
| 216 | #define IVPU_TEST_MODE_CLK_RELINQ_DISABLE BIT(11) |
| 217 | #define IVPU_TEST_MODE_CLK_RELINQ_ENABLE BIT(12) |
| 218 | #define IVPU_TEST_MODE_D0I2_DISABLE BIT(13) |
| 219 | extern int ivpu_test_mode; |
| 220 | |
| 221 | struct ivpu_file_priv *ivpu_file_priv_get(struct ivpu_file_priv *file_priv); |
| 222 | void ivpu_file_priv_put(struct ivpu_file_priv **link); |
| 223 | |
| 224 | int ivpu_boot(struct ivpu_device *vdev); |
| 225 | int ivpu_shutdown(struct ivpu_device *vdev); |
| 226 | void ivpu_prepare_for_reset(struct ivpu_device *vdev); |
| 227 | bool ivpu_is_capable(struct ivpu_device *vdev, u32 capability); |
| 228 | |
| 229 | static inline u8 ivpu_revision(struct ivpu_device *vdev) |
| 230 | { |
| 231 | return to_pci_dev(vdev->drm.dev)->revision; |
| 232 | } |
| 233 | |
| 234 | static inline u16 ivpu_device_id(struct ivpu_device *vdev) |
| 235 | { |
| 236 | return to_pci_dev(vdev->drm.dev)->device; |
| 237 | } |
| 238 | |
| 239 | static inline int ivpu_hw_ip_gen(struct ivpu_device *vdev) |
| 240 | { |
| 241 | switch (ivpu_device_id(vdev)) { |
| 242 | case PCI_DEVICE_ID_MTL: |
| 243 | case PCI_DEVICE_ID_ARL: |
| 244 | return IVPU_HW_IP_37XX; |
| 245 | case PCI_DEVICE_ID_LNL: |
| 246 | return IVPU_HW_IP_40XX; |
| 247 | case PCI_DEVICE_ID_PTL_P: |
| 248 | case PCI_DEVICE_ID_WCL: |
| 249 | return IVPU_HW_IP_50XX; |
| 250 | case PCI_DEVICE_ID_NVL: |
| 251 | return IVPU_HW_IP_60XX; |
| 252 | default: |
| 253 | dump_stack(); |
| 254 | ivpu_err(vdev, "Unknown NPU IP generation\n" ); |
| 255 | return 0; |
| 256 | } |
| 257 | } |
| 258 | |
| 259 | static inline int ivpu_hw_btrs_gen(struct ivpu_device *vdev) |
| 260 | { |
| 261 | switch (ivpu_device_id(vdev)) { |
| 262 | case PCI_DEVICE_ID_MTL: |
| 263 | case PCI_DEVICE_ID_ARL: |
| 264 | return IVPU_HW_BTRS_MTL; |
| 265 | case PCI_DEVICE_ID_LNL: |
| 266 | case PCI_DEVICE_ID_PTL_P: |
| 267 | case PCI_DEVICE_ID_WCL: |
| 268 | case PCI_DEVICE_ID_NVL: |
| 269 | return IVPU_HW_BTRS_LNL; |
| 270 | default: |
| 271 | dump_stack(); |
| 272 | ivpu_err(vdev, "Unknown buttress generation\n" ); |
| 273 | return 0; |
| 274 | } |
| 275 | } |
| 276 | |
| 277 | static inline struct ivpu_device *to_ivpu_device(struct drm_device *dev) |
| 278 | { |
| 279 | return container_of(dev, struct ivpu_device, drm); |
| 280 | } |
| 281 | |
| 282 | static inline u32 ivpu_get_context_count(struct ivpu_device *vdev) |
| 283 | { |
| 284 | struct xa_limit ctx_limit = vdev->context_xa_limit; |
| 285 | |
| 286 | return (ctx_limit.max - ctx_limit.min + 1); |
| 287 | } |
| 288 | |
| 289 | static inline u32 ivpu_get_platform(struct ivpu_device *vdev) |
| 290 | { |
| 291 | WARN_ON_ONCE(vdev->platform == IVPU_PLATFORM_INVALID); |
| 292 | return vdev->platform; |
| 293 | } |
| 294 | |
| 295 | static inline bool ivpu_is_silicon(struct ivpu_device *vdev) |
| 296 | { |
| 297 | return ivpu_get_platform(vdev) == IVPU_PLATFORM_SILICON; |
| 298 | } |
| 299 | |
| 300 | static inline bool ivpu_is_simics(struct ivpu_device *vdev) |
| 301 | { |
| 302 | return ivpu_get_platform(vdev) == IVPU_PLATFORM_SIMICS; |
| 303 | } |
| 304 | |
| 305 | static inline bool ivpu_is_fpga(struct ivpu_device *vdev) |
| 306 | { |
| 307 | return ivpu_get_platform(vdev) == IVPU_PLATFORM_FPGA || |
| 308 | ivpu_get_platform(vdev) == IVPU_PLATFORM_HSLE; |
| 309 | } |
| 310 | |
| 311 | static inline bool ivpu_is_force_snoop_enabled(struct ivpu_device *vdev) |
| 312 | { |
| 313 | return ivpu_force_snoop; |
| 314 | } |
| 315 | |
| 316 | #endif /* __IVPU_DRV_H__ */ |
| 317 | |