Lines Matching +full:max +full:- +full:outbound +full:- +full:regions
1 /* SPDX-License-Identifier: GPL-2.0
3 * Copyright 2016-2022 HabanaLabs, Ltd.
19 #include <linux/dma-direction.h>
28 #include <linux/io-64-nonatomic-lo-hi.h>
30 #include <linux/dma-buf.h>
42 * bits[63:59] - Encode mmap type
43 * bits[45:0] - mmap offset value
48 #define HL_MMAP_TYPE_SHIFT (59 - PAGE_SHIFT)
107 * enum hl_mmu_page_table_location - mmu page table location
108 * @MMU_DR_PGT: page-table is located on device DRAM.
109 * @MMU_HR_PGT: page-table is located on host memory.
110 * @MMU_NUM_PGT_LOCATIONS: number of page-table locations currently supported.
113 MMU_DR_PGT = 0, /* device-dram-resident MMU PGT */
132 #define IS_POWER_OF_2(n) (n != 0 && ((n & (n - 1)) == 0))
160 * - HL_DRV_RESET_HARD
164 * - HL_DRV_RESET_FROM_RESET_THR
165 * Set if the caller is the hard-reset thread
167 * - HL_DRV_RESET_HEARTBEAT
170 * - HL_DRV_RESET_TDR
173 * - HL_DRV_RESET_DEV_RELEASE
176 * - HL_DRV_RESET_BYPASS_REQ_TO_FW
180 * - HL_DRV_RESET_FW_FATAL_ERR
183 * - HL_DRV_RESET_DELAY
186 * - HL_DRV_RESET_FROM_WD_THR
212 #define HL_BLOCK_GLBL_SEC_SIZE (HL_BLOCK_SIZE - HL_BLOCK_GLBL_SEC_OFFS)
223 * struct iterate_module_ctx - HW module iterator
226 * @rc: return code for optional use of iterator/iterator-caller
234 * @offset: current HW module instance offset from the 1-st HW module instance
235 * in the 1-st block
251 * struct hl_gen_wait_properties - properties for generating a wait CB
271 * struct pgt_info - MMU hop page info.
272 * @node: hash linked-list node for the pgts on host (shadow pgts for device resident MMU and
299 * enum hl_pci_match_mode - pci match mode per region
309 * enum hl_fw_component - F/W components to read version through registers.
321 * enum hl_fw_types - F/W types present in the system
325 * @FW_TYPE_PREBOOT_CPU: Indicates pre-loaded CPUs are present in the system
339 * enum hl_queue_type - Supported QUEUE types.
370 * struct hl_inbound_pci_region - inbound region descriptor
386 * struct hl_outbound_pci_region - outbound region descriptor
396 * enum queue_cb_alloc_flags - Indicates queue support for CBs that
407 * struct hl_hw_sob - H/W SOB info.
431 * struct hw_queue_properties - queue information.
453 * enum vm_type - virtual memory mapping request information.
463 * enum mmu_op_flags - mmu operation relevant information.
478 * enum hl_device_hw_state - H/W device state. use this to understand whether
492 * struct hl_mmu_properties - ASIC specific MMU address translation properties.
524 * struct hl_hints_range - hint addresses reserved va range.
534 * struct asic_fixed_properties - ASIC specific immutable properties.
538 * @cpucp_info: received various information from CPU-CP regarding the H/W, e.g.
540 * @uboot_ver: F/W U-boot version.
552 * @sram_user_base_address - SRAM physical start address for user access.
558 * @max_power_default: max power of the device after reset.
579 * @max_freq_value: current max clk frequency.
629 * 1- enabled, 0- isolated.
631 * 1- faulty cluster, 0- good cluster.
633 * 1- enabled, 0- isolated.
655 * @eq_interrupt_id: interrupt id for EQ, uses to synchronize EQ interrupts in hard-reset.
660 * @completion_mode: 0 - job based completion, 1 - cs based completion
661 * @mme_master_slave_mode: 0 - Each MME works independently, 1 - MME works
679 * @supports_compute_reset: is a reset which is not a hard-reset supported by this asic.
682 * in inference ASICs, as there is no real-world
683 * use-case of doing soft-reset in training (due
686 * @configurable_stop_on_err: is stop-on-error option configurable via debugfs.
687 * @set_max_power_on_device_init: true if need to set max power in F/W on device init.
814 * struct hl_fence - software synchronization primitive
837 * struct hl_cs_compl - command submission completion object.
844 * @type: type of the CS - signal/wait.
868 * struct hl_ts_buff - describes a timestamp buffer.
882 * struct hl_mem_mgr - describes unified memory manager for mappable memory chunks.
894 * struct hl_mmap_mem_buf_behavior - describes unified memory manager buffer behavior
913 * struct hl_mmap_mem_buf - describes a single unified memory buffer
937 * struct hl_cb - describes a Command Buffer.
995 /* Host <-> CPU-CP shared memory size */
999 * struct hl_sync_stream_properties -
1026 * struct hl_encaps_signals_mgr - describes sync stream encapsulated signals
1037 * struct hl_hw_queue - describes a H/W transport queue.
1072 * struct hl_cq - describes a completion queue
1101 * struct hl_user_interrupt - holds user interrupt information
1119 * struct timestamp_reg_free_node - holds the timestamp registration free objects node
1130 /* struct timestamp_reg_work_obj - holds the timestamp registration free objects job
1143 /* struct timestamp_reg_info - holds the timestamp registration related data.
1163 * struct hl_user_pending_interrupt - holds a context to a user thread
1181 * struct hl_eq - describes the event queue (single one per device)
1202 * struct hl_dec - describes a decoder sw instance.
1216 * enum hl_asic_type - supported ASIC types.
1218 * @ASIC_GOYA: Goya device (HL-1000).
1219 * @ASIC_GAUDI: Gaudi device (HL-2000).
1220 * @ASIC_GAUDI_SEC: Gaudi secured device (HL-2000).
1236 * enum hl_pm_mng_profile - power management profile.
1248 * enum hl_pll_frequency - PLL frequency.
1286 * struct pci_mem_region - describe memory region in a PCI bar
1304 * struct static_fw_load_mgr - static FW load manager
1305 * @preboot_version_max_off: max offset to preboot version
1306 * @boot_fit_version_max_off: max offset to boot fit version
1307 * @kmd_msg_to_cpu_reg: register address for KDM->CPU messages
1336 * struct fw_response - FW response to LKD command
1348 * struct dynamic_fw_load_mgr - dynamic FW load manager
1366 * struct pre_fw_load_props - needed properties for pre-FW load
1384 * struct fw_image_props - properties of FW image
1396 * struct fw_load_mgr - manager FW loading process
1429 * struct engines_data - asic engines data
1441 * struct hl_asic_funcs - ASIC specific functions that are can be called from
1445 * @late_init: sets up late driver/hw state (post hw_init) - Optional.
1446 * @late_fini: tears down what was done in late_init (pre hw_fini) - Optional.
1458 * @pqe_write: Write the PQ entry to the PQ. This is ASIC-specific
1482 * @hl_dma_unmap_sgtable: DMA unmap scatter-gather table.
1484 * @asic_dma_map_sgtable: DMA map scatter-gather table.
1492 * @handle_eqe: handle event queue entry (IRQ) from CPU-CP.
1498 * @mmu_invalidate_cache_range: flush specific MMU STLB cache lines with ASID-VA-size mask.
1499 * @mmu_prefetch_cache_range: pre-fetch specific MMU STLB cache lines with ASID-VA-size mask.
1500 * @send_heartbeat: send is-alive packet to CPU-CP and verify response.
1522 * @pre_schedule_cs: Perform pre-CS-scheduling operations.
1540 * @descramble_addr: Routine to de-scramble the address prior of
1550 * after every hard-reset of the device
1552 * @get_msi_info: Retrieve asic-specific MSI ID of the f/w async event
1555 * @init_firmware_preload_params: initialize pre FW-load parameters.
1560 * @set_pci_memory_regions: setting properties of PCI memory regions
1721 * enum hl_va_range_type - virtual address range type.
1734 * struct hl_va_range - virtual addresses range.
1750 * struct hl_cs_counters_atomic - command submission counters
1755 * @max_cs_in_flight_drop_cnt: dropped due to maximum CS in-flight
1768 * struct hl_dmabuf_priv - a dma-buf private object.
1769 * @dmabuf: pointer to dma-buf object.
1770 * @ctx: pointer to the dma-buf owner's context.
1771 * @phys_pg_pack: pointer to physical page pack if the dma-buf was exported
1775 * if phys_pg_pack is NULL (dma-buf was exported from address).
1789 * struct hl_cs_outcome - represents a single completed CS outcome
1805 * struct hl_cs_outcome_store - represents a limited store of completed CS outcomes
1809 * @nodes_pool: a static pool of pre-allocated outcome objects
1821 * struct hl_ctx - user/kernel context.
1825 * @hr_mmu_phys_hash: if host-resident MMU is used, holds a mapping from
1826 * MMU-hop-page physical address to its host-resident
1891 * struct hl_ctx_mgr - for handling multiple contexts.
1906 * struct hl_userptr - memory mapping chunk information
1908 * @job_node: linked-list node for hanging the object on the Job's list.
1911 * @sgt: pointer to the scatter-gather table that holds the pages.
1915 * @addr: user-space virtual address of the start of the memory area.
1934 * struct hl_cs - command submission.
2014 * struct hl_cs_job - command submission job.
2021 * @userptr_list: linked-list of userptr mappings that belong to this job and
2034 * handle to a kernel-allocated CB object, false
2064 * struct hl_cs_parser - command submission parser properties.
2068 * @job_userptr_list: linked-list of userptr mappings that belong to the related
2078 * handle to a kernel-allocated CB object, false
2109 * struct hl_vm_hash_node - hash element from virtual address to virtual
2127 * struct hl_vm_hw_block_list_node - list element from user virtual address to
2133 * @mapped_size: size of the block which is mapped. May change if partial un-mappings are done.
2146 * struct hl_vm_phys_pg_pack - physical page pack.
2181 * struct hl_vm_va_block - virtual range block information.
2195 * struct hl_vm - virtual memory manager for MMU.
2217 * struct hl_debug_params - Coresight debug parameters.
2235 * struct hl_notifier_event - holds the notifier data structure
2251 * struct hl_fpriv - process information stored in FD private data.
2287 * struct hl_info_list - debugfs file ops.
2300 * struct hl_debugfs_entry - debugfs dentry wrapper.
2310 * struct hl_dbg_device_entry - ASIC specific debugfs manager.
2374 * struct hl_hw_obj_name_entry - single hw object name, member of
2427 * struct hl_mon_state_dump - represents a state dump of a single monitor
2445 * struct hl_sync_to_engine_map_entry - sync object id to engine mapping entry
2458 * struct hl_sync_to_engine_map - maps sync object id to associated engine id
2467 * struct hl_state_dump_specs_funcs - virtual functions used by the state dump
2489 * struct hl_state_dump_specs - defines ASIC known hw objects names
2525 #define RREG32(reg) hdev->asic_funcs->rreg(hdev, (reg))
2526 #define WREG32(reg, v) hdev->asic_funcs->wreg(hdev, (reg), (v))
2528 hdev->asic_funcs->rreg(hdev, (reg)))
2588 __rc ? __rc : ((cond) ? 0 : -ETIMEDOUT); \
2602 * TODO: use read from PCI bar in other places in the code (SW-91406)
2616 __rc = -EINVAL; \
2618 __elem_bitmask = BIT_ULL(arr_size) - 1; \
2642 __rc ? __rc : ((__elem_bitmask == 0) ? 0 : -ETIMEDOUT); \
2665 * might need to be byte-swapped before returning value to caller.
2691 (cond) ? 0 : -ETIMEDOUT; \
2698 p->address = base; \
2699 p->size = sz; \
2714 * struct hl_device_reset_work - reset work wrapper.
2726 * struct hl_mmu_hr_pgt_priv - used for holding per-device mmu host-resident
2727 * page-table internal information.
2728 * @mmu_pgt_pool: pool of page tables used by a host-resident MMU for
2730 * @mmu_asid_hop0: per-ASID array of host-resident hop0 tables.
2738 * struct hl_mmu_dr_pgt_priv - used for holding per-device mmu device-resident
2739 * page-table internal information.
2749 * struct hl_mmu_priv - used for holding per-device mmu internal information.
2750 * @dr: information on the device-resident MMU, when exists.
2751 * @hr: information on the host-resident MMU, when exists.
2759 * struct hl_mmu_per_hop_info - A structure describing one TLB HOP and its entry
2773 * struct hl_mmu_hop_info - A structure describing the TLB hops and their
2774 * hop-entries that were created in order to translate a virtual address to a
2777 * address replaces the original virtual-address when mapped
2779 * @unscrambled_paddr: The un-scrambled physical address.
2780 * @hop_info: Array holding the per-hop information used for the translation.
2793 * struct hl_hr_mmu_funcs - Device related host resident MMU functions.
2809 * struct hl_mmu_funcs - Device related MMU functions.
2819 * @get_tlb_info: returns the list of hops and hop-entries used that were
2840 * struct hl_prefetch_work - prefetch work structure handler
2843 * @va: virtual address to pre-fetch.
2844 * @size: pre-fetch size.
2864 * struct multi_cs_completion - multi CS wait completion.
2867 * @timestamp: timestamp for the multi-CS completion
2868 * @stream_master_qid_map: bitmap of all stream masters on which the multi-CS
2881 * struct multi_cs_data - internal data for multi CS call
2888 * @completion_bitmap: bitmap of completed CSs (1- completed, otherwise 0)
2890 * @gone_cs: indication of gone CS (1- there was gone CS, otherwise 0)
2891 * @update_ts: update timestamp. 1- update the timestamp, otherwise 0.
2907 * struct hl_clk_throttle_timestamp - current/last clock throttling timestamp
2917 * struct hl_clk_throttle - keeps current/last clock throttling timestamps
2933 * struct user_mapped_block - describes a hw block allowed to be mmapped by user
2943 * struct cs_timeout_info - info of last CS timeout occurred.
2945 * @write_enable: if set writing to CS parameters in the structure is enabled. otherwise - disabled,
2958 * struct undefined_opcode_info - info about last undefined opcode error
2962 * filled with values, it means the execution was in Lower-CP.
2965 * @cb_addr_streams_len: num of streams - actual len of cb_addr_streams array.
2967 * in Upper-CP (specific stream) and equal to 4 in case
2968 * of undefined opcode in Lower-CP.
2969 * @engine_id: engine-id that the error occurred on
2971 * MAX_QMAN_STREAMS_INFO it means the error occurred on a Lower-CP.
2988 * struct page_fault_info - page fault information.
2992 * @page_fault_detected: if set as 1, then a page-fault was discovered for the
2993 * first time after the driver has finished booting-up.
2994 * Since we're looking for the page-fault's root cause,
2995 * we don't care of the others that might follow it-
3008 * struct razwi_info - RAZWI information.
3011 * first time after the driver has finished booting-up.
3013 * we don't care of the others that might follow it-
3024 * struct hw_err_info - HW error information.
3027 * first time after the driver has finished booting-up.
3028 * currently we assume that only fatal events (that require hard-reset) are
3041 * struct fw_err_info - FW error information.
3044 * first time after the driver has finished booting-up.
3045 * currently we assume that only fatal events (that require hard-reset) are
3058 * struct hl_error_info - holds information collected during an error.
3076 * struct hl_reset_info - holds current device reset information.
3083 * @in_compute_reset: Device is currently in reset but not in hard-reset.
3114 * struct hl_device - habanalabs device structure.
3125 * @work_heartbeat: delayed work for CPU-CP is-alive check.
3149 * @prefetch_wq: work queue for MMU pre-fetch operations.
3156 * @event_queue: event queue for IRQ from CPU-CP.
3158 * @cpu_accessible_dma_mem: Host <-> CPU-CP shared memory CPU address.
3159 * @cpu_accessible_dma_address: Host <-> CPU-CP shared memory DMA address.
3160 * @cpu_accessible_dma_pool: Host <-> CPU-CP shared memory pool.
3163 * @send_cpu_message_lock: enforces only one message in Host <-> CPU-CP queue.
3192 * @mmu_priv: device-specific MMU data.
3193 * @mmu_func: device-related MMU functions.
3196 * @pci_mem_region: array of memory regions in the PCI
3198 * @multi_cs_completion: array of multi-CS completion.
3207 * @fw_sw_sub_minor_ver: the sub-minor of current loaded preboot SW version.
3211 * @max_power: the max power of the device, as configured by the sysadmin. This
3212 * value is saved so in case of hard-reset, the driver will restore
3213 * this value and update the F/W after the re-initialization
3232 * drams are binned-out
3234 * tpc engines are binned-out
3235 * @dmabuf_export_cnt: number of dma-buf exporting.
3241 * indicates which decoder engines are binned-out
3243 * indicates which edma engines are binned-out
3246 * which indicates which rotator engines are binned-out(Gaudi3 and above).
3250 * @cpu_pci_msb_addr: 50-bit extension bits for the device CPU's 40-bit
3260 * such as context-switch, context close, page free, etc.
3298 * @hard_reset_on_fw_events: Whether to do device hard-reset when a fatal event is received from
3455 /* Parameters for bring-up to be upstreamed */
3469 * struct hl_cs_encaps_sig_handle - encapsulated signals handle structure
3494 * struct hl_info_fw_err_info - firmware error information structure
3512 * typedef hl_ioctl_t - typedef for ioctl function in the driver
3522 * struct hl_ioctl_desc - describes an IOCTL entry of the driver.
3533 if (hdev->fw_sw_major_ver < fw_sw_major) in hl_is_fw_sw_ver_below()
3535 if (hdev->fw_sw_major_ver > fw_sw_major) in hl_is_fw_sw_ver_below()
3537 if (hdev->fw_sw_minor_ver < fw_sw_minor) in hl_is_fw_sw_ver_below()
3547 * hl_get_sg_info() - get number of pages and the DMA address from SG list.
3559 return ((((*dma_addr) & (PAGE_SIZE - 1)) + sg_dma_len(sg)) + in hl_get_sg_info()
3560 (PAGE_SIZE - 1)) >> PAGE_SHIFT; in hl_get_sg_info()
3564 * hl_mem_area_inside_range() - Checks whether address+size are inside a range.
3586 * hl_mem_area_crosses_range() - Checks whether address+size crossing a range.
3598 u64 end_address = address + size - 1; in hl_mem_area_crosses_range()
3644 #define hl_pi_2_offset(pi) ((pi) & (HL_QUEUE_LENGTH - 1))