Linux Audio

Check our new training course

Loading...
Note: File does not exist in v3.1.
   1/* SPDX-License-Identifier: GPL-2.0
   2 *
   3 * Copyright 2016-2019 HabanaLabs, Ltd.
   4 * All Rights Reserved.
   5 *
   6 */
   7
   8#ifndef HABANALABSP_H_
   9#define HABANALABSP_H_
  10
  11#include "include/armcp_if.h"
  12#include "include/qman_if.h"
  13
  14#include <linux/cdev.h>
  15#include <linux/iopoll.h>
  16#include <linux/irqreturn.h>
  17#include <linux/dma-fence.h>
  18#include <linux/dma-direction.h>
  19#include <linux/scatterlist.h>
  20#include <linux/hashtable.h>
  21
  22#define HL_NAME				"habanalabs"
  23
  24#define HL_MMAP_CB_MASK			(0x8000000000000000ull >> PAGE_SHIFT)
  25
  26#define HL_PENDING_RESET_PER_SEC	5
  27
  28#define HL_DEVICE_TIMEOUT_USEC		1000000 /* 1 s */
  29
  30#define HL_HEARTBEAT_PER_USEC		5000000 /* 5 s */
  31
  32#define HL_PLL_LOW_JOB_FREQ_USEC	5000000 /* 5 s */
  33
  34#define HL_ARMCP_INFO_TIMEOUT_USEC	10000000 /* 10s */
  35#define HL_ARMCP_EEPROM_TIMEOUT_USEC	10000000 /* 10s */
  36
  37#define HL_PCI_ELBI_TIMEOUT_MSEC	10 /* 10ms */
  38
  39#define HL_SIM_MAX_TIMEOUT_US		10000000 /* 10s */
  40
  41#define HL_MAX_QUEUES			128
  42
  43#define HL_MAX_JOBS_PER_CS		64
  44
  45/* MUST BE POWER OF 2 and larger than 1 */
  46#define HL_MAX_PENDING_CS		64
  47
  48#define HL_IDLE_BUSY_TS_ARR_SIZE	4096
  49
  50/* Memory */
  51#define MEM_HASH_TABLE_BITS		7 /* 1 << 7 buckets */
  52
  53/* MMU */
  54#define MMU_HASH_TABLE_BITS		7 /* 1 << 7 buckets */
  55
  56/**
  57 * struct pgt_info - MMU hop page info.
  58 * @node: hash linked-list node for the pgts shadow hash of pgts.
  59 * @phys_addr: physical address of the pgt.
  60 * @shadow_addr: shadow hop in the host.
  61 * @ctx: pointer to the owner ctx.
  62 * @num_of_ptes: indicates how many ptes are used in the pgt.
  63 *
  64 * The MMU page tables hierarchy is placed on the DRAM. When a new level (hop)
  65 * is needed during mapping, a new page is allocated and this structure holds
  66 * its essential information. During unmapping, if no valid PTEs remained in the
  67 * page, it is freed with its pgt_info structure.
  68 */
  69struct pgt_info {
  70	struct hlist_node	node;
  71	u64			phys_addr;
  72	u64			shadow_addr;
  73	struct hl_ctx		*ctx;
  74	int			num_of_ptes;
  75};
  76
  77struct hl_device;
  78struct hl_fpriv;
  79
  80/**
  81 * enum hl_queue_type - Supported QUEUE types.
  82 * @QUEUE_TYPE_NA: queue is not available.
  83 * @QUEUE_TYPE_EXT: external queue which is a DMA channel that may access the
  84 *                  host.
  85 * @QUEUE_TYPE_INT: internal queue that performs DMA inside the device's
  86 *			memories and/or operates the compute engines.
  87 * @QUEUE_TYPE_CPU: S/W queue for communication with the device's CPU.
  88 */
  89enum hl_queue_type {
  90	QUEUE_TYPE_NA,
  91	QUEUE_TYPE_EXT,
  92	QUEUE_TYPE_INT,
  93	QUEUE_TYPE_CPU
  94};
  95
  96/**
  97 * struct hw_queue_properties - queue information.
  98 * @type: queue type.
  99 * @driver_only: true if only the driver is allowed to send a job to this queue,
 100 *               false otherwise.
 101 */
 102struct hw_queue_properties {
 103	enum hl_queue_type	type;
 104	u8			driver_only;
 105};
 106
 107/**
 108 * enum vm_type_t - virtual memory mapping request information.
 109 * @VM_TYPE_USERPTR: mapping of user memory to device virtual address.
 110 * @VM_TYPE_PHYS_PACK: mapping of DRAM memory to device virtual address.
 111 */
 112enum vm_type_t {
 113	VM_TYPE_USERPTR,
 114	VM_TYPE_PHYS_PACK
 115};
 116
 117/**
 118 * enum hl_device_hw_state - H/W device state. use this to understand whether
 119 *                           to do reset before hw_init or not
 120 * @HL_DEVICE_HW_STATE_CLEAN: H/W state is clean. i.e. after hard reset
 121 * @HL_DEVICE_HW_STATE_DIRTY: H/W state is dirty. i.e. we started to execute
 122 *                            hw_init
 123 */
 124enum hl_device_hw_state {
 125	HL_DEVICE_HW_STATE_CLEAN = 0,
 126	HL_DEVICE_HW_STATE_DIRTY
 127};
 128
 129/**
 130 * struct asic_fixed_properties - ASIC specific immutable properties.
 131 * @hw_queues_props: H/W queues properties.
 132 * @armcp_info: received various information from ArmCP regarding the H/W, e.g.
 133 *		available sensors.
 134 * @uboot_ver: F/W U-boot version.
 135 * @preboot_ver: F/W Preboot version.
 136 * @sram_base_address: SRAM physical start address.
 137 * @sram_end_address: SRAM physical end address.
 138 * @sram_user_base_address - SRAM physical start address for user access.
 139 * @dram_base_address: DRAM physical start address.
 140 * @dram_end_address: DRAM physical end address.
 141 * @dram_user_base_address: DRAM physical start address for user access.
 142 * @dram_size: DRAM total size.
 143 * @dram_pci_bar_size: size of PCI bar towards DRAM.
 144 * @max_power_default: max power of the device after reset
 145 * @va_space_host_start_address: base address of virtual memory range for
 146 *                               mapping host memory.
 147 * @va_space_host_end_address: end address of virtual memory range for
 148 *                             mapping host memory.
 149 * @va_space_dram_start_address: base address of virtual memory range for
 150 *                               mapping DRAM memory.
 151 * @va_space_dram_end_address: end address of virtual memory range for
 152 *                             mapping DRAM memory.
 153 * @dram_size_for_default_page_mapping: DRAM size needed to map to avoid page
 154 *                                      fault.
 155 * @pcie_dbi_base_address: Base address of the PCIE_DBI block.
 156 * @pcie_aux_dbi_reg_addr: Address of the PCIE_AUX DBI register.
 157 * @mmu_pgt_addr: base physical address in DRAM of MMU page tables.
 158 * @mmu_dram_default_page_addr: DRAM default page physical address.
 159 * @mmu_pgt_size: MMU page tables total size.
 160 * @mmu_pte_size: PTE size in MMU page tables.
 161 * @mmu_hop_table_size: MMU hop table size.
 162 * @mmu_hop0_tables_total_size: total size of MMU hop0 tables.
 163 * @dram_page_size: page size for MMU DRAM allocation.
 164 * @cfg_size: configuration space size on SRAM.
 165 * @sram_size: total size of SRAM.
 166 * @max_asid: maximum number of open contexts (ASIDs).
 167 * @num_of_events: number of possible internal H/W IRQs.
 168 * @psoc_pci_pll_nr: PCI PLL NR value.
 169 * @psoc_pci_pll_nf: PCI PLL NF value.
 170 * @psoc_pci_pll_od: PCI PLL OD value.
 171 * @psoc_pci_pll_div_factor: PCI PLL DIV FACTOR 1 value.
 172 * @completion_queues_count: number of completion queues.
 173 * @high_pll: high PLL frequency used by the device.
 174 * @cb_pool_cb_cnt: number of CBs in the CB pool.
 175 * @cb_pool_cb_size: size of each CB in the CB pool.
 176 * @tpc_enabled_mask: which TPCs are enabled.
 177 */
 178struct asic_fixed_properties {
 179	struct hw_queue_properties	hw_queues_props[HL_MAX_QUEUES];
 180	struct armcp_info	armcp_info;
 181	char			uboot_ver[VERSION_MAX_LEN];
 182	char			preboot_ver[VERSION_MAX_LEN];
 183	u64			sram_base_address;
 184	u64			sram_end_address;
 185	u64			sram_user_base_address;
 186	u64			dram_base_address;
 187	u64			dram_end_address;
 188	u64			dram_user_base_address;
 189	u64			dram_size;
 190	u64			dram_pci_bar_size;
 191	u64			max_power_default;
 192	u64			va_space_host_start_address;
 193	u64			va_space_host_end_address;
 194	u64			va_space_dram_start_address;
 195	u64			va_space_dram_end_address;
 196	u64			dram_size_for_default_page_mapping;
 197	u64			pcie_dbi_base_address;
 198	u64			pcie_aux_dbi_reg_addr;
 199	u64			mmu_pgt_addr;
 200	u64			mmu_dram_default_page_addr;
 201	u32			mmu_pgt_size;
 202	u32			mmu_pte_size;
 203	u32			mmu_hop_table_size;
 204	u32			mmu_hop0_tables_total_size;
 205	u32			dram_page_size;
 206	u32			cfg_size;
 207	u32			sram_size;
 208	u32			max_asid;
 209	u32			num_of_events;
 210	u32			psoc_pci_pll_nr;
 211	u32			psoc_pci_pll_nf;
 212	u32			psoc_pci_pll_od;
 213	u32			psoc_pci_pll_div_factor;
 214	u32			high_pll;
 215	u32			cb_pool_cb_cnt;
 216	u32			cb_pool_cb_size;
 217	u8			completion_queues_count;
 218	u8			tpc_enabled_mask;
 219};
 220
 221/**
 222 * struct hl_dma_fence - wrapper for fence object used by command submissions.
 223 * @base_fence: kernel fence object.
 224 * @lock: spinlock to protect fence.
 225 * @hdev: habanalabs device structure.
 226 * @cs_seq: command submission sequence number.
 227 */
 228struct hl_dma_fence {
 229	struct dma_fence	base_fence;
 230	spinlock_t		lock;
 231	struct hl_device	*hdev;
 232	u64			cs_seq;
 233};
 234
 235/*
 236 * Command Buffers
 237 */
 238
 239#define HL_MAX_CB_SIZE		0x200000	/* 2MB */
 240
 241/**
 242 * struct hl_cb_mgr - describes a Command Buffer Manager.
 243 * @cb_lock: protects cb_handles.
 244 * @cb_handles: an idr to hold all command buffer handles.
 245 */
 246struct hl_cb_mgr {
 247	spinlock_t		cb_lock;
 248	struct idr		cb_handles; /* protected by cb_lock */
 249};
 250
 251/**
 252 * struct hl_cb - describes a Command Buffer.
 253 * @refcount: reference counter for usage of the CB.
 254 * @hdev: pointer to device this CB belongs to.
 255 * @lock: spinlock to protect mmap/cs flows.
 256 * @debugfs_list: node in debugfs list of command buffers.
 257 * @pool_list: node in pool list of command buffers.
 258 * @kernel_address: Holds the CB's kernel virtual address.
 259 * @bus_address: Holds the CB's DMA address.
 260 * @mmap_size: Holds the CB's size that was mmaped.
 261 * @size: holds the CB's size.
 262 * @id: the CB's ID.
 263 * @cs_cnt: holds number of CS that this CB participates in.
 264 * @ctx_id: holds the ID of the owner's context.
 265 * @mmap: true if the CB is currently mmaped to user.
 266 * @is_pool: true if CB was acquired from the pool, false otherwise.
 267 */
 268struct hl_cb {
 269	struct kref		refcount;
 270	struct hl_device	*hdev;
 271	spinlock_t		lock;
 272	struct list_head	debugfs_list;
 273	struct list_head	pool_list;
 274	u64			kernel_address;
 275	dma_addr_t		bus_address;
 276	u32			mmap_size;
 277	u32			size;
 278	u32			id;
 279	u32			cs_cnt;
 280	u32			ctx_id;
 281	u8			mmap;
 282	u8			is_pool;
 283};
 284
 285
 286/*
 287 * QUEUES
 288 */
 289
 290struct hl_cs_job;
 291
 292/*
 293 * Currently, there are two limitations on the maximum length of a queue:
 294 *
 295 * 1. The memory footprint of the queue. The current allocated space for the
 296 *    queue is PAGE_SIZE. Because each entry in the queue is HL_BD_SIZE,
 297 *    the maximum length of the queue can be PAGE_SIZE / HL_BD_SIZE,
 298 *    which currently is 4096/16 = 256 entries.
 299 *
 300 *    To increase that, we need either to decrease the size of the
 301 *    BD (difficult), or allocate more than a single page (easier).
 302 *
 303 * 2. Because the size of the JOB handle field in the BD CTL / completion queue
 304 *    is 10-bit, we can have up to 1024 open jobs per hardware queue.
 305 *    Therefore, each queue can hold up to 1024 entries.
 306 *
 307 * HL_QUEUE_LENGTH is in units of struct hl_bd.
 308 * HL_QUEUE_LENGTH * sizeof(struct hl_bd) should be <= HL_PAGE_SIZE
 309 */
 310
 311#define HL_PAGE_SIZE			4096 /* minimum page size */
 312/* Must be power of 2 (HL_PAGE_SIZE / HL_BD_SIZE) */
 313#define HL_QUEUE_LENGTH			256
 314#define HL_QUEUE_SIZE_IN_BYTES		(HL_QUEUE_LENGTH * HL_BD_SIZE)
 315
 316/*
 317 * HL_CQ_LENGTH is in units of struct hl_cq_entry.
 318 * HL_CQ_LENGTH should be <= HL_PAGE_SIZE
 319 */
 320#define HL_CQ_LENGTH			HL_QUEUE_LENGTH
 321#define HL_CQ_SIZE_IN_BYTES		(HL_CQ_LENGTH * HL_CQ_ENTRY_SIZE)
 322
 323/* Must be power of 2 (HL_PAGE_SIZE / HL_EQ_ENTRY_SIZE) */
 324#define HL_EQ_LENGTH			64
 325#define HL_EQ_SIZE_IN_BYTES		(HL_EQ_LENGTH * HL_EQ_ENTRY_SIZE)
 326
 327/* Host <-> ArmCP shared memory size */
 328#define HL_CPU_ACCESSIBLE_MEM_SIZE	SZ_2M
 329
 330/**
 331 * struct hl_hw_queue - describes a H/W transport queue.
 332 * @shadow_queue: pointer to a shadow queue that holds pointers to jobs.
 333 * @queue_type: type of queue.
 334 * @kernel_address: holds the queue's kernel virtual address.
 335 * @bus_address: holds the queue's DMA address.
 336 * @pi: holds the queue's pi value.
 337 * @ci: holds the queue's ci value, AS CALCULATED BY THE DRIVER (not real ci).
 338 * @hw_queue_id: the id of the H/W queue.
 339 * @int_queue_len: length of internal queue (number of entries).
 340 * @valid: is the queue valid (we have array of 32 queues, not all of them
 341 *		exists).
 342 */
 343struct hl_hw_queue {
 344	struct hl_cs_job	**shadow_queue;
 345	enum hl_queue_type	queue_type;
 346	u64			kernel_address;
 347	dma_addr_t		bus_address;
 348	u32			pi;
 349	u32			ci;
 350	u32			hw_queue_id;
 351	u16			int_queue_len;
 352	u8			valid;
 353};
 354
 355/**
 356 * struct hl_cq - describes a completion queue
 357 * @hdev: pointer to the device structure
 358 * @kernel_address: holds the queue's kernel virtual address
 359 * @bus_address: holds the queue's DMA address
 360 * @hw_queue_id: the id of the matching H/W queue
 361 * @ci: ci inside the queue
 362 * @pi: pi inside the queue
 363 * @free_slots_cnt: counter of free slots in queue
 364 */
 365struct hl_cq {
 366	struct hl_device	*hdev;
 367	u64			kernel_address;
 368	dma_addr_t		bus_address;
 369	u32			hw_queue_id;
 370	u32			ci;
 371	u32			pi;
 372	atomic_t		free_slots_cnt;
 373};
 374
 375/**
 376 * struct hl_eq - describes the event queue (single one per device)
 377 * @hdev: pointer to the device structure
 378 * @kernel_address: holds the queue's kernel virtual address
 379 * @bus_address: holds the queue's DMA address
 380 * @ci: ci inside the queue
 381 */
 382struct hl_eq {
 383	struct hl_device	*hdev;
 384	u64			kernel_address;
 385	dma_addr_t		bus_address;
 386	u32			ci;
 387};
 388
 389
 390/*
 391 * ASICs
 392 */
 393
 394/**
 395 * enum hl_asic_type - supported ASIC types.
 396 * @ASIC_INVALID: Invalid ASIC type.
 397 * @ASIC_GOYA: Goya device.
 398 */
 399enum hl_asic_type {
 400	ASIC_INVALID,
 401	ASIC_GOYA
 402};
 403
 404struct hl_cs_parser;
 405
 406/**
 407 * enum hl_pm_mng_profile - power management profile.
 408 * @PM_AUTO: internal clock is set by the Linux driver.
 409 * @PM_MANUAL: internal clock is set by the user.
 410 * @PM_LAST: last power management type.
 411 */
 412enum hl_pm_mng_profile {
 413	PM_AUTO = 1,
 414	PM_MANUAL,
 415	PM_LAST
 416};
 417
 418/**
 419 * enum hl_pll_frequency - PLL frequency.
 420 * @PLL_HIGH: high frequency.
 421 * @PLL_LOW: low frequency.
 422 * @PLL_LAST: last frequency values that were configured by the user.
 423 */
 424enum hl_pll_frequency {
 425	PLL_HIGH = 1,
 426	PLL_LOW,
 427	PLL_LAST
 428};
 429
 430/**
 431 * struct hl_asic_funcs - ASIC specific functions that are can be called from
 432 *                        common code.
 433 * @early_init: sets up early driver state (pre sw_init), doesn't configure H/W.
 434 * @early_fini: tears down what was done in early_init.
 435 * @late_init: sets up late driver/hw state (post hw_init) - Optional.
 436 * @late_fini: tears down what was done in late_init (pre hw_fini) - Optional.
 437 * @sw_init: sets up driver state, does not configure H/W.
 438 * @sw_fini: tears down driver state, does not configure H/W.
 439 * @hw_init: sets up the H/W state.
 440 * @hw_fini: tears down the H/W state.
 441 * @halt_engines: halt engines, needed for reset sequence. This also disables
 442 *                interrupts from the device. Should be called before
 443 *                hw_fini and before CS rollback.
 444 * @suspend: handles IP specific H/W or SW changes for suspend.
 445 * @resume: handles IP specific H/W or SW changes for resume.
 446 * @cb_mmap: maps a CB.
 447 * @ring_doorbell: increment PI on a given QMAN.
 448 * @pqe_write: Write the PQ entry to the PQ. This is ASIC-specific
 449 *             function because the PQs are located in different memory areas
 450 *             per ASIC (SRAM, DRAM, Host memory) and therefore, the method of
 451 *             writing the PQE must match the destination memory area
 452 *             properties.
 453 * @asic_dma_alloc_coherent: Allocate coherent DMA memory by calling
 454 *                           dma_alloc_coherent(). This is ASIC function because
 455 *                           its implementation is not trivial when the driver
 456 *                           is loaded in simulation mode (not upstreamed).
 457 * @asic_dma_free_coherent:  Free coherent DMA memory by calling
 458 *                           dma_free_coherent(). This is ASIC function because
 459 *                           its implementation is not trivial when the driver
 460 *                           is loaded in simulation mode (not upstreamed).
 461 * @get_int_queue_base: get the internal queue base address.
 462 * @test_queues: run simple test on all queues for sanity check.
 463 * @asic_dma_pool_zalloc: small DMA allocation of coherent memory from DMA pool.
 464 *                        size of allocation is HL_DMA_POOL_BLK_SIZE.
 465 * @asic_dma_pool_free: free small DMA allocation from pool.
 466 * @cpu_accessible_dma_pool_alloc: allocate CPU PQ packet from DMA pool.
 467 * @cpu_accessible_dma_pool_free: free CPU PQ packet from DMA pool.
 468 * @hl_dma_unmap_sg: DMA unmap scatter-gather list.
 469 * @cs_parser: parse Command Submission.
 470 * @asic_dma_map_sg: DMA map scatter-gather list.
 471 * @get_dma_desc_list_size: get number of LIN_DMA packets required for CB.
 472 * @add_end_of_cb_packets: Add packets to the end of CB, if device requires it.
 473 * @update_eq_ci: update event queue CI.
 474 * @context_switch: called upon ASID context switch.
 475 * @restore_phase_topology: clear all SOBs amd MONs.
 476 * @debugfs_read32: debug interface for reading u32 from DRAM/SRAM.
 477 * @debugfs_write32: debug interface for writing u32 to DRAM/SRAM.
 478 * @add_device_attr: add ASIC specific device attributes.
 479 * @handle_eqe: handle event queue entry (IRQ) from ArmCP.
 480 * @set_pll_profile: change PLL profile (manual/automatic).
 481 * @get_events_stat: retrieve event queue entries histogram.
 482 * @read_pte: read MMU page table entry from DRAM.
 483 * @write_pte: write MMU page table entry to DRAM.
 484 * @mmu_invalidate_cache: flush MMU STLB cache, either with soft (L1 only) or
 485 *                        hard (L0 & L1) flush.
 486 * @mmu_invalidate_cache_range: flush specific MMU STLB cache lines with
 487 *                              ASID-VA-size mask.
 488 * @send_heartbeat: send is-alive packet to ArmCP and verify response.
 489 * @debug_coresight: perform certain actions on Coresight for debugging.
 490 * @is_device_idle: return true if device is idle, false otherwise.
 491 * @soft_reset_late_init: perform certain actions needed after soft reset.
 492 * @hw_queues_lock: acquire H/W queues lock.
 493 * @hw_queues_unlock: release H/W queues lock.
 494 * @get_pci_id: retrieve PCI ID.
 495 * @get_eeprom_data: retrieve EEPROM data from F/W.
 496 * @send_cpu_message: send buffer to ArmCP.
 497 * @get_hw_state: retrieve the H/W state
 498 * @pci_bars_map: Map PCI BARs.
 499 * @set_dram_bar_base: Set DRAM BAR to map specific device address. Returns
 500 *                     old address the bar pointed to or U64_MAX for failure
 501 * @init_iatu: Initialize the iATU unit inside the PCI controller.
 502 * @rreg: Read a register. Needed for simulator support.
 503 * @wreg: Write a register. Needed for simulator support.
 504 * @halt_coresight: stop the ETF and ETR traces.
 505 */
 506struct hl_asic_funcs {
 507	int (*early_init)(struct hl_device *hdev);
 508	int (*early_fini)(struct hl_device *hdev);
 509	int (*late_init)(struct hl_device *hdev);
 510	void (*late_fini)(struct hl_device *hdev);
 511	int (*sw_init)(struct hl_device *hdev);
 512	int (*sw_fini)(struct hl_device *hdev);
 513	int (*hw_init)(struct hl_device *hdev);
 514	void (*hw_fini)(struct hl_device *hdev, bool hard_reset);
 515	void (*halt_engines)(struct hl_device *hdev, bool hard_reset);
 516	int (*suspend)(struct hl_device *hdev);
 517	int (*resume)(struct hl_device *hdev);
 518	int (*cb_mmap)(struct hl_device *hdev, struct vm_area_struct *vma,
 519			u64 kaddress, phys_addr_t paddress, u32 size);
 520	void (*ring_doorbell)(struct hl_device *hdev, u32 hw_queue_id, u32 pi);
 521	void (*pqe_write)(struct hl_device *hdev, __le64 *pqe,
 522			struct hl_bd *bd);
 523	void* (*asic_dma_alloc_coherent)(struct hl_device *hdev, size_t size,
 524					dma_addr_t *dma_handle, gfp_t flag);
 525	void (*asic_dma_free_coherent)(struct hl_device *hdev, size_t size,
 526					void *cpu_addr, dma_addr_t dma_handle);
 527	void* (*get_int_queue_base)(struct hl_device *hdev, u32 queue_id,
 528				dma_addr_t *dma_handle, u16 *queue_len);
 529	int (*test_queues)(struct hl_device *hdev);
 530	void* (*asic_dma_pool_zalloc)(struct hl_device *hdev, size_t size,
 531				gfp_t mem_flags, dma_addr_t *dma_handle);
 532	void (*asic_dma_pool_free)(struct hl_device *hdev, void *vaddr,
 533				dma_addr_t dma_addr);
 534	void* (*cpu_accessible_dma_pool_alloc)(struct hl_device *hdev,
 535				size_t size, dma_addr_t *dma_handle);
 536	void (*cpu_accessible_dma_pool_free)(struct hl_device *hdev,
 537				size_t size, void *vaddr);
 538	void (*hl_dma_unmap_sg)(struct hl_device *hdev,
 539				struct scatterlist *sgl, int nents,
 540				enum dma_data_direction dir);
 541	int (*cs_parser)(struct hl_device *hdev, struct hl_cs_parser *parser);
 542	int (*asic_dma_map_sg)(struct hl_device *hdev,
 543				struct scatterlist *sgl, int nents,
 544				enum dma_data_direction dir);
 545	u32 (*get_dma_desc_list_size)(struct hl_device *hdev,
 546					struct sg_table *sgt);
 547	void (*add_end_of_cb_packets)(struct hl_device *hdev,
 548					u64 kernel_address, u32 len,
 549					u64 cq_addr, u32 cq_val, u32 msix_num);
 550	void (*update_eq_ci)(struct hl_device *hdev, u32 val);
 551	int (*context_switch)(struct hl_device *hdev, u32 asid);
 552	void (*restore_phase_topology)(struct hl_device *hdev);
 553	int (*debugfs_read32)(struct hl_device *hdev, u64 addr, u32 *val);
 554	int (*debugfs_write32)(struct hl_device *hdev, u64 addr, u32 val);
 555	void (*add_device_attr)(struct hl_device *hdev,
 556				struct attribute_group *dev_attr_grp);
 557	void (*handle_eqe)(struct hl_device *hdev,
 558				struct hl_eq_entry *eq_entry);
 559	void (*set_pll_profile)(struct hl_device *hdev,
 560			enum hl_pll_frequency freq);
 561	void* (*get_events_stat)(struct hl_device *hdev, bool aggregate,
 562				u32 *size);
 563	u64 (*read_pte)(struct hl_device *hdev, u64 addr);
 564	void (*write_pte)(struct hl_device *hdev, u64 addr, u64 val);
 565	void (*mmu_invalidate_cache)(struct hl_device *hdev, bool is_hard);
 566	void (*mmu_invalidate_cache_range)(struct hl_device *hdev, bool is_hard,
 567			u32 asid, u64 va, u64 size);
 568	int (*send_heartbeat)(struct hl_device *hdev);
 569	int (*debug_coresight)(struct hl_device *hdev, void *data);
 570	bool (*is_device_idle)(struct hl_device *hdev, u32 *mask,
 571				struct seq_file *s);
 572	int (*soft_reset_late_init)(struct hl_device *hdev);
 573	void (*hw_queues_lock)(struct hl_device *hdev);
 574	void (*hw_queues_unlock)(struct hl_device *hdev);
 575	u32 (*get_pci_id)(struct hl_device *hdev);
 576	int (*get_eeprom_data)(struct hl_device *hdev, void *data,
 577				size_t max_size);
 578	int (*send_cpu_message)(struct hl_device *hdev, u32 *msg,
 579				u16 len, u32 timeout, long *result);
 580	enum hl_device_hw_state (*get_hw_state)(struct hl_device *hdev);
 581	int (*pci_bars_map)(struct hl_device *hdev);
 582	u64 (*set_dram_bar_base)(struct hl_device *hdev, u64 addr);
 583	int (*init_iatu)(struct hl_device *hdev);
 584	u32 (*rreg)(struct hl_device *hdev, u32 reg);
 585	void (*wreg)(struct hl_device *hdev, u32 reg, u32 val);
 586	void (*halt_coresight)(struct hl_device *hdev);
 587};
 588
 589
 590/*
 591 * CONTEXTS
 592 */
 593
 594#define HL_KERNEL_ASID_ID	0
 595
 596/**
 597 * struct hl_va_range - virtual addresses range.
 598 * @lock: protects the virtual addresses list.
 599 * @list: list of virtual addresses blocks available for mappings.
 600 * @start_addr: range start address.
 601 * @end_addr: range end address.
 602 */
 603struct hl_va_range {
 604	struct mutex		lock;
 605	struct list_head	list;
 606	u64			start_addr;
 607	u64			end_addr;
 608};
 609
 610/**
 611 * struct hl_ctx - user/kernel context.
 612 * @mem_hash: holds mapping from virtual address to virtual memory area
 613 *		descriptor (hl_vm_phys_pg_list or hl_userptr).
 614 * @mmu_phys_hash: holds a mapping from physical address to pgt_info structure.
 615 * @mmu_shadow_hash: holds a mapping from shadow address to pgt_info structure.
 616 * @hpriv: pointer to the private (Kernel Driver) data of the process (fd).
 617 * @hdev: pointer to the device structure.
 618 * @refcount: reference counter for the context. Context is released only when
 619 *		this hits 0l. It is incremented on CS and CS_WAIT.
 620 * @cs_pending: array of DMA fence objects representing pending CS.
 621 * @host_va_range: holds available virtual addresses for host mappings.
 622 * @dram_va_range: holds available virtual addresses for DRAM mappings.
 623 * @mem_hash_lock: protects the mem_hash.
 624 * @mmu_lock: protects the MMU page tables. Any change to the PGT, modifing the
 625 *            MMU hash or walking the PGT requires talking this lock
 626 * @debugfs_list: node in debugfs list of contexts.
 627 * @cs_sequence: sequence number for CS. Value is assigned to a CS and passed
 628 *			to user so user could inquire about CS. It is used as
 629 *			index to cs_pending array.
 630 * @dram_default_hops: array that holds all hops addresses needed for default
 631 *                     DRAM mapping.
 632 * @cs_lock: spinlock to protect cs_sequence.
 633 * @dram_phys_mem: amount of used physical DRAM memory by this context.
 634 * @thread_ctx_switch_token: token to prevent multiple threads of the same
 635 *				context	from running the context switch phase.
 636 *				Only a single thread should run it.
 637 * @thread_ctx_switch_wait_token: token to prevent the threads that didn't run
 638 *				the context switch phase from moving to their
 639 *				execution phase before the context switch phase
 640 *				has finished.
 641 * @asid: context's unique address space ID in the device's MMU.
 642 * @handle: context's opaque handle for user
 643 */
 644struct hl_ctx {
 645	DECLARE_HASHTABLE(mem_hash, MEM_HASH_TABLE_BITS);
 646	DECLARE_HASHTABLE(mmu_phys_hash, MMU_HASH_TABLE_BITS);
 647	DECLARE_HASHTABLE(mmu_shadow_hash, MMU_HASH_TABLE_BITS);
 648	struct hl_fpriv		*hpriv;
 649	struct hl_device	*hdev;
 650	struct kref		refcount;
 651	struct dma_fence	*cs_pending[HL_MAX_PENDING_CS];
 652	struct hl_va_range	host_va_range;
 653	struct hl_va_range	dram_va_range;
 654	struct mutex		mem_hash_lock;
 655	struct mutex		mmu_lock;
 656	struct list_head	debugfs_list;
 657	u64			cs_sequence;
 658	u64			*dram_default_hops;
 659	spinlock_t		cs_lock;
 660	atomic64_t		dram_phys_mem;
 661	atomic_t		thread_ctx_switch_token;
 662	u32			thread_ctx_switch_wait_token;
 663	u32			asid;
 664	u32			handle;
 665};
 666
 667/**
 668 * struct hl_ctx_mgr - for handling multiple contexts.
 669 * @ctx_lock: protects ctx_handles.
 670 * @ctx_handles: idr to hold all ctx handles.
 671 */
 672struct hl_ctx_mgr {
 673	struct mutex		ctx_lock;
 674	struct idr		ctx_handles;
 675};
 676
 677
 678
 679/*
 680 * COMMAND SUBMISSIONS
 681 */
 682
 683/**
 684 * struct hl_userptr - memory mapping chunk information
 685 * @vm_type: type of the VM.
 686 * @job_node: linked-list node for hanging the object on the Job's list.
 687 * @vec: pointer to the frame vector.
 688 * @sgt: pointer to the scatter-gather table that holds the pages.
 689 * @dir: for DMA unmapping, the direction must be supplied, so save it.
 690 * @debugfs_list: node in debugfs list of command submissions.
 691 * @addr: user-space virtual pointer to the start of the memory area.
 692 * @size: size of the memory area to pin & map.
 693 * @dma_mapped: true if the SG was mapped to DMA addresses, false otherwise.
 694 */
 695struct hl_userptr {
 696	enum vm_type_t		vm_type; /* must be first */
 697	struct list_head	job_node;
 698	struct frame_vector	*vec;
 699	struct sg_table		*sgt;
 700	enum dma_data_direction dir;
 701	struct list_head	debugfs_list;
 702	u64			addr;
 703	u32			size;
 704	u8			dma_mapped;
 705};
 706
 707/**
 708 * struct hl_cs - command submission.
 709 * @jobs_in_queue_cnt: per each queue, maintain counter of submitted jobs.
 710 * @ctx: the context this CS belongs to.
 711 * @job_list: list of the CS's jobs in the various queues.
 712 * @job_lock: spinlock for the CS's jobs list. Needed for free_job.
 713 * @refcount: reference counter for usage of the CS.
 714 * @fence: pointer to the fence object of this CS.
 715 * @work_tdr: delayed work node for TDR.
 716 * @mirror_node : node in device mirror list of command submissions.
 717 * @debugfs_list: node in debugfs list of command submissions.
 718 * @sequence: the sequence number of this CS.
 719 * @submitted: true if CS was submitted to H/W.
 720 * @completed: true if CS was completed by device.
 721 * @timedout : true if CS was timedout.
 722 * @tdr_active: true if TDR was activated for this CS (to prevent
 723 *		double TDR activation).
 724 * @aborted: true if CS was aborted due to some device error.
 725 */
 726struct hl_cs {
 727	u8			jobs_in_queue_cnt[HL_MAX_QUEUES];
 728	struct hl_ctx		*ctx;
 729	struct list_head	job_list;
 730	spinlock_t		job_lock;
 731	struct kref		refcount;
 732	struct dma_fence	*fence;
 733	struct delayed_work	work_tdr;
 734	struct list_head	mirror_node;
 735	struct list_head	debugfs_list;
 736	u64			sequence;
 737	u8			submitted;
 738	u8			completed;
 739	u8			timedout;
 740	u8			tdr_active;
 741	u8			aborted;
 742};
 743
 744/**
 745 * struct hl_cs_job - command submission job.
 746 * @cs_node: the node to hang on the CS jobs list.
 747 * @cs: the CS this job belongs to.
 748 * @user_cb: the CB we got from the user.
 749 * @patched_cb: in case of patching, this is internal CB which is submitted on
 750 *		the queue instead of the CB we got from the IOCTL.
 751 * @finish_work: workqueue object to run when job is completed.
 752 * @userptr_list: linked-list of userptr mappings that belong to this job and
 753 *			wait for completion.
 754 * @debugfs_list: node in debugfs list of command submission jobs.
 755 * @id: the id of this job inside a CS.
 756 * @hw_queue_id: the id of the H/W queue this job is submitted to.
 757 * @user_cb_size: the actual size of the CB we got from the user.
 758 * @job_cb_size: the actual size of the CB that we put on the queue.
 759 * @ext_queue: whether the job is for external queue or internal queue.
 760 */
 761struct hl_cs_job {
 762	struct list_head	cs_node;
 763	struct hl_cs		*cs;
 764	struct hl_cb		*user_cb;
 765	struct hl_cb		*patched_cb;
 766	struct work_struct	finish_work;
 767	struct list_head	userptr_list;
 768	struct list_head	debugfs_list;
 769	u32			id;
 770	u32			hw_queue_id;
 771	u32			user_cb_size;
 772	u32			job_cb_size;
 773	u8			ext_queue;
 774};
 775
 776/**
 777 * struct hl_cs_parser - command submission paerser properties.
 778 * @user_cb: the CB we got from the user.
 779 * @patched_cb: in case of patching, this is internal CB which is submitted on
 780 *		the queue instead of the CB we got from the IOCTL.
 781 * @job_userptr_list: linked-list of userptr mappings that belong to the related
 782 *			job and wait for completion.
 783 * @cs_sequence: the sequence number of the related CS.
 784 * @ctx_id: the ID of the context the related CS belongs to.
 785 * @hw_queue_id: the id of the H/W queue this job is submitted to.
 786 * @user_cb_size: the actual size of the CB we got from the user.
 787 * @patched_cb_size: the size of the CB after parsing.
 788 * @ext_queue: whether the job is for external queue or internal queue.
 789 * @job_id: the id of the related job inside the related CS.
 790 */
 791struct hl_cs_parser {
 792	struct hl_cb		*user_cb;
 793	struct hl_cb		*patched_cb;
 794	struct list_head	*job_userptr_list;
 795	u64			cs_sequence;
 796	u32			ctx_id;
 797	u32			hw_queue_id;
 798	u32			user_cb_size;
 799	u32			patched_cb_size;
 800	u8			ext_queue;
 801	u8			job_id;
 802};
 803
 804
 805/*
 806 * MEMORY STRUCTURE
 807 */
 808
 809/**
 810 * struct hl_vm_hash_node - hash element from virtual address to virtual
 811 *				memory area descriptor (hl_vm_phys_pg_list or
 812 *				hl_userptr).
 813 * @node: node to hang on the hash table in context object.
 814 * @vaddr: key virtual address.
 815 * @ptr: value pointer (hl_vm_phys_pg_list or hl_userptr).
 816 */
 817struct hl_vm_hash_node {
 818	struct hlist_node	node;
 819	u64			vaddr;
 820	void			*ptr;
 821};
 822
 823/**
 824 * struct hl_vm_phys_pg_pack - physical page pack.
 825 * @vm_type: describes the type of the virtual area descriptor.
 826 * @pages: the physical page array.
 827 * @npages: num physical pages in the pack.
 828 * @total_size: total size of all the pages in this list.
 829 * @mapping_cnt: number of shared mappings.
 830 * @asid: the context related to this list.
 831 * @page_size: size of each page in the pack.
 832 * @flags: HL_MEM_* flags related to this list.
 833 * @handle: the provided handle related to this list.
 834 * @offset: offset from the first page.
 835 * @contiguous: is contiguous physical memory.
 836 * @created_from_userptr: is product of host virtual address.
 837 */
 838struct hl_vm_phys_pg_pack {
 839	enum vm_type_t		vm_type; /* must be first */
 840	u64			*pages;
 841	u64			npages;
 842	u64			total_size;
 843	atomic_t		mapping_cnt;
 844	u32			asid;
 845	u32			page_size;
 846	u32			flags;
 847	u32			handle;
 848	u32			offset;
 849	u8			contiguous;
 850	u8			created_from_userptr;
 851};
 852
 853/**
 854 * struct hl_vm_va_block - virtual range block information.
 855 * @node: node to hang on the virtual range list in context object.
 856 * @start: virtual range start address.
 857 * @end: virtual range end address.
 858 * @size: virtual range size.
 859 */
 860struct hl_vm_va_block {
 861	struct list_head	node;
 862	u64			start;
 863	u64			end;
 864	u64			size;
 865};
 866
 867/**
 868 * struct hl_vm - virtual memory manager for MMU.
 869 * @dram_pg_pool: pool for DRAM physical pages of 2MB.
 870 * @dram_pg_pool_refcount: reference counter for the pool usage.
 871 * @idr_lock: protects the phys_pg_list_handles.
 872 * @phys_pg_pack_handles: idr to hold all device allocations handles.
 873 * @init_done: whether initialization was done. We need this because VM
 874 *		initialization might be skipped during device initialization.
 875 */
 876struct hl_vm {
 877	struct gen_pool		*dram_pg_pool;
 878	struct kref		dram_pg_pool_refcount;
 879	spinlock_t		idr_lock;
 880	struct idr		phys_pg_pack_handles;
 881	u8			init_done;
 882};
 883
 884
 885/*
 886 * DEBUG, PROFILING STRUCTURE
 887 */
 888
 889/**
 890 * struct hl_debug_params - Coresight debug parameters.
 891 * @input: pointer to component specific input parameters.
 892 * @output: pointer to component specific output parameters.
 893 * @output_size: size of output buffer.
 894 * @reg_idx: relevant register ID.
 895 * @op: component operation to execute.
 896 * @enable: true if to enable component debugging, false otherwise.
 897 */
 898struct hl_debug_params {
 899	void *input;
 900	void *output;
 901	u32 output_size;
 902	u32 reg_idx;
 903	u32 op;
 904	bool enable;
 905};
 906
 907/*
 908 * FILE PRIVATE STRUCTURE
 909 */
 910
 911/**
 912 * struct hl_fpriv - process information stored in FD private data.
 913 * @hdev: habanalabs device structure.
 914 * @filp: pointer to the given file structure.
 915 * @taskpid: current process ID.
 916 * @ctx: current executing context. TODO: remove for multiple ctx per process
 917 * @ctx_mgr: context manager to handle multiple context for this FD.
 918 * @cb_mgr: command buffer manager to handle multiple buffers for this FD.
 919 * @debugfs_list: list of relevant ASIC debugfs.
 920 * @dev_node: node in the device list of file private data
 921 * @refcount: number of related contexts.
 922 * @restore_phase_mutex: lock for context switch and restore phase.
 923 * @is_control: true for control device, false otherwise
 924 */
 925struct hl_fpriv {
 926	struct hl_device	*hdev;
 927	struct file		*filp;
 928	struct pid		*taskpid;
 929	struct hl_ctx		*ctx;
 930	struct hl_ctx_mgr	ctx_mgr;
 931	struct hl_cb_mgr	cb_mgr;
 932	struct list_head	debugfs_list;
 933	struct list_head	dev_node;
 934	struct kref		refcount;
 935	struct mutex		restore_phase_mutex;
 936	u8			is_control;
 937};
 938
 939
 940/*
 941 * DebugFS
 942 */
 943
 944/**
 945 * struct hl_info_list - debugfs file ops.
 946 * @name: file name.
 947 * @show: function to output information.
 948 * @write: function to write to the file.
 949 */
 950struct hl_info_list {
 951	const char	*name;
 952	int		(*show)(struct seq_file *s, void *data);
 953	ssize_t		(*write)(struct file *file, const char __user *buf,
 954				size_t count, loff_t *f_pos);
 955};
 956
 957/**
 958 * struct hl_debugfs_entry - debugfs dentry wrapper.
 959 * @dent: base debugfs entry structure.
 960 * @info_ent: dentry realted ops.
 961 * @dev_entry: ASIC specific debugfs manager.
 962 */
 963struct hl_debugfs_entry {
 964	struct dentry			*dent;
 965	const struct hl_info_list	*info_ent;
 966	struct hl_dbg_device_entry	*dev_entry;
 967};
 968
 969/**
 970 * struct hl_dbg_device_entry - ASIC specific debugfs manager.
 971 * @root: root dentry.
 972 * @hdev: habanalabs device structure.
 973 * @entry_arr: array of available hl_debugfs_entry.
 974 * @file_list: list of available debugfs files.
 975 * @file_mutex: protects file_list.
 976 * @cb_list: list of available CBs.
 977 * @cb_spinlock: protects cb_list.
 978 * @cs_list: list of available CSs.
 979 * @cs_spinlock: protects cs_list.
 980 * @cs_job_list: list of available CB jobs.
 981 * @cs_job_spinlock: protects cs_job_list.
 982 * @userptr_list: list of available userptrs (virtual memory chunk descriptor).
 983 * @userptr_spinlock: protects userptr_list.
 984 * @ctx_mem_hash_list: list of available contexts with MMU mappings.
 985 * @ctx_mem_hash_spinlock: protects cb_list.
 986 * @addr: next address to read/write from/to in read/write32.
 987 * @mmu_addr: next virtual address to translate to physical address in mmu_show.
 988 * @mmu_asid: ASID to use while translating in mmu_show.
 989 * @i2c_bus: generic u8 debugfs file for bus value to use in i2c_data_read.
 990 * @i2c_bus: generic u8 debugfs file for address value to use in i2c_data_read.
 991 * @i2c_bus: generic u8 debugfs file for register value to use in i2c_data_read.
 992 */
 993struct hl_dbg_device_entry {
 994	struct dentry			*root;
 995	struct hl_device		*hdev;
 996	struct hl_debugfs_entry		*entry_arr;
 997	struct list_head		file_list;
 998	struct mutex			file_mutex;
 999	struct list_head		cb_list;
1000	spinlock_t			cb_spinlock;
1001	struct list_head		cs_list;
1002	spinlock_t			cs_spinlock;
1003	struct list_head		cs_job_list;
1004	spinlock_t			cs_job_spinlock;
1005	struct list_head		userptr_list;
1006	spinlock_t			userptr_spinlock;
1007	struct list_head		ctx_mem_hash_list;
1008	spinlock_t			ctx_mem_hash_spinlock;
1009	u64				addr;
1010	u64				mmu_addr;
1011	u32				mmu_asid;
1012	u8				i2c_bus;
1013	u8				i2c_addr;
1014	u8				i2c_reg;
1015};
1016
1017
1018/*
1019 * DEVICES
1020 */
1021
1022/* Theoretical limit only. A single host can only contain up to 4 or 8 PCIe
1023 * x16 cards. In extreme cases, there are hosts that can accommodate 16 cards.
1024 */
1025#define HL_MAX_MINORS	256
1026
1027/*
1028 * Registers read & write functions.
1029 */
1030
1031u32 hl_rreg(struct hl_device *hdev, u32 reg);
1032void hl_wreg(struct hl_device *hdev, u32 reg, u32 val);
1033
1034#define RREG32(reg) hdev->asic_funcs->rreg(hdev, (reg))
1035#define WREG32(reg, v) hdev->asic_funcs->wreg(hdev, (reg), (v))
1036#define DREG32(reg) pr_info("REGISTER: " #reg " : 0x%08X\n",	\
1037			hdev->asic_funcs->rreg(hdev, (reg)))
1038
1039#define WREG32_P(reg, val, mask)				\
1040	do {							\
1041		u32 tmp_ = RREG32(reg);				\
1042		tmp_ &= (mask);					\
1043		tmp_ |= ((val) & ~(mask));			\
1044		WREG32(reg, tmp_);				\
1045	} while (0)
1046#define WREG32_AND(reg, and) WREG32_P(reg, 0, and)
1047#define WREG32_OR(reg, or) WREG32_P(reg, or, ~(or))
1048
1049#define REG_FIELD_SHIFT(reg, field) reg##_##field##_SHIFT
1050#define REG_FIELD_MASK(reg, field) reg##_##field##_MASK
1051#define WREG32_FIELD(reg, field, val)	\
1052	WREG32(mm##reg, (RREG32(mm##reg) & ~REG_FIELD_MASK(reg, field)) | \
1053			(val) << REG_FIELD_SHIFT(reg, field))
1054
1055/* Timeout should be longer when working with simulator but cap the
1056 * increased timeout to some maximum
1057 */
1058#define hl_poll_timeout(hdev, addr, val, cond, sleep_us, timeout_us) \
1059({ \
1060	ktime_t __timeout; \
1061	if (hdev->pdev) \
1062		__timeout = ktime_add_us(ktime_get(), timeout_us); \
1063	else \
1064		__timeout = ktime_add_us(ktime_get(),\
1065				min((u64)(timeout_us * 10), \
1066					(u64) HL_SIM_MAX_TIMEOUT_US)); \
1067	might_sleep_if(sleep_us); \
1068	for (;;) { \
1069		(val) = RREG32(addr); \
1070		if (cond) \
1071			break; \
1072		if (timeout_us && ktime_compare(ktime_get(), __timeout) > 0) { \
1073			(val) = RREG32(addr); \
1074			break; \
1075		} \
1076		if (sleep_us) \
1077			usleep_range((sleep_us >> 2) + 1, sleep_us); \
1078	} \
1079	(cond) ? 0 : -ETIMEDOUT; \
1080})
1081
1082/*
1083 * address in this macro points always to a memory location in the
1084 * host's (server's) memory. That location is updated asynchronously
1085 * either by the direct access of the device or by another core.
1086 *
1087 * To work both in LE and BE architectures, we need to distinguish between the
1088 * two states (device or another core updates the memory location). Therefore,
1089 * if mem_written_by_device is true, the host memory being polled will be
1090 * updated directly by the device. If false, the host memory being polled will
1091 * be updated by host CPU. Required so host knows whether or not the memory
1092 * might need to be byte-swapped before returning value to caller.
1093 */
1094#define hl_poll_timeout_memory(hdev, addr, val, cond, sleep_us, timeout_us, \
1095				mem_written_by_device) \
1096({ \
1097	ktime_t __timeout; \
1098	if (hdev->pdev) \
1099		__timeout = ktime_add_us(ktime_get(), timeout_us); \
1100	else \
1101		__timeout = ktime_add_us(ktime_get(),\
1102				min((u64)(timeout_us * 10), \
1103					(u64) HL_SIM_MAX_TIMEOUT_US)); \
1104	might_sleep_if(sleep_us); \
1105	for (;;) { \
1106		/* Verify we read updates done by other cores or by device */ \
1107		mb(); \
1108		(val) = *((u32 *) (uintptr_t) (addr)); \
1109		if (mem_written_by_device) \
1110			(val) = le32_to_cpu(*(__le32 *) &(val)); \
1111		if (cond) \
1112			break; \
1113		if (timeout_us && ktime_compare(ktime_get(), __timeout) > 0) { \
1114			(val) = *((u32 *) (uintptr_t) (addr)); \
1115			if (mem_written_by_device) \
1116				(val) = le32_to_cpu(*(__le32 *) &(val)); \
1117			break; \
1118		} \
1119		if (sleep_us) \
1120			usleep_range((sleep_us >> 2) + 1, sleep_us); \
1121	} \
1122	(cond) ? 0 : -ETIMEDOUT; \
1123})
1124
1125#define hl_poll_timeout_device_memory(hdev, addr, val, cond, sleep_us, \
1126					timeout_us) \
1127({ \
1128	ktime_t __timeout; \
1129	if (hdev->pdev) \
1130		__timeout = ktime_add_us(ktime_get(), timeout_us); \
1131	else \
1132		__timeout = ktime_add_us(ktime_get(),\
1133				min((u64)(timeout_us * 10), \
1134					(u64) HL_SIM_MAX_TIMEOUT_US)); \
1135	might_sleep_if(sleep_us); \
1136	for (;;) { \
1137		(val) = readl(addr); \
1138		if (cond) \
1139			break; \
1140		if (timeout_us && ktime_compare(ktime_get(), __timeout) > 0) { \
1141			(val) = readl(addr); \
1142			break; \
1143		} \
1144		if (sleep_us) \
1145			usleep_range((sleep_us >> 2) + 1, sleep_us); \
1146	} \
1147	(cond) ? 0 : -ETIMEDOUT; \
1148})
1149
1150struct hwmon_chip_info;
1151
1152/**
1153 * struct hl_device_reset_work - reset workqueue task wrapper.
1154 * @reset_work: reset work to be done.
1155 * @hdev: habanalabs device structure.
1156 */
1157struct hl_device_reset_work {
1158	struct work_struct		reset_work;
1159	struct hl_device		*hdev;
1160};
1161
1162/**
1163 * struct hl_device_idle_busy_ts - used for calculating device utilization rate.
1164 * @idle_to_busy_ts: timestamp where device changed from idle to busy.
1165 * @busy_to_idle_ts: timestamp where device changed from busy to idle.
1166 */
1167struct hl_device_idle_busy_ts {
1168	ktime_t				idle_to_busy_ts;
1169	ktime_t				busy_to_idle_ts;
1170};
1171
1172/**
1173 * struct hl_device - habanalabs device structure.
1174 * @pdev: pointer to PCI device, can be NULL in case of simulator device.
1175 * @pcie_bar: array of available PCIe bars.
1176 * @rmmio: configuration area address on SRAM.
1177 * @cdev: related char device.
1178 * @cdev_ctrl: char device for control operations only (INFO IOCTL)
1179 * @dev: related kernel basic device structure.
1180 * @dev_ctrl: related kernel device structure for the control device
1181 * @work_freq: delayed work to lower device frequency if possible.
1182 * @work_heartbeat: delayed work for ArmCP is-alive check.
1183 * @asic_name: ASIC specific nmae.
1184 * @asic_type: ASIC specific type.
1185 * @completion_queue: array of hl_cq.
1186 * @cq_wq: work queue of completion queues for executing work in process context
1187 * @eq_wq: work queue of event queue for executing work in process context.
1188 * @kernel_ctx: Kernel driver context structure.
1189 * @kernel_queues: array of hl_hw_queue.
1190 * @hw_queues_mirror_list: CS mirror list for TDR.
1191 * @hw_queues_mirror_lock: protects hw_queues_mirror_list.
1192 * @kernel_cb_mgr: command buffer manager for creating/destroying/handling CGs.
1193 * @event_queue: event queue for IRQ from ArmCP.
1194 * @dma_pool: DMA pool for small allocations.
1195 * @cpu_accessible_dma_mem: Host <-> ArmCP shared memory CPU address.
1196 * @cpu_accessible_dma_address: Host <-> ArmCP shared memory DMA address.
1197 * @cpu_accessible_dma_pool: Host <-> ArmCP shared memory pool.
1198 * @asid_bitmap: holds used/available ASIDs.
1199 * @asid_mutex: protects asid_bitmap.
1200 * @send_cpu_message_lock: enforces only one message in Host <-> ArmCP queue.
1201 * @debug_lock: protects critical section of setting debug mode for device
1202 * @asic_prop: ASIC specific immutable properties.
1203 * @asic_funcs: ASIC specific functions.
1204 * @asic_specific: ASIC specific information to use only from ASIC files.
1205 * @mmu_pgt_pool: pool of available MMU hops.
1206 * @vm: virtual memory manager for MMU.
1207 * @mmu_cache_lock: protects MMU cache invalidation as it can serve one context.
1208 * @mmu_shadow_hop0: shadow mapping of the MMU hop 0 zone.
1209 * @hwmon_dev: H/W monitor device.
1210 * @pm_mng_profile: current power management profile.
1211 * @hl_chip_info: ASIC's sensors information.
1212 * @hl_debugfs: device's debugfs manager.
1213 * @cb_pool: list of preallocated CBs.
1214 * @cb_pool_lock: protects the CB pool.
1215 * @fpriv_list: list of file private data structures. Each structure is created
1216 *              when a user opens the device
1217 * @fpriv_list_lock: protects the fpriv_list
1218 * @compute_ctx: current compute context executing.
1219 * @idle_busy_ts_arr: array to hold time stamps of transitions from idle to busy
1220 *                    and vice-versa
1221 * @dram_used_mem: current DRAM memory consumption.
1222 * @timeout_jiffies: device CS timeout value.
1223 * @max_power: the max power of the device, as configured by the sysadmin. This
1224 *             value is saved so in case of hard-reset, the driver will restore
1225 *             this value and update the F/W after the re-initialization
1226 * @in_reset: is device in reset flow.
1227 * @curr_pll_profile: current PLL profile.
1228 * @cs_active_cnt: number of active command submissions on this device (active
1229 *                 means already in H/W queues)
1230 * @major: habanalabs kernel driver major.
1231 * @high_pll: high PLL profile frequency.
1232 * @soft_reset_cnt: number of soft reset since the driver was loaded.
1233 * @hard_reset_cnt: number of hard reset since the driver was loaded.
1234 * @idle_busy_ts_idx: index of current entry in idle_busy_ts_arr
1235 * @id: device minor.
1236 * @id_control: minor of the control device
1237 * @disabled: is device disabled.
1238 * @late_init_done: is late init stage was done during initialization.
1239 * @hwmon_initialized: is H/W monitor sensors was initialized.
1240 * @hard_reset_pending: is there a hard reset work pending.
1241 * @heartbeat: is heartbeat sanity check towards ArmCP enabled.
1242 * @reset_on_lockup: true if a reset should be done in case of stuck CS, false
1243 *                   otherwise.
1244 * @dram_supports_virtual_memory: is MMU enabled towards DRAM.
1245 * @dram_default_page_mapping: is DRAM default page mapping enabled.
1246 * @init_done: is the initialization of the device done.
1247 * @mmu_enable: is MMU enabled.
1248 * @device_cpu_disabled: is the device CPU disabled (due to timeouts)
1249 * @dma_mask: the dma mask that was set for this device
1250 * @in_debug: is device under debug. This, together with fpriv_list, enforces
1251 *            that only a single user is configuring the debug infrastructure.
1252 * @cdev_sysfs_created: were char devices and sysfs nodes created.
1253 */
1254struct hl_device {
1255	struct pci_dev			*pdev;
1256	void __iomem			*pcie_bar[6];
1257	void __iomem			*rmmio;
1258	struct cdev			cdev;
1259	struct cdev			cdev_ctrl;
1260	struct device			*dev;
1261	struct device			*dev_ctrl;
1262	struct delayed_work		work_freq;
1263	struct delayed_work		work_heartbeat;
1264	char				asic_name[16];
1265	enum hl_asic_type		asic_type;
1266	struct hl_cq			*completion_queue;
1267	struct workqueue_struct		*cq_wq;
1268	struct workqueue_struct		*eq_wq;
1269	struct hl_ctx			*kernel_ctx;
1270	struct hl_hw_queue		*kernel_queues;
1271	struct list_head		hw_queues_mirror_list;
1272	spinlock_t			hw_queues_mirror_lock;
1273	struct hl_cb_mgr		kernel_cb_mgr;
1274	struct hl_eq			event_queue;
1275	struct dma_pool			*dma_pool;
1276	void				*cpu_accessible_dma_mem;
1277	dma_addr_t			cpu_accessible_dma_address;
1278	struct gen_pool			*cpu_accessible_dma_pool;
1279	unsigned long			*asid_bitmap;
1280	struct mutex			asid_mutex;
1281	struct mutex			send_cpu_message_lock;
1282	struct mutex			debug_lock;
1283	struct asic_fixed_properties	asic_prop;
1284	const struct hl_asic_funcs	*asic_funcs;
1285	void				*asic_specific;
1286	struct gen_pool			*mmu_pgt_pool;
1287	struct hl_vm			vm;
1288	struct mutex			mmu_cache_lock;
1289	void				*mmu_shadow_hop0;
1290	struct device			*hwmon_dev;
1291	enum hl_pm_mng_profile		pm_mng_profile;
1292	struct hwmon_chip_info		*hl_chip_info;
1293
1294	struct hl_dbg_device_entry	hl_debugfs;
1295
1296	struct list_head		cb_pool;
1297	spinlock_t			cb_pool_lock;
1298
1299	struct list_head		fpriv_list;
1300	struct mutex			fpriv_list_lock;
1301
1302	struct hl_ctx			*compute_ctx;
1303
1304	struct hl_device_idle_busy_ts	*idle_busy_ts_arr;
1305
1306	atomic64_t			dram_used_mem;
1307	u64				timeout_jiffies;
1308	u64				max_power;
1309	atomic_t			in_reset;
1310	enum hl_pll_frequency		curr_pll_profile;
1311	int				cs_active_cnt;
1312	u32				major;
1313	u32				high_pll;
1314	u32				soft_reset_cnt;
1315	u32				hard_reset_cnt;
1316	u32				idle_busy_ts_idx;
1317	u16				id;
1318	u16				id_control;
1319	u8				disabled;
1320	u8				late_init_done;
1321	u8				hwmon_initialized;
1322	u8				hard_reset_pending;
1323	u8				heartbeat;
1324	u8				reset_on_lockup;
1325	u8				dram_supports_virtual_memory;
1326	u8				dram_default_page_mapping;
1327	u8				init_done;
1328	u8				device_cpu_disabled;
1329	u8				dma_mask;
1330	u8				in_debug;
1331	u8				cdev_sysfs_created;
1332
1333	/* Parameters for bring-up */
1334	u8				mmu_enable;
1335	u8				cpu_enable;
1336	u8				reset_pcilink;
1337	u8				cpu_queues_enable;
1338	u8				fw_loading;
1339	u8				pldm;
1340};
1341
1342
1343/*
1344 * IOCTLs
1345 */
1346
1347/**
1348 * typedef hl_ioctl_t - typedef for ioctl function in the driver
1349 * @hpriv: pointer to the FD's private data, which contains state of
1350 *		user process
1351 * @data: pointer to the input/output arguments structure of the IOCTL
1352 *
1353 * Return: 0 for success, negative value for error
1354 */
1355typedef int hl_ioctl_t(struct hl_fpriv *hpriv, void *data);
1356
1357/**
1358 * struct hl_ioctl_desc - describes an IOCTL entry of the driver.
1359 * @cmd: the IOCTL code as created by the kernel macros.
1360 * @func: pointer to the driver's function that should be called for this IOCTL.
1361 */
1362struct hl_ioctl_desc {
1363	unsigned int cmd;
1364	hl_ioctl_t *func;
1365};
1366
1367
1368/*
1369 * Kernel module functions that can be accessed by entire module
1370 */
1371
1372/**
1373 * hl_mem_area_inside_range() - Checks whether address+size are inside a range.
1374 * @address: The start address of the area we want to validate.
1375 * @size: The size in bytes of the area we want to validate.
1376 * @range_start_address: The start address of the valid range.
1377 * @range_end_address: The end address of the valid range.
1378 *
1379 * Return: true if the area is inside the valid range, false otherwise.
1380 */
1381static inline bool hl_mem_area_inside_range(u64 address, u32 size,
1382				u64 range_start_address, u64 range_end_address)
1383{
1384	u64 end_address = address + size;
1385
1386	if ((address >= range_start_address) &&
1387			(end_address <= range_end_address) &&
1388			(end_address > address))
1389		return true;
1390
1391	return false;
1392}
1393
1394/**
1395 * hl_mem_area_crosses_range() - Checks whether address+size crossing a range.
1396 * @address: The start address of the area we want to validate.
1397 * @size: The size in bytes of the area we want to validate.
1398 * @range_start_address: The start address of the valid range.
1399 * @range_end_address: The end address of the valid range.
1400 *
1401 * Return: true if the area overlaps part or all of the valid range,
1402 *		false otherwise.
1403 */
1404static inline bool hl_mem_area_crosses_range(u64 address, u32 size,
1405				u64 range_start_address, u64 range_end_address)
1406{
1407	u64 end_address = address + size;
1408
1409	if ((address >= range_start_address) &&
1410			(address < range_end_address))
1411		return true;
1412
1413	if ((end_address >= range_start_address) &&
1414			(end_address < range_end_address))
1415		return true;
1416
1417	if ((address < range_start_address) &&
1418			(end_address >= range_end_address))
1419		return true;
1420
1421	return false;
1422}
1423
1424int hl_device_open(struct inode *inode, struct file *filp);
1425int hl_device_open_ctrl(struct inode *inode, struct file *filp);
1426bool hl_device_disabled_or_in_reset(struct hl_device *hdev);
1427enum hl_device_status hl_device_status(struct hl_device *hdev);
1428int hl_device_set_debug_mode(struct hl_device *hdev, bool enable);
1429int create_hdev(struct hl_device **dev, struct pci_dev *pdev,
1430		enum hl_asic_type asic_type, int minor);
1431void destroy_hdev(struct hl_device *hdev);
1432int hl_hw_queues_create(struct hl_device *hdev);
1433void hl_hw_queues_destroy(struct hl_device *hdev);
1434int hl_hw_queue_send_cb_no_cmpl(struct hl_device *hdev, u32 hw_queue_id,
1435				u32 cb_size, u64 cb_ptr);
1436int hl_hw_queue_schedule_cs(struct hl_cs *cs);
1437u32 hl_hw_queue_add_ptr(u32 ptr, u16 val);
1438void hl_hw_queue_inc_ci_kernel(struct hl_device *hdev, u32 hw_queue_id);
1439void hl_int_hw_queue_update_ci(struct hl_cs *cs);
1440void hl_hw_queue_reset(struct hl_device *hdev, bool hard_reset);
1441
1442#define hl_queue_inc_ptr(p)		hl_hw_queue_add_ptr(p, 1)
1443#define hl_pi_2_offset(pi)		((pi) & (HL_QUEUE_LENGTH - 1))
1444
1445int hl_cq_init(struct hl_device *hdev, struct hl_cq *q, u32 hw_queue_id);
1446void hl_cq_fini(struct hl_device *hdev, struct hl_cq *q);
1447int hl_eq_init(struct hl_device *hdev, struct hl_eq *q);
1448void hl_eq_fini(struct hl_device *hdev, struct hl_eq *q);
1449void hl_cq_reset(struct hl_device *hdev, struct hl_cq *q);
1450void hl_eq_reset(struct hl_device *hdev, struct hl_eq *q);
1451irqreturn_t hl_irq_handler_cq(int irq, void *arg);
1452irqreturn_t hl_irq_handler_eq(int irq, void *arg);
1453u32 hl_cq_inc_ptr(u32 ptr);
1454
1455int hl_asid_init(struct hl_device *hdev);
1456void hl_asid_fini(struct hl_device *hdev);
1457unsigned long hl_asid_alloc(struct hl_device *hdev);
1458void hl_asid_free(struct hl_device *hdev, unsigned long asid);
1459
1460int hl_ctx_create(struct hl_device *hdev, struct hl_fpriv *hpriv);
1461void hl_ctx_free(struct hl_device *hdev, struct hl_ctx *ctx);
1462int hl_ctx_init(struct hl_device *hdev, struct hl_ctx *ctx, bool is_kernel_ctx);
1463void hl_ctx_do_release(struct kref *ref);
1464void hl_ctx_get(struct hl_device *hdev,	struct hl_ctx *ctx);
1465int hl_ctx_put(struct hl_ctx *ctx);
1466struct dma_fence *hl_ctx_get_fence(struct hl_ctx *ctx, u64 seq);
1467void hl_ctx_mgr_init(struct hl_ctx_mgr *mgr);
1468void hl_ctx_mgr_fini(struct hl_device *hdev, struct hl_ctx_mgr *mgr);
1469
1470int hl_device_init(struct hl_device *hdev, struct class *hclass);
1471void hl_device_fini(struct hl_device *hdev);
1472int hl_device_suspend(struct hl_device *hdev);
1473int hl_device_resume(struct hl_device *hdev);
1474int hl_device_reset(struct hl_device *hdev, bool hard_reset,
1475			bool from_hard_reset_thread);
1476void hl_hpriv_get(struct hl_fpriv *hpriv);
1477void hl_hpriv_put(struct hl_fpriv *hpriv);
1478int hl_device_set_frequency(struct hl_device *hdev, enum hl_pll_frequency freq);
1479uint32_t hl_device_utilization(struct hl_device *hdev, uint32_t period_ms);
1480
1481int hl_build_hwmon_channel_info(struct hl_device *hdev,
1482		struct armcp_sensor *sensors_arr);
1483
1484int hl_sysfs_init(struct hl_device *hdev);
1485void hl_sysfs_fini(struct hl_device *hdev);
1486
1487int hl_hwmon_init(struct hl_device *hdev);
1488void hl_hwmon_fini(struct hl_device *hdev);
1489
1490int hl_cb_create(struct hl_device *hdev, struct hl_cb_mgr *mgr, u32 cb_size,
1491		u64 *handle, int ctx_id);
1492int hl_cb_destroy(struct hl_device *hdev, struct hl_cb_mgr *mgr, u64 cb_handle);
1493int hl_cb_mmap(struct hl_fpriv *hpriv, struct vm_area_struct *vma);
1494struct hl_cb *hl_cb_get(struct hl_device *hdev,	struct hl_cb_mgr *mgr,
1495			u32 handle);
1496void hl_cb_put(struct hl_cb *cb);
1497void hl_cb_mgr_init(struct hl_cb_mgr *mgr);
1498void hl_cb_mgr_fini(struct hl_device *hdev, struct hl_cb_mgr *mgr);
1499struct hl_cb *hl_cb_kernel_create(struct hl_device *hdev, u32 cb_size);
1500int hl_cb_pool_init(struct hl_device *hdev);
1501int hl_cb_pool_fini(struct hl_device *hdev);
1502
1503void hl_cs_rollback_all(struct hl_device *hdev);
1504struct hl_cs_job *hl_cs_allocate_job(struct hl_device *hdev, bool ext_queue);
1505
1506void goya_set_asic_funcs(struct hl_device *hdev);
1507
1508int hl_vm_ctx_init(struct hl_ctx *ctx);
1509void hl_vm_ctx_fini(struct hl_ctx *ctx);
1510
1511int hl_vm_init(struct hl_device *hdev);
1512void hl_vm_fini(struct hl_device *hdev);
1513
1514int hl_pin_host_memory(struct hl_device *hdev, u64 addr, u64 size,
1515			struct hl_userptr *userptr);
1516int hl_unpin_host_memory(struct hl_device *hdev, struct hl_userptr *userptr);
1517void hl_userptr_delete_list(struct hl_device *hdev,
1518				struct list_head *userptr_list);
1519bool hl_userptr_is_pinned(struct hl_device *hdev, u64 addr, u32 size,
1520				struct list_head *userptr_list,
1521				struct hl_userptr **userptr);
1522
1523int hl_mmu_init(struct hl_device *hdev);
1524void hl_mmu_fini(struct hl_device *hdev);
1525int hl_mmu_ctx_init(struct hl_ctx *ctx);
1526void hl_mmu_ctx_fini(struct hl_ctx *ctx);
1527int hl_mmu_map(struct hl_ctx *ctx, u64 virt_addr, u64 phys_addr, u32 page_size);
1528int hl_mmu_unmap(struct hl_ctx *ctx, u64 virt_addr, u32 page_size);
1529void hl_mmu_swap_out(struct hl_ctx *ctx);
1530void hl_mmu_swap_in(struct hl_ctx *ctx);
1531
1532int hl_fw_push_fw_to_device(struct hl_device *hdev, const char *fw_name,
1533				void __iomem *dst);
1534int hl_fw_send_pci_access_msg(struct hl_device *hdev, u32 opcode);
1535int hl_fw_send_cpu_message(struct hl_device *hdev, u32 hw_queue_id, u32 *msg,
1536				u16 len, u32 timeout, long *result);
1537int hl_fw_test_cpu_queue(struct hl_device *hdev);
1538void *hl_fw_cpu_accessible_dma_pool_alloc(struct hl_device *hdev, size_t size,
1539						dma_addr_t *dma_handle);
1540void hl_fw_cpu_accessible_dma_pool_free(struct hl_device *hdev, size_t size,
1541					void *vaddr);
1542int hl_fw_send_heartbeat(struct hl_device *hdev);
1543int hl_fw_armcp_info_get(struct hl_device *hdev);
1544int hl_fw_get_eeprom_data(struct hl_device *hdev, void *data, size_t max_size);
1545
1546int hl_pci_bars_map(struct hl_device *hdev, const char * const name[3],
1547			bool is_wc[3]);
1548int hl_pci_iatu_write(struct hl_device *hdev, u32 addr, u32 data);
1549int hl_pci_set_dram_bar_base(struct hl_device *hdev, u8 inbound_region, u8 bar,
1550				u64 addr);
1551int hl_pci_init_iatu(struct hl_device *hdev, u64 sram_base_address,
1552			u64 dram_base_address, u64 host_phys_base_address,
1553			u64 host_phys_size);
1554int hl_pci_init(struct hl_device *hdev, u8 dma_mask);
1555void hl_pci_fini(struct hl_device *hdev);
1556int hl_pci_set_dma_mask(struct hl_device *hdev, u8 dma_mask);
1557
1558long hl_get_frequency(struct hl_device *hdev, u32 pll_index, bool curr);
1559void hl_set_frequency(struct hl_device *hdev, u32 pll_index, u64 freq);
1560long hl_get_temperature(struct hl_device *hdev, int sensor_index, u32 attr);
1561long hl_get_voltage(struct hl_device *hdev, int sensor_index, u32 attr);
1562long hl_get_current(struct hl_device *hdev, int sensor_index, u32 attr);
1563long hl_get_fan_speed(struct hl_device *hdev, int sensor_index, u32 attr);
1564long hl_get_pwm_info(struct hl_device *hdev, int sensor_index, u32 attr);
1565void hl_set_pwm_info(struct hl_device *hdev, int sensor_index, u32 attr,
1566			long value);
1567u64 hl_get_max_power(struct hl_device *hdev);
1568void hl_set_max_power(struct hl_device *hdev, u64 value);
1569
1570#ifdef CONFIG_DEBUG_FS
1571
1572void hl_debugfs_init(void);
1573void hl_debugfs_fini(void);
1574void hl_debugfs_add_device(struct hl_device *hdev);
1575void hl_debugfs_remove_device(struct hl_device *hdev);
1576void hl_debugfs_add_file(struct hl_fpriv *hpriv);
1577void hl_debugfs_remove_file(struct hl_fpriv *hpriv);
1578void hl_debugfs_add_cb(struct hl_cb *cb);
1579void hl_debugfs_remove_cb(struct hl_cb *cb);
1580void hl_debugfs_add_cs(struct hl_cs *cs);
1581void hl_debugfs_remove_cs(struct hl_cs *cs);
1582void hl_debugfs_add_job(struct hl_device *hdev, struct hl_cs_job *job);
1583void hl_debugfs_remove_job(struct hl_device *hdev, struct hl_cs_job *job);
1584void hl_debugfs_add_userptr(struct hl_device *hdev, struct hl_userptr *userptr);
1585void hl_debugfs_remove_userptr(struct hl_device *hdev,
1586				struct hl_userptr *userptr);
1587void hl_debugfs_add_ctx_mem_hash(struct hl_device *hdev, struct hl_ctx *ctx);
1588void hl_debugfs_remove_ctx_mem_hash(struct hl_device *hdev, struct hl_ctx *ctx);
1589
1590#else
1591
1592static inline void __init hl_debugfs_init(void)
1593{
1594}
1595
1596static inline void hl_debugfs_fini(void)
1597{
1598}
1599
1600static inline void hl_debugfs_add_device(struct hl_device *hdev)
1601{
1602}
1603
1604static inline void hl_debugfs_remove_device(struct hl_device *hdev)
1605{
1606}
1607
1608static inline void hl_debugfs_add_file(struct hl_fpriv *hpriv)
1609{
1610}
1611
1612static inline void hl_debugfs_remove_file(struct hl_fpriv *hpriv)
1613{
1614}
1615
1616static inline void hl_debugfs_add_cb(struct hl_cb *cb)
1617{
1618}
1619
1620static inline void hl_debugfs_remove_cb(struct hl_cb *cb)
1621{
1622}
1623
1624static inline void hl_debugfs_add_cs(struct hl_cs *cs)
1625{
1626}
1627
1628static inline void hl_debugfs_remove_cs(struct hl_cs *cs)
1629{
1630}
1631
1632static inline void hl_debugfs_add_job(struct hl_device *hdev,
1633					struct hl_cs_job *job)
1634{
1635}
1636
1637static inline void hl_debugfs_remove_job(struct hl_device *hdev,
1638					struct hl_cs_job *job)
1639{
1640}
1641
1642static inline void hl_debugfs_add_userptr(struct hl_device *hdev,
1643					struct hl_userptr *userptr)
1644{
1645}
1646
1647static inline void hl_debugfs_remove_userptr(struct hl_device *hdev,
1648					struct hl_userptr *userptr)
1649{
1650}
1651
1652static inline void hl_debugfs_add_ctx_mem_hash(struct hl_device *hdev,
1653					struct hl_ctx *ctx)
1654{
1655}
1656
1657static inline void hl_debugfs_remove_ctx_mem_hash(struct hl_device *hdev,
1658					struct hl_ctx *ctx)
1659{
1660}
1661
1662#endif
1663
1664/* IOCTLs */
1665long hl_ioctl(struct file *filep, unsigned int cmd, unsigned long arg);
1666long hl_ioctl_control(struct file *filep, unsigned int cmd, unsigned long arg);
1667int hl_cb_ioctl(struct hl_fpriv *hpriv, void *data);
1668int hl_cs_ioctl(struct hl_fpriv *hpriv, void *data);
1669int hl_cs_wait_ioctl(struct hl_fpriv *hpriv, void *data);
1670int hl_mem_ioctl(struct hl_fpriv *hpriv, void *data);
1671
1672#endif /* HABANALABSP_H_ */