io_uring: punt final io_ring_ctx wait-and-free to workqueue
[linux/fpc-iii.git] / include / uapi / misc / habanalabs.h
blob4faa2c9767e588018b5bdcbb60c80dfab6ccb844
1 /* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note
3 * Copyright 2016-2019 HabanaLabs, Ltd.
4 * All Rights Reserved.
6 */
8 #ifndef HABANALABS_H_
9 #define HABANALABS_H_
11 #include <linux/types.h>
12 #include <linux/ioctl.h>
15 * Defines that are asic-specific but constitutes as ABI between kernel driver
16 * and userspace
18 #define GOYA_KMD_SRAM_RESERVED_SIZE_FROM_START 0x8000 /* 32KB */
21 * Queue Numbering
23 * The external queues (PCI DMA channels) MUST be before the internal queues
24 * and each group (PCI DMA channels and internal) must be contiguous inside
25 * itself but there can be a gap between the two groups (although not
26 * recommended)
29 enum goya_queue_id {
30 GOYA_QUEUE_ID_DMA_0 = 0,
31 GOYA_QUEUE_ID_DMA_1 = 1,
32 GOYA_QUEUE_ID_DMA_2 = 2,
33 GOYA_QUEUE_ID_DMA_3 = 3,
34 GOYA_QUEUE_ID_DMA_4 = 4,
35 GOYA_QUEUE_ID_CPU_PQ = 5,
36 GOYA_QUEUE_ID_MME = 6, /* Internal queues start here */
37 GOYA_QUEUE_ID_TPC0 = 7,
38 GOYA_QUEUE_ID_TPC1 = 8,
39 GOYA_QUEUE_ID_TPC2 = 9,
40 GOYA_QUEUE_ID_TPC3 = 10,
41 GOYA_QUEUE_ID_TPC4 = 11,
42 GOYA_QUEUE_ID_TPC5 = 12,
43 GOYA_QUEUE_ID_TPC6 = 13,
44 GOYA_QUEUE_ID_TPC7 = 14,
45 GOYA_QUEUE_ID_SIZE
49 * Engine Numbering
51 * Used in the "busy_engines_mask" field in `struct hl_info_hw_idle'
54 enum goya_engine_id {
55 GOYA_ENGINE_ID_DMA_0 = 0,
56 GOYA_ENGINE_ID_DMA_1,
57 GOYA_ENGINE_ID_DMA_2,
58 GOYA_ENGINE_ID_DMA_3,
59 GOYA_ENGINE_ID_DMA_4,
60 GOYA_ENGINE_ID_MME_0,
61 GOYA_ENGINE_ID_TPC_0,
62 GOYA_ENGINE_ID_TPC_1,
63 GOYA_ENGINE_ID_TPC_2,
64 GOYA_ENGINE_ID_TPC_3,
65 GOYA_ENGINE_ID_TPC_4,
66 GOYA_ENGINE_ID_TPC_5,
67 GOYA_ENGINE_ID_TPC_6,
68 GOYA_ENGINE_ID_TPC_7,
69 GOYA_ENGINE_ID_SIZE
72 enum hl_device_status {
73 HL_DEVICE_STATUS_OPERATIONAL,
74 HL_DEVICE_STATUS_IN_RESET,
75 HL_DEVICE_STATUS_MALFUNCTION
78 /* Opcode for management ioctl
80 * HW_IP_INFO - Receive information about different IP blocks in the
81 * device.
82 * HL_INFO_HW_EVENTS - Receive an array describing how many times each event
83 * occurred since the last hard reset.
84 * HL_INFO_DRAM_USAGE - Retrieve the dram usage inside the device and of the
85 * specific context. This is relevant only for devices
86 * where the dram is managed by the kernel driver
87 * HL_INFO_HW_IDLE - Retrieve information about the idle status of each
88 * internal engine.
89 * HL_INFO_DEVICE_STATUS - Retrieve the device's status. This opcode doesn't
90 * require an open context.
91 * HL_INFO_DEVICE_UTILIZATION - Retrieve the total utilization of the device
92 * over the last period specified by the user.
93 * The period can be between 100ms to 1s, in
94 * resolution of 100ms. The return value is a
95 * percentage of the utilization rate.
96 * HL_INFO_HW_EVENTS_AGGREGATE - Receive an array describing how many times each
97 * event occurred since the driver was loaded.
98 * HL_INFO_CLK_RATE - Retrieve the current and maximum clock rate
99 * of the device in MHz. The maximum clock rate is
100 * configurable via sysfs parameter
101 * HL_INFO_RESET_COUNT - Retrieve the counts of the soft and hard reset
102 * operations performed on the device since the last
103 * time the driver was loaded.
105 #define HL_INFO_HW_IP_INFO 0
106 #define HL_INFO_HW_EVENTS 1
107 #define HL_INFO_DRAM_USAGE 2
108 #define HL_INFO_HW_IDLE 3
109 #define HL_INFO_DEVICE_STATUS 4
110 #define HL_INFO_DEVICE_UTILIZATION 6
111 #define HL_INFO_HW_EVENTS_AGGREGATE 7
112 #define HL_INFO_CLK_RATE 8
113 #define HL_INFO_RESET_COUNT 9
115 #define HL_INFO_VERSION_MAX_LEN 128
116 #define HL_INFO_CARD_NAME_MAX_LEN 16
118 struct hl_info_hw_ip_info {
119 __u64 sram_base_address;
120 __u64 dram_base_address;
121 __u64 dram_size;
122 __u32 sram_size;
123 __u32 num_of_events;
124 __u32 device_id; /* PCI Device ID */
125 __u32 reserved[3];
126 __u32 armcp_cpld_version;
127 __u32 psoc_pci_pll_nr;
128 __u32 psoc_pci_pll_nf;
129 __u32 psoc_pci_pll_od;
130 __u32 psoc_pci_pll_div_factor;
131 __u8 tpc_enabled_mask;
132 __u8 dram_enabled;
133 __u8 pad[2];
134 __u8 armcp_version[HL_INFO_VERSION_MAX_LEN];
135 __u8 card_name[HL_INFO_CARD_NAME_MAX_LEN];
138 struct hl_info_dram_usage {
139 __u64 dram_free_mem;
140 __u64 ctx_dram_mem;
143 struct hl_info_hw_idle {
144 __u32 is_idle;
146 * Bitmask of busy engines.
147 * Bits definition is according to `enum <chip>_enging_id'.
149 __u32 busy_engines_mask;
152 struct hl_info_device_status {
153 __u32 status;
154 __u32 pad;
157 struct hl_info_device_utilization {
158 __u32 utilization;
159 __u32 pad;
162 struct hl_info_clk_rate {
163 __u32 cur_clk_rate_mhz;
164 __u32 max_clk_rate_mhz;
167 struct hl_info_reset_count {
168 __u32 hard_reset_cnt;
169 __u32 soft_reset_cnt;
172 struct hl_info_args {
173 /* Location of relevant struct in userspace */
174 __u64 return_pointer;
176 * The size of the return value. Just like "size" in "snprintf",
177 * it limits how many bytes the kernel can write
179 * For hw_events array, the size should be
180 * hl_info_hw_ip_info.num_of_events * sizeof(__u32)
182 __u32 return_size;
184 /* HL_INFO_* */
185 __u32 op;
187 union {
188 /* Context ID - Currently not in use */
189 __u32 ctx_id;
190 /* Period value for utilization rate (100ms - 1000ms, in 100ms
191 * resolution.
193 __u32 period_ms;
196 __u32 pad;
199 /* Opcode to create a new command buffer */
200 #define HL_CB_OP_CREATE 0
201 /* Opcode to destroy previously created command buffer */
202 #define HL_CB_OP_DESTROY 1
204 #define HL_MAX_CB_SIZE 0x200000 /* 2MB */
206 struct hl_cb_in {
207 /* Handle of CB or 0 if we want to create one */
208 __u64 cb_handle;
209 /* HL_CB_OP_* */
210 __u32 op;
211 /* Size of CB. Maximum size is HL_MAX_CB_SIZE. The minimum size that
212 * will be allocated, regardless of this parameter's value, is PAGE_SIZE
214 __u32 cb_size;
215 /* Context ID - Currently not in use */
216 __u32 ctx_id;
217 __u32 pad;
220 struct hl_cb_out {
221 /* Handle of CB */
222 __u64 cb_handle;
225 union hl_cb_args {
226 struct hl_cb_in in;
227 struct hl_cb_out out;
231 * This structure size must always be fixed to 64-bytes for backward
232 * compatibility
234 struct hl_cs_chunk {
236 * For external queue, this represents a Handle of CB on the Host
237 * For internal queue, this represents an SRAM or DRAM address of the
238 * internal CB
240 __u64 cb_handle;
241 /* Index of queue to put the CB on */
242 __u32 queue_index;
244 * Size of command buffer with valid packets
245 * Can be smaller then actual CB size
247 __u32 cb_size;
248 /* HL_CS_CHUNK_FLAGS_* */
249 __u32 cs_chunk_flags;
250 /* Align structure to 64 bytes */
251 __u32 pad[11];
254 #define HL_CS_FLAGS_FORCE_RESTORE 0x1
256 #define HL_CS_STATUS_SUCCESS 0
258 #define HL_MAX_JOBS_PER_CS 512
260 struct hl_cs_in {
261 /* this holds address of array of hl_cs_chunk for restore phase */
262 __u64 chunks_restore;
263 /* this holds address of array of hl_cs_chunk for execution phase */
264 __u64 chunks_execute;
265 /* this holds address of array of hl_cs_chunk for store phase -
266 * Currently not in use
268 __u64 chunks_store;
269 /* Number of chunks in restore phase array. Maximum number is
270 * HL_MAX_JOBS_PER_CS
272 __u32 num_chunks_restore;
273 /* Number of chunks in execution array. Maximum number is
274 * HL_MAX_JOBS_PER_CS
276 __u32 num_chunks_execute;
277 /* Number of chunks in restore phase array - Currently not in use */
278 __u32 num_chunks_store;
279 /* HL_CS_FLAGS_* */
280 __u32 cs_flags;
281 /* Context ID - Currently not in use */
282 __u32 ctx_id;
285 struct hl_cs_out {
287 * seq holds the sequence number of the CS to pass to wait ioctl. All
288 * values are valid except for 0 and ULLONG_MAX
290 __u64 seq;
291 /* HL_CS_STATUS_* */
292 __u32 status;
293 __u32 pad;
296 union hl_cs_args {
297 struct hl_cs_in in;
298 struct hl_cs_out out;
301 struct hl_wait_cs_in {
302 /* Command submission sequence number */
303 __u64 seq;
304 /* Absolute timeout to wait in microseconds */
305 __u64 timeout_us;
306 /* Context ID - Currently not in use */
307 __u32 ctx_id;
308 __u32 pad;
311 #define HL_WAIT_CS_STATUS_COMPLETED 0
312 #define HL_WAIT_CS_STATUS_BUSY 1
313 #define HL_WAIT_CS_STATUS_TIMEDOUT 2
314 #define HL_WAIT_CS_STATUS_ABORTED 3
315 #define HL_WAIT_CS_STATUS_INTERRUPTED 4
317 struct hl_wait_cs_out {
318 /* HL_WAIT_CS_STATUS_* */
319 __u32 status;
320 __u32 pad;
323 union hl_wait_cs_args {
324 struct hl_wait_cs_in in;
325 struct hl_wait_cs_out out;
328 /* Opcode to alloc device memory */
329 #define HL_MEM_OP_ALLOC 0
330 /* Opcode to free previously allocated device memory */
331 #define HL_MEM_OP_FREE 1
332 /* Opcode to map host memory */
333 #define HL_MEM_OP_MAP 2
334 /* Opcode to unmap previously mapped host memory */
335 #define HL_MEM_OP_UNMAP 3
337 /* Memory flags */
338 #define HL_MEM_CONTIGUOUS 0x1
339 #define HL_MEM_SHARED 0x2
340 #define HL_MEM_USERPTR 0x4
342 struct hl_mem_in {
343 union {
344 /* HL_MEM_OP_ALLOC- allocate device memory */
345 struct {
346 /* Size to alloc */
347 __u64 mem_size;
348 } alloc;
350 /* HL_MEM_OP_FREE - free device memory */
351 struct {
352 /* Handle returned from HL_MEM_OP_ALLOC */
353 __u64 handle;
354 } free;
356 /* HL_MEM_OP_MAP - map device memory */
357 struct {
359 * Requested virtual address of mapped memory.
360 * The driver will try to map the requested region to
361 * this hint address, as long as the address is valid
362 * and not already mapped. The user should check the
363 * returned address of the IOCTL to make sure he got
364 * the hint address. Passing 0 here means that the
365 * driver will choose the address itself.
367 __u64 hint_addr;
368 /* Handle returned from HL_MEM_OP_ALLOC */
369 __u64 handle;
370 } map_device;
372 /* HL_MEM_OP_MAP - map host memory */
373 struct {
374 /* Address of allocated host memory */
375 __u64 host_virt_addr;
377 * Requested virtual address of mapped memory.
378 * The driver will try to map the requested region to
379 * this hint address, as long as the address is valid
380 * and not already mapped. The user should check the
381 * returned address of the IOCTL to make sure he got
382 * the hint address. Passing 0 here means that the
383 * driver will choose the address itself.
385 __u64 hint_addr;
386 /* Size of allocated host memory */
387 __u64 mem_size;
388 } map_host;
390 /* HL_MEM_OP_UNMAP - unmap host memory */
391 struct {
392 /* Virtual address returned from HL_MEM_OP_MAP */
393 __u64 device_virt_addr;
394 } unmap;
397 /* HL_MEM_OP_* */
398 __u32 op;
399 /* HL_MEM_* flags */
400 __u32 flags;
401 /* Context ID - Currently not in use */
402 __u32 ctx_id;
403 __u32 pad;
406 struct hl_mem_out {
407 union {
409 * Used for HL_MEM_OP_MAP as the virtual address that was
410 * assigned in the device VA space.
411 * A value of 0 means the requested operation failed.
413 __u64 device_virt_addr;
416 * Used for HL_MEM_OP_ALLOC. This is the assigned
417 * handle for the allocated memory
419 __u64 handle;
423 union hl_mem_args {
424 struct hl_mem_in in;
425 struct hl_mem_out out;
428 #define HL_DEBUG_MAX_AUX_VALUES 10
430 struct hl_debug_params_etr {
431 /* Address in memory to allocate buffer */
432 __u64 buffer_address;
434 /* Size of buffer to allocate */
435 __u64 buffer_size;
437 /* Sink operation mode: SW fifo, HW fifo, Circular buffer */
438 __u32 sink_mode;
439 __u32 pad;
442 struct hl_debug_params_etf {
443 /* Address in memory to allocate buffer */
444 __u64 buffer_address;
446 /* Size of buffer to allocate */
447 __u64 buffer_size;
449 /* Sink operation mode: SW fifo, HW fifo, Circular buffer */
450 __u32 sink_mode;
451 __u32 pad;
454 struct hl_debug_params_stm {
455 /* Two bit masks for HW event and Stimulus Port */
456 __u64 he_mask;
457 __u64 sp_mask;
459 /* Trace source ID */
460 __u32 id;
462 /* Frequency for the timestamp register */
463 __u32 frequency;
466 struct hl_debug_params_bmon {
467 /* Two address ranges that the user can request to filter */
468 __u64 start_addr0;
469 __u64 addr_mask0;
471 __u64 start_addr1;
472 __u64 addr_mask1;
474 /* Capture window configuration */
475 __u32 bw_win;
476 __u32 win_capture;
478 /* Trace source ID */
479 __u32 id;
480 __u32 pad;
483 struct hl_debug_params_spmu {
484 /* Event types selection */
485 __u64 event_types[HL_DEBUG_MAX_AUX_VALUES];
487 /* Number of event types selection */
488 __u32 event_types_num;
489 __u32 pad;
492 /* Opcode for ETR component */
493 #define HL_DEBUG_OP_ETR 0
494 /* Opcode for ETF component */
495 #define HL_DEBUG_OP_ETF 1
496 /* Opcode for STM component */
497 #define HL_DEBUG_OP_STM 2
498 /* Opcode for FUNNEL component */
499 #define HL_DEBUG_OP_FUNNEL 3
500 /* Opcode for BMON component */
501 #define HL_DEBUG_OP_BMON 4
502 /* Opcode for SPMU component */
503 #define HL_DEBUG_OP_SPMU 5
504 /* Opcode for timestamp (deprecated) */
505 #define HL_DEBUG_OP_TIMESTAMP 6
506 /* Opcode for setting the device into or out of debug mode. The enable
507 * variable should be 1 for enabling debug mode and 0 for disabling it
509 #define HL_DEBUG_OP_SET_MODE 7
511 struct hl_debug_args {
513 * Pointer to user input structure.
514 * This field is relevant to specific opcodes.
516 __u64 input_ptr;
517 /* Pointer to user output structure */
518 __u64 output_ptr;
519 /* Size of user input structure */
520 __u32 input_size;
521 /* Size of user output structure */
522 __u32 output_size;
523 /* HL_DEBUG_OP_* */
524 __u32 op;
526 * Register index in the component, taken from the debug_regs_index enum
527 * in the various ASIC header files
529 __u32 reg_idx;
530 /* Enable/disable */
531 __u32 enable;
532 /* Context ID - Currently not in use */
533 __u32 ctx_id;
537 * Various information operations such as:
538 * - H/W IP information
539 * - Current dram usage
541 * The user calls this IOCTL with an opcode that describes the required
542 * information. The user should supply a pointer to a user-allocated memory
543 * chunk, which will be filled by the driver with the requested information.
545 * The user supplies the maximum amount of size to copy into the user's memory,
546 * in order to prevent data corruption in case of differences between the
547 * definitions of structures in kernel and userspace, e.g. in case of old
548 * userspace and new kernel driver
550 #define HL_IOCTL_INFO \
551 _IOWR('H', 0x01, struct hl_info_args)
554 * Command Buffer
555 * - Request a Command Buffer
556 * - Destroy a Command Buffer
558 * The command buffers are memory blocks that reside in DMA-able address
559 * space and are physically contiguous so they can be accessed by the device
560 * directly. They are allocated using the coherent DMA API.
562 * When creating a new CB, the IOCTL returns a handle of it, and the user-space
563 * process needs to use that handle to mmap the buffer so it can access them.
566 #define HL_IOCTL_CB \
567 _IOWR('H', 0x02, union hl_cb_args)
570 * Command Submission
572 * To submit work to the device, the user need to call this IOCTL with a set
573 * of JOBS. That set of JOBS constitutes a CS object.
574 * Each JOB will be enqueued on a specific queue, according to the user's input.
575 * There can be more then one JOB per queue.
577 * The CS IOCTL will receive three sets of JOBS. One set is for "restore" phase,
578 * a second set is for "execution" phase and a third set is for "store" phase.
579 * The JOBS on the "restore" phase are enqueued only after context-switch
580 * (or if its the first CS for this context). The user can also order the
581 * driver to run the "restore" phase explicitly
583 * There are two types of queues - external and internal. External queues
584 * are DMA queues which transfer data from/to the Host. All other queues are
585 * internal. The driver will get completion notifications from the device only
586 * on JOBS which are enqueued in the external queues.
588 * For jobs on external queues, the user needs to create command buffers
589 * through the CB ioctl and give the CB's handle to the CS ioctl. For jobs on
590 * internal queues, the user needs to prepare a "command buffer" with packets
591 * on either the SRAM or DRAM, and give the device address of that buffer to
592 * the CS ioctl.
594 * This IOCTL is asynchronous in regard to the actual execution of the CS. This
595 * means it returns immediately after ALL the JOBS were enqueued on their
596 * relevant queues. Therefore, the user mustn't assume the CS has been completed
597 * or has even started to execute.
599 * Upon successful enqueue, the IOCTL returns a sequence number which the user
600 * can use with the "Wait for CS" IOCTL to check whether the handle's CS
601 * external JOBS have been completed. Note that if the CS has internal JOBS
602 * which can execute AFTER the external JOBS have finished, the driver might
603 * report that the CS has finished executing BEFORE the internal JOBS have
604 * actually finish executing.
606 * Even though the sequence number increments per CS, the user can NOT
607 * automatically assume that if CS with sequence number N finished, then CS
608 * with sequence number N-1 also finished. The user can make this assumption if
609 * and only if CS N and CS N-1 are exactly the same (same CBs for the same
610 * queues).
612 #define HL_IOCTL_CS \
613 _IOWR('H', 0x03, union hl_cs_args)
616 * Wait for Command Submission
618 * The user can call this IOCTL with a handle it received from the CS IOCTL
619 * to wait until the handle's CS has finished executing. The user will wait
620 * inside the kernel until the CS has finished or until the user-requested
621 * timeout has expired.
623 * The return value of the IOCTL is a standard Linux error code. The possible
624 * values are:
626 * EINTR - Kernel waiting has been interrupted, e.g. due to OS signal
627 * that the user process received
628 * ETIMEDOUT - The CS has caused a timeout on the device
629 * EIO - The CS was aborted (usually because the device was reset)
630 * ENODEV - The device wants to do hard-reset (so user need to close FD)
632 * The driver also returns a custom define inside the IOCTL which can be:
634 * HL_WAIT_CS_STATUS_COMPLETED - The CS has been completed successfully (0)
635 * HL_WAIT_CS_STATUS_BUSY - The CS is still executing (0)
636 * HL_WAIT_CS_STATUS_TIMEDOUT - The CS has caused a timeout on the device
637 * (ETIMEDOUT)
638 * HL_WAIT_CS_STATUS_ABORTED - The CS was aborted, usually because the
639 * device was reset (EIO)
640 * HL_WAIT_CS_STATUS_INTERRUPTED - Waiting for the CS was interrupted (EINTR)
644 #define HL_IOCTL_WAIT_CS \
645 _IOWR('H', 0x04, union hl_wait_cs_args)
648 * Memory
649 * - Map host memory to device MMU
650 * - Unmap host memory from device MMU
652 * This IOCTL allows the user to map host memory to the device MMU
654 * For host memory, the IOCTL doesn't allocate memory. The user is supposed
655 * to allocate the memory in user-space (malloc/new). The driver pins the
656 * physical pages (up to the allowed limit by the OS), assigns a virtual
657 * address in the device VA space and initializes the device MMU.
659 * There is an option for the user to specify the requested virtual address.
662 #define HL_IOCTL_MEMORY \
663 _IOWR('H', 0x05, union hl_mem_args)
666 * Debug
667 * - Enable/disable the ETR/ETF/FUNNEL/STM/BMON/SPMU debug traces
669 * This IOCTL allows the user to get debug traces from the chip.
671 * Before the user can send configuration requests of the various
672 * debug/profile engines, it needs to set the device into debug mode.
673 * This is because the debug/profile infrastructure is shared component in the
674 * device and we can't allow multiple users to access it at the same time.
676 * Once a user set the device into debug mode, the driver won't allow other
677 * users to "work" with the device, i.e. open a FD. If there are multiple users
678 * opened on the device, the driver won't allow any user to debug the device.
680 * For each configuration request, the user needs to provide the register index
681 * and essential data such as buffer address and size.
683 * Once the user has finished using the debug/profile engines, he should
684 * set the device into non-debug mode, i.e. disable debug mode.
686 * The driver can decide to "kick out" the user if he abuses this interface.
689 #define HL_IOCTL_DEBUG \
690 _IOWR('H', 0x06, struct hl_debug_args)
692 #define HL_COMMAND_START 0x01
693 #define HL_COMMAND_END 0x07
695 #endif /* HABANALABS_H_ */