1 /* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note
3 * Copyright 2016-2018 HabanaLabs, Ltd.
11 #include <linux/types.h>
12 #include <linux/ioctl.h>
15 * Defines that are asic-specific but constitutes as ABI between kernel driver
18 #define GOYA_KMD_SRAM_RESERVED_SIZE_FROM_START 0x8000 /* 32KB */
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
30 GOYA_QUEUE_ID_DMA_0 = 0,
36 GOYA_QUEUE_ID_MME, /* Internal queues start here */
51 * Used in the "busy_engines_mask" field in `struct hl_info_hw_idle'
55 GOYA_ENGINE_ID_DMA_0 = 0,
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 */
79 #define HL_INFO_HW_IP_INFO 0
80 #define HL_INFO_HW_EVENTS 1
81 #define HL_INFO_DRAM_USAGE 2
82 #define HL_INFO_HW_IDLE 3
83 #define HL_INFO_DEVICE_STATUS 4
85 #define HL_INFO_VERSION_MAX_LEN 128
87 struct hl_info_hw_ip_info {
88 __u64 sram_base_address;
89 __u64 dram_base_address;
93 __u32 device_id; /* PCI Device ID */
95 __u32 armcp_cpld_version;
96 __u32 psoc_pci_pll_nr;
97 __u32 psoc_pci_pll_nf;
98 __u32 psoc_pci_pll_od;
99 __u32 psoc_pci_pll_div_factor;
100 __u8 tpc_enabled_mask;
103 __u8 armcp_version[HL_INFO_VERSION_MAX_LEN];
106 struct hl_info_dram_usage {
111 struct hl_info_hw_idle {
114 * Bitmask of busy engines.
115 * Bits definition is according to `enum <chip>_enging_id'.
117 __u32 busy_engines_mask;
120 struct hl_info_device_status {
125 struct hl_info_args {
126 /* Location of relevant struct in userspace */
127 __u64 return_pointer;
129 * The size of the return value. Just like "size" in "snprintf",
130 * it limits how many bytes the kernel can write
132 * For hw_events array, the size should be
133 * hl_info_hw_ip_info.num_of_events * sizeof(__u32)
140 /* Context ID - Currently not in use */
145 /* Opcode to create a new command buffer */
146 #define HL_CB_OP_CREATE 0
147 /* Opcode to destroy previously created command buffer */
148 #define HL_CB_OP_DESTROY 1
151 /* Handle of CB or 0 if we want to create one */
155 /* Size of CB. Maximum size is 2MB. The minimum size that will be
156 * allocated, regardless of this parameter's value, is PAGE_SIZE
159 /* Context ID - Currently not in use */
171 struct hl_cb_out out;
175 * This structure size must always be fixed to 64-bytes for backward
180 * For external queue, this represents a Handle of CB on the Host
181 * For internal queue, this represents an SRAM or DRAM address of the
185 /* Index of queue to put the CB on */
188 * Size of command buffer with valid packets
189 * Can be smaller then actual CB size
192 /* HL_CS_CHUNK_FLAGS_* */
193 __u32 cs_chunk_flags;
194 /* Align structure to 64 bytes */
198 #define HL_CS_FLAGS_FORCE_RESTORE 0x1
200 #define HL_CS_STATUS_SUCCESS 0
203 /* this holds address of array of hl_cs_chunk for restore phase */
204 __u64 chunks_restore;
205 /* this holds address of array of hl_cs_chunk for execution phase */
206 __u64 chunks_execute;
207 /* this holds address of array of hl_cs_chunk for store phase -
208 * Currently not in use
211 /* Number of chunks in restore phase array */
212 __u32 num_chunks_restore;
213 /* Number of chunks in execution array */
214 __u32 num_chunks_execute;
215 /* Number of chunks in restore phase array - Currently not in use */
216 __u32 num_chunks_store;
219 /* Context ID - Currently not in use */
225 * seq holds the sequence number of the CS to pass to wait ioctl. All
226 * values are valid except for 0 and ULLONG_MAX
236 struct hl_cs_out out;
239 struct hl_wait_cs_in {
240 /* Command submission sequence number */
242 /* Absolute timeout to wait in microseconds */
244 /* Context ID - Currently not in use */
249 #define HL_WAIT_CS_STATUS_COMPLETED 0
250 #define HL_WAIT_CS_STATUS_BUSY 1
251 #define HL_WAIT_CS_STATUS_TIMEDOUT 2
252 #define HL_WAIT_CS_STATUS_ABORTED 3
253 #define HL_WAIT_CS_STATUS_INTERRUPTED 4
255 struct hl_wait_cs_out {
256 /* HL_WAIT_CS_STATUS_* */
261 union hl_wait_cs_args {
262 struct hl_wait_cs_in in;
263 struct hl_wait_cs_out out;
266 /* Opcode to alloc device memory */
267 #define HL_MEM_OP_ALLOC 0
268 /* Opcode to free previously allocated device memory */
269 #define HL_MEM_OP_FREE 1
270 /* Opcode to map host memory */
271 #define HL_MEM_OP_MAP 2
272 /* Opcode to unmap previously mapped host memory */
273 #define HL_MEM_OP_UNMAP 3
276 #define HL_MEM_CONTIGUOUS 0x1
277 #define HL_MEM_SHARED 0x2
278 #define HL_MEM_USERPTR 0x4
282 /* HL_MEM_OP_ALLOC- allocate device memory */
288 /* HL_MEM_OP_FREE - free device memory */
290 /* Handle returned from HL_MEM_OP_ALLOC */
294 /* HL_MEM_OP_MAP - map device memory */
297 * Requested virtual address of mapped memory.
298 * KMD will try to map the requested region to this
299 * hint address, as long as the address is valid and
300 * not already mapped. The user should check the
301 * returned address of the IOCTL to make sure he got
302 * the hint address. Passing 0 here means that KMD
303 * will choose the address itself.
306 /* Handle returned from HL_MEM_OP_ALLOC */
310 /* HL_MEM_OP_MAP - map host memory */
312 /* Address of allocated host memory */
313 __u64 host_virt_addr;
315 * Requested virtual address of mapped memory.
316 * KMD will try to map the requested region to this
317 * hint address, as long as the address is valid and
318 * not already mapped. The user should check the
319 * returned address of the IOCTL to make sure he got
320 * the hint address. Passing 0 here means that KMD
321 * will choose the address itself.
324 /* Size of allocated host memory */
328 /* HL_MEM_OP_UNMAP - unmap host memory */
330 /* Virtual address returned from HL_MEM_OP_MAP */
331 __u64 device_virt_addr;
339 /* Context ID - Currently not in use */
347 * Used for HL_MEM_OP_MAP as the virtual address that was
348 * assigned in the device VA space.
349 * A value of 0 means the requested operation failed.
351 __u64 device_virt_addr;
354 * Used for HL_MEM_OP_ALLOC. This is the assigned
355 * handle for the allocated memory
363 struct hl_mem_out out;
366 #define HL_DEBUG_MAX_AUX_VALUES 10
368 struct hl_debug_params_etr {
369 /* Address in memory to allocate buffer */
370 __u64 buffer_address;
372 /* Size of buffer to allocate */
375 /* Sink operation mode: SW fifo, HW fifo, Circular buffer */
380 struct hl_debug_params_etf {
381 /* Address in memory to allocate buffer */
382 __u64 buffer_address;
384 /* Size of buffer to allocate */
387 /* Sink operation mode: SW fifo, HW fifo, Circular buffer */
392 struct hl_debug_params_stm {
393 /* Two bit masks for HW event and Stimulus Port */
397 /* Trace source ID */
400 /* Frequency for the timestamp register */
404 struct hl_debug_params_bmon {
405 /* Two address ranges that the user can request to filter */
412 /* Capture window configuration */
416 /* Trace source ID */
421 struct hl_debug_params_spmu {
422 /* Event types selection */
423 __u64 event_types[HL_DEBUG_MAX_AUX_VALUES];
425 /* Number of event types selection */
426 __u32 event_types_num;
430 /* Opcode for ETR component */
431 #define HL_DEBUG_OP_ETR 0
432 /* Opcode for ETF component */
433 #define HL_DEBUG_OP_ETF 1
434 /* Opcode for STM component */
435 #define HL_DEBUG_OP_STM 2
436 /* Opcode for FUNNEL component */
437 #define HL_DEBUG_OP_FUNNEL 3
438 /* Opcode for BMON component */
439 #define HL_DEBUG_OP_BMON 4
440 /* Opcode for SPMU component */
441 #define HL_DEBUG_OP_SPMU 5
442 /* Opcode for timestamp */
443 #define HL_DEBUG_OP_TIMESTAMP 6
444 /* Opcode for setting the device into or out of debug mode. The enable
445 * variable should be 1 for enabling debug mode and 0 for disabling it
447 #define HL_DEBUG_OP_SET_MODE 7
449 struct hl_debug_args {
451 * Pointer to user input structure.
452 * This field is relevant to specific opcodes.
455 /* Pointer to user output structure */
457 /* Size of user input structure */
459 /* Size of user output structure */
464 * Register index in the component, taken from the debug_regs_index enum
465 * in the various ASIC header files
470 /* Context ID - Currently not in use */
475 * Various information operations such as:
476 * - H/W IP information
477 * - Current dram usage
479 * The user calls this IOCTL with an opcode that describes the required
480 * information. The user should supply a pointer to a user-allocated memory
481 * chunk, which will be filled by the driver with the requested information.
483 * The user supplies the maximum amount of size to copy into the user's memory,
484 * in order to prevent data corruption in case of differences between the
485 * definitions of structures in kernel and userspace, e.g. in case of old
486 * userspace and new kernel driver
488 #define HL_IOCTL_INFO \
489 _IOWR('H', 0x01, struct hl_info_args)
493 * - Request a Command Buffer
494 * - Destroy a Command Buffer
496 * The command buffers are memory blocks that reside in DMA-able address
497 * space and are physically contiguous so they can be accessed by the device
498 * directly. They are allocated using the coherent DMA API.
500 * When creating a new CB, the IOCTL returns a handle of it, and the user-space
501 * process needs to use that handle to mmap the buffer so it can access them.
504 #define HL_IOCTL_CB \
505 _IOWR('H', 0x02, union hl_cb_args)
510 * To submit work to the device, the user need to call this IOCTL with a set
511 * of JOBS. That set of JOBS constitutes a CS object.
512 * Each JOB will be enqueued on a specific queue, according to the user's input.
513 * There can be more then one JOB per queue.
515 * The CS IOCTL will receive three sets of JOBS. One set is for "restore" phase,
516 * a second set is for "execution" phase and a third set is for "store" phase.
517 * The JOBS on the "restore" phase are enqueued only after context-switch
518 * (or if its the first CS for this context). The user can also order the
519 * driver to run the "restore" phase explicitly
521 * There are two types of queues - external and internal. External queues
522 * are DMA queues which transfer data from/to the Host. All other queues are
523 * internal. The driver will get completion notifications from the device only
524 * on JOBS which are enqueued in the external queues.
526 * For jobs on external queues, the user needs to create command buffers
527 * through the CB ioctl and give the CB's handle to the CS ioctl. For jobs on
528 * internal queues, the user needs to prepare a "command buffer" with packets
529 * on either the SRAM or DRAM, and give the device address of that buffer to
532 * This IOCTL is asynchronous in regard to the actual execution of the CS. This
533 * means it returns immediately after ALL the JOBS were enqueued on their
534 * relevant queues. Therefore, the user mustn't assume the CS has been completed
535 * or has even started to execute.
537 * Upon successful enqueue, the IOCTL returns a sequence number which the user
538 * can use with the "Wait for CS" IOCTL to check whether the handle's CS
539 * external JOBS have been completed. Note that if the CS has internal JOBS
540 * which can execute AFTER the external JOBS have finished, the driver might
541 * report that the CS has finished executing BEFORE the internal JOBS have
542 * actually finish executing.
544 * Even though the sequence number increments per CS, the user can NOT
545 * automatically assume that if CS with sequence number N finished, then CS
546 * with sequence number N-1 also finished. The user can make this assumption if
547 * and only if CS N and CS N-1 are exactly the same (same CBs for the same
550 #define HL_IOCTL_CS \
551 _IOWR('H', 0x03, union hl_cs_args)
554 * Wait for Command Submission
556 * The user can call this IOCTL with a handle it received from the CS IOCTL
557 * to wait until the handle's CS has finished executing. The user will wait
558 * inside the kernel until the CS has finished or until the user-requeusted
559 * timeout has expired.
561 * The return value of the IOCTL is a standard Linux error code. The possible
564 * EINTR - Kernel waiting has been interrupted, e.g. due to OS signal
565 * that the user process received
566 * ETIMEDOUT - The CS has caused a timeout on the device
567 * EIO - The CS was aborted (usually because the device was reset)
568 * ENODEV - The device wants to do hard-reset (so user need to close FD)
570 * The driver also returns a custom define inside the IOCTL which can be:
572 * HL_WAIT_CS_STATUS_COMPLETED - The CS has been completed successfully (0)
573 * HL_WAIT_CS_STATUS_BUSY - The CS is still executing (0)
574 * HL_WAIT_CS_STATUS_TIMEDOUT - The CS has caused a timeout on the device
576 * HL_WAIT_CS_STATUS_ABORTED - The CS was aborted, usually because the
577 * device was reset (EIO)
578 * HL_WAIT_CS_STATUS_INTERRUPTED - Waiting for the CS was interrupted (EINTR)
582 #define HL_IOCTL_WAIT_CS \
583 _IOWR('H', 0x04, union hl_wait_cs_args)
587 * - Map host memory to device MMU
588 * - Unmap host memory from device MMU
590 * This IOCTL allows the user to map host memory to the device MMU
592 * For host memory, the IOCTL doesn't allocate memory. The user is supposed
593 * to allocate the memory in user-space (malloc/new). The driver pins the
594 * physical pages (up to the allowed limit by the OS), assigns a virtual
595 * address in the device VA space and initializes the device MMU.
597 * There is an option for the user to specify the requested virtual address.
600 #define HL_IOCTL_MEMORY \
601 _IOWR('H', 0x05, union hl_mem_args)
605 * - Enable/disable the ETR/ETF/FUNNEL/STM/BMON/SPMU debug traces
607 * This IOCTL allows the user to get debug traces from the chip.
609 * Before the user can send configuration requests of the various
610 * debug/profile engines, it needs to set the device into debug mode.
611 * This is because the debug/profile infrastructure is shared component in the
612 * device and we can't allow multiple users to access it at the same time.
614 * Once a user set the device into debug mode, the driver won't allow other
615 * users to "work" with the device, i.e. open a FD. If there are multiple users
616 * opened on the device, the driver won't allow any user to debug the device.
618 * For each configuration request, the user needs to provide the register index
619 * and essential data such as buffer address and size.
621 * Once the user has finished using the debug/profile engines, he should
622 * set the device into non-debug mode, i.e. disable debug mode.
624 * The driver can decide to "kick out" the user if he abuses this interface.
627 #define HL_IOCTL_DEBUG \
628 _IOWR('H', 0x06, struct hl_debug_args)
630 #define HL_COMMAND_START 0x01
631 #define HL_COMMAND_END 0x07
633 #endif /* HABANALABS_H_ */