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 (DMA channels + CPU) MUST be before the internal queues
24 * and each group (DMA channels + CPU 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,
48 /* Opcode for management ioctl */
49 #define HL_INFO_HW_IP_INFO 0
50 #define HL_INFO_HW_EVENTS 1
51 #define HL_INFO_DRAM_USAGE 2
52 #define HL_INFO_HW_IDLE 3
54 #define HL_INFO_VERSION_MAX_LEN 128
56 struct hl_info_hw_ip_info {
57 __u64 sram_base_address;
58 __u64 dram_base_address;
62 __u32 device_id; /* PCI Device ID */
64 __u32 armcp_cpld_version;
65 __u32 psoc_pci_pll_nr;
66 __u32 psoc_pci_pll_nf;
67 __u32 psoc_pci_pll_od;
68 __u32 psoc_pci_pll_div_factor;
69 __u8 tpc_enabled_mask;
72 __u8 armcp_version[HL_INFO_VERSION_MAX_LEN];
75 struct hl_info_dram_usage {
80 struct hl_info_hw_idle {
86 /* Location of relevant struct in userspace */
89 * The size of the return value. Just like "size" in "snprintf",
90 * it limits how many bytes the kernel can write
92 * For hw_events array, the size should be
93 * hl_info_hw_ip_info.num_of_events * sizeof(__u32)
100 /* Context ID - Currently not in use */
105 /* Opcode to create a new command buffer */
106 #define HL_CB_OP_CREATE 0
107 /* Opcode to destroy previously created command buffer */
108 #define HL_CB_OP_DESTROY 1
111 /* Handle of CB or 0 if we want to create one */
115 /* Size of CB. Maximum size is 2MB. The minimum size that will be
116 * allocated, regardless of this parameter's value, is PAGE_SIZE
119 /* Context ID - Currently not in use */
131 struct hl_cb_out out;
135 * This structure size must always be fixed to 64-bytes for backward
140 * For external queue, this represents a Handle of CB on the Host
141 * For internal queue, this represents an SRAM or DRAM address of the
145 /* Index of queue to put the CB on */
148 * Size of command buffer with valid packets
149 * Can be smaller then actual CB size
152 /* HL_CS_CHUNK_FLAGS_* */
153 __u32 cs_chunk_flags;
154 /* Align structure to 64 bytes */
158 #define HL_CS_FLAGS_FORCE_RESTORE 0x1
160 #define HL_CS_STATUS_SUCCESS 0
163 /* this holds address of array of hl_cs_chunk for restore phase */
164 __u64 chunks_restore;
165 /* this holds address of array of hl_cs_chunk for execution phase */
166 __u64 chunks_execute;
167 /* this holds address of array of hl_cs_chunk for store phase -
168 * Currently not in use
171 /* Number of chunks in restore phase array */
172 __u32 num_chunks_restore;
173 /* Number of chunks in execution array */
174 __u32 num_chunks_execute;
175 /* Number of chunks in restore phase array - Currently not in use */
176 __u32 num_chunks_store;
179 /* Context ID - Currently not in use */
184 /* this holds the sequence number of the CS to pass to wait ioctl */
193 struct hl_cs_out out;
196 struct hl_wait_cs_in {
197 /* Command submission sequence number */
199 /* Absolute timeout to wait in microseconds */
201 /* Context ID - Currently not in use */
206 #define HL_WAIT_CS_STATUS_COMPLETED 0
207 #define HL_WAIT_CS_STATUS_BUSY 1
208 #define HL_WAIT_CS_STATUS_TIMEDOUT 2
209 #define HL_WAIT_CS_STATUS_ABORTED 3
210 #define HL_WAIT_CS_STATUS_INTERRUPTED 4
212 struct hl_wait_cs_out {
213 /* HL_WAIT_CS_STATUS_* */
218 union hl_wait_cs_args {
219 struct hl_wait_cs_in in;
220 struct hl_wait_cs_out out;
223 /* Opcode to alloc device memory */
224 #define HL_MEM_OP_ALLOC 0
225 /* Opcode to free previously allocated device memory */
226 #define HL_MEM_OP_FREE 1
227 /* Opcode to map host memory */
228 #define HL_MEM_OP_MAP 2
229 /* Opcode to unmap previously mapped host memory */
230 #define HL_MEM_OP_UNMAP 3
233 #define HL_MEM_CONTIGUOUS 0x1
234 #define HL_MEM_SHARED 0x2
235 #define HL_MEM_USERPTR 0x4
239 /* HL_MEM_OP_ALLOC- allocate device memory */
245 /* HL_MEM_OP_FREE - free device memory */
247 /* Handle returned from HL_MEM_OP_ALLOC */
251 /* HL_MEM_OP_MAP - map device memory */
254 * Requested virtual address of mapped memory.
255 * KMD will try to map the requested region to this
256 * hint address, as long as the address is valid and
257 * not already mapped. The user should check the
258 * returned address of the IOCTL to make sure he got
259 * the hint address. Passing 0 here means that KMD
260 * will choose the address itself.
263 /* Handle returned from HL_MEM_OP_ALLOC */
267 /* HL_MEM_OP_MAP - map host memory */
269 /* Address of allocated host memory */
270 __u64 host_virt_addr;
272 * Requested virtual address of mapped memory.
273 * KMD will try to map the requested region to this
274 * hint address, as long as the address is valid and
275 * not already mapped. The user should check the
276 * returned address of the IOCTL to make sure he got
277 * the hint address. Passing 0 here means that KMD
278 * will choose the address itself.
281 /* Size of allocated host memory */
285 /* HL_MEM_OP_UNMAP - unmap host memory */
287 /* Virtual address returned from HL_MEM_OP_MAP */
288 __u64 device_virt_addr;
296 /* Context ID - Currently not in use */
304 * Used for HL_MEM_OP_MAP as the virtual address that was
305 * assigned in the device VA space.
306 * A value of 0 means the requested operation failed.
308 __u64 device_virt_addr;
311 * Used for HL_MEM_OP_ALLOC. This is the assigned
312 * handle for the allocated memory
320 struct hl_mem_out out;
324 * Various information operations such as:
325 * - H/W IP information
326 * - Current dram usage
328 * The user calls this IOCTL with an opcode that describes the required
329 * information. The user should supply a pointer to a user-allocated memory
330 * chunk, which will be filled by the driver with the requested information.
332 * The user supplies the maximum amount of size to copy into the user's memory,
333 * in order to prevent data corruption in case of differences between the
334 * definitions of structures in kernel and userspace, e.g. in case of old
335 * userspace and new kernel driver
337 #define HL_IOCTL_INFO \
338 _IOWR('H', 0x01, struct hl_info_args)
342 * - Request a Command Buffer
343 * - Destroy a Command Buffer
345 * The command buffers are memory blocks that reside in DMA-able address
346 * space and are physically contiguous so they can be accessed by the device
347 * directly. They are allocated using the coherent DMA API.
349 * When creating a new CB, the IOCTL returns a handle of it, and the user-space
350 * process needs to use that handle to mmap the buffer so it can access them.
353 #define HL_IOCTL_CB \
354 _IOWR('H', 0x02, union hl_cb_args)
359 * To submit work to the device, the user need to call this IOCTL with a set
360 * of JOBS. That set of JOBS constitutes a CS object.
361 * Each JOB will be enqueued on a specific queue, according to the user's input.
362 * There can be more then one JOB per queue.
364 * There are two types of queues - external and internal. External queues
365 * are DMA queues which transfer data from/to the Host. All other queues are
366 * internal. The driver will get completion notifications from the device only
367 * on JOBS which are enqueued in the external queues.
369 * For jobs on external queues, the user needs to create command buffers
370 * through the CB ioctl and give the CB's handle to the CS ioctl. For jobs on
371 * internal queues, the user needs to prepare a "command buffer" with packets
372 * on either the SRAM or DRAM, and give the device address of that buffer to
375 * This IOCTL is asynchronous in regard to the actual execution of the CS. This
376 * means it returns immediately after ALL the JOBS were enqueued on their
377 * relevant queues. Therefore, the user mustn't assume the CS has been completed
378 * or has even started to execute.
380 * Upon successful enqueue, the IOCTL returns an opaque handle which the user
381 * can use with the "Wait for CS" IOCTL to check whether the handle's CS
382 * external JOBS have been completed. Note that if the CS has internal JOBS
383 * which can execute AFTER the external JOBS have finished, the driver might
384 * report that the CS has finished executing BEFORE the internal JOBS have
385 * actually finish executing.
387 * The CS IOCTL will receive three sets of JOBS. One set is for "restore" phase,
388 * a second set is for "execution" phase and a third set is for "store" phase.
389 * The JOBS on the "restore" phase are enqueued only after context-switch
390 * (or if its the first CS for this context). The user can also order the
391 * driver to run the "restore" phase explicitly
394 #define HL_IOCTL_CS \
395 _IOWR('H', 0x03, union hl_cs_args)
398 * Wait for Command Submission
400 * The user can call this IOCTL with a handle it received from the CS IOCTL
401 * to wait until the handle's CS has finished executing. The user will wait
402 * inside the kernel until the CS has finished or until the user-requeusted
403 * timeout has expired.
405 * The return value of the IOCTL is a standard Linux error code. The possible
408 * EINTR - Kernel waiting has been interrupted, e.g. due to OS signal
409 * that the user process received
410 * ETIMEDOUT - The CS has caused a timeout on the device
411 * EIO - The CS was aborted (usually because the device was reset)
412 * ENODEV - The device wants to do hard-reset (so user need to close FD)
414 * The driver also returns a custom define inside the IOCTL which can be:
416 * HL_WAIT_CS_STATUS_COMPLETED - The CS has been completed successfully (0)
417 * HL_WAIT_CS_STATUS_BUSY - The CS is still executing (0)
418 * HL_WAIT_CS_STATUS_TIMEDOUT - The CS has caused a timeout on the device
420 * HL_WAIT_CS_STATUS_ABORTED - The CS was aborted, usually because the
421 * device was reset (EIO)
422 * HL_WAIT_CS_STATUS_INTERRUPTED - Waiting for the CS was interrupted (EINTR)
426 #define HL_IOCTL_WAIT_CS \
427 _IOWR('H', 0x04, union hl_wait_cs_args)
431 * - Map host memory to device MMU
432 * - Unmap host memory from device MMU
434 * This IOCTL allows the user to map host memory to the device MMU
436 * For host memory, the IOCTL doesn't allocate memory. The user is supposed
437 * to allocate the memory in user-space (malloc/new). The driver pins the
438 * physical pages (up to the allowed limit by the OS), assigns a virtual
439 * address in the device VA space and initializes the device MMU.
441 * There is an option for the user to specify the requested virtual address.
444 #define HL_IOCTL_MEMORY \
445 _IOWR('H', 0x05, union hl_mem_args)
447 #define HL_COMMAND_START 0x01
448 #define HL_COMMAND_END 0x06
450 #endif /* HABANALABS_H_ */