1 // SPDX-License-Identifier: GPL-2.0
4 * PCIe r6.0, sec 6.30 DOE
6 * Copyright (C) 2021 Huawei
7 * Jonathan Cameron <Jonathan.Cameron@huawei.com>
9 * Copyright (C) 2022 Intel Corporation
10 * Ira Weiny <ira.weiny@intel.com>
13 #define dev_fmt(fmt) "DOE: " fmt
15 #include <linux/bitfield.h>
16 #include <linux/delay.h>
17 #include <linux/jiffies.h>
18 #include <linux/mutex.h>
19 #include <linux/pci.h>
20 #include <linux/pci-doe.h>
21 #include <linux/workqueue.h>
25 #define PCI_DOE_PROTOCOL_DISCOVERY 0
27 /* Timeout of 1 second from 6.30.2 Operation, PCI Spec r6.0 */
28 #define PCI_DOE_TIMEOUT HZ
29 #define PCI_DOE_POLL_INTERVAL (PCI_DOE_TIMEOUT / 128)
31 #define PCI_DOE_FLAG_CANCEL 0
32 #define PCI_DOE_FLAG_DEAD 1
34 /* Max data object length is 2^18 dwords */
35 #define PCI_DOE_MAX_LENGTH (1 << 18)
38 * struct pci_doe_mb - State for a single DOE mailbox
40 * This state is used to manage a single DOE mailbox capability. All fields
41 * should be considered opaque to the consumers and the structure passed into
42 * the helpers below after being created by pci_doe_create_mb().
44 * @pdev: PCI device this mailbox belongs to
45 * @cap_offset: Capability offset
46 * @prots: Array of protocols supported (encoded as long values)
47 * @wq: Wait queue for work item
48 * @work_queue: Queue of pci_doe_work items
49 * @flags: Bit array of PCI_DOE_FLAG_* flags
57 struct workqueue_struct *work_queue;
61 struct pci_doe_protocol {
67 * struct pci_doe_task - represents a single query/response
70 * @request_pl: The request payload
71 * @request_pl_sz: Size of the request payload (bytes)
72 * @response_pl: The response payload
73 * @response_pl_sz: Size of the response payload (bytes)
74 * @rv: Return value. Length of received response or error (bytes)
75 * @complete: Called when task is complete
76 * @private: Private data for the consumer
77 * @work: Used internally by the mailbox
78 * @doe_mb: Used internally by the mailbox
81 struct pci_doe_protocol prot;
82 const __le32 *request_pl;
85 size_t response_pl_sz;
87 void (*complete)(struct pci_doe_task *task);
90 /* initialized by pci_doe_submit_task() */
91 struct work_struct work;
92 struct pci_doe_mb *doe_mb;
95 static int pci_doe_wait(struct pci_doe_mb *doe_mb, unsigned long timeout)
97 if (wait_event_timeout(doe_mb->wq,
98 test_bit(PCI_DOE_FLAG_CANCEL, &doe_mb->flags),
104 static void pci_doe_write_ctrl(struct pci_doe_mb *doe_mb, u32 val)
106 struct pci_dev *pdev = doe_mb->pdev;
107 int offset = doe_mb->cap_offset;
109 pci_write_config_dword(pdev, offset + PCI_DOE_CTRL, val);
112 static int pci_doe_abort(struct pci_doe_mb *doe_mb)
114 struct pci_dev *pdev = doe_mb->pdev;
115 int offset = doe_mb->cap_offset;
116 unsigned long timeout_jiffies;
118 pci_dbg(pdev, "[%x] Issuing Abort\n", offset);
120 timeout_jiffies = jiffies + PCI_DOE_TIMEOUT;
121 pci_doe_write_ctrl(doe_mb, PCI_DOE_CTRL_ABORT);
127 rc = pci_doe_wait(doe_mb, PCI_DOE_POLL_INTERVAL);
130 pci_read_config_dword(pdev, offset + PCI_DOE_STATUS, &val);
133 if (!FIELD_GET(PCI_DOE_STATUS_ERROR, val) &&
134 !FIELD_GET(PCI_DOE_STATUS_BUSY, val))
137 } while (!time_after(jiffies, timeout_jiffies));
139 /* Abort has timed out and the MB is dead */
140 pci_err(pdev, "[%x] ABORT timed out\n", offset);
144 static int pci_doe_send_req(struct pci_doe_mb *doe_mb,
145 struct pci_doe_task *task)
147 struct pci_dev *pdev = doe_mb->pdev;
148 int offset = doe_mb->cap_offset;
149 size_t length, remainder;
154 * Check the DOE busy bit is not set. If it is set, this could indicate
155 * someone other than Linux (e.g. firmware) is using the mailbox. Note
156 * it is expected that firmware and OS will negotiate access rights via
157 * an, as yet to be defined, method.
159 pci_read_config_dword(pdev, offset + PCI_DOE_STATUS, &val);
160 if (FIELD_GET(PCI_DOE_STATUS_BUSY, val))
163 if (FIELD_GET(PCI_DOE_STATUS_ERROR, val))
166 /* Length is 2 DW of header + length of payload in DW */
167 length = 2 + DIV_ROUND_UP(task->request_pl_sz, sizeof(__le32));
168 if (length > PCI_DOE_MAX_LENGTH)
170 if (length == PCI_DOE_MAX_LENGTH)
173 /* Write DOE Header */
174 val = FIELD_PREP(PCI_DOE_DATA_OBJECT_HEADER_1_VID, task->prot.vid) |
175 FIELD_PREP(PCI_DOE_DATA_OBJECT_HEADER_1_TYPE, task->prot.type);
176 pci_write_config_dword(pdev, offset + PCI_DOE_WRITE, val);
177 pci_write_config_dword(pdev, offset + PCI_DOE_WRITE,
178 FIELD_PREP(PCI_DOE_DATA_OBJECT_HEADER_2_LENGTH,
182 for (i = 0; i < task->request_pl_sz / sizeof(__le32); i++)
183 pci_write_config_dword(pdev, offset + PCI_DOE_WRITE,
184 le32_to_cpu(task->request_pl[i]));
186 /* Write last payload dword */
187 remainder = task->request_pl_sz % sizeof(__le32);
190 memcpy(&val, &task->request_pl[i], remainder);
192 pci_write_config_dword(pdev, offset + PCI_DOE_WRITE, val);
195 pci_doe_write_ctrl(doe_mb, PCI_DOE_CTRL_GO);
200 static bool pci_doe_data_obj_ready(struct pci_doe_mb *doe_mb)
202 struct pci_dev *pdev = doe_mb->pdev;
203 int offset = doe_mb->cap_offset;
206 pci_read_config_dword(pdev, offset + PCI_DOE_STATUS, &val);
207 if (FIELD_GET(PCI_DOE_STATUS_DATA_OBJECT_READY, val))
212 static int pci_doe_recv_resp(struct pci_doe_mb *doe_mb, struct pci_doe_task *task)
214 size_t length, payload_length, remainder, received;
215 struct pci_dev *pdev = doe_mb->pdev;
216 int offset = doe_mb->cap_offset;
220 /* Read the first dword to get the protocol */
221 pci_read_config_dword(pdev, offset + PCI_DOE_READ, &val);
222 if ((FIELD_GET(PCI_DOE_DATA_OBJECT_HEADER_1_VID, val) != task->prot.vid) ||
223 (FIELD_GET(PCI_DOE_DATA_OBJECT_HEADER_1_TYPE, val) != task->prot.type)) {
224 dev_err_ratelimited(&pdev->dev, "[%x] expected [VID, Protocol] = [%04x, %02x], got [%04x, %02x]\n",
225 doe_mb->cap_offset, task->prot.vid, task->prot.type,
226 FIELD_GET(PCI_DOE_DATA_OBJECT_HEADER_1_VID, val),
227 FIELD_GET(PCI_DOE_DATA_OBJECT_HEADER_1_TYPE, val));
231 pci_write_config_dword(pdev, offset + PCI_DOE_READ, 0);
232 /* Read the second dword to get the length */
233 pci_read_config_dword(pdev, offset + PCI_DOE_READ, &val);
234 pci_write_config_dword(pdev, offset + PCI_DOE_READ, 0);
236 length = FIELD_GET(PCI_DOE_DATA_OBJECT_HEADER_2_LENGTH, val);
237 /* A value of 0x0 indicates max data object length */
239 length = PCI_DOE_MAX_LENGTH;
243 /* First 2 dwords have already been read */
245 received = task->response_pl_sz;
246 payload_length = DIV_ROUND_UP(task->response_pl_sz, sizeof(__le32));
247 remainder = task->response_pl_sz % sizeof(__le32);
249 /* remainder signifies number of data bytes in last payload dword */
251 remainder = sizeof(__le32);
253 if (length < payload_length) {
254 received = length * sizeof(__le32);
255 payload_length = length;
256 remainder = sizeof(__le32);
259 if (payload_length) {
260 /* Read all payload dwords except the last */
261 for (; i < payload_length - 1; i++) {
262 pci_read_config_dword(pdev, offset + PCI_DOE_READ,
264 task->response_pl[i] = cpu_to_le32(val);
265 pci_write_config_dword(pdev, offset + PCI_DOE_READ, 0);
268 /* Read last payload dword */
269 pci_read_config_dword(pdev, offset + PCI_DOE_READ, &val);
271 memcpy(&task->response_pl[i], &val, remainder);
272 /* Prior to the last ack, ensure Data Object Ready */
273 if (!pci_doe_data_obj_ready(doe_mb))
275 pci_write_config_dword(pdev, offset + PCI_DOE_READ, 0);
279 /* Flush excess length */
280 for (; i < length; i++) {
281 pci_read_config_dword(pdev, offset + PCI_DOE_READ, &val);
282 pci_write_config_dword(pdev, offset + PCI_DOE_READ, 0);
285 /* Final error check to pick up on any since Data Object Ready */
286 pci_read_config_dword(pdev, offset + PCI_DOE_STATUS, &val);
287 if (FIELD_GET(PCI_DOE_STATUS_ERROR, val))
293 static void signal_task_complete(struct pci_doe_task *task, int rv)
296 destroy_work_on_stack(&task->work);
297 task->complete(task);
300 static void signal_task_abort(struct pci_doe_task *task, int rv)
302 struct pci_doe_mb *doe_mb = task->doe_mb;
303 struct pci_dev *pdev = doe_mb->pdev;
305 if (pci_doe_abort(doe_mb)) {
307 * If the device can't process an abort; set the mailbox dead
308 * - no more submissions
310 pci_err(pdev, "[%x] Abort failed marking mailbox dead\n",
312 set_bit(PCI_DOE_FLAG_DEAD, &doe_mb->flags);
314 signal_task_complete(task, rv);
317 static void doe_statemachine_work(struct work_struct *work)
319 struct pci_doe_task *task = container_of(work, struct pci_doe_task,
321 struct pci_doe_mb *doe_mb = task->doe_mb;
322 struct pci_dev *pdev = doe_mb->pdev;
323 int offset = doe_mb->cap_offset;
324 unsigned long timeout_jiffies;
328 if (test_bit(PCI_DOE_FLAG_DEAD, &doe_mb->flags)) {
329 signal_task_complete(task, -EIO);
334 rc = pci_doe_send_req(doe_mb, task);
337 * The specification does not provide any guidance on how to
338 * resolve conflicting requests from other entities.
339 * Furthermore, it is likely that busy will not be detected
340 * most of the time. Flag any detection of status busy with an
344 dev_err_ratelimited(&pdev->dev, "[%x] busy detected; another entity is sending conflicting requests\n",
346 signal_task_abort(task, rc);
350 timeout_jiffies = jiffies + PCI_DOE_TIMEOUT;
351 /* Poll for response */
353 pci_read_config_dword(pdev, offset + PCI_DOE_STATUS, &val);
354 if (FIELD_GET(PCI_DOE_STATUS_ERROR, val)) {
355 signal_task_abort(task, -EIO);
359 if (!FIELD_GET(PCI_DOE_STATUS_DATA_OBJECT_READY, val)) {
360 if (time_after(jiffies, timeout_jiffies)) {
361 signal_task_abort(task, -EIO);
364 rc = pci_doe_wait(doe_mb, PCI_DOE_POLL_INTERVAL);
366 signal_task_abort(task, rc);
372 rc = pci_doe_recv_resp(doe_mb, task);
374 signal_task_abort(task, rc);
378 signal_task_complete(task, rc);
381 static void pci_doe_task_complete(struct pci_doe_task *task)
383 complete(task->private);
386 static int pci_doe_discovery(struct pci_doe_mb *doe_mb, u8 *index, u16 *vid,
389 u32 request_pl = FIELD_PREP(PCI_DOE_DATA_OBJECT_DISC_REQ_3_INDEX,
391 __le32 request_pl_le = cpu_to_le32(request_pl);
392 __le32 response_pl_le;
396 rc = pci_doe(doe_mb, PCI_VENDOR_ID_PCI_SIG, PCI_DOE_PROTOCOL_DISCOVERY,
397 &request_pl_le, sizeof(request_pl_le),
398 &response_pl_le, sizeof(response_pl_le));
402 if (rc != sizeof(response_pl_le))
405 response_pl = le32_to_cpu(response_pl_le);
406 *vid = FIELD_GET(PCI_DOE_DATA_OBJECT_DISC_RSP_3_VID, response_pl);
407 *protocol = FIELD_GET(PCI_DOE_DATA_OBJECT_DISC_RSP_3_PROTOCOL,
409 *index = FIELD_GET(PCI_DOE_DATA_OBJECT_DISC_RSP_3_NEXT_INDEX,
415 static void *pci_doe_xa_prot_entry(u16 vid, u8 prot)
417 return xa_mk_value((vid << 8) | prot);
420 static int pci_doe_cache_protocols(struct pci_doe_mb *doe_mb)
430 rc = pci_doe_discovery(doe_mb, &index, &vid, &prot);
434 pci_dbg(doe_mb->pdev,
435 "[%x] Found protocol %d vid: %x prot: %x\n",
436 doe_mb->cap_offset, xa_idx, vid, prot);
438 rc = xa_insert(&doe_mb->prots, xa_idx++,
439 pci_doe_xa_prot_entry(vid, prot), GFP_KERNEL);
447 static void pci_doe_cancel_tasks(struct pci_doe_mb *doe_mb)
449 /* Stop all pending work items from starting */
450 set_bit(PCI_DOE_FLAG_DEAD, &doe_mb->flags);
452 /* Cancel an in progress work item, if necessary */
453 set_bit(PCI_DOE_FLAG_CANCEL, &doe_mb->flags);
454 wake_up(&doe_mb->wq);
458 * pci_doe_create_mb() - Create a DOE mailbox object
460 * @pdev: PCI device to create the DOE mailbox for
461 * @cap_offset: Offset of the DOE mailbox
463 * Create a single mailbox object to manage the mailbox protocol at the
464 * cap_offset specified.
466 * RETURNS: created mailbox object on success
467 * ERR_PTR(-errno) on failure
469 static struct pci_doe_mb *pci_doe_create_mb(struct pci_dev *pdev,
472 struct pci_doe_mb *doe_mb;
475 doe_mb = kzalloc(sizeof(*doe_mb), GFP_KERNEL);
477 return ERR_PTR(-ENOMEM);
480 doe_mb->cap_offset = cap_offset;
481 init_waitqueue_head(&doe_mb->wq);
482 xa_init(&doe_mb->prots);
484 doe_mb->work_queue = alloc_ordered_workqueue("%s %s DOE [%x]", 0,
485 dev_bus_name(&pdev->dev),
488 if (!doe_mb->work_queue) {
489 pci_err(pdev, "[%x] failed to allocate work queue\n",
495 /* Reset the mailbox by issuing an abort */
496 rc = pci_doe_abort(doe_mb);
498 pci_err(pdev, "[%x] failed to reset mailbox with abort command : %d\n",
499 doe_mb->cap_offset, rc);
504 * The state machine and the mailbox should be in sync now;
505 * Use the mailbox to query protocols.
507 rc = pci_doe_cache_protocols(doe_mb);
509 pci_err(pdev, "[%x] failed to cache protocols : %d\n",
510 doe_mb->cap_offset, rc);
517 pci_doe_cancel_tasks(doe_mb);
518 xa_destroy(&doe_mb->prots);
520 destroy_workqueue(doe_mb->work_queue);
527 * pci_doe_destroy_mb() - Destroy a DOE mailbox object
529 * @doe_mb: DOE mailbox
531 * Destroy all internal data structures created for the DOE mailbox.
533 static void pci_doe_destroy_mb(struct pci_doe_mb *doe_mb)
535 pci_doe_cancel_tasks(doe_mb);
536 xa_destroy(&doe_mb->prots);
537 destroy_workqueue(doe_mb->work_queue);
542 * pci_doe_supports_prot() - Return if the DOE instance supports the given
544 * @doe_mb: DOE mailbox capability to query
545 * @vid: Protocol Vendor ID
546 * @type: Protocol type
548 * RETURNS: True if the DOE mailbox supports the protocol specified
550 static bool pci_doe_supports_prot(struct pci_doe_mb *doe_mb, u16 vid, u8 type)
555 /* The discovery protocol must always be supported */
556 if (vid == PCI_VENDOR_ID_PCI_SIG && type == PCI_DOE_PROTOCOL_DISCOVERY)
559 xa_for_each(&doe_mb->prots, index, entry)
560 if (entry == pci_doe_xa_prot_entry(vid, type))
567 * pci_doe_submit_task() - Submit a task to be processed by the state machine
569 * @doe_mb: DOE mailbox capability to submit to
570 * @task: task to be queued
572 * Submit a DOE task (request/response) to the DOE mailbox to be processed.
573 * Returns upon queueing the task object. If the queue is full this function
574 * will sleep until there is room in the queue.
576 * task->complete will be called when the state machine is done processing this
579 * @task must be allocated on the stack.
581 * Excess data will be discarded.
583 * RETURNS: 0 when task has been successfully queued, -ERRNO on error
585 static int pci_doe_submit_task(struct pci_doe_mb *doe_mb,
586 struct pci_doe_task *task)
588 if (!pci_doe_supports_prot(doe_mb, task->prot.vid, task->prot.type))
591 if (test_bit(PCI_DOE_FLAG_DEAD, &doe_mb->flags))
594 task->doe_mb = doe_mb;
595 INIT_WORK_ONSTACK(&task->work, doe_statemachine_work);
596 queue_work(doe_mb->work_queue, &task->work);
601 * pci_doe() - Perform Data Object Exchange
603 * @doe_mb: DOE Mailbox
605 * @type: Data Object Type
606 * @request: Request payload
607 * @request_sz: Size of request payload (bytes)
608 * @response: Response payload
609 * @response_sz: Size of response payload (bytes)
611 * Submit @request to @doe_mb and store the @response.
612 * The DOE exchange is performed synchronously and may therefore sleep.
614 * Payloads are treated as opaque byte streams which are transmitted verbatim,
615 * without byte-swapping. If payloads contain little-endian register values,
616 * the caller is responsible for conversion with cpu_to_le32() / le32_to_cpu().
618 * For convenience, arbitrary payload sizes are allowed even though PCIe r6.0
619 * sec 6.30.1 specifies the Data Object Header 2 "Length" in dwords. The last
620 * (partial) dword is copied with byte granularity and padded with zeroes if
621 * necessary. Callers are thus relieved of using dword-sized bounce buffers.
623 * RETURNS: Length of received response or negative errno.
624 * Received data in excess of @response_sz is discarded.
625 * The length may be smaller than @response_sz and the caller
626 * is responsible for checking that.
628 int pci_doe(struct pci_doe_mb *doe_mb, u16 vendor, u8 type,
629 const void *request, size_t request_sz,
630 void *response, size_t response_sz)
632 DECLARE_COMPLETION_ONSTACK(c);
633 struct pci_doe_task task = {
636 .request_pl = request,
637 .request_pl_sz = request_sz,
638 .response_pl = response,
639 .response_pl_sz = response_sz,
640 .complete = pci_doe_task_complete,
645 rc = pci_doe_submit_task(doe_mb, &task);
649 wait_for_completion(&c);
653 EXPORT_SYMBOL_GPL(pci_doe);
656 * pci_find_doe_mailbox() - Find Data Object Exchange mailbox
660 * @type: Data Object Type
662 * Find first DOE mailbox of a PCI device which supports the given protocol.
664 * RETURNS: Pointer to the DOE mailbox or NULL if none was found.
666 struct pci_doe_mb *pci_find_doe_mailbox(struct pci_dev *pdev, u16 vendor,
669 struct pci_doe_mb *doe_mb;
672 xa_for_each(&pdev->doe_mbs, index, doe_mb)
673 if (pci_doe_supports_prot(doe_mb, vendor, type))
678 EXPORT_SYMBOL_GPL(pci_find_doe_mailbox);
680 void pci_doe_init(struct pci_dev *pdev)
682 struct pci_doe_mb *doe_mb;
686 xa_init(&pdev->doe_mbs);
688 while ((offset = pci_find_next_ext_capability(pdev, offset,
689 PCI_EXT_CAP_ID_DOE))) {
690 doe_mb = pci_doe_create_mb(pdev, offset);
691 if (IS_ERR(doe_mb)) {
692 pci_err(pdev, "[%x] failed to create mailbox: %ld\n",
693 offset, PTR_ERR(doe_mb));
697 rc = xa_insert(&pdev->doe_mbs, offset, doe_mb, GFP_KERNEL);
699 pci_err(pdev, "[%x] failed to insert mailbox: %d\n",
701 pci_doe_destroy_mb(doe_mb);
706 void pci_doe_destroy(struct pci_dev *pdev)
708 struct pci_doe_mb *doe_mb;
711 xa_for_each(&pdev->doe_mbs, index, doe_mb)
712 pci_doe_destroy_mb(doe_mb);
714 xa_destroy(&pdev->doe_mbs);
717 void pci_doe_disconnected(struct pci_dev *pdev)
719 struct pci_doe_mb *doe_mb;
722 xa_for_each(&pdev->doe_mbs, index, doe_mb)
723 pci_doe_cancel_tasks(doe_mb);