xref: /openbmc/linux/drivers/crypto/hisilicon/qm.c (revision 884caada)
1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright (c) 2019 HiSilicon Limited. */
3 #include <asm/page.h>
4 #include <linux/bitmap.h>
5 #include <linux/debugfs.h>
6 #include <linux/dma-mapping.h>
7 #include <linux/io.h>
8 #include <linux/irqreturn.h>
9 #include <linux/log2.h>
10 #include <linux/seq_file.h>
11 #include <linux/slab.h>
12 #include "qm.h"
13 
14 /* eq/aeq irq enable */
15 #define QM_VF_AEQ_INT_SOURCE		0x0
16 #define QM_VF_AEQ_INT_MASK		0x4
17 #define QM_VF_EQ_INT_SOURCE		0x8
18 #define QM_VF_EQ_INT_MASK		0xc
19 #define QM_IRQ_NUM_V1			1
20 #define QM_IRQ_NUM_PF_V2		4
21 #define QM_IRQ_NUM_VF_V2		2
22 
23 #define QM_EQ_EVENT_IRQ_VECTOR		0
24 #define QM_AEQ_EVENT_IRQ_VECTOR		1
25 #define QM_ABNORMAL_EVENT_IRQ_VECTOR	3
26 
27 /* mailbox */
28 #define QM_MB_CMD_SQC			0x0
29 #define QM_MB_CMD_CQC			0x1
30 #define QM_MB_CMD_EQC			0x2
31 #define QM_MB_CMD_AEQC			0x3
32 #define QM_MB_CMD_SQC_BT		0x4
33 #define QM_MB_CMD_CQC_BT		0x5
34 #define QM_MB_CMD_SQC_VFT_V2		0x6
35 
36 #define QM_MB_CMD_SEND_BASE		0x300
37 #define QM_MB_EVENT_SHIFT		8
38 #define QM_MB_BUSY_SHIFT		13
39 #define QM_MB_OP_SHIFT			14
40 #define QM_MB_CMD_DATA_ADDR_L		0x304
41 #define QM_MB_CMD_DATA_ADDR_H		0x308
42 
43 /* sqc shift */
44 #define QM_SQ_HOP_NUM_SHIFT		0
45 #define QM_SQ_PAGE_SIZE_SHIFT		4
46 #define QM_SQ_BUF_SIZE_SHIFT		8
47 #define QM_SQ_SQE_SIZE_SHIFT		12
48 #define QM_SQ_PRIORITY_SHIFT		0
49 #define QM_SQ_ORDERS_SHIFT		4
50 #define QM_SQ_TYPE_SHIFT		8
51 
52 #define QM_SQ_TYPE_MASK			GENMASK(3, 0)
53 
54 /* cqc shift */
55 #define QM_CQ_HOP_NUM_SHIFT		0
56 #define QM_CQ_PAGE_SIZE_SHIFT		4
57 #define QM_CQ_BUF_SIZE_SHIFT		8
58 #define QM_CQ_CQE_SIZE_SHIFT		12
59 #define QM_CQ_PHASE_SHIFT		0
60 #define QM_CQ_FLAG_SHIFT		1
61 
62 #define QM_CQE_PHASE(cqe)		((cqe)->w7 & 0x1)
63 #define QM_QC_CQE_SIZE			4
64 
65 /* eqc shift */
66 #define QM_EQE_AEQE_SIZE		(2UL << 12)
67 #define QM_EQC_PHASE_SHIFT		16
68 
69 #define QM_EQE_PHASE(eqe)		(((eqe)->dw0 >> 16) & 0x1)
70 #define QM_EQE_CQN_MASK			GENMASK(15, 0)
71 
72 #define QM_AEQE_PHASE(aeqe)		(((aeqe)->dw0 >> 16) & 0x1)
73 #define QM_AEQE_TYPE_SHIFT		17
74 
75 #define QM_DOORBELL_CMD_SQ		0
76 #define QM_DOORBELL_CMD_CQ		1
77 #define QM_DOORBELL_CMD_EQ		2
78 #define QM_DOORBELL_CMD_AEQ		3
79 
80 #define QM_DOORBELL_BASE_V1		0x340
81 #define QM_DB_CMD_SHIFT_V1		16
82 #define QM_DB_INDEX_SHIFT_V1		32
83 #define QM_DB_PRIORITY_SHIFT_V1		48
84 #define QM_DOORBELL_SQ_CQ_BASE_V2	0x1000
85 #define QM_DOORBELL_EQ_AEQ_BASE_V2	0x2000
86 #define QM_DB_CMD_SHIFT_V2		12
87 #define QM_DB_RAND_SHIFT_V2		16
88 #define QM_DB_INDEX_SHIFT_V2		32
89 #define QM_DB_PRIORITY_SHIFT_V2		48
90 
91 #define QM_MEM_START_INIT		0x100040
92 #define QM_MEM_INIT_DONE		0x100044
93 #define QM_VFT_CFG_RDY			0x10006c
94 #define QM_VFT_CFG_OP_WR		0x100058
95 #define QM_VFT_CFG_TYPE			0x10005c
96 #define QM_SQC_VFT			0x0
97 #define QM_CQC_VFT			0x1
98 #define QM_VFT_CFG			0x100060
99 #define QM_VFT_CFG_OP_ENABLE		0x100054
100 
101 #define QM_VFT_CFG_DATA_L		0x100064
102 #define QM_VFT_CFG_DATA_H		0x100068
103 #define QM_SQC_VFT_BUF_SIZE		(7ULL << 8)
104 #define QM_SQC_VFT_SQC_SIZE		(5ULL << 12)
105 #define QM_SQC_VFT_INDEX_NUMBER		(1ULL << 16)
106 #define QM_SQC_VFT_START_SQN_SHIFT	28
107 #define QM_SQC_VFT_VALID		(1ULL << 44)
108 #define QM_SQC_VFT_SQN_SHIFT		45
109 #define QM_CQC_VFT_BUF_SIZE		(7ULL << 8)
110 #define QM_CQC_VFT_SQC_SIZE		(5ULL << 12)
111 #define QM_CQC_VFT_INDEX_NUMBER		(1ULL << 16)
112 #define QM_CQC_VFT_VALID		(1ULL << 28)
113 
114 #define QM_SQC_VFT_BASE_SHIFT_V2	28
115 #define QM_SQC_VFT_BASE_MASK_V2		GENMASK(5, 0)
116 #define QM_SQC_VFT_NUM_SHIFT_V2		45
117 #define QM_SQC_VFT_NUM_MASK_v2		GENMASK(9, 0)
118 
119 #define QM_DFX_CNT_CLR_CE		0x100118
120 
121 #define QM_ABNORMAL_INT_SOURCE		0x100000
122 #define QM_ABNORMAL_INT_MASK		0x100004
123 #define QM_ABNORMAL_INT_MASK_VALUE	0x1fff
124 #define QM_ABNORMAL_INT_STATUS		0x100008
125 #define QM_ABNORMAL_INF00		0x100010
126 #define QM_FIFO_OVERFLOW_TYPE		0xc0
127 #define QM_FIFO_OVERFLOW_TYPE_SHIFT	6
128 #define QM_FIFO_OVERFLOW_VF		0x3f
129 #define QM_ABNORMAL_INF01		0x100014
130 #define QM_DB_TIMEOUT_TYPE		0xc0
131 #define QM_DB_TIMEOUT_TYPE_SHIFT	6
132 #define QM_DB_TIMEOUT_VF		0x3f
133 #define QM_RAS_CE_ENABLE		0x1000ec
134 #define QM_RAS_FE_ENABLE		0x1000f0
135 #define QM_RAS_NFE_ENABLE		0x1000f4
136 #define QM_RAS_CE_THRESHOLD		0x1000f8
137 #define QM_RAS_CE_TIMES_PER_IRQ		1
138 #define QM_RAS_MSI_INT_SEL		0x1040f4
139 
140 #define QM_CACHE_WB_START		0x204
141 #define QM_CACHE_WB_DONE		0x208
142 
143 #define PCI_BAR_2			2
144 #define QM_SQE_DATA_ALIGN_MASK		GENMASK(6, 0)
145 #define QMC_ALIGN(sz)			ALIGN(sz, 32)
146 
147 #define QM_DBG_TMP_BUF_LEN		22
148 
149 #define QM_MK_CQC_DW3_V1(hop_num, pg_sz, buf_sz, cqe_sz) \
150 	(((hop_num) << QM_CQ_HOP_NUM_SHIFT)	| \
151 	((pg_sz) << QM_CQ_PAGE_SIZE_SHIFT)	| \
152 	((buf_sz) << QM_CQ_BUF_SIZE_SHIFT)	| \
153 	((cqe_sz) << QM_CQ_CQE_SIZE_SHIFT))
154 
155 #define QM_MK_CQC_DW3_V2(cqe_sz) \
156 	((QM_Q_DEPTH - 1) | ((cqe_sz) << QM_CQ_CQE_SIZE_SHIFT))
157 
158 #define QM_MK_SQC_W13(priority, orders, alg_type) \
159 	(((priority) << QM_SQ_PRIORITY_SHIFT)	| \
160 	((orders) << QM_SQ_ORDERS_SHIFT)	| \
161 	(((alg_type) & QM_SQ_TYPE_MASK) << QM_SQ_TYPE_SHIFT))
162 
163 #define QM_MK_SQC_DW3_V1(hop_num, pg_sz, buf_sz, sqe_sz) \
164 	(((hop_num) << QM_SQ_HOP_NUM_SHIFT)	| \
165 	((pg_sz) << QM_SQ_PAGE_SIZE_SHIFT)	| \
166 	((buf_sz) << QM_SQ_BUF_SIZE_SHIFT)	| \
167 	((u32)ilog2(sqe_sz) << QM_SQ_SQE_SIZE_SHIFT))
168 
169 #define QM_MK_SQC_DW3_V2(sqe_sz) \
170 	((QM_Q_DEPTH - 1) | ((u32)ilog2(sqe_sz) << QM_SQ_SQE_SIZE_SHIFT))
171 
172 #define INIT_QC_COMMON(qc, base, pasid) do {	\
173 	(qc)->head = 0;				\
174 	(qc)->tail = 0;				\
175 	(qc)->base_l = lower_32_bits(base);	\
176 	(qc)->base_h = upper_32_bits(base);	\
177 	(qc)->dw3 = 0;				\
178 	(qc)->w8 = 0;				\
179 	(qc)->rsvd0 = 0;			\
180 	(qc)->pasid = pasid;			\
181 	(qc)->w11 = 0;				\
182 	(qc)->rsvd1 = 0;			\
183 } while (0)
184 
185 enum vft_type {
186 	SQC_VFT = 0,
187 	CQC_VFT,
188 };
189 
190 struct qm_cqe {
191 	__le32 rsvd0;
192 	__le16 cmd_id;
193 	__le16 rsvd1;
194 	__le16 sq_head;
195 	__le16 sq_num;
196 	__le16 rsvd2;
197 	__le16 w7;
198 };
199 
200 struct qm_eqe {
201 	__le32 dw0;
202 };
203 
204 struct qm_aeqe {
205 	__le32 dw0;
206 };
207 
208 struct qm_sqc {
209 	__le16 head;
210 	__le16 tail;
211 	__le32 base_l;
212 	__le32 base_h;
213 	__le32 dw3;
214 	__le16 w8;
215 	__le16 rsvd0;
216 	__le16 pasid;
217 	__le16 w11;
218 	__le16 cq_num;
219 	__le16 w13;
220 	__le32 rsvd1;
221 };
222 
223 struct qm_cqc {
224 	__le16 head;
225 	__le16 tail;
226 	__le32 base_l;
227 	__le32 base_h;
228 	__le32 dw3;
229 	__le16 w8;
230 	__le16 rsvd0;
231 	__le16 pasid;
232 	__le16 w11;
233 	__le32 dw6;
234 	__le32 rsvd1;
235 };
236 
237 struct qm_eqc {
238 	__le16 head;
239 	__le16 tail;
240 	__le32 base_l;
241 	__le32 base_h;
242 	__le32 dw3;
243 	__le32 rsvd[2];
244 	__le32 dw6;
245 };
246 
247 struct qm_aeqc {
248 	__le16 head;
249 	__le16 tail;
250 	__le32 base_l;
251 	__le32 base_h;
252 	__le32 dw3;
253 	__le32 rsvd[2];
254 	__le32 dw6;
255 };
256 
257 struct qm_mailbox {
258 	__le16 w0;
259 	__le16 queue_num;
260 	__le32 base_l;
261 	__le32 base_h;
262 	__le32 rsvd;
263 };
264 
265 struct qm_doorbell {
266 	__le16 queue_num;
267 	__le16 cmd;
268 	__le16 index;
269 	__le16 priority;
270 };
271 
272 struct hisi_qm_hw_ops {
273 	int (*get_vft)(struct hisi_qm *qm, u32 *base, u32 *number);
274 	void (*qm_db)(struct hisi_qm *qm, u16 qn,
275 		      u8 cmd, u16 index, u8 priority);
276 	u32 (*get_irq_num)(struct hisi_qm *qm);
277 	int (*debug_init)(struct hisi_qm *qm);
278 	void (*hw_error_init)(struct hisi_qm *qm, u32 ce, u32 nfe, u32 fe,
279 			      u32 msi);
280 	pci_ers_result_t (*hw_error_handle)(struct hisi_qm *qm);
281 };
282 
283 static const char * const qm_debug_file_name[] = {
284 	[CURRENT_Q]    = "current_q",
285 	[CLEAR_ENABLE] = "clear_enable",
286 };
287 
288 struct hisi_qm_hw_error {
289 	u32 int_msk;
290 	const char *msg;
291 };
292 
293 static const struct hisi_qm_hw_error qm_hw_error[] = {
294 	{ .int_msk = BIT(0), .msg = "qm_axi_rresp" },
295 	{ .int_msk = BIT(1), .msg = "qm_axi_bresp" },
296 	{ .int_msk = BIT(2), .msg = "qm_ecc_mbit" },
297 	{ .int_msk = BIT(3), .msg = "qm_ecc_1bit" },
298 	{ .int_msk = BIT(4), .msg = "qm_acc_get_task_timeout" },
299 	{ .int_msk = BIT(5), .msg = "qm_acc_do_task_timeout" },
300 	{ .int_msk = BIT(6), .msg = "qm_acc_wb_not_ready_timeout" },
301 	{ .int_msk = BIT(7), .msg = "qm_sq_cq_vf_invalid" },
302 	{ .int_msk = BIT(8), .msg = "qm_cq_vf_invalid" },
303 	{ .int_msk = BIT(9), .msg = "qm_sq_vf_invalid" },
304 	{ .int_msk = BIT(10), .msg = "qm_db_timeout" },
305 	{ .int_msk = BIT(11), .msg = "qm_of_fifo_of" },
306 	{ .int_msk = BIT(12), .msg = "qm_db_random_invalid" },
307 	{ /* sentinel */ }
308 };
309 
310 static const char * const qm_db_timeout[] = {
311 	"sq", "cq", "eq", "aeq",
312 };
313 
314 static const char * const qm_fifo_overflow[] = {
315 	"cq", "eq", "aeq",
316 };
317 
318 /* return 0 mailbox ready, -ETIMEDOUT hardware timeout */
319 static int qm_wait_mb_ready(struct hisi_qm *qm)
320 {
321 	u32 val;
322 
323 	return readl_relaxed_poll_timeout(qm->io_base + QM_MB_CMD_SEND_BASE,
324 					  val, !((val >> QM_MB_BUSY_SHIFT) &
325 					  0x1), 10, 1000);
326 }
327 
328 /* 128 bit should be written to hardware at one time to trigger a mailbox */
329 static void qm_mb_write(struct hisi_qm *qm, const void *src)
330 {
331 	void __iomem *fun_base = qm->io_base + QM_MB_CMD_SEND_BASE;
332 	unsigned long tmp0 = 0, tmp1 = 0;
333 
334 	asm volatile("ldp %0, %1, %3\n"
335 		     "stp %0, %1, %2\n"
336 		     "dsb sy\n"
337 		     : "=&r" (tmp0),
338 		       "=&r" (tmp1),
339 		       "+Q" (*((char *)fun_base))
340 		     : "Q" (*((char *)src))
341 		     : "memory");
342 }
343 
344 static int qm_mb(struct hisi_qm *qm, u8 cmd, dma_addr_t dma_addr, u16 queue,
345 		 bool op)
346 {
347 	struct qm_mailbox mailbox;
348 	int ret = 0;
349 
350 	dev_dbg(&qm->pdev->dev, "QM mailbox request to q%u: %u-%llx\n",
351 		queue, cmd, (unsigned long long)dma_addr);
352 
353 	mailbox.w0 = cmd |
354 		     (op ? 0x1 << QM_MB_OP_SHIFT : 0) |
355 		     (0x1 << QM_MB_BUSY_SHIFT);
356 	mailbox.queue_num = queue;
357 	mailbox.base_l = lower_32_bits(dma_addr);
358 	mailbox.base_h = upper_32_bits(dma_addr);
359 	mailbox.rsvd = 0;
360 
361 	mutex_lock(&qm->mailbox_lock);
362 
363 	if (unlikely(qm_wait_mb_ready(qm))) {
364 		ret = -EBUSY;
365 		dev_err(&qm->pdev->dev, "QM mailbox is busy to start!\n");
366 		goto busy_unlock;
367 	}
368 
369 	qm_mb_write(qm, &mailbox);
370 
371 	if (unlikely(qm_wait_mb_ready(qm))) {
372 		ret = -EBUSY;
373 		dev_err(&qm->pdev->dev, "QM mailbox operation timeout!\n");
374 		goto busy_unlock;
375 	}
376 
377 busy_unlock:
378 	mutex_unlock(&qm->mailbox_lock);
379 
380 	return ret;
381 }
382 
383 static void qm_db_v1(struct hisi_qm *qm, u16 qn, u8 cmd, u16 index, u8 priority)
384 {
385 	u64 doorbell;
386 
387 	doorbell = qn | ((u64)cmd << QM_DB_CMD_SHIFT_V1) |
388 		   ((u64)index << QM_DB_INDEX_SHIFT_V1)  |
389 		   ((u64)priority << QM_DB_PRIORITY_SHIFT_V1);
390 
391 	writeq(doorbell, qm->io_base + QM_DOORBELL_BASE_V1);
392 }
393 
394 static void qm_db_v2(struct hisi_qm *qm, u16 qn, u8 cmd, u16 index, u8 priority)
395 {
396 	u64 doorbell;
397 	u64 dbase;
398 	u16 randata = 0;
399 
400 	if (cmd == QM_DOORBELL_CMD_SQ || cmd == QM_DOORBELL_CMD_CQ)
401 		dbase = QM_DOORBELL_SQ_CQ_BASE_V2;
402 	else
403 		dbase = QM_DOORBELL_EQ_AEQ_BASE_V2;
404 
405 	doorbell = qn | ((u64)cmd << QM_DB_CMD_SHIFT_V2) |
406 		   ((u64)randata << QM_DB_RAND_SHIFT_V2) |
407 		   ((u64)index << QM_DB_INDEX_SHIFT_V2)	 |
408 		   ((u64)priority << QM_DB_PRIORITY_SHIFT_V2);
409 
410 	writeq(doorbell, qm->io_base + dbase);
411 }
412 
413 static void qm_db(struct hisi_qm *qm, u16 qn, u8 cmd, u16 index, u8 priority)
414 {
415 	dev_dbg(&qm->pdev->dev, "QM doorbell request: qn=%u, cmd=%u, index=%u\n",
416 		qn, cmd, index);
417 
418 	qm->ops->qm_db(qm, qn, cmd, index, priority);
419 }
420 
421 static int qm_dev_mem_reset(struct hisi_qm *qm)
422 {
423 	u32 val;
424 
425 	writel(0x1, qm->io_base + QM_MEM_START_INIT);
426 	return readl_relaxed_poll_timeout(qm->io_base + QM_MEM_INIT_DONE, val,
427 					  val & BIT(0), 10, 1000);
428 }
429 
430 static u32 qm_get_irq_num_v1(struct hisi_qm *qm)
431 {
432 	return QM_IRQ_NUM_V1;
433 }
434 
435 static u32 qm_get_irq_num_v2(struct hisi_qm *qm)
436 {
437 	if (qm->fun_type == QM_HW_PF)
438 		return QM_IRQ_NUM_PF_V2;
439 	else
440 		return QM_IRQ_NUM_VF_V2;
441 }
442 
443 static struct hisi_qp *qm_to_hisi_qp(struct hisi_qm *qm, struct qm_eqe *eqe)
444 {
445 	u16 cqn = eqe->dw0 & QM_EQE_CQN_MASK;
446 
447 	return qm->qp_array[cqn];
448 }
449 
450 static void qm_cq_head_update(struct hisi_qp *qp)
451 {
452 	if (qp->qp_status.cq_head == QM_Q_DEPTH - 1) {
453 		qp->qp_status.cqc_phase = !qp->qp_status.cqc_phase;
454 		qp->qp_status.cq_head = 0;
455 	} else {
456 		qp->qp_status.cq_head++;
457 	}
458 }
459 
460 static void qm_poll_qp(struct hisi_qp *qp, struct hisi_qm *qm)
461 {
462 	struct qm_cqe *cqe = qp->cqe + qp->qp_status.cq_head;
463 
464 	if (qp->req_cb) {
465 		while (QM_CQE_PHASE(cqe) == qp->qp_status.cqc_phase) {
466 			dma_rmb();
467 			qp->req_cb(qp, qp->sqe + qm->sqe_size * cqe->sq_head);
468 			qm_cq_head_update(qp);
469 			cqe = qp->cqe + qp->qp_status.cq_head;
470 			qm_db(qm, qp->qp_id, QM_DOORBELL_CMD_CQ,
471 			      qp->qp_status.cq_head, 0);
472 			atomic_dec(&qp->qp_status.used);
473 		}
474 
475 		/* set c_flag */
476 		qm_db(qm, qp->qp_id, QM_DOORBELL_CMD_CQ,
477 		      qp->qp_status.cq_head, 1);
478 	}
479 }
480 
481 static void qm_qp_work_func(struct work_struct *work)
482 {
483 	struct hisi_qp *qp;
484 
485 	qp = container_of(work, struct hisi_qp, work);
486 	qm_poll_qp(qp, qp->qm);
487 }
488 
489 static irqreturn_t qm_irq_handler(int irq, void *data)
490 {
491 	struct hisi_qm *qm = data;
492 	struct qm_eqe *eqe = qm->eqe + qm->status.eq_head;
493 	struct hisi_qp *qp;
494 	int eqe_num = 0;
495 
496 	while (QM_EQE_PHASE(eqe) == qm->status.eqc_phase) {
497 		eqe_num++;
498 		qp = qm_to_hisi_qp(qm, eqe);
499 		if (qp)
500 			queue_work(qp->wq, &qp->work);
501 
502 		if (qm->status.eq_head == QM_Q_DEPTH - 1) {
503 			qm->status.eqc_phase = !qm->status.eqc_phase;
504 			eqe = qm->eqe;
505 			qm->status.eq_head = 0;
506 		} else {
507 			eqe++;
508 			qm->status.eq_head++;
509 		}
510 
511 		if (eqe_num == QM_Q_DEPTH / 2 - 1) {
512 			eqe_num = 0;
513 			qm_db(qm, 0, QM_DOORBELL_CMD_EQ, qm->status.eq_head, 0);
514 		}
515 	}
516 
517 	qm_db(qm, 0, QM_DOORBELL_CMD_EQ, qm->status.eq_head, 0);
518 
519 	return IRQ_HANDLED;
520 }
521 
522 static irqreturn_t qm_irq(int irq, void *data)
523 {
524 	struct hisi_qm *qm = data;
525 
526 	if (readl(qm->io_base + QM_VF_EQ_INT_SOURCE))
527 		return qm_irq_handler(irq, data);
528 
529 	dev_err(&qm->pdev->dev, "invalid int source\n");
530 	qm_db(qm, 0, QM_DOORBELL_CMD_EQ, qm->status.eq_head, 0);
531 
532 	return IRQ_NONE;
533 }
534 
535 static irqreturn_t qm_aeq_irq(int irq, void *data)
536 {
537 	struct hisi_qm *qm = data;
538 	struct qm_aeqe *aeqe = qm->aeqe + qm->status.aeq_head;
539 	u32 type;
540 
541 	if (!readl(qm->io_base + QM_VF_AEQ_INT_SOURCE))
542 		return IRQ_NONE;
543 
544 	while (QM_AEQE_PHASE(aeqe) == qm->status.aeqc_phase) {
545 		type = aeqe->dw0 >> QM_AEQE_TYPE_SHIFT;
546 		if (type < ARRAY_SIZE(qm_fifo_overflow))
547 			dev_err(&qm->pdev->dev, "%s overflow\n",
548 				qm_fifo_overflow[type]);
549 		else
550 			dev_err(&qm->pdev->dev, "unknown error type %d\n",
551 				type);
552 
553 		if (qm->status.aeq_head == QM_Q_DEPTH - 1) {
554 			qm->status.aeqc_phase = !qm->status.aeqc_phase;
555 			aeqe = qm->aeqe;
556 			qm->status.aeq_head = 0;
557 		} else {
558 			aeqe++;
559 			qm->status.aeq_head++;
560 		}
561 
562 		qm_db(qm, 0, QM_DOORBELL_CMD_AEQ, qm->status.aeq_head, 0);
563 	}
564 
565 	return IRQ_HANDLED;
566 }
567 
568 static irqreturn_t qm_abnormal_irq(int irq, void *data)
569 {
570 	const struct hisi_qm_hw_error *err = qm_hw_error;
571 	struct hisi_qm *qm = data;
572 	struct device *dev = &qm->pdev->dev;
573 	u32 error_status, tmp;
574 
575 	/* read err sts */
576 	tmp = readl(qm->io_base + QM_ABNORMAL_INT_STATUS);
577 	error_status = qm->msi_mask & tmp;
578 
579 	while (err->msg) {
580 		if (err->int_msk & error_status)
581 			dev_err(dev, "%s [error status=0x%x] found\n",
582 				err->msg, err->int_msk);
583 
584 		err++;
585 	}
586 
587 	/* clear err sts */
588 	writel(error_status, qm->io_base + QM_ABNORMAL_INT_SOURCE);
589 
590 	return IRQ_HANDLED;
591 }
592 
593 static int qm_irq_register(struct hisi_qm *qm)
594 {
595 	struct pci_dev *pdev = qm->pdev;
596 	int ret;
597 
598 	ret = request_irq(pci_irq_vector(pdev, QM_EQ_EVENT_IRQ_VECTOR),
599 			  qm_irq, IRQF_SHARED, qm->dev_name, qm);
600 	if (ret)
601 		return ret;
602 
603 	if (qm->ver == QM_HW_V2) {
604 		ret = request_irq(pci_irq_vector(pdev, QM_AEQ_EVENT_IRQ_VECTOR),
605 				  qm_aeq_irq, IRQF_SHARED, qm->dev_name, qm);
606 		if (ret)
607 			goto err_aeq_irq;
608 
609 		if (qm->fun_type == QM_HW_PF) {
610 			ret = request_irq(pci_irq_vector(pdev,
611 					  QM_ABNORMAL_EVENT_IRQ_VECTOR),
612 					  qm_abnormal_irq, IRQF_SHARED,
613 					  qm->dev_name, qm);
614 			if (ret)
615 				goto err_abonormal_irq;
616 		}
617 	}
618 
619 	return 0;
620 
621 err_abonormal_irq:
622 	free_irq(pci_irq_vector(pdev, QM_AEQ_EVENT_IRQ_VECTOR), qm);
623 err_aeq_irq:
624 	free_irq(pci_irq_vector(pdev, QM_EQ_EVENT_IRQ_VECTOR), qm);
625 	return ret;
626 }
627 
628 static void qm_irq_unregister(struct hisi_qm *qm)
629 {
630 	struct pci_dev *pdev = qm->pdev;
631 
632 	free_irq(pci_irq_vector(pdev, QM_EQ_EVENT_IRQ_VECTOR), qm);
633 
634 	if (qm->ver == QM_HW_V2) {
635 		free_irq(pci_irq_vector(pdev, QM_AEQ_EVENT_IRQ_VECTOR), qm);
636 
637 		if (qm->fun_type == QM_HW_PF)
638 			free_irq(pci_irq_vector(pdev,
639 				 QM_ABNORMAL_EVENT_IRQ_VECTOR), qm);
640 	}
641 }
642 
643 static void qm_init_qp_status(struct hisi_qp *qp)
644 {
645 	struct hisi_qp_status *qp_status = &qp->qp_status;
646 
647 	qp_status->sq_tail = 0;
648 	qp_status->cq_head = 0;
649 	qp_status->cqc_phase = 1;
650 	qp_status->flags = 0;
651 }
652 
653 static void qm_vft_data_cfg(struct hisi_qm *qm, enum vft_type type, u32 base,
654 			    u32 number)
655 {
656 	u64 tmp = 0;
657 
658 	if (number > 0) {
659 		switch (type) {
660 		case SQC_VFT:
661 			switch (qm->ver) {
662 			case QM_HW_V1:
663 				tmp = QM_SQC_VFT_BUF_SIZE	|
664 				      QM_SQC_VFT_SQC_SIZE	|
665 				      QM_SQC_VFT_INDEX_NUMBER	|
666 				      QM_SQC_VFT_VALID		|
667 				      (u64)base << QM_SQC_VFT_START_SQN_SHIFT;
668 				break;
669 			case QM_HW_V2:
670 				tmp = (u64)base << QM_SQC_VFT_START_SQN_SHIFT |
671 				      QM_SQC_VFT_VALID |
672 				      (u64)(number - 1) << QM_SQC_VFT_SQN_SHIFT;
673 				break;
674 			case QM_HW_UNKNOWN:
675 				break;
676 			}
677 			break;
678 		case CQC_VFT:
679 			switch (qm->ver) {
680 			case QM_HW_V1:
681 				tmp = QM_CQC_VFT_BUF_SIZE	|
682 				      QM_CQC_VFT_SQC_SIZE	|
683 				      QM_CQC_VFT_INDEX_NUMBER	|
684 				      QM_CQC_VFT_VALID;
685 				break;
686 			case QM_HW_V2:
687 				tmp = QM_CQC_VFT_VALID;
688 				break;
689 			case QM_HW_UNKNOWN:
690 				break;
691 			}
692 			break;
693 		}
694 	}
695 
696 	writel(lower_32_bits(tmp), qm->io_base + QM_VFT_CFG_DATA_L);
697 	writel(upper_32_bits(tmp), qm->io_base + QM_VFT_CFG_DATA_H);
698 }
699 
700 static int qm_set_vft_common(struct hisi_qm *qm, enum vft_type type,
701 			     u32 fun_num, u32 base, u32 number)
702 {
703 	unsigned int val;
704 	int ret;
705 
706 	ret = readl_relaxed_poll_timeout(qm->io_base + QM_VFT_CFG_RDY, val,
707 					 val & BIT(0), 10, 1000);
708 	if (ret)
709 		return ret;
710 
711 	writel(0x0, qm->io_base + QM_VFT_CFG_OP_WR);
712 	writel(type, qm->io_base + QM_VFT_CFG_TYPE);
713 	writel(fun_num, qm->io_base + QM_VFT_CFG);
714 
715 	qm_vft_data_cfg(qm, type, base, number);
716 
717 	writel(0x0, qm->io_base + QM_VFT_CFG_RDY);
718 	writel(0x1, qm->io_base + QM_VFT_CFG_OP_ENABLE);
719 
720 	return readl_relaxed_poll_timeout(qm->io_base + QM_VFT_CFG_RDY, val,
721 					  val & BIT(0), 10, 1000);
722 }
723 
724 /* The config should be conducted after qm_dev_mem_reset() */
725 static int qm_set_sqc_cqc_vft(struct hisi_qm *qm, u32 fun_num, u32 base,
726 			      u32 number)
727 {
728 	int ret, i;
729 
730 	for (i = SQC_VFT; i <= CQC_VFT; i++) {
731 		ret = qm_set_vft_common(qm, i, fun_num, base, number);
732 		if (ret)
733 			return ret;
734 	}
735 
736 	return 0;
737 }
738 
739 static int qm_get_vft_v2(struct hisi_qm *qm, u32 *base, u32 *number)
740 {
741 	u64 sqc_vft;
742 	int ret;
743 
744 	ret = qm_mb(qm, QM_MB_CMD_SQC_VFT_V2, 0, 0, 1);
745 	if (ret)
746 		return ret;
747 
748 	sqc_vft = readl(qm->io_base + QM_MB_CMD_DATA_ADDR_L) |
749 		  ((u64)readl(qm->io_base + QM_MB_CMD_DATA_ADDR_H) << 32);
750 	*base = QM_SQC_VFT_BASE_MASK_V2 & (sqc_vft >> QM_SQC_VFT_BASE_SHIFT_V2);
751 	*number = (QM_SQC_VFT_NUM_MASK_v2 &
752 		   (sqc_vft >> QM_SQC_VFT_NUM_SHIFT_V2)) + 1;
753 
754 	return 0;
755 }
756 
757 static struct hisi_qm *file_to_qm(struct debugfs_file *file)
758 {
759 	struct qm_debug *debug = file->debug;
760 
761 	return container_of(debug, struct hisi_qm, debug);
762 }
763 
764 static u32 current_q_read(struct debugfs_file *file)
765 {
766 	struct hisi_qm *qm = file_to_qm(file);
767 
768 	return readl(qm->io_base + QM_DFX_SQE_CNT_VF_SQN) >> QM_DFX_QN_SHIFT;
769 }
770 
771 static int current_q_write(struct debugfs_file *file, u32 val)
772 {
773 	struct hisi_qm *qm = file_to_qm(file);
774 	u32 tmp;
775 
776 	if (val >= qm->debug.curr_qm_qp_num)
777 		return -EINVAL;
778 
779 	tmp = val << QM_DFX_QN_SHIFT |
780 	      (readl(qm->io_base + QM_DFX_SQE_CNT_VF_SQN) & CURRENT_FUN_MASK);
781 	writel(tmp, qm->io_base + QM_DFX_SQE_CNT_VF_SQN);
782 
783 	tmp = val << QM_DFX_QN_SHIFT |
784 	      (readl(qm->io_base + QM_DFX_CQE_CNT_VF_CQN) & CURRENT_FUN_MASK);
785 	writel(tmp, qm->io_base + QM_DFX_CQE_CNT_VF_CQN);
786 
787 	return 0;
788 }
789 
790 static u32 clear_enable_read(struct debugfs_file *file)
791 {
792 	struct hisi_qm *qm = file_to_qm(file);
793 
794 	return readl(qm->io_base + QM_DFX_CNT_CLR_CE);
795 }
796 
797 /* rd_clr_ctrl 1 enable read clear, otherwise 0 disable it */
798 static int clear_enable_write(struct debugfs_file *file, u32 rd_clr_ctrl)
799 {
800 	struct hisi_qm *qm = file_to_qm(file);
801 
802 	if (rd_clr_ctrl > 1)
803 		return -EINVAL;
804 
805 	writel(rd_clr_ctrl, qm->io_base + QM_DFX_CNT_CLR_CE);
806 
807 	return 0;
808 }
809 
810 static ssize_t qm_debug_read(struct file *filp, char __user *buf,
811 			     size_t count, loff_t *pos)
812 {
813 	struct debugfs_file *file = filp->private_data;
814 	enum qm_debug_file index = file->index;
815 	char tbuf[QM_DBG_TMP_BUF_LEN];
816 	u32 val;
817 	int ret;
818 
819 	mutex_lock(&file->lock);
820 	switch (index) {
821 	case CURRENT_Q:
822 		val = current_q_read(file);
823 		break;
824 	case CLEAR_ENABLE:
825 		val = clear_enable_read(file);
826 		break;
827 	default:
828 		mutex_unlock(&file->lock);
829 		return -EINVAL;
830 	}
831 	mutex_unlock(&file->lock);
832 	ret = sprintf(tbuf, "%u\n", val);
833 	return simple_read_from_buffer(buf, count, pos, tbuf, ret);
834 }
835 
836 static ssize_t qm_debug_write(struct file *filp, const char __user *buf,
837 			      size_t count, loff_t *pos)
838 {
839 	struct debugfs_file *file = filp->private_data;
840 	enum qm_debug_file index = file->index;
841 	unsigned long val;
842 	char tbuf[QM_DBG_TMP_BUF_LEN];
843 	int len, ret;
844 
845 	if (*pos != 0)
846 		return 0;
847 
848 	if (count >= QM_DBG_TMP_BUF_LEN)
849 		return -ENOSPC;
850 
851 	len = simple_write_to_buffer(tbuf, QM_DBG_TMP_BUF_LEN - 1, pos, buf,
852 				     count);
853 	if (len < 0)
854 		return len;
855 
856 	tbuf[len] = '\0';
857 	if (kstrtoul(tbuf, 0, &val))
858 		return -EFAULT;
859 
860 	mutex_lock(&file->lock);
861 	switch (index) {
862 	case CURRENT_Q:
863 		ret = current_q_write(file, val);
864 		if (ret)
865 			goto err_input;
866 		break;
867 	case CLEAR_ENABLE:
868 		ret = clear_enable_write(file, val);
869 		if (ret)
870 			goto err_input;
871 		break;
872 	default:
873 		ret = -EINVAL;
874 		goto err_input;
875 	}
876 	mutex_unlock(&file->lock);
877 
878 	return count;
879 
880 err_input:
881 	mutex_unlock(&file->lock);
882 	return ret;
883 }
884 
885 static const struct file_operations qm_debug_fops = {
886 	.owner = THIS_MODULE,
887 	.open = simple_open,
888 	.read = qm_debug_read,
889 	.write = qm_debug_write,
890 };
891 
892 struct qm_dfx_registers {
893 	char  *reg_name;
894 	u64   reg_offset;
895 };
896 
897 #define CNT_CYC_REGS_NUM		10
898 static struct qm_dfx_registers qm_dfx_regs[] = {
899 	/* XXX_CNT are reading clear register */
900 	{"QM_ECC_1BIT_CNT               ",  0x104000ull},
901 	{"QM_ECC_MBIT_CNT               ",  0x104008ull},
902 	{"QM_DFX_MB_CNT                 ",  0x104018ull},
903 	{"QM_DFX_DB_CNT                 ",  0x104028ull},
904 	{"QM_DFX_SQE_CNT                ",  0x104038ull},
905 	{"QM_DFX_CQE_CNT                ",  0x104048ull},
906 	{"QM_DFX_SEND_SQE_TO_ACC_CNT    ",  0x104050ull},
907 	{"QM_DFX_WB_SQE_FROM_ACC_CNT    ",  0x104058ull},
908 	{"QM_DFX_ACC_FINISH_CNT         ",  0x104060ull},
909 	{"QM_DFX_CQE_ERR_CNT            ",  0x1040b4ull},
910 	{"QM_DFX_FUNS_ACTIVE_ST         ",  0x200ull},
911 	{"QM_ECC_1BIT_INF               ",  0x104004ull},
912 	{"QM_ECC_MBIT_INF               ",  0x10400cull},
913 	{"QM_DFX_ACC_RDY_VLD0           ",  0x1040a0ull},
914 	{"QM_DFX_ACC_RDY_VLD1           ",  0x1040a4ull},
915 	{"QM_DFX_AXI_RDY_VLD            ",  0x1040a8ull},
916 	{"QM_DFX_FF_ST0                 ",  0x1040c8ull},
917 	{"QM_DFX_FF_ST1                 ",  0x1040ccull},
918 	{"QM_DFX_FF_ST2                 ",  0x1040d0ull},
919 	{"QM_DFX_FF_ST3                 ",  0x1040d4ull},
920 	{"QM_DFX_FF_ST4                 ",  0x1040d8ull},
921 	{"QM_DFX_FF_ST5                 ",  0x1040dcull},
922 	{"QM_DFX_FF_ST6                 ",  0x1040e0ull},
923 	{"QM_IN_IDLE_ST                 ",  0x1040e4ull},
924 	{ NULL, 0}
925 };
926 
927 static struct qm_dfx_registers qm_vf_dfx_regs[] = {
928 	{"QM_DFX_FUNS_ACTIVE_ST         ",  0x200ull},
929 	{ NULL, 0}
930 };
931 
932 static int qm_regs_show(struct seq_file *s, void *unused)
933 {
934 	struct hisi_qm *qm = s->private;
935 	struct qm_dfx_registers *regs;
936 	u32 val;
937 
938 	if (qm->fun_type == QM_HW_PF)
939 		regs = qm_dfx_regs;
940 	else
941 		regs = qm_vf_dfx_regs;
942 
943 	while (regs->reg_name) {
944 		val = readl(qm->io_base + regs->reg_offset);
945 		seq_printf(s, "%s= 0x%08x\n", regs->reg_name, val);
946 		regs++;
947 	}
948 
949 	return 0;
950 }
951 
952 static int qm_regs_open(struct inode *inode, struct file *file)
953 {
954 	return single_open(file, qm_regs_show, inode->i_private);
955 }
956 
957 static const struct file_operations qm_regs_fops = {
958 	.owner = THIS_MODULE,
959 	.open = qm_regs_open,
960 	.read = seq_read,
961 	.release = single_release,
962 };
963 
964 static int qm_create_debugfs_file(struct hisi_qm *qm, enum qm_debug_file index)
965 {
966 	struct dentry *qm_d = qm->debug.qm_d, *tmp;
967 	struct debugfs_file *file = qm->debug.files + index;
968 
969 	tmp = debugfs_create_file(qm_debug_file_name[index], 0600, qm_d, file,
970 				  &qm_debug_fops);
971 	if (IS_ERR(tmp))
972 		return -ENOENT;
973 
974 	file->index = index;
975 	mutex_init(&file->lock);
976 	file->debug = &qm->debug;
977 
978 	return 0;
979 }
980 
981 static void qm_hw_error_init_v1(struct hisi_qm *qm, u32 ce, u32 nfe, u32 fe,
982 				u32 msi)
983 {
984 	dev_info(&qm->pdev->dev,
985 		 "QM v%d does not support hw error handle\n", qm->ver);
986 
987 	writel(QM_ABNORMAL_INT_MASK_VALUE, qm->io_base + QM_ABNORMAL_INT_MASK);
988 }
989 
990 static void qm_hw_error_init_v2(struct hisi_qm *qm, u32 ce, u32 nfe, u32 fe,
991 				u32 msi)
992 {
993 	u32 irq_enable = ce | nfe | fe | msi;
994 	u32 irq_unmask = ~irq_enable;
995 
996 	qm->error_mask = ce | nfe | fe;
997 	qm->msi_mask = msi;
998 
999 	/* configure error type */
1000 	writel(ce, qm->io_base + QM_RAS_CE_ENABLE);
1001 	writel(QM_RAS_CE_TIMES_PER_IRQ, qm->io_base + QM_RAS_CE_THRESHOLD);
1002 	writel(nfe, qm->io_base + QM_RAS_NFE_ENABLE);
1003 	writel(fe, qm->io_base + QM_RAS_FE_ENABLE);
1004 
1005 	/* use RAS irq default, so only set QM_RAS_MSI_INT_SEL for MSI */
1006 	writel(msi, qm->io_base + QM_RAS_MSI_INT_SEL);
1007 
1008 	irq_unmask &= readl(qm->io_base + QM_ABNORMAL_INT_MASK);
1009 	writel(irq_unmask, qm->io_base + QM_ABNORMAL_INT_MASK);
1010 }
1011 
1012 static void qm_log_hw_error(struct hisi_qm *qm, u32 error_status)
1013 {
1014 	const struct hisi_qm_hw_error *err = qm_hw_error;
1015 	struct device *dev = &qm->pdev->dev;
1016 	u32 reg_val, type, vf_num;
1017 
1018 	while (err->msg) {
1019 		if (err->int_msk & error_status) {
1020 			dev_err(dev, "%s [error status=0x%x] found\n",
1021 				err->msg, err->int_msk);
1022 
1023 			if (error_status & QM_DB_TIMEOUT) {
1024 				reg_val = readl(qm->io_base +
1025 						QM_ABNORMAL_INF01);
1026 				type = (reg_val & QM_DB_TIMEOUT_TYPE) >>
1027 				       QM_DB_TIMEOUT_TYPE_SHIFT;
1028 				vf_num = reg_val & QM_DB_TIMEOUT_VF;
1029 				dev_err(dev, "qm %s doorbell timeout in function %u\n",
1030 					qm_db_timeout[type], vf_num);
1031 			}
1032 
1033 			if (error_status & QM_OF_FIFO_OF) {
1034 				reg_val = readl(qm->io_base +
1035 						QM_ABNORMAL_INF00);
1036 				type = (reg_val & QM_FIFO_OVERFLOW_TYPE) >>
1037 				       QM_FIFO_OVERFLOW_TYPE_SHIFT;
1038 				vf_num = reg_val & QM_FIFO_OVERFLOW_VF;
1039 
1040 				if (type < ARRAY_SIZE(qm_fifo_overflow))
1041 					dev_err(dev, "qm %s fifo overflow in function %u\n",
1042 						qm_fifo_overflow[type],
1043 						vf_num);
1044 				else
1045 					dev_err(dev, "unknown error type\n");
1046 			}
1047 		}
1048 		err++;
1049 	}
1050 }
1051 
1052 static pci_ers_result_t qm_hw_error_handle_v2(struct hisi_qm *qm)
1053 {
1054 	u32 error_status, tmp;
1055 
1056 	/* read err sts */
1057 	tmp = readl(qm->io_base + QM_ABNORMAL_INT_STATUS);
1058 	error_status = qm->error_mask & tmp;
1059 
1060 	if (error_status) {
1061 		qm_log_hw_error(qm, error_status);
1062 
1063 		/* clear err sts */
1064 		writel(error_status, qm->io_base + QM_ABNORMAL_INT_SOURCE);
1065 
1066 		return PCI_ERS_RESULT_NEED_RESET;
1067 	}
1068 
1069 	return PCI_ERS_RESULT_RECOVERED;
1070 }
1071 
1072 static const struct hisi_qm_hw_ops qm_hw_ops_v1 = {
1073 	.qm_db = qm_db_v1,
1074 	.get_irq_num = qm_get_irq_num_v1,
1075 	.hw_error_init = qm_hw_error_init_v1,
1076 };
1077 
1078 static const struct hisi_qm_hw_ops qm_hw_ops_v2 = {
1079 	.get_vft = qm_get_vft_v2,
1080 	.qm_db = qm_db_v2,
1081 	.get_irq_num = qm_get_irq_num_v2,
1082 	.hw_error_init = qm_hw_error_init_v2,
1083 	.hw_error_handle = qm_hw_error_handle_v2,
1084 };
1085 
1086 static void *qm_get_avail_sqe(struct hisi_qp *qp)
1087 {
1088 	struct hisi_qp_status *qp_status = &qp->qp_status;
1089 	u16 sq_tail = qp_status->sq_tail;
1090 
1091 	if (unlikely(atomic_read(&qp->qp_status.used) == QM_Q_DEPTH))
1092 		return NULL;
1093 
1094 	return qp->sqe + sq_tail * qp->qm->sqe_size;
1095 }
1096 
1097 /**
1098  * hisi_qm_create_qp() - Create a queue pair from qm.
1099  * @qm: The qm we create a qp from.
1100  * @alg_type: Accelerator specific algorithm type in sqc.
1101  *
1102  * return created qp, -EBUSY if all qps in qm allocated, -ENOMEM if allocating
1103  * qp memory fails.
1104  */
1105 struct hisi_qp *hisi_qm_create_qp(struct hisi_qm *qm, u8 alg_type)
1106 {
1107 	struct device *dev = &qm->pdev->dev;
1108 	struct hisi_qp *qp;
1109 	int qp_id, ret;
1110 
1111 	qp = kzalloc(sizeof(*qp), GFP_KERNEL);
1112 	if (!qp)
1113 		return ERR_PTR(-ENOMEM);
1114 
1115 	write_lock(&qm->qps_lock);
1116 
1117 	qp_id = find_first_zero_bit(qm->qp_bitmap, qm->qp_num);
1118 	if (qp_id >= qm->qp_num) {
1119 		write_unlock(&qm->qps_lock);
1120 		dev_info(&qm->pdev->dev, "QM all queues are busy!\n");
1121 		ret = -EBUSY;
1122 		goto err_free_qp;
1123 	}
1124 	set_bit(qp_id, qm->qp_bitmap);
1125 	qm->qp_array[qp_id] = qp;
1126 
1127 	write_unlock(&qm->qps_lock);
1128 
1129 	qp->qm = qm;
1130 
1131 	if (qm->use_dma_api) {
1132 		qp->qdma.size = qm->sqe_size * QM_Q_DEPTH +
1133 				sizeof(struct qm_cqe) * QM_Q_DEPTH;
1134 		qp->qdma.va = dma_alloc_coherent(dev, qp->qdma.size,
1135 						 &qp->qdma.dma, GFP_KERNEL);
1136 		if (!qp->qdma.va) {
1137 			ret = -ENOMEM;
1138 			goto err_clear_bit;
1139 		}
1140 
1141 		dev_dbg(dev, "allocate qp dma buf(va=%pK, dma=%pad, size=%zx)\n",
1142 			qp->qdma.va, &qp->qdma.dma, qp->qdma.size);
1143 	}
1144 
1145 	qp->qp_id = qp_id;
1146 	qp->alg_type = alg_type;
1147 	INIT_WORK(&qp->work, qm_qp_work_func);
1148 	qp->wq = alloc_workqueue("hisi_qm", WQ_UNBOUND | WQ_HIGHPRI |
1149 				 WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM, 0);
1150 	if (!qp->wq) {
1151 		ret = -EFAULT;
1152 		goto err_free_qp_mem;
1153 	}
1154 
1155 	return qp;
1156 
1157 err_free_qp_mem:
1158 	if (qm->use_dma_api)
1159 		dma_free_coherent(dev, qp->qdma.size, qp->qdma.va,
1160 				  qp->qdma.dma);
1161 err_clear_bit:
1162 	write_lock(&qm->qps_lock);
1163 	qm->qp_array[qp_id] = NULL;
1164 	clear_bit(qp_id, qm->qp_bitmap);
1165 	write_unlock(&qm->qps_lock);
1166 err_free_qp:
1167 	kfree(qp);
1168 	return ERR_PTR(ret);
1169 }
1170 EXPORT_SYMBOL_GPL(hisi_qm_create_qp);
1171 
1172 /**
1173  * hisi_qm_release_qp() - Release a qp back to its qm.
1174  * @qp: The qp we want to release.
1175  *
1176  * This function releases the resource of a qp.
1177  */
1178 void hisi_qm_release_qp(struct hisi_qp *qp)
1179 {
1180 	struct hisi_qm *qm = qp->qm;
1181 	struct qm_dma *qdma = &qp->qdma;
1182 	struct device *dev = &qm->pdev->dev;
1183 
1184 	if (qm->use_dma_api && qdma->va)
1185 		dma_free_coherent(dev, qdma->size, qdma->va, qdma->dma);
1186 
1187 	write_lock(&qm->qps_lock);
1188 	qm->qp_array[qp->qp_id] = NULL;
1189 	clear_bit(qp->qp_id, qm->qp_bitmap);
1190 	write_unlock(&qm->qps_lock);
1191 
1192 	kfree(qp);
1193 }
1194 EXPORT_SYMBOL_GPL(hisi_qm_release_qp);
1195 
1196 static int qm_qp_ctx_cfg(struct hisi_qp *qp, int qp_id, int pasid)
1197 {
1198 	struct hisi_qm *qm = qp->qm;
1199 	struct device *dev = &qm->pdev->dev;
1200 	enum qm_hw_ver ver = qm->ver;
1201 	struct qm_sqc *sqc;
1202 	struct qm_cqc *cqc;
1203 	dma_addr_t sqc_dma;
1204 	dma_addr_t cqc_dma;
1205 	int ret;
1206 
1207 	qm_init_qp_status(qp);
1208 
1209 	sqc = kzalloc(sizeof(struct qm_sqc), GFP_KERNEL);
1210 	if (!sqc)
1211 		return -ENOMEM;
1212 	sqc_dma = dma_map_single(dev, sqc, sizeof(struct qm_sqc),
1213 				 DMA_TO_DEVICE);
1214 	if (dma_mapping_error(dev, sqc_dma)) {
1215 		kfree(sqc);
1216 		return -ENOMEM;
1217 	}
1218 
1219 	INIT_QC_COMMON(sqc, qp->sqe_dma, pasid);
1220 	if (ver == QM_HW_V1) {
1221 		sqc->dw3 = QM_MK_SQC_DW3_V1(0, 0, 0, qm->sqe_size);
1222 		sqc->w8 = QM_Q_DEPTH - 1;
1223 	} else if (ver == QM_HW_V2) {
1224 		sqc->dw3 = QM_MK_SQC_DW3_V2(qm->sqe_size);
1225 		sqc->w8 = 0; /* rand_qc */
1226 	}
1227 	sqc->cq_num = qp_id;
1228 	sqc->w13 = QM_MK_SQC_W13(0, 1, qp->alg_type);
1229 
1230 	ret = qm_mb(qm, QM_MB_CMD_SQC, sqc_dma, qp_id, 0);
1231 	dma_unmap_single(dev, sqc_dma, sizeof(struct qm_sqc), DMA_TO_DEVICE);
1232 	kfree(sqc);
1233 	if (ret)
1234 		return ret;
1235 
1236 	cqc = kzalloc(sizeof(struct qm_cqc), GFP_KERNEL);
1237 	if (!cqc)
1238 		return -ENOMEM;
1239 	cqc_dma = dma_map_single(dev, cqc, sizeof(struct qm_cqc),
1240 				 DMA_TO_DEVICE);
1241 	if (dma_mapping_error(dev, cqc_dma)) {
1242 		kfree(cqc);
1243 		return -ENOMEM;
1244 	}
1245 
1246 	INIT_QC_COMMON(cqc, qp->cqe_dma, pasid);
1247 	if (ver == QM_HW_V1) {
1248 		cqc->dw3 = QM_MK_CQC_DW3_V1(0, 0, 0, 4);
1249 		cqc->w8 = QM_Q_DEPTH - 1;
1250 	} else if (ver == QM_HW_V2) {
1251 		cqc->dw3 = QM_MK_CQC_DW3_V2(4);
1252 		cqc->w8 = 0;
1253 	}
1254 	cqc->dw6 = 1 << QM_CQ_PHASE_SHIFT | 1 << QM_CQ_FLAG_SHIFT;
1255 
1256 	ret = qm_mb(qm, QM_MB_CMD_CQC, cqc_dma, qp_id, 0);
1257 	dma_unmap_single(dev, cqc_dma, sizeof(struct qm_cqc), DMA_TO_DEVICE);
1258 	kfree(cqc);
1259 
1260 	return ret;
1261 }
1262 
1263 /**
1264  * hisi_qm_start_qp() - Start a qp into running.
1265  * @qp: The qp we want to start to run.
1266  * @arg: Accelerator specific argument.
1267  *
1268  * After this function, qp can receive request from user. Return qp_id if
1269  * successful, Return -EBUSY if failed.
1270  */
1271 int hisi_qm_start_qp(struct hisi_qp *qp, unsigned long arg)
1272 {
1273 	struct hisi_qm *qm = qp->qm;
1274 	struct device *dev = &qm->pdev->dev;
1275 	enum qm_hw_ver ver = qm->ver;
1276 	int qp_id = qp->qp_id;
1277 	int pasid = arg;
1278 	size_t off = 0;
1279 	int ret;
1280 
1281 #define QP_INIT_BUF(qp, type, size) do { \
1282 	(qp)->type = ((qp)->qdma.va + (off)); \
1283 	(qp)->type##_dma = (qp)->qdma.dma + (off); \
1284 	off += (size); \
1285 } while (0)
1286 
1287 	if (!qp->qdma.dma) {
1288 		dev_err(dev, "cannot get qm dma buffer\n");
1289 		return -EINVAL;
1290 	}
1291 
1292 	/* sq need 128 bytes alignment */
1293 	if (qp->qdma.dma & QM_SQE_DATA_ALIGN_MASK) {
1294 		dev_err(dev, "qm sq is not aligned to 128 byte\n");
1295 		return -EINVAL;
1296 	}
1297 
1298 	QP_INIT_BUF(qp, sqe, qm->sqe_size * QM_Q_DEPTH);
1299 	QP_INIT_BUF(qp, cqe, sizeof(struct qm_cqe) * QM_Q_DEPTH);
1300 
1301 	dev_dbg(dev, "init qp buffer(v%d):\n"
1302 		     " sqe	(%pK, %lx)\n"
1303 		     " cqe	(%pK, %lx)\n",
1304 		     ver, qp->sqe, (unsigned long)qp->sqe_dma,
1305 		     qp->cqe, (unsigned long)qp->cqe_dma);
1306 
1307 	ret = qm_qp_ctx_cfg(qp, qp_id, pasid);
1308 	if (ret)
1309 		return ret;
1310 
1311 	dev_dbg(dev, "queue %d started\n", qp_id);
1312 
1313 	return qp_id;
1314 }
1315 EXPORT_SYMBOL_GPL(hisi_qm_start_qp);
1316 
1317 /**
1318  * hisi_qm_stop_qp() - Stop a qp in qm.
1319  * @qp: The qp we want to stop.
1320  *
1321  * This function is reverse of hisi_qm_start_qp. Return 0 if successful.
1322  */
1323 int hisi_qm_stop_qp(struct hisi_qp *qp)
1324 {
1325 	struct device *dev = &qp->qm->pdev->dev;
1326 	int i = 0;
1327 
1328 	/* it is stopped */
1329 	if (test_bit(QP_STOP, &qp->qp_status.flags))
1330 		return 0;
1331 
1332 	while (atomic_read(&qp->qp_status.used)) {
1333 		i++;
1334 		msleep(20);
1335 		if (i == 10) {
1336 			dev_err(dev, "Cannot drain out data for stopping, Force to stop!\n");
1337 			return 0;
1338 		}
1339 	}
1340 
1341 	set_bit(QP_STOP, &qp->qp_status.flags);
1342 
1343 	dev_dbg(dev, "stop queue %u!", qp->qp_id);
1344 
1345 	return 0;
1346 }
1347 EXPORT_SYMBOL_GPL(hisi_qm_stop_qp);
1348 
1349 /**
1350  * hisi_qp_send() - Queue up a task in the hardware queue.
1351  * @qp: The qp in which to put the message.
1352  * @msg: The message.
1353  *
1354  * This function will return -EBUSY if qp is currently full, and -EAGAIN
1355  * if qp related qm is resetting.
1356  */
1357 int hisi_qp_send(struct hisi_qp *qp, const void *msg)
1358 {
1359 	struct hisi_qp_status *qp_status = &qp->qp_status;
1360 	u16 sq_tail = qp_status->sq_tail;
1361 	u16 sq_tail_next = (sq_tail + 1) % QM_Q_DEPTH;
1362 	void *sqe = qm_get_avail_sqe(qp);
1363 
1364 	if (unlikely(test_bit(QP_STOP, &qp->qp_status.flags))) {
1365 		dev_info(&qp->qm->pdev->dev, "QP is stopped or resetting\n");
1366 		return -EAGAIN;
1367 	}
1368 
1369 	if (!sqe)
1370 		return -EBUSY;
1371 
1372 	memcpy(sqe, msg, qp->qm->sqe_size);
1373 
1374 	qm_db(qp->qm, qp->qp_id, QM_DOORBELL_CMD_SQ, sq_tail_next, 0);
1375 	atomic_inc(&qp->qp_status.used);
1376 	qp_status->sq_tail = sq_tail_next;
1377 
1378 	return 0;
1379 }
1380 EXPORT_SYMBOL_GPL(hisi_qp_send);
1381 
1382 static void hisi_qm_cache_wb(struct hisi_qm *qm)
1383 {
1384 	unsigned int val;
1385 
1386 	if (qm->ver == QM_HW_V2) {
1387 		writel(0x1, qm->io_base + QM_CACHE_WB_START);
1388 		if (readl_relaxed_poll_timeout(qm->io_base + QM_CACHE_WB_DONE,
1389 					       val, val & BIT(0), 10, 1000))
1390 			dev_err(&qm->pdev->dev, "QM writeback sqc cache fail!\n");
1391 	}
1392 }
1393 
1394 /**
1395  * hisi_qm_init() - Initialize configures about qm.
1396  * @qm: The qm needing init.
1397  *
1398  * This function init qm, then we can call hisi_qm_start to put qm into work.
1399  */
1400 int hisi_qm_init(struct hisi_qm *qm)
1401 {
1402 	struct pci_dev *pdev = qm->pdev;
1403 	struct device *dev = &pdev->dev;
1404 	unsigned int num_vec;
1405 	int ret;
1406 
1407 	switch (qm->ver) {
1408 	case QM_HW_V1:
1409 		qm->ops = &qm_hw_ops_v1;
1410 		break;
1411 	case QM_HW_V2:
1412 		qm->ops = &qm_hw_ops_v2;
1413 		break;
1414 	default:
1415 		return -EINVAL;
1416 	}
1417 
1418 	ret = pci_enable_device_mem(pdev);
1419 	if (ret < 0) {
1420 		dev_err(&pdev->dev, "Failed to enable device mem!\n");
1421 		return ret;
1422 	}
1423 
1424 	ret = pci_request_mem_regions(pdev, qm->dev_name);
1425 	if (ret < 0) {
1426 		dev_err(&pdev->dev, "Failed to request mem regions!\n");
1427 		goto err_disable_pcidev;
1428 	}
1429 
1430 	qm->io_base = ioremap(pci_resource_start(pdev, PCI_BAR_2),
1431 			      pci_resource_len(qm->pdev, PCI_BAR_2));
1432 	if (!qm->io_base) {
1433 		ret = -EIO;
1434 		goto err_release_mem_regions;
1435 	}
1436 
1437 	ret = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(64));
1438 	if (ret < 0)
1439 		goto err_iounmap;
1440 	pci_set_master(pdev);
1441 
1442 	if (!qm->ops->get_irq_num) {
1443 		ret = -EOPNOTSUPP;
1444 		goto err_iounmap;
1445 	}
1446 	num_vec = qm->ops->get_irq_num(qm);
1447 	ret = pci_alloc_irq_vectors(pdev, num_vec, num_vec, PCI_IRQ_MSI);
1448 	if (ret < 0) {
1449 		dev_err(dev, "Failed to enable MSI vectors!\n");
1450 		goto err_iounmap;
1451 	}
1452 
1453 	ret = qm_irq_register(qm);
1454 	if (ret)
1455 		goto err_free_irq_vectors;
1456 
1457 	mutex_init(&qm->mailbox_lock);
1458 	rwlock_init(&qm->qps_lock);
1459 
1460 	dev_dbg(dev, "init qm %s with %s\n", pdev->is_physfn ? "pf" : "vf",
1461 		qm->use_dma_api ? "dma api" : "iommu api");
1462 
1463 	return 0;
1464 
1465 err_free_irq_vectors:
1466 	pci_free_irq_vectors(pdev);
1467 err_iounmap:
1468 	iounmap(qm->io_base);
1469 err_release_mem_regions:
1470 	pci_release_mem_regions(pdev);
1471 err_disable_pcidev:
1472 	pci_disable_device(pdev);
1473 
1474 	return ret;
1475 }
1476 EXPORT_SYMBOL_GPL(hisi_qm_init);
1477 
1478 /**
1479  * hisi_qm_uninit() - Uninitialize qm.
1480  * @qm: The qm needed uninit.
1481  *
1482  * This function uninits qm related device resources.
1483  */
1484 void hisi_qm_uninit(struct hisi_qm *qm)
1485 {
1486 	struct pci_dev *pdev = qm->pdev;
1487 	struct device *dev = &pdev->dev;
1488 
1489 	if (qm->use_dma_api && qm->qdma.va) {
1490 		hisi_qm_cache_wb(qm);
1491 		dma_free_coherent(dev, qm->qdma.size,
1492 				  qm->qdma.va, qm->qdma.dma);
1493 		memset(&qm->qdma, 0, sizeof(qm->qdma));
1494 	}
1495 
1496 	qm_irq_unregister(qm);
1497 	pci_free_irq_vectors(pdev);
1498 	iounmap(qm->io_base);
1499 	pci_release_mem_regions(pdev);
1500 	pci_disable_device(pdev);
1501 }
1502 EXPORT_SYMBOL_GPL(hisi_qm_uninit);
1503 
1504 /**
1505  * hisi_qm_get_vft() - Get vft from a qm.
1506  * @qm: The qm we want to get its vft.
1507  * @base: The base number of queue in vft.
1508  * @number: The number of queues in vft.
1509  *
1510  * We can allocate multiple queues to a qm by configuring virtual function
1511  * table. We get related configures by this function. Normally, we call this
1512  * function in VF driver to get the queue information.
1513  *
1514  * qm hw v1 does not support this interface.
1515  */
1516 int hisi_qm_get_vft(struct hisi_qm *qm, u32 *base, u32 *number)
1517 {
1518 	if (!base || !number)
1519 		return -EINVAL;
1520 
1521 	if (!qm->ops->get_vft) {
1522 		dev_err(&qm->pdev->dev, "Don't support vft read!\n");
1523 		return -EINVAL;
1524 	}
1525 
1526 	return qm->ops->get_vft(qm, base, number);
1527 }
1528 EXPORT_SYMBOL_GPL(hisi_qm_get_vft);
1529 
1530 /**
1531  * hisi_qm_set_vft() - Set "virtual function table" for a qm.
1532  * @fun_num: Number of operated function.
1533  * @qm: The qm in which to set vft, alway in a PF.
1534  * @base: The base number of queue in vft.
1535  * @number: The number of queues in vft. 0 means invalid vft.
1536  *
1537  * This function is alway called in PF driver, it is used to assign queues
1538  * among PF and VFs.
1539  *
1540  * Assign queues A~B to PF: hisi_qm_set_vft(qm, 0, A, B - A + 1)
1541  * Assign queues A~B to VF: hisi_qm_set_vft(qm, 2, A, B - A + 1)
1542  * (VF function number 0x2)
1543  */
1544 int hisi_qm_set_vft(struct hisi_qm *qm, u32 fun_num, u32 base,
1545 		    u32 number)
1546 {
1547 	u32 max_q_num = qm->ctrl_qp_num;
1548 
1549 	if (base >= max_q_num || number > max_q_num ||
1550 	    (base + number) > max_q_num)
1551 		return -EINVAL;
1552 
1553 	return qm_set_sqc_cqc_vft(qm, fun_num, base, number);
1554 }
1555 EXPORT_SYMBOL_GPL(hisi_qm_set_vft);
1556 
1557 static void qm_init_eq_aeq_status(struct hisi_qm *qm)
1558 {
1559 	struct hisi_qm_status *status = &qm->status;
1560 
1561 	status->eq_head = 0;
1562 	status->aeq_head = 0;
1563 	status->eqc_phase = 1;
1564 	status->aeqc_phase = 1;
1565 }
1566 
1567 static int qm_eq_ctx_cfg(struct hisi_qm *qm)
1568 {
1569 	struct device *dev = &qm->pdev->dev;
1570 	struct qm_eqc *eqc;
1571 	struct qm_aeqc *aeqc;
1572 	dma_addr_t eqc_dma;
1573 	dma_addr_t aeqc_dma;
1574 	int ret;
1575 
1576 	qm_init_eq_aeq_status(qm);
1577 
1578 	eqc = kzalloc(sizeof(struct qm_eqc), GFP_KERNEL);
1579 	if (!eqc)
1580 		return -ENOMEM;
1581 	eqc_dma = dma_map_single(dev, eqc, sizeof(struct qm_eqc),
1582 				 DMA_TO_DEVICE);
1583 	if (dma_mapping_error(dev, eqc_dma)) {
1584 		kfree(eqc);
1585 		return -ENOMEM;
1586 	}
1587 
1588 	eqc->base_l = lower_32_bits(qm->eqe_dma);
1589 	eqc->base_h = upper_32_bits(qm->eqe_dma);
1590 	if (qm->ver == QM_HW_V1)
1591 		eqc->dw3 = QM_EQE_AEQE_SIZE;
1592 	eqc->dw6 = (QM_Q_DEPTH - 1) | (1 << QM_EQC_PHASE_SHIFT);
1593 	ret = qm_mb(qm, QM_MB_CMD_EQC, eqc_dma, 0, 0);
1594 	dma_unmap_single(dev, eqc_dma, sizeof(struct qm_eqc), DMA_TO_DEVICE);
1595 	kfree(eqc);
1596 	if (ret)
1597 		return ret;
1598 
1599 	aeqc = kzalloc(sizeof(struct qm_aeqc), GFP_KERNEL);
1600 	if (!aeqc)
1601 		return -ENOMEM;
1602 	aeqc_dma = dma_map_single(dev, aeqc, sizeof(struct qm_aeqc),
1603 				  DMA_TO_DEVICE);
1604 	if (dma_mapping_error(dev, aeqc_dma)) {
1605 		kfree(aeqc);
1606 		return -ENOMEM;
1607 	}
1608 
1609 	aeqc->base_l = lower_32_bits(qm->aeqe_dma);
1610 	aeqc->base_h = upper_32_bits(qm->aeqe_dma);
1611 	aeqc->dw6 = (QM_Q_DEPTH - 1) | (1 << QM_EQC_PHASE_SHIFT);
1612 
1613 	ret = qm_mb(qm, QM_MB_CMD_AEQC, aeqc_dma, 0, 0);
1614 	dma_unmap_single(dev, aeqc_dma, sizeof(struct qm_aeqc), DMA_TO_DEVICE);
1615 	kfree(aeqc);
1616 
1617 	return ret;
1618 }
1619 
1620 static int __hisi_qm_start(struct hisi_qm *qm)
1621 {
1622 	struct pci_dev *pdev = qm->pdev;
1623 	struct device *dev = &pdev->dev;
1624 	size_t off = 0;
1625 	int ret;
1626 
1627 #define QM_INIT_BUF(qm, type, num) do { \
1628 	(qm)->type = ((qm)->qdma.va + (off)); \
1629 	(qm)->type##_dma = (qm)->qdma.dma + (off); \
1630 	off += QMC_ALIGN(sizeof(struct qm_##type) * (num)); \
1631 } while (0)
1632 
1633 	WARN_ON(!qm->qdma.dma);
1634 
1635 	if (qm->qp_num == 0)
1636 		return -EINVAL;
1637 
1638 	if (qm->fun_type == QM_HW_PF) {
1639 		ret = qm_dev_mem_reset(qm);
1640 		if (ret)
1641 			return ret;
1642 
1643 		ret = hisi_qm_set_vft(qm, 0, qm->qp_base, qm->qp_num);
1644 		if (ret)
1645 			return ret;
1646 	}
1647 
1648 	QM_INIT_BUF(qm, eqe, QM_Q_DEPTH);
1649 	QM_INIT_BUF(qm, aeqe, QM_Q_DEPTH);
1650 	QM_INIT_BUF(qm, sqc, qm->qp_num);
1651 	QM_INIT_BUF(qm, cqc, qm->qp_num);
1652 
1653 	dev_dbg(dev, "init qm buffer:\n"
1654 		     " eqe	(%pK, %lx)\n"
1655 		     " aeqe	(%pK, %lx)\n"
1656 		     " sqc	(%pK, %lx)\n"
1657 		     " cqc	(%pK, %lx)\n",
1658 		     qm->eqe, (unsigned long)qm->eqe_dma,
1659 		     qm->aeqe, (unsigned long)qm->aeqe_dma,
1660 		     qm->sqc, (unsigned long)qm->sqc_dma,
1661 		     qm->cqc, (unsigned long)qm->cqc_dma);
1662 
1663 	ret = qm_eq_ctx_cfg(qm);
1664 	if (ret)
1665 		return ret;
1666 
1667 	ret = qm_mb(qm, QM_MB_CMD_SQC_BT, qm->sqc_dma, 0, 0);
1668 	if (ret)
1669 		return ret;
1670 
1671 	ret = qm_mb(qm, QM_MB_CMD_CQC_BT, qm->cqc_dma, 0, 0);
1672 	if (ret)
1673 		return ret;
1674 
1675 	writel(0x0, qm->io_base + QM_VF_EQ_INT_MASK);
1676 	writel(0x0, qm->io_base + QM_VF_AEQ_INT_MASK);
1677 
1678 	return 0;
1679 }
1680 
1681 /**
1682  * hisi_qm_start() - start qm
1683  * @qm: The qm to be started.
1684  *
1685  * This function starts a qm, then we can allocate qp from this qm.
1686  */
1687 int hisi_qm_start(struct hisi_qm *qm)
1688 {
1689 	struct device *dev = &qm->pdev->dev;
1690 
1691 	dev_dbg(dev, "qm start with %d queue pairs\n", qm->qp_num);
1692 
1693 	if (!qm->qp_num) {
1694 		dev_err(dev, "qp_num should not be 0\n");
1695 		return -EINVAL;
1696 	}
1697 
1698 	if (!qm->qp_bitmap) {
1699 		qm->qp_bitmap = devm_kcalloc(dev, BITS_TO_LONGS(qm->qp_num),
1700 					     sizeof(long), GFP_KERNEL);
1701 		qm->qp_array = devm_kcalloc(dev, qm->qp_num,
1702 					    sizeof(struct hisi_qp *),
1703 					    GFP_KERNEL);
1704 		if (!qm->qp_bitmap || !qm->qp_array)
1705 			return -ENOMEM;
1706 	}
1707 
1708 	if (!qm->use_dma_api) {
1709 		dev_dbg(&qm->pdev->dev, "qm delay start\n");
1710 		return 0;
1711 	} else if (!qm->qdma.va) {
1712 		qm->qdma.size = QMC_ALIGN(sizeof(struct qm_eqe) * QM_Q_DEPTH) +
1713 				QMC_ALIGN(sizeof(struct qm_aeqe) * QM_Q_DEPTH) +
1714 				QMC_ALIGN(sizeof(struct qm_sqc) * qm->qp_num) +
1715 				QMC_ALIGN(sizeof(struct qm_cqc) * qm->qp_num);
1716 		qm->qdma.va = dma_alloc_coherent(dev, qm->qdma.size,
1717 						 &qm->qdma.dma, GFP_KERNEL);
1718 		dev_dbg(dev, "allocate qm dma buf(va=%pK, dma=%pad, size=%zx)\n",
1719 			qm->qdma.va, &qm->qdma.dma, qm->qdma.size);
1720 		if (!qm->qdma.va)
1721 			return -ENOMEM;
1722 	}
1723 
1724 	return __hisi_qm_start(qm);
1725 }
1726 EXPORT_SYMBOL_GPL(hisi_qm_start);
1727 
1728 /**
1729  * hisi_qm_stop() - Stop a qm.
1730  * @qm: The qm which will be stopped.
1731  *
1732  * This function stops qm and its qps, then qm can not accept request.
1733  * Related resources are not released at this state, we can use hisi_qm_start
1734  * to let qm start again.
1735  */
1736 int hisi_qm_stop(struct hisi_qm *qm)
1737 {
1738 	struct device *dev;
1739 	struct hisi_qp *qp;
1740 	int ret = 0, i;
1741 
1742 	if (!qm || !qm->pdev) {
1743 		WARN_ON(1);
1744 		return -EINVAL;
1745 	}
1746 
1747 	dev = &qm->pdev->dev;
1748 
1749 	/* Mask eq and aeq irq */
1750 	writel(0x1, qm->io_base + QM_VF_EQ_INT_MASK);
1751 	writel(0x1, qm->io_base + QM_VF_AEQ_INT_MASK);
1752 
1753 	/* Stop all qps belong to this qm */
1754 	for (i = 0; i < qm->qp_num; i++) {
1755 		qp = qm->qp_array[i];
1756 		if (qp) {
1757 			ret = hisi_qm_stop_qp(qp);
1758 			if (ret < 0) {
1759 				dev_err(dev, "Failed to stop qp%d!\n", i);
1760 				return -EBUSY;
1761 			}
1762 		}
1763 	}
1764 
1765 	if (qm->fun_type == QM_HW_PF) {
1766 		ret = hisi_qm_set_vft(qm, 0, 0, 0);
1767 		if (ret < 0)
1768 			dev_err(dev, "Failed to set vft!\n");
1769 	}
1770 
1771 	return ret;
1772 }
1773 EXPORT_SYMBOL_GPL(hisi_qm_stop);
1774 
1775 /**
1776  * hisi_qm_debug_init() - Initialize qm related debugfs files.
1777  * @qm: The qm for which we want to add debugfs files.
1778  *
1779  * Create qm related debugfs files.
1780  */
1781 int hisi_qm_debug_init(struct hisi_qm *qm)
1782 {
1783 	struct dentry *qm_d, *qm_regs;
1784 	int i, ret;
1785 
1786 	qm_d = debugfs_create_dir("qm", qm->debug.debug_root);
1787 	if (IS_ERR(qm_d))
1788 		return -ENOENT;
1789 	qm->debug.qm_d = qm_d;
1790 
1791 	/* only show this in PF */
1792 	if (qm->fun_type == QM_HW_PF)
1793 		for (i = CURRENT_Q; i < DEBUG_FILE_NUM; i++)
1794 			if (qm_create_debugfs_file(qm, i)) {
1795 				ret = -ENOENT;
1796 				goto failed_to_create;
1797 			}
1798 
1799 	qm_regs = debugfs_create_file("qm_regs", 0444, qm->debug.qm_d, qm,
1800 				      &qm_regs_fops);
1801 	if (IS_ERR(qm_regs)) {
1802 		ret = -ENOENT;
1803 		goto failed_to_create;
1804 	}
1805 
1806 	return 0;
1807 
1808 failed_to_create:
1809 	debugfs_remove_recursive(qm_d);
1810 	return ret;
1811 }
1812 EXPORT_SYMBOL_GPL(hisi_qm_debug_init);
1813 
1814 /**
1815  * hisi_qm_debug_regs_clear() - clear qm debug related registers.
1816  * @qm: The qm for which we want to clear its debug registers.
1817  */
1818 void hisi_qm_debug_regs_clear(struct hisi_qm *qm)
1819 {
1820 	struct qm_dfx_registers *regs;
1821 	int i;
1822 
1823 	/* clear current_q */
1824 	writel(0x0, qm->io_base + QM_DFX_SQE_CNT_VF_SQN);
1825 	writel(0x0, qm->io_base + QM_DFX_CQE_CNT_VF_CQN);
1826 
1827 	/*
1828 	 * these registers are reading and clearing, so clear them after
1829 	 * reading them.
1830 	 */
1831 	writel(0x1, qm->io_base + QM_DFX_CNT_CLR_CE);
1832 
1833 	regs = qm_dfx_regs;
1834 	for (i = 0; i < CNT_CYC_REGS_NUM; i++) {
1835 		readl(qm->io_base + regs->reg_offset);
1836 		regs++;
1837 	}
1838 
1839 	writel(0x0, qm->io_base + QM_DFX_CNT_CLR_CE);
1840 }
1841 EXPORT_SYMBOL_GPL(hisi_qm_debug_regs_clear);
1842 
1843 /**
1844  * hisi_qm_hw_error_init() - Configure qm hardware error report method.
1845  * @qm: The qm which we want to configure.
1846  * @ce: Bit mask of correctable error configure.
1847  * @nfe: Bit mask of non-fatal error configure.
1848  * @fe: Bit mask of fatal error configure.
1849  * @msi: Bit mask of error reported by message signal interrupt.
1850  *
1851  * Hardware errors of qm can be reported either by RAS interrupts which will
1852  * be handled by UEFI and then PCIe AER or by device MSI. User can configure
1853  * each error to use either of above two methods. For RAS interrupts, we can
1854  * configure an error as one of correctable error, non-fatal error or
1855  * fatal error.
1856  *
1857  * Bits indicating errors can be configured to ce, nfe, fe and msi to enable
1858  * related report methods. Error report will be masked if related error bit
1859  * does not configure.
1860  */
1861 void hisi_qm_hw_error_init(struct hisi_qm *qm, u32 ce, u32 nfe, u32 fe,
1862 			   u32 msi)
1863 {
1864 	if (!qm->ops->hw_error_init) {
1865 		dev_err(&qm->pdev->dev, "QM version %d doesn't support hw error handling!\n",
1866 			qm->ver);
1867 		return;
1868 	}
1869 
1870 	qm->ops->hw_error_init(qm, ce, nfe, fe, msi);
1871 }
1872 EXPORT_SYMBOL_GPL(hisi_qm_hw_error_init);
1873 
1874 /**
1875  * hisi_qm_hw_error_handle() - Handle qm non-fatal hardware errors.
1876  * @qm: The qm which has non-fatal hardware errors.
1877  *
1878  * Accelerators use this function to handle qm non-fatal hardware errors.
1879  */
1880 int hisi_qm_hw_error_handle(struct hisi_qm *qm)
1881 {
1882 	if (!qm->ops->hw_error_handle) {
1883 		dev_err(&qm->pdev->dev, "QM version %d doesn't support hw error report!\n",
1884 			qm->ver);
1885 		return PCI_ERS_RESULT_NONE;
1886 	}
1887 
1888 	return qm->ops->hw_error_handle(qm);
1889 }
1890 EXPORT_SYMBOL_GPL(hisi_qm_hw_error_handle);
1891 
1892 /**
1893  * hisi_qm_get_hw_version() - Get hardware version of a qm.
1894  * @pdev: The device which hardware version we want to get.
1895  *
1896  * This function gets the hardware version of a qm. Return QM_HW_UNKNOWN
1897  * if the hardware version is not supported.
1898  */
1899 enum qm_hw_ver hisi_qm_get_hw_version(struct pci_dev *pdev)
1900 {
1901 	switch (pdev->revision) {
1902 	case QM_HW_V1:
1903 	case QM_HW_V2:
1904 		return pdev->revision;
1905 	default:
1906 		return QM_HW_UNKNOWN;
1907 	}
1908 }
1909 EXPORT_SYMBOL_GPL(hisi_qm_get_hw_version);
1910 
1911 MODULE_LICENSE("GPL v2");
1912 MODULE_AUTHOR("Zhou Wang <wangzhou1@hisilicon.com>");
1913 MODULE_DESCRIPTION("HiSilicon Accelerator queue manager driver");
1914