xref: /openbmc/linux/drivers/crypto/hisilicon/qm.c (revision f125e2d4)
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)		(le16_to_cpu((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)		((le32_to_cpu((eqe)->dw0) >> 16) & 0x1)
70 #define QM_EQE_CQN_MASK			GENMASK(15, 0)
71 
72 #define QM_AEQE_PHASE(aeqe)		((le32_to_cpu((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 = cpu_to_le32(lower_32_bits(base));	\
176 	(qc)->base_h = cpu_to_le32(upper_32_bits(base));	\
177 	(qc)->dw3 = 0;						\
178 	(qc)->w8 = 0;						\
179 	(qc)->rsvd0 = 0;					\
180 	(qc)->pasid = cpu_to_le16(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 	if (!IS_ENABLED(CONFIG_ARM64)) {
335 		memcpy_toio(fun_base, src, 16);
336 		wmb();
337 		return;
338 	}
339 
340 	asm volatile("ldp %0, %1, %3\n"
341 		     "stp %0, %1, %2\n"
342 		     "dsb sy\n"
343 		     : "=&r" (tmp0),
344 		       "=&r" (tmp1),
345 		       "+Q" (*((char __iomem *)fun_base))
346 		     : "Q" (*((char *)src))
347 		     : "memory");
348 }
349 
350 static int qm_mb(struct hisi_qm *qm, u8 cmd, dma_addr_t dma_addr, u16 queue,
351 		 bool op)
352 {
353 	struct qm_mailbox mailbox;
354 	int ret = 0;
355 
356 	dev_dbg(&qm->pdev->dev, "QM mailbox request to q%u: %u-%llx\n",
357 		queue, cmd, (unsigned long long)dma_addr);
358 
359 	mailbox.w0 = cpu_to_le16(cmd |
360 		     (op ? 0x1 << QM_MB_OP_SHIFT : 0) |
361 		     (0x1 << QM_MB_BUSY_SHIFT));
362 	mailbox.queue_num = cpu_to_le16(queue);
363 	mailbox.base_l = cpu_to_le32(lower_32_bits(dma_addr));
364 	mailbox.base_h = cpu_to_le32(upper_32_bits(dma_addr));
365 	mailbox.rsvd = 0;
366 
367 	mutex_lock(&qm->mailbox_lock);
368 
369 	if (unlikely(qm_wait_mb_ready(qm))) {
370 		ret = -EBUSY;
371 		dev_err(&qm->pdev->dev, "QM mailbox is busy to start!\n");
372 		goto busy_unlock;
373 	}
374 
375 	qm_mb_write(qm, &mailbox);
376 
377 	if (unlikely(qm_wait_mb_ready(qm))) {
378 		ret = -EBUSY;
379 		dev_err(&qm->pdev->dev, "QM mailbox operation timeout!\n");
380 		goto busy_unlock;
381 	}
382 
383 busy_unlock:
384 	mutex_unlock(&qm->mailbox_lock);
385 
386 	return ret;
387 }
388 
389 static void qm_db_v1(struct hisi_qm *qm, u16 qn, u8 cmd, u16 index, u8 priority)
390 {
391 	u64 doorbell;
392 
393 	doorbell = qn | ((u64)cmd << QM_DB_CMD_SHIFT_V1) |
394 		   ((u64)index << QM_DB_INDEX_SHIFT_V1)  |
395 		   ((u64)priority << QM_DB_PRIORITY_SHIFT_V1);
396 
397 	writeq(doorbell, qm->io_base + QM_DOORBELL_BASE_V1);
398 }
399 
400 static void qm_db_v2(struct hisi_qm *qm, u16 qn, u8 cmd, u16 index, u8 priority)
401 {
402 	u64 doorbell;
403 	u64 dbase;
404 	u16 randata = 0;
405 
406 	if (cmd == QM_DOORBELL_CMD_SQ || cmd == QM_DOORBELL_CMD_CQ)
407 		dbase = QM_DOORBELL_SQ_CQ_BASE_V2;
408 	else
409 		dbase = QM_DOORBELL_EQ_AEQ_BASE_V2;
410 
411 	doorbell = qn | ((u64)cmd << QM_DB_CMD_SHIFT_V2) |
412 		   ((u64)randata << QM_DB_RAND_SHIFT_V2) |
413 		   ((u64)index << QM_DB_INDEX_SHIFT_V2)	 |
414 		   ((u64)priority << QM_DB_PRIORITY_SHIFT_V2);
415 
416 	writeq(doorbell, qm->io_base + dbase);
417 }
418 
419 static void qm_db(struct hisi_qm *qm, u16 qn, u8 cmd, u16 index, u8 priority)
420 {
421 	dev_dbg(&qm->pdev->dev, "QM doorbell request: qn=%u, cmd=%u, index=%u\n",
422 		qn, cmd, index);
423 
424 	qm->ops->qm_db(qm, qn, cmd, index, priority);
425 }
426 
427 static int qm_dev_mem_reset(struct hisi_qm *qm)
428 {
429 	u32 val;
430 
431 	writel(0x1, qm->io_base + QM_MEM_START_INIT);
432 	return readl_relaxed_poll_timeout(qm->io_base + QM_MEM_INIT_DONE, val,
433 					  val & BIT(0), 10, 1000);
434 }
435 
436 static u32 qm_get_irq_num_v1(struct hisi_qm *qm)
437 {
438 	return QM_IRQ_NUM_V1;
439 }
440 
441 static u32 qm_get_irq_num_v2(struct hisi_qm *qm)
442 {
443 	if (qm->fun_type == QM_HW_PF)
444 		return QM_IRQ_NUM_PF_V2;
445 	else
446 		return QM_IRQ_NUM_VF_V2;
447 }
448 
449 static struct hisi_qp *qm_to_hisi_qp(struct hisi_qm *qm, struct qm_eqe *eqe)
450 {
451 	u16 cqn = le32_to_cpu(eqe->dw0) & QM_EQE_CQN_MASK;
452 
453 	return qm->qp_array[cqn];
454 }
455 
456 static void qm_cq_head_update(struct hisi_qp *qp)
457 {
458 	if (qp->qp_status.cq_head == QM_Q_DEPTH - 1) {
459 		qp->qp_status.cqc_phase = !qp->qp_status.cqc_phase;
460 		qp->qp_status.cq_head = 0;
461 	} else {
462 		qp->qp_status.cq_head++;
463 	}
464 }
465 
466 static void qm_poll_qp(struct hisi_qp *qp, struct hisi_qm *qm)
467 {
468 	struct qm_cqe *cqe = qp->cqe + qp->qp_status.cq_head;
469 
470 	if (qp->req_cb) {
471 		while (QM_CQE_PHASE(cqe) == qp->qp_status.cqc_phase) {
472 			dma_rmb();
473 			qp->req_cb(qp, qp->sqe + qm->sqe_size *
474 				   le16_to_cpu(cqe->sq_head));
475 			qm_cq_head_update(qp);
476 			cqe = qp->cqe + qp->qp_status.cq_head;
477 			qm_db(qm, qp->qp_id, QM_DOORBELL_CMD_CQ,
478 			      qp->qp_status.cq_head, 0);
479 			atomic_dec(&qp->qp_status.used);
480 		}
481 
482 		/* set c_flag */
483 		qm_db(qm, qp->qp_id, QM_DOORBELL_CMD_CQ,
484 		      qp->qp_status.cq_head, 1);
485 	}
486 }
487 
488 static void qm_qp_work_func(struct work_struct *work)
489 {
490 	struct hisi_qp *qp;
491 
492 	qp = container_of(work, struct hisi_qp, work);
493 	qm_poll_qp(qp, qp->qm);
494 }
495 
496 static irqreturn_t qm_irq_handler(int irq, void *data)
497 {
498 	struct hisi_qm *qm = data;
499 	struct qm_eqe *eqe = qm->eqe + qm->status.eq_head;
500 	struct hisi_qp *qp;
501 	int eqe_num = 0;
502 
503 	while (QM_EQE_PHASE(eqe) == qm->status.eqc_phase) {
504 		eqe_num++;
505 		qp = qm_to_hisi_qp(qm, eqe);
506 		if (qp)
507 			queue_work(qp->wq, &qp->work);
508 
509 		if (qm->status.eq_head == QM_Q_DEPTH - 1) {
510 			qm->status.eqc_phase = !qm->status.eqc_phase;
511 			eqe = qm->eqe;
512 			qm->status.eq_head = 0;
513 		} else {
514 			eqe++;
515 			qm->status.eq_head++;
516 		}
517 
518 		if (eqe_num == QM_Q_DEPTH / 2 - 1) {
519 			eqe_num = 0;
520 			qm_db(qm, 0, QM_DOORBELL_CMD_EQ, qm->status.eq_head, 0);
521 		}
522 	}
523 
524 	qm_db(qm, 0, QM_DOORBELL_CMD_EQ, qm->status.eq_head, 0);
525 
526 	return IRQ_HANDLED;
527 }
528 
529 static irqreturn_t qm_irq(int irq, void *data)
530 {
531 	struct hisi_qm *qm = data;
532 
533 	if (readl(qm->io_base + QM_VF_EQ_INT_SOURCE))
534 		return qm_irq_handler(irq, data);
535 
536 	dev_err(&qm->pdev->dev, "invalid int source\n");
537 	qm_db(qm, 0, QM_DOORBELL_CMD_EQ, qm->status.eq_head, 0);
538 
539 	return IRQ_NONE;
540 }
541 
542 static irqreturn_t qm_aeq_irq(int irq, void *data)
543 {
544 	struct hisi_qm *qm = data;
545 	struct qm_aeqe *aeqe = qm->aeqe + qm->status.aeq_head;
546 	u32 type;
547 
548 	if (!readl(qm->io_base + QM_VF_AEQ_INT_SOURCE))
549 		return IRQ_NONE;
550 
551 	while (QM_AEQE_PHASE(aeqe) == qm->status.aeqc_phase) {
552 		type = le32_to_cpu(aeqe->dw0) >> QM_AEQE_TYPE_SHIFT;
553 		if (type < ARRAY_SIZE(qm_fifo_overflow))
554 			dev_err(&qm->pdev->dev, "%s overflow\n",
555 				qm_fifo_overflow[type]);
556 		else
557 			dev_err(&qm->pdev->dev, "unknown error type %d\n",
558 				type);
559 
560 		if (qm->status.aeq_head == QM_Q_DEPTH - 1) {
561 			qm->status.aeqc_phase = !qm->status.aeqc_phase;
562 			aeqe = qm->aeqe;
563 			qm->status.aeq_head = 0;
564 		} else {
565 			aeqe++;
566 			qm->status.aeq_head++;
567 		}
568 
569 		qm_db(qm, 0, QM_DOORBELL_CMD_AEQ, qm->status.aeq_head, 0);
570 	}
571 
572 	return IRQ_HANDLED;
573 }
574 
575 static irqreturn_t qm_abnormal_irq(int irq, void *data)
576 {
577 	const struct hisi_qm_hw_error *err = qm_hw_error;
578 	struct hisi_qm *qm = data;
579 	struct device *dev = &qm->pdev->dev;
580 	u32 error_status, tmp;
581 
582 	/* read err sts */
583 	tmp = readl(qm->io_base + QM_ABNORMAL_INT_STATUS);
584 	error_status = qm->msi_mask & tmp;
585 
586 	while (err->msg) {
587 		if (err->int_msk & error_status)
588 			dev_err(dev, "%s [error status=0x%x] found\n",
589 				err->msg, err->int_msk);
590 
591 		err++;
592 	}
593 
594 	/* clear err sts */
595 	writel(error_status, qm->io_base + QM_ABNORMAL_INT_SOURCE);
596 
597 	return IRQ_HANDLED;
598 }
599 
600 static int qm_irq_register(struct hisi_qm *qm)
601 {
602 	struct pci_dev *pdev = qm->pdev;
603 	int ret;
604 
605 	ret = request_irq(pci_irq_vector(pdev, QM_EQ_EVENT_IRQ_VECTOR),
606 			  qm_irq, IRQF_SHARED, qm->dev_name, qm);
607 	if (ret)
608 		return ret;
609 
610 	if (qm->ver == QM_HW_V2) {
611 		ret = request_irq(pci_irq_vector(pdev, QM_AEQ_EVENT_IRQ_VECTOR),
612 				  qm_aeq_irq, IRQF_SHARED, qm->dev_name, qm);
613 		if (ret)
614 			goto err_aeq_irq;
615 
616 		if (qm->fun_type == QM_HW_PF) {
617 			ret = request_irq(pci_irq_vector(pdev,
618 					  QM_ABNORMAL_EVENT_IRQ_VECTOR),
619 					  qm_abnormal_irq, IRQF_SHARED,
620 					  qm->dev_name, qm);
621 			if (ret)
622 				goto err_abonormal_irq;
623 		}
624 	}
625 
626 	return 0;
627 
628 err_abonormal_irq:
629 	free_irq(pci_irq_vector(pdev, QM_AEQ_EVENT_IRQ_VECTOR), qm);
630 err_aeq_irq:
631 	free_irq(pci_irq_vector(pdev, QM_EQ_EVENT_IRQ_VECTOR), qm);
632 	return ret;
633 }
634 
635 static void qm_irq_unregister(struct hisi_qm *qm)
636 {
637 	struct pci_dev *pdev = qm->pdev;
638 
639 	free_irq(pci_irq_vector(pdev, QM_EQ_EVENT_IRQ_VECTOR), qm);
640 
641 	if (qm->ver == QM_HW_V2) {
642 		free_irq(pci_irq_vector(pdev, QM_AEQ_EVENT_IRQ_VECTOR), qm);
643 
644 		if (qm->fun_type == QM_HW_PF)
645 			free_irq(pci_irq_vector(pdev,
646 				 QM_ABNORMAL_EVENT_IRQ_VECTOR), qm);
647 	}
648 }
649 
650 static void qm_init_qp_status(struct hisi_qp *qp)
651 {
652 	struct hisi_qp_status *qp_status = &qp->qp_status;
653 
654 	qp_status->sq_tail = 0;
655 	qp_status->cq_head = 0;
656 	qp_status->cqc_phase = true;
657 	qp_status->flags = 0;
658 }
659 
660 static void qm_vft_data_cfg(struct hisi_qm *qm, enum vft_type type, u32 base,
661 			    u32 number)
662 {
663 	u64 tmp = 0;
664 
665 	if (number > 0) {
666 		switch (type) {
667 		case SQC_VFT:
668 			switch (qm->ver) {
669 			case QM_HW_V1:
670 				tmp = QM_SQC_VFT_BUF_SIZE	|
671 				      QM_SQC_VFT_SQC_SIZE	|
672 				      QM_SQC_VFT_INDEX_NUMBER	|
673 				      QM_SQC_VFT_VALID		|
674 				      (u64)base << QM_SQC_VFT_START_SQN_SHIFT;
675 				break;
676 			case QM_HW_V2:
677 				tmp = (u64)base << QM_SQC_VFT_START_SQN_SHIFT |
678 				      QM_SQC_VFT_VALID |
679 				      (u64)(number - 1) << QM_SQC_VFT_SQN_SHIFT;
680 				break;
681 			case QM_HW_UNKNOWN:
682 				break;
683 			}
684 			break;
685 		case CQC_VFT:
686 			switch (qm->ver) {
687 			case QM_HW_V1:
688 				tmp = QM_CQC_VFT_BUF_SIZE	|
689 				      QM_CQC_VFT_SQC_SIZE	|
690 				      QM_CQC_VFT_INDEX_NUMBER	|
691 				      QM_CQC_VFT_VALID;
692 				break;
693 			case QM_HW_V2:
694 				tmp = QM_CQC_VFT_VALID;
695 				break;
696 			case QM_HW_UNKNOWN:
697 				break;
698 			}
699 			break;
700 		}
701 	}
702 
703 	writel(lower_32_bits(tmp), qm->io_base + QM_VFT_CFG_DATA_L);
704 	writel(upper_32_bits(tmp), qm->io_base + QM_VFT_CFG_DATA_H);
705 }
706 
707 static int qm_set_vft_common(struct hisi_qm *qm, enum vft_type type,
708 			     u32 fun_num, u32 base, u32 number)
709 {
710 	unsigned int val;
711 	int ret;
712 
713 	ret = readl_relaxed_poll_timeout(qm->io_base + QM_VFT_CFG_RDY, val,
714 					 val & BIT(0), 10, 1000);
715 	if (ret)
716 		return ret;
717 
718 	writel(0x0, qm->io_base + QM_VFT_CFG_OP_WR);
719 	writel(type, qm->io_base + QM_VFT_CFG_TYPE);
720 	writel(fun_num, qm->io_base + QM_VFT_CFG);
721 
722 	qm_vft_data_cfg(qm, type, base, number);
723 
724 	writel(0x0, qm->io_base + QM_VFT_CFG_RDY);
725 	writel(0x1, qm->io_base + QM_VFT_CFG_OP_ENABLE);
726 
727 	return readl_relaxed_poll_timeout(qm->io_base + QM_VFT_CFG_RDY, val,
728 					  val & BIT(0), 10, 1000);
729 }
730 
731 /* The config should be conducted after qm_dev_mem_reset() */
732 static int qm_set_sqc_cqc_vft(struct hisi_qm *qm, u32 fun_num, u32 base,
733 			      u32 number)
734 {
735 	int ret, i;
736 
737 	for (i = SQC_VFT; i <= CQC_VFT; i++) {
738 		ret = qm_set_vft_common(qm, i, fun_num, base, number);
739 		if (ret)
740 			return ret;
741 	}
742 
743 	return 0;
744 }
745 
746 static int qm_get_vft_v2(struct hisi_qm *qm, u32 *base, u32 *number)
747 {
748 	u64 sqc_vft;
749 	int ret;
750 
751 	ret = qm_mb(qm, QM_MB_CMD_SQC_VFT_V2, 0, 0, 1);
752 	if (ret)
753 		return ret;
754 
755 	sqc_vft = readl(qm->io_base + QM_MB_CMD_DATA_ADDR_L) |
756 		  ((u64)readl(qm->io_base + QM_MB_CMD_DATA_ADDR_H) << 32);
757 	*base = QM_SQC_VFT_BASE_MASK_V2 & (sqc_vft >> QM_SQC_VFT_BASE_SHIFT_V2);
758 	*number = (QM_SQC_VFT_NUM_MASK_v2 &
759 		   (sqc_vft >> QM_SQC_VFT_NUM_SHIFT_V2)) + 1;
760 
761 	return 0;
762 }
763 
764 static struct hisi_qm *file_to_qm(struct debugfs_file *file)
765 {
766 	struct qm_debug *debug = file->debug;
767 
768 	return container_of(debug, struct hisi_qm, debug);
769 }
770 
771 static u32 current_q_read(struct debugfs_file *file)
772 {
773 	struct hisi_qm *qm = file_to_qm(file);
774 
775 	return readl(qm->io_base + QM_DFX_SQE_CNT_VF_SQN) >> QM_DFX_QN_SHIFT;
776 }
777 
778 static int current_q_write(struct debugfs_file *file, u32 val)
779 {
780 	struct hisi_qm *qm = file_to_qm(file);
781 	u32 tmp;
782 
783 	if (val >= qm->debug.curr_qm_qp_num)
784 		return -EINVAL;
785 
786 	tmp = val << QM_DFX_QN_SHIFT |
787 	      (readl(qm->io_base + QM_DFX_SQE_CNT_VF_SQN) & CURRENT_FUN_MASK);
788 	writel(tmp, qm->io_base + QM_DFX_SQE_CNT_VF_SQN);
789 
790 	tmp = val << QM_DFX_QN_SHIFT |
791 	      (readl(qm->io_base + QM_DFX_CQE_CNT_VF_CQN) & CURRENT_FUN_MASK);
792 	writel(tmp, qm->io_base + QM_DFX_CQE_CNT_VF_CQN);
793 
794 	return 0;
795 }
796 
797 static u32 clear_enable_read(struct debugfs_file *file)
798 {
799 	struct hisi_qm *qm = file_to_qm(file);
800 
801 	return readl(qm->io_base + QM_DFX_CNT_CLR_CE);
802 }
803 
804 /* rd_clr_ctrl 1 enable read clear, otherwise 0 disable it */
805 static int clear_enable_write(struct debugfs_file *file, u32 rd_clr_ctrl)
806 {
807 	struct hisi_qm *qm = file_to_qm(file);
808 
809 	if (rd_clr_ctrl > 1)
810 		return -EINVAL;
811 
812 	writel(rd_clr_ctrl, qm->io_base + QM_DFX_CNT_CLR_CE);
813 
814 	return 0;
815 }
816 
817 static ssize_t qm_debug_read(struct file *filp, char __user *buf,
818 			     size_t count, loff_t *pos)
819 {
820 	struct debugfs_file *file = filp->private_data;
821 	enum qm_debug_file index = file->index;
822 	char tbuf[QM_DBG_TMP_BUF_LEN];
823 	u32 val;
824 	int ret;
825 
826 	mutex_lock(&file->lock);
827 	switch (index) {
828 	case CURRENT_Q:
829 		val = current_q_read(file);
830 		break;
831 	case CLEAR_ENABLE:
832 		val = clear_enable_read(file);
833 		break;
834 	default:
835 		mutex_unlock(&file->lock);
836 		return -EINVAL;
837 	}
838 	mutex_unlock(&file->lock);
839 	ret = sprintf(tbuf, "%u\n", val);
840 	return simple_read_from_buffer(buf, count, pos, tbuf, ret);
841 }
842 
843 static ssize_t qm_debug_write(struct file *filp, const char __user *buf,
844 			      size_t count, loff_t *pos)
845 {
846 	struct debugfs_file *file = filp->private_data;
847 	enum qm_debug_file index = file->index;
848 	unsigned long val;
849 	char tbuf[QM_DBG_TMP_BUF_LEN];
850 	int len, ret;
851 
852 	if (*pos != 0)
853 		return 0;
854 
855 	if (count >= QM_DBG_TMP_BUF_LEN)
856 		return -ENOSPC;
857 
858 	len = simple_write_to_buffer(tbuf, QM_DBG_TMP_BUF_LEN - 1, pos, buf,
859 				     count);
860 	if (len < 0)
861 		return len;
862 
863 	tbuf[len] = '\0';
864 	if (kstrtoul(tbuf, 0, &val))
865 		return -EFAULT;
866 
867 	mutex_lock(&file->lock);
868 	switch (index) {
869 	case CURRENT_Q:
870 		ret = current_q_write(file, val);
871 		if (ret)
872 			goto err_input;
873 		break;
874 	case CLEAR_ENABLE:
875 		ret = clear_enable_write(file, val);
876 		if (ret)
877 			goto err_input;
878 		break;
879 	default:
880 		ret = -EINVAL;
881 		goto err_input;
882 	}
883 	mutex_unlock(&file->lock);
884 
885 	return count;
886 
887 err_input:
888 	mutex_unlock(&file->lock);
889 	return ret;
890 }
891 
892 static const struct file_operations qm_debug_fops = {
893 	.owner = THIS_MODULE,
894 	.open = simple_open,
895 	.read = qm_debug_read,
896 	.write = qm_debug_write,
897 };
898 
899 struct qm_dfx_registers {
900 	char  *reg_name;
901 	u64   reg_offset;
902 };
903 
904 #define CNT_CYC_REGS_NUM		10
905 static struct qm_dfx_registers qm_dfx_regs[] = {
906 	/* XXX_CNT are reading clear register */
907 	{"QM_ECC_1BIT_CNT               ",  0x104000ull},
908 	{"QM_ECC_MBIT_CNT               ",  0x104008ull},
909 	{"QM_DFX_MB_CNT                 ",  0x104018ull},
910 	{"QM_DFX_DB_CNT                 ",  0x104028ull},
911 	{"QM_DFX_SQE_CNT                ",  0x104038ull},
912 	{"QM_DFX_CQE_CNT                ",  0x104048ull},
913 	{"QM_DFX_SEND_SQE_TO_ACC_CNT    ",  0x104050ull},
914 	{"QM_DFX_WB_SQE_FROM_ACC_CNT    ",  0x104058ull},
915 	{"QM_DFX_ACC_FINISH_CNT         ",  0x104060ull},
916 	{"QM_DFX_CQE_ERR_CNT            ",  0x1040b4ull},
917 	{"QM_DFX_FUNS_ACTIVE_ST         ",  0x200ull},
918 	{"QM_ECC_1BIT_INF               ",  0x104004ull},
919 	{"QM_ECC_MBIT_INF               ",  0x10400cull},
920 	{"QM_DFX_ACC_RDY_VLD0           ",  0x1040a0ull},
921 	{"QM_DFX_ACC_RDY_VLD1           ",  0x1040a4ull},
922 	{"QM_DFX_AXI_RDY_VLD            ",  0x1040a8ull},
923 	{"QM_DFX_FF_ST0                 ",  0x1040c8ull},
924 	{"QM_DFX_FF_ST1                 ",  0x1040ccull},
925 	{"QM_DFX_FF_ST2                 ",  0x1040d0ull},
926 	{"QM_DFX_FF_ST3                 ",  0x1040d4ull},
927 	{"QM_DFX_FF_ST4                 ",  0x1040d8ull},
928 	{"QM_DFX_FF_ST5                 ",  0x1040dcull},
929 	{"QM_DFX_FF_ST6                 ",  0x1040e0ull},
930 	{"QM_IN_IDLE_ST                 ",  0x1040e4ull},
931 	{ NULL, 0}
932 };
933 
934 static struct qm_dfx_registers qm_vf_dfx_regs[] = {
935 	{"QM_DFX_FUNS_ACTIVE_ST         ",  0x200ull},
936 	{ NULL, 0}
937 };
938 
939 static int qm_regs_show(struct seq_file *s, void *unused)
940 {
941 	struct hisi_qm *qm = s->private;
942 	struct qm_dfx_registers *regs;
943 	u32 val;
944 
945 	if (qm->fun_type == QM_HW_PF)
946 		regs = qm_dfx_regs;
947 	else
948 		regs = qm_vf_dfx_regs;
949 
950 	while (regs->reg_name) {
951 		val = readl(qm->io_base + regs->reg_offset);
952 		seq_printf(s, "%s= 0x%08x\n", regs->reg_name, val);
953 		regs++;
954 	}
955 
956 	return 0;
957 }
958 
959 static int qm_regs_open(struct inode *inode, struct file *file)
960 {
961 	return single_open(file, qm_regs_show, inode->i_private);
962 }
963 
964 static const struct file_operations qm_regs_fops = {
965 	.owner = THIS_MODULE,
966 	.open = qm_regs_open,
967 	.read = seq_read,
968 	.release = single_release,
969 };
970 
971 static int qm_create_debugfs_file(struct hisi_qm *qm, enum qm_debug_file index)
972 {
973 	struct dentry *qm_d = qm->debug.qm_d;
974 	struct debugfs_file *file = qm->debug.files + index;
975 
976 	debugfs_create_file(qm_debug_file_name[index], 0600, qm_d, file,
977 			    &qm_debug_fops);
978 
979 	file->index = index;
980 	mutex_init(&file->lock);
981 	file->debug = &qm->debug;
982 
983 	return 0;
984 }
985 
986 static void qm_hw_error_init_v1(struct hisi_qm *qm, u32 ce, u32 nfe, u32 fe,
987 				u32 msi)
988 {
989 	writel(QM_ABNORMAL_INT_MASK_VALUE, qm->io_base + QM_ABNORMAL_INT_MASK);
990 }
991 
992 static void qm_hw_error_init_v2(struct hisi_qm *qm, u32 ce, u32 nfe, u32 fe,
993 				u32 msi)
994 {
995 	u32 irq_enable = ce | nfe | fe | msi;
996 	u32 irq_unmask = ~irq_enable;
997 
998 	qm->error_mask = ce | nfe | fe;
999 	qm->msi_mask = msi;
1000 
1001 	/* configure error type */
1002 	writel(ce, qm->io_base + QM_RAS_CE_ENABLE);
1003 	writel(QM_RAS_CE_TIMES_PER_IRQ, qm->io_base + QM_RAS_CE_THRESHOLD);
1004 	writel(nfe, qm->io_base + QM_RAS_NFE_ENABLE);
1005 	writel(fe, qm->io_base + QM_RAS_FE_ENABLE);
1006 
1007 	/* use RAS irq default, so only set QM_RAS_MSI_INT_SEL for MSI */
1008 	writel(msi, qm->io_base + QM_RAS_MSI_INT_SEL);
1009 
1010 	irq_unmask &= readl(qm->io_base + QM_ABNORMAL_INT_MASK);
1011 	writel(irq_unmask, qm->io_base + QM_ABNORMAL_INT_MASK);
1012 }
1013 
1014 static void qm_log_hw_error(struct hisi_qm *qm, u32 error_status)
1015 {
1016 	const struct hisi_qm_hw_error *err = qm_hw_error;
1017 	struct device *dev = &qm->pdev->dev;
1018 	u32 reg_val, type, vf_num;
1019 
1020 	while (err->msg) {
1021 		if (err->int_msk & error_status) {
1022 			dev_err(dev, "%s [error status=0x%x] found\n",
1023 				err->msg, err->int_msk);
1024 
1025 			if (error_status & QM_DB_TIMEOUT) {
1026 				reg_val = readl(qm->io_base +
1027 						QM_ABNORMAL_INF01);
1028 				type = (reg_val & QM_DB_TIMEOUT_TYPE) >>
1029 				       QM_DB_TIMEOUT_TYPE_SHIFT;
1030 				vf_num = reg_val & QM_DB_TIMEOUT_VF;
1031 				dev_err(dev, "qm %s doorbell timeout in function %u\n",
1032 					qm_db_timeout[type], vf_num);
1033 			}
1034 
1035 			if (error_status & QM_OF_FIFO_OF) {
1036 				reg_val = readl(qm->io_base +
1037 						QM_ABNORMAL_INF00);
1038 				type = (reg_val & QM_FIFO_OVERFLOW_TYPE) >>
1039 				       QM_FIFO_OVERFLOW_TYPE_SHIFT;
1040 				vf_num = reg_val & QM_FIFO_OVERFLOW_VF;
1041 
1042 				if (type < ARRAY_SIZE(qm_fifo_overflow))
1043 					dev_err(dev, "qm %s fifo overflow in function %u\n",
1044 						qm_fifo_overflow[type],
1045 						vf_num);
1046 				else
1047 					dev_err(dev, "unknown error type\n");
1048 			}
1049 		}
1050 		err++;
1051 	}
1052 }
1053 
1054 static pci_ers_result_t qm_hw_error_handle_v2(struct hisi_qm *qm)
1055 {
1056 	u32 error_status, tmp;
1057 
1058 	/* read err sts */
1059 	tmp = readl(qm->io_base + QM_ABNORMAL_INT_STATUS);
1060 	error_status = qm->error_mask & tmp;
1061 
1062 	if (error_status) {
1063 		qm_log_hw_error(qm, error_status);
1064 
1065 		/* clear err sts */
1066 		writel(error_status, qm->io_base + QM_ABNORMAL_INT_SOURCE);
1067 
1068 		return PCI_ERS_RESULT_NEED_RESET;
1069 	}
1070 
1071 	return PCI_ERS_RESULT_RECOVERED;
1072 }
1073 
1074 static const struct hisi_qm_hw_ops qm_hw_ops_v1 = {
1075 	.qm_db = qm_db_v1,
1076 	.get_irq_num = qm_get_irq_num_v1,
1077 	.hw_error_init = qm_hw_error_init_v1,
1078 };
1079 
1080 static const struct hisi_qm_hw_ops qm_hw_ops_v2 = {
1081 	.get_vft = qm_get_vft_v2,
1082 	.qm_db = qm_db_v2,
1083 	.get_irq_num = qm_get_irq_num_v2,
1084 	.hw_error_init = qm_hw_error_init_v2,
1085 	.hw_error_handle = qm_hw_error_handle_v2,
1086 };
1087 
1088 static void *qm_get_avail_sqe(struct hisi_qp *qp)
1089 {
1090 	struct hisi_qp_status *qp_status = &qp->qp_status;
1091 	u16 sq_tail = qp_status->sq_tail;
1092 
1093 	if (unlikely(atomic_read(&qp->qp_status.used) == QM_Q_DEPTH))
1094 		return NULL;
1095 
1096 	return qp->sqe + sq_tail * qp->qm->sqe_size;
1097 }
1098 
1099 /**
1100  * hisi_qm_create_qp() - Create a queue pair from qm.
1101  * @qm: The qm we create a qp from.
1102  * @alg_type: Accelerator specific algorithm type in sqc.
1103  *
1104  * return created qp, -EBUSY if all qps in qm allocated, -ENOMEM if allocating
1105  * qp memory fails.
1106  */
1107 struct hisi_qp *hisi_qm_create_qp(struct hisi_qm *qm, u8 alg_type)
1108 {
1109 	struct device *dev = &qm->pdev->dev;
1110 	struct hisi_qp *qp;
1111 	int qp_id, ret;
1112 
1113 	qp = kzalloc(sizeof(*qp), GFP_KERNEL);
1114 	if (!qp)
1115 		return ERR_PTR(-ENOMEM);
1116 
1117 	write_lock(&qm->qps_lock);
1118 
1119 	qp_id = find_first_zero_bit(qm->qp_bitmap, qm->qp_num);
1120 	if (qp_id >= qm->qp_num) {
1121 		write_unlock(&qm->qps_lock);
1122 		dev_info(&qm->pdev->dev, "QM all queues are busy!\n");
1123 		ret = -EBUSY;
1124 		goto err_free_qp;
1125 	}
1126 	set_bit(qp_id, qm->qp_bitmap);
1127 	qm->qp_array[qp_id] = qp;
1128 	qm->qp_in_used++;
1129 
1130 	write_unlock(&qm->qps_lock);
1131 
1132 	qp->qm = qm;
1133 
1134 	if (qm->use_dma_api) {
1135 		qp->qdma.size = qm->sqe_size * QM_Q_DEPTH +
1136 				sizeof(struct qm_cqe) * QM_Q_DEPTH;
1137 		qp->qdma.va = dma_alloc_coherent(dev, qp->qdma.size,
1138 						 &qp->qdma.dma, GFP_KERNEL);
1139 		if (!qp->qdma.va) {
1140 			ret = -ENOMEM;
1141 			goto err_clear_bit;
1142 		}
1143 
1144 		dev_dbg(dev, "allocate qp dma buf(va=%pK, dma=%pad, size=%zx)\n",
1145 			qp->qdma.va, &qp->qdma.dma, qp->qdma.size);
1146 	}
1147 
1148 	qp->qp_id = qp_id;
1149 	qp->alg_type = alg_type;
1150 	INIT_WORK(&qp->work, qm_qp_work_func);
1151 	qp->wq = alloc_workqueue("hisi_qm", WQ_UNBOUND | WQ_HIGHPRI |
1152 				 WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM, 0);
1153 	if (!qp->wq) {
1154 		ret = -EFAULT;
1155 		goto err_free_qp_mem;
1156 	}
1157 
1158 	return qp;
1159 
1160 err_free_qp_mem:
1161 	if (qm->use_dma_api)
1162 		dma_free_coherent(dev, qp->qdma.size, qp->qdma.va,
1163 				  qp->qdma.dma);
1164 err_clear_bit:
1165 	write_lock(&qm->qps_lock);
1166 	qm->qp_array[qp_id] = NULL;
1167 	clear_bit(qp_id, qm->qp_bitmap);
1168 	write_unlock(&qm->qps_lock);
1169 err_free_qp:
1170 	kfree(qp);
1171 	return ERR_PTR(ret);
1172 }
1173 EXPORT_SYMBOL_GPL(hisi_qm_create_qp);
1174 
1175 /**
1176  * hisi_qm_release_qp() - Release a qp back to its qm.
1177  * @qp: The qp we want to release.
1178  *
1179  * This function releases the resource of a qp.
1180  */
1181 void hisi_qm_release_qp(struct hisi_qp *qp)
1182 {
1183 	struct hisi_qm *qm = qp->qm;
1184 	struct qm_dma *qdma = &qp->qdma;
1185 	struct device *dev = &qm->pdev->dev;
1186 
1187 	if (qm->use_dma_api && qdma->va)
1188 		dma_free_coherent(dev, qdma->size, qdma->va, qdma->dma);
1189 
1190 	write_lock(&qm->qps_lock);
1191 	qm->qp_array[qp->qp_id] = NULL;
1192 	clear_bit(qp->qp_id, qm->qp_bitmap);
1193 	qm->qp_in_used--;
1194 	write_unlock(&qm->qps_lock);
1195 
1196 	kfree(qp);
1197 }
1198 EXPORT_SYMBOL_GPL(hisi_qm_release_qp);
1199 
1200 static int qm_qp_ctx_cfg(struct hisi_qp *qp, int qp_id, int pasid)
1201 {
1202 	struct hisi_qm *qm = qp->qm;
1203 	struct device *dev = &qm->pdev->dev;
1204 	enum qm_hw_ver ver = qm->ver;
1205 	struct qm_sqc *sqc;
1206 	struct qm_cqc *cqc;
1207 	dma_addr_t sqc_dma;
1208 	dma_addr_t cqc_dma;
1209 	int ret;
1210 
1211 	qm_init_qp_status(qp);
1212 
1213 	sqc = kzalloc(sizeof(struct qm_sqc), GFP_KERNEL);
1214 	if (!sqc)
1215 		return -ENOMEM;
1216 	sqc_dma = dma_map_single(dev, sqc, sizeof(struct qm_sqc),
1217 				 DMA_TO_DEVICE);
1218 	if (dma_mapping_error(dev, sqc_dma)) {
1219 		kfree(sqc);
1220 		return -ENOMEM;
1221 	}
1222 
1223 	INIT_QC_COMMON(sqc, qp->sqe_dma, pasid);
1224 	if (ver == QM_HW_V1) {
1225 		sqc->dw3 = cpu_to_le32(QM_MK_SQC_DW3_V1(0, 0, 0, qm->sqe_size));
1226 		sqc->w8 = cpu_to_le16(QM_Q_DEPTH - 1);
1227 	} else if (ver == QM_HW_V2) {
1228 		sqc->dw3 = cpu_to_le32(QM_MK_SQC_DW3_V2(qm->sqe_size));
1229 		sqc->w8 = 0; /* rand_qc */
1230 	}
1231 	sqc->cq_num = cpu_to_le16(qp_id);
1232 	sqc->w13 = cpu_to_le16(QM_MK_SQC_W13(0, 1, qp->alg_type));
1233 
1234 	ret = qm_mb(qm, QM_MB_CMD_SQC, sqc_dma, qp_id, 0);
1235 	dma_unmap_single(dev, sqc_dma, sizeof(struct qm_sqc), DMA_TO_DEVICE);
1236 	kfree(sqc);
1237 	if (ret)
1238 		return ret;
1239 
1240 	cqc = kzalloc(sizeof(struct qm_cqc), GFP_KERNEL);
1241 	if (!cqc)
1242 		return -ENOMEM;
1243 	cqc_dma = dma_map_single(dev, cqc, sizeof(struct qm_cqc),
1244 				 DMA_TO_DEVICE);
1245 	if (dma_mapping_error(dev, cqc_dma)) {
1246 		kfree(cqc);
1247 		return -ENOMEM;
1248 	}
1249 
1250 	INIT_QC_COMMON(cqc, qp->cqe_dma, pasid);
1251 	if (ver == QM_HW_V1) {
1252 		cqc->dw3 = cpu_to_le32(QM_MK_CQC_DW3_V1(0, 0, 0, 4));
1253 		cqc->w8 = cpu_to_le16(QM_Q_DEPTH - 1);
1254 	} else if (ver == QM_HW_V2) {
1255 		cqc->dw3 = cpu_to_le32(QM_MK_CQC_DW3_V2(4));
1256 		cqc->w8 = 0;
1257 	}
1258 	cqc->dw6 = cpu_to_le32(1 << QM_CQ_PHASE_SHIFT | 1 << QM_CQ_FLAG_SHIFT);
1259 
1260 	ret = qm_mb(qm, QM_MB_CMD_CQC, cqc_dma, qp_id, 0);
1261 	dma_unmap_single(dev, cqc_dma, sizeof(struct qm_cqc), DMA_TO_DEVICE);
1262 	kfree(cqc);
1263 
1264 	return ret;
1265 }
1266 
1267 /**
1268  * hisi_qm_start_qp() - Start a qp into running.
1269  * @qp: The qp we want to start to run.
1270  * @arg: Accelerator specific argument.
1271  *
1272  * After this function, qp can receive request from user. Return qp_id if
1273  * successful, Return -EBUSY if failed.
1274  */
1275 int hisi_qm_start_qp(struct hisi_qp *qp, unsigned long arg)
1276 {
1277 	struct hisi_qm *qm = qp->qm;
1278 	struct device *dev = &qm->pdev->dev;
1279 	enum qm_hw_ver ver = qm->ver;
1280 	int qp_id = qp->qp_id;
1281 	int pasid = arg;
1282 	size_t off = 0;
1283 	int ret;
1284 
1285 #define QP_INIT_BUF(qp, type, size) do { \
1286 	(qp)->type = ((qp)->qdma.va + (off)); \
1287 	(qp)->type##_dma = (qp)->qdma.dma + (off); \
1288 	off += (size); \
1289 } while (0)
1290 
1291 	if (!qp->qdma.dma) {
1292 		dev_err(dev, "cannot get qm dma buffer\n");
1293 		return -EINVAL;
1294 	}
1295 
1296 	/* sq need 128 bytes alignment */
1297 	if (qp->qdma.dma & QM_SQE_DATA_ALIGN_MASK) {
1298 		dev_err(dev, "qm sq is not aligned to 128 byte\n");
1299 		return -EINVAL;
1300 	}
1301 
1302 	QP_INIT_BUF(qp, sqe, qm->sqe_size * QM_Q_DEPTH);
1303 	QP_INIT_BUF(qp, cqe, sizeof(struct qm_cqe) * QM_Q_DEPTH);
1304 
1305 	dev_dbg(dev, "init qp buffer(v%d):\n"
1306 		     " sqe	(%pK, %lx)\n"
1307 		     " cqe	(%pK, %lx)\n",
1308 		     ver, qp->sqe, (unsigned long)qp->sqe_dma,
1309 		     qp->cqe, (unsigned long)qp->cqe_dma);
1310 
1311 	ret = qm_qp_ctx_cfg(qp, qp_id, pasid);
1312 	if (ret)
1313 		return ret;
1314 
1315 	dev_dbg(dev, "queue %d started\n", qp_id);
1316 
1317 	return qp_id;
1318 }
1319 EXPORT_SYMBOL_GPL(hisi_qm_start_qp);
1320 
1321 /**
1322  * hisi_qm_stop_qp() - Stop a qp in qm.
1323  * @qp: The qp we want to stop.
1324  *
1325  * This function is reverse of hisi_qm_start_qp. Return 0 if successful.
1326  */
1327 int hisi_qm_stop_qp(struct hisi_qp *qp)
1328 {
1329 	struct device *dev = &qp->qm->pdev->dev;
1330 	int i = 0;
1331 
1332 	/* it is stopped */
1333 	if (test_bit(QP_STOP, &qp->qp_status.flags))
1334 		return 0;
1335 
1336 	while (atomic_read(&qp->qp_status.used)) {
1337 		i++;
1338 		msleep(20);
1339 		if (i == 10) {
1340 			dev_err(dev, "Cannot drain out data for stopping, Force to stop!\n");
1341 			return 0;
1342 		}
1343 	}
1344 
1345 	set_bit(QP_STOP, &qp->qp_status.flags);
1346 
1347 	dev_dbg(dev, "stop queue %u!", qp->qp_id);
1348 
1349 	return 0;
1350 }
1351 EXPORT_SYMBOL_GPL(hisi_qm_stop_qp);
1352 
1353 /**
1354  * hisi_qp_send() - Queue up a task in the hardware queue.
1355  * @qp: The qp in which to put the message.
1356  * @msg: The message.
1357  *
1358  * This function will return -EBUSY if qp is currently full, and -EAGAIN
1359  * if qp related qm is resetting.
1360  */
1361 int hisi_qp_send(struct hisi_qp *qp, const void *msg)
1362 {
1363 	struct hisi_qp_status *qp_status = &qp->qp_status;
1364 	u16 sq_tail = qp_status->sq_tail;
1365 	u16 sq_tail_next = (sq_tail + 1) % QM_Q_DEPTH;
1366 	void *sqe = qm_get_avail_sqe(qp);
1367 
1368 	if (unlikely(test_bit(QP_STOP, &qp->qp_status.flags))) {
1369 		dev_info(&qp->qm->pdev->dev, "QP is stopped or resetting\n");
1370 		return -EAGAIN;
1371 	}
1372 
1373 	if (!sqe)
1374 		return -EBUSY;
1375 
1376 	memcpy(sqe, msg, qp->qm->sqe_size);
1377 
1378 	qm_db(qp->qm, qp->qp_id, QM_DOORBELL_CMD_SQ, sq_tail_next, 0);
1379 	atomic_inc(&qp->qp_status.used);
1380 	qp_status->sq_tail = sq_tail_next;
1381 
1382 	return 0;
1383 }
1384 EXPORT_SYMBOL_GPL(hisi_qp_send);
1385 
1386 static void hisi_qm_cache_wb(struct hisi_qm *qm)
1387 {
1388 	unsigned int val;
1389 
1390 	if (qm->ver == QM_HW_V2) {
1391 		writel(0x1, qm->io_base + QM_CACHE_WB_START);
1392 		if (readl_relaxed_poll_timeout(qm->io_base + QM_CACHE_WB_DONE,
1393 					       val, val & BIT(0), 10, 1000))
1394 			dev_err(&qm->pdev->dev, "QM writeback sqc cache fail!\n");
1395 	}
1396 }
1397 
1398 /**
1399  * hisi_qm_get_free_qp_num() - Get free number of qp in qm.
1400  * @qm: The qm which want to get free qp.
1401  *
1402  * This function return free number of qp in qm.
1403  */
1404 int hisi_qm_get_free_qp_num(struct hisi_qm *qm)
1405 {
1406 	int ret;
1407 
1408 	read_lock(&qm->qps_lock);
1409 	ret = qm->qp_num - qm->qp_in_used;
1410 	read_unlock(&qm->qps_lock);
1411 
1412 	return ret;
1413 }
1414 EXPORT_SYMBOL_GPL(hisi_qm_get_free_qp_num);
1415 
1416 /**
1417  * hisi_qm_init() - Initialize configures about qm.
1418  * @qm: The qm needing init.
1419  *
1420  * This function init qm, then we can call hisi_qm_start to put qm into work.
1421  */
1422 int hisi_qm_init(struct hisi_qm *qm)
1423 {
1424 	struct pci_dev *pdev = qm->pdev;
1425 	struct device *dev = &pdev->dev;
1426 	unsigned int num_vec;
1427 	int ret;
1428 
1429 	switch (qm->ver) {
1430 	case QM_HW_V1:
1431 		qm->ops = &qm_hw_ops_v1;
1432 		break;
1433 	case QM_HW_V2:
1434 		qm->ops = &qm_hw_ops_v2;
1435 		break;
1436 	default:
1437 		return -EINVAL;
1438 	}
1439 
1440 	ret = pci_enable_device_mem(pdev);
1441 	if (ret < 0) {
1442 		dev_err(&pdev->dev, "Failed to enable device mem!\n");
1443 		return ret;
1444 	}
1445 
1446 	ret = pci_request_mem_regions(pdev, qm->dev_name);
1447 	if (ret < 0) {
1448 		dev_err(&pdev->dev, "Failed to request mem regions!\n");
1449 		goto err_disable_pcidev;
1450 	}
1451 
1452 	qm->io_base = ioremap(pci_resource_start(pdev, PCI_BAR_2),
1453 			      pci_resource_len(qm->pdev, PCI_BAR_2));
1454 	if (!qm->io_base) {
1455 		ret = -EIO;
1456 		goto err_release_mem_regions;
1457 	}
1458 
1459 	ret = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(64));
1460 	if (ret < 0)
1461 		goto err_iounmap;
1462 	pci_set_master(pdev);
1463 
1464 	if (!qm->ops->get_irq_num) {
1465 		ret = -EOPNOTSUPP;
1466 		goto err_iounmap;
1467 	}
1468 	num_vec = qm->ops->get_irq_num(qm);
1469 	ret = pci_alloc_irq_vectors(pdev, num_vec, num_vec, PCI_IRQ_MSI);
1470 	if (ret < 0) {
1471 		dev_err(dev, "Failed to enable MSI vectors!\n");
1472 		goto err_iounmap;
1473 	}
1474 
1475 	ret = qm_irq_register(qm);
1476 	if (ret)
1477 		goto err_free_irq_vectors;
1478 
1479 	qm->qp_in_used = 0;
1480 	mutex_init(&qm->mailbox_lock);
1481 	rwlock_init(&qm->qps_lock);
1482 
1483 	dev_dbg(dev, "init qm %s with %s\n", pdev->is_physfn ? "pf" : "vf",
1484 		qm->use_dma_api ? "dma api" : "iommu api");
1485 
1486 	return 0;
1487 
1488 err_free_irq_vectors:
1489 	pci_free_irq_vectors(pdev);
1490 err_iounmap:
1491 	iounmap(qm->io_base);
1492 err_release_mem_regions:
1493 	pci_release_mem_regions(pdev);
1494 err_disable_pcidev:
1495 	pci_disable_device(pdev);
1496 
1497 	return ret;
1498 }
1499 EXPORT_SYMBOL_GPL(hisi_qm_init);
1500 
1501 /**
1502  * hisi_qm_uninit() - Uninitialize qm.
1503  * @qm: The qm needed uninit.
1504  *
1505  * This function uninits qm related device resources.
1506  */
1507 void hisi_qm_uninit(struct hisi_qm *qm)
1508 {
1509 	struct pci_dev *pdev = qm->pdev;
1510 	struct device *dev = &pdev->dev;
1511 
1512 	if (qm->use_dma_api && qm->qdma.va) {
1513 		hisi_qm_cache_wb(qm);
1514 		dma_free_coherent(dev, qm->qdma.size,
1515 				  qm->qdma.va, qm->qdma.dma);
1516 		memset(&qm->qdma, 0, sizeof(qm->qdma));
1517 	}
1518 
1519 	qm_irq_unregister(qm);
1520 	pci_free_irq_vectors(pdev);
1521 	iounmap(qm->io_base);
1522 	pci_release_mem_regions(pdev);
1523 	pci_disable_device(pdev);
1524 }
1525 EXPORT_SYMBOL_GPL(hisi_qm_uninit);
1526 
1527 /**
1528  * hisi_qm_get_vft() - Get vft from a qm.
1529  * @qm: The qm we want to get its vft.
1530  * @base: The base number of queue in vft.
1531  * @number: The number of queues in vft.
1532  *
1533  * We can allocate multiple queues to a qm by configuring virtual function
1534  * table. We get related configures by this function. Normally, we call this
1535  * function in VF driver to get the queue information.
1536  *
1537  * qm hw v1 does not support this interface.
1538  */
1539 int hisi_qm_get_vft(struct hisi_qm *qm, u32 *base, u32 *number)
1540 {
1541 	if (!base || !number)
1542 		return -EINVAL;
1543 
1544 	if (!qm->ops->get_vft) {
1545 		dev_err(&qm->pdev->dev, "Don't support vft read!\n");
1546 		return -EINVAL;
1547 	}
1548 
1549 	return qm->ops->get_vft(qm, base, number);
1550 }
1551 EXPORT_SYMBOL_GPL(hisi_qm_get_vft);
1552 
1553 /**
1554  * hisi_qm_set_vft() - Set "virtual function table" for a qm.
1555  * @fun_num: Number of operated function.
1556  * @qm: The qm in which to set vft, alway in a PF.
1557  * @base: The base number of queue in vft.
1558  * @number: The number of queues in vft. 0 means invalid vft.
1559  *
1560  * This function is alway called in PF driver, it is used to assign queues
1561  * among PF and VFs.
1562  *
1563  * Assign queues A~B to PF: hisi_qm_set_vft(qm, 0, A, B - A + 1)
1564  * Assign queues A~B to VF: hisi_qm_set_vft(qm, 2, A, B - A + 1)
1565  * (VF function number 0x2)
1566  */
1567 int hisi_qm_set_vft(struct hisi_qm *qm, u32 fun_num, u32 base,
1568 		    u32 number)
1569 {
1570 	u32 max_q_num = qm->ctrl_qp_num;
1571 
1572 	if (base >= max_q_num || number > max_q_num ||
1573 	    (base + number) > max_q_num)
1574 		return -EINVAL;
1575 
1576 	return qm_set_sqc_cqc_vft(qm, fun_num, base, number);
1577 }
1578 EXPORT_SYMBOL_GPL(hisi_qm_set_vft);
1579 
1580 static void qm_init_eq_aeq_status(struct hisi_qm *qm)
1581 {
1582 	struct hisi_qm_status *status = &qm->status;
1583 
1584 	status->eq_head = 0;
1585 	status->aeq_head = 0;
1586 	status->eqc_phase = true;
1587 	status->aeqc_phase = true;
1588 }
1589 
1590 static int qm_eq_ctx_cfg(struct hisi_qm *qm)
1591 {
1592 	struct device *dev = &qm->pdev->dev;
1593 	struct qm_eqc *eqc;
1594 	struct qm_aeqc *aeqc;
1595 	dma_addr_t eqc_dma;
1596 	dma_addr_t aeqc_dma;
1597 	int ret;
1598 
1599 	qm_init_eq_aeq_status(qm);
1600 
1601 	eqc = kzalloc(sizeof(struct qm_eqc), GFP_KERNEL);
1602 	if (!eqc)
1603 		return -ENOMEM;
1604 	eqc_dma = dma_map_single(dev, eqc, sizeof(struct qm_eqc),
1605 				 DMA_TO_DEVICE);
1606 	if (dma_mapping_error(dev, eqc_dma)) {
1607 		kfree(eqc);
1608 		return -ENOMEM;
1609 	}
1610 
1611 	eqc->base_l = cpu_to_le32(lower_32_bits(qm->eqe_dma));
1612 	eqc->base_h = cpu_to_le32(upper_32_bits(qm->eqe_dma));
1613 	if (qm->ver == QM_HW_V1)
1614 		eqc->dw3 = cpu_to_le32(QM_EQE_AEQE_SIZE);
1615 	eqc->dw6 = cpu_to_le32((QM_Q_DEPTH - 1) | (1 << QM_EQC_PHASE_SHIFT));
1616 	ret = qm_mb(qm, QM_MB_CMD_EQC, eqc_dma, 0, 0);
1617 	dma_unmap_single(dev, eqc_dma, sizeof(struct qm_eqc), DMA_TO_DEVICE);
1618 	kfree(eqc);
1619 	if (ret)
1620 		return ret;
1621 
1622 	aeqc = kzalloc(sizeof(struct qm_aeqc), GFP_KERNEL);
1623 	if (!aeqc)
1624 		return -ENOMEM;
1625 	aeqc_dma = dma_map_single(dev, aeqc, sizeof(struct qm_aeqc),
1626 				  DMA_TO_DEVICE);
1627 	if (dma_mapping_error(dev, aeqc_dma)) {
1628 		kfree(aeqc);
1629 		return -ENOMEM;
1630 	}
1631 
1632 	aeqc->base_l = cpu_to_le32(lower_32_bits(qm->aeqe_dma));
1633 	aeqc->base_h = cpu_to_le32(upper_32_bits(qm->aeqe_dma));
1634 	aeqc->dw6 = cpu_to_le32((QM_Q_DEPTH - 1) | (1 << QM_EQC_PHASE_SHIFT));
1635 
1636 	ret = qm_mb(qm, QM_MB_CMD_AEQC, aeqc_dma, 0, 0);
1637 	dma_unmap_single(dev, aeqc_dma, sizeof(struct qm_aeqc), DMA_TO_DEVICE);
1638 	kfree(aeqc);
1639 
1640 	return ret;
1641 }
1642 
1643 static int __hisi_qm_start(struct hisi_qm *qm)
1644 {
1645 	struct pci_dev *pdev = qm->pdev;
1646 	struct device *dev = &pdev->dev;
1647 	size_t off = 0;
1648 	int ret;
1649 
1650 #define QM_INIT_BUF(qm, type, num) do { \
1651 	(qm)->type = ((qm)->qdma.va + (off)); \
1652 	(qm)->type##_dma = (qm)->qdma.dma + (off); \
1653 	off += QMC_ALIGN(sizeof(struct qm_##type) * (num)); \
1654 } while (0)
1655 
1656 	WARN_ON(!qm->qdma.dma);
1657 
1658 	if (qm->qp_num == 0)
1659 		return -EINVAL;
1660 
1661 	if (qm->fun_type == QM_HW_PF) {
1662 		ret = qm_dev_mem_reset(qm);
1663 		if (ret)
1664 			return ret;
1665 
1666 		ret = hisi_qm_set_vft(qm, 0, qm->qp_base, qm->qp_num);
1667 		if (ret)
1668 			return ret;
1669 	}
1670 
1671 	QM_INIT_BUF(qm, eqe, QM_Q_DEPTH);
1672 	QM_INIT_BUF(qm, aeqe, QM_Q_DEPTH);
1673 	QM_INIT_BUF(qm, sqc, qm->qp_num);
1674 	QM_INIT_BUF(qm, cqc, qm->qp_num);
1675 
1676 	dev_dbg(dev, "init qm buffer:\n"
1677 		     " eqe	(%pK, %lx)\n"
1678 		     " aeqe	(%pK, %lx)\n"
1679 		     " sqc	(%pK, %lx)\n"
1680 		     " cqc	(%pK, %lx)\n",
1681 		     qm->eqe, (unsigned long)qm->eqe_dma,
1682 		     qm->aeqe, (unsigned long)qm->aeqe_dma,
1683 		     qm->sqc, (unsigned long)qm->sqc_dma,
1684 		     qm->cqc, (unsigned long)qm->cqc_dma);
1685 
1686 	ret = qm_eq_ctx_cfg(qm);
1687 	if (ret)
1688 		return ret;
1689 
1690 	ret = qm_mb(qm, QM_MB_CMD_SQC_BT, qm->sqc_dma, 0, 0);
1691 	if (ret)
1692 		return ret;
1693 
1694 	ret = qm_mb(qm, QM_MB_CMD_CQC_BT, qm->cqc_dma, 0, 0);
1695 	if (ret)
1696 		return ret;
1697 
1698 	writel(0x0, qm->io_base + QM_VF_EQ_INT_MASK);
1699 	writel(0x0, qm->io_base + QM_VF_AEQ_INT_MASK);
1700 
1701 	return 0;
1702 }
1703 
1704 /**
1705  * hisi_qm_start() - start qm
1706  * @qm: The qm to be started.
1707  *
1708  * This function starts a qm, then we can allocate qp from this qm.
1709  */
1710 int hisi_qm_start(struct hisi_qm *qm)
1711 {
1712 	struct device *dev = &qm->pdev->dev;
1713 
1714 	dev_dbg(dev, "qm start with %d queue pairs\n", qm->qp_num);
1715 
1716 	if (!qm->qp_num) {
1717 		dev_err(dev, "qp_num should not be 0\n");
1718 		return -EINVAL;
1719 	}
1720 
1721 	if (!qm->qp_bitmap) {
1722 		qm->qp_bitmap = devm_kcalloc(dev, BITS_TO_LONGS(qm->qp_num),
1723 					     sizeof(long), GFP_KERNEL);
1724 		qm->qp_array = devm_kcalloc(dev, qm->qp_num,
1725 					    sizeof(struct hisi_qp *),
1726 					    GFP_KERNEL);
1727 		if (!qm->qp_bitmap || !qm->qp_array)
1728 			return -ENOMEM;
1729 	}
1730 
1731 	if (!qm->use_dma_api) {
1732 		dev_dbg(&qm->pdev->dev, "qm delay start\n");
1733 		return 0;
1734 	} else if (!qm->qdma.va) {
1735 		qm->qdma.size = QMC_ALIGN(sizeof(struct qm_eqe) * QM_Q_DEPTH) +
1736 				QMC_ALIGN(sizeof(struct qm_aeqe) * QM_Q_DEPTH) +
1737 				QMC_ALIGN(sizeof(struct qm_sqc) * qm->qp_num) +
1738 				QMC_ALIGN(sizeof(struct qm_cqc) * qm->qp_num);
1739 		qm->qdma.va = dma_alloc_coherent(dev, qm->qdma.size,
1740 						 &qm->qdma.dma, GFP_KERNEL);
1741 		dev_dbg(dev, "allocate qm dma buf(va=%pK, dma=%pad, size=%zx)\n",
1742 			qm->qdma.va, &qm->qdma.dma, qm->qdma.size);
1743 		if (!qm->qdma.va)
1744 			return -ENOMEM;
1745 	}
1746 
1747 	return __hisi_qm_start(qm);
1748 }
1749 EXPORT_SYMBOL_GPL(hisi_qm_start);
1750 
1751 /**
1752  * hisi_qm_stop() - Stop a qm.
1753  * @qm: The qm which will be stopped.
1754  *
1755  * This function stops qm and its qps, then qm can not accept request.
1756  * Related resources are not released at this state, we can use hisi_qm_start
1757  * to let qm start again.
1758  */
1759 int hisi_qm_stop(struct hisi_qm *qm)
1760 {
1761 	struct device *dev;
1762 	struct hisi_qp *qp;
1763 	int ret = 0, i;
1764 
1765 	if (!qm || !qm->pdev) {
1766 		WARN_ON(1);
1767 		return -EINVAL;
1768 	}
1769 
1770 	dev = &qm->pdev->dev;
1771 
1772 	/* Mask eq and aeq irq */
1773 	writel(0x1, qm->io_base + QM_VF_EQ_INT_MASK);
1774 	writel(0x1, qm->io_base + QM_VF_AEQ_INT_MASK);
1775 
1776 	/* Stop all qps belong to this qm */
1777 	for (i = 0; i < qm->qp_num; i++) {
1778 		qp = qm->qp_array[i];
1779 		if (qp) {
1780 			ret = hisi_qm_stop_qp(qp);
1781 			if (ret < 0) {
1782 				dev_err(dev, "Failed to stop qp%d!\n", i);
1783 				return -EBUSY;
1784 			}
1785 		}
1786 	}
1787 
1788 	if (qm->fun_type == QM_HW_PF) {
1789 		ret = hisi_qm_set_vft(qm, 0, 0, 0);
1790 		if (ret < 0)
1791 			dev_err(dev, "Failed to set vft!\n");
1792 	}
1793 
1794 	return ret;
1795 }
1796 EXPORT_SYMBOL_GPL(hisi_qm_stop);
1797 
1798 /**
1799  * hisi_qm_debug_init() - Initialize qm related debugfs files.
1800  * @qm: The qm for which we want to add debugfs files.
1801  *
1802  * Create qm related debugfs files.
1803  */
1804 int hisi_qm_debug_init(struct hisi_qm *qm)
1805 {
1806 	struct dentry *qm_d;
1807 	int i, ret;
1808 
1809 	qm_d = debugfs_create_dir("qm", qm->debug.debug_root);
1810 	qm->debug.qm_d = qm_d;
1811 
1812 	/* only show this in PF */
1813 	if (qm->fun_type == QM_HW_PF)
1814 		for (i = CURRENT_Q; i < DEBUG_FILE_NUM; i++)
1815 			if (qm_create_debugfs_file(qm, i)) {
1816 				ret = -ENOENT;
1817 				goto failed_to_create;
1818 			}
1819 
1820 	debugfs_create_file("qm_regs", 0444, qm->debug.qm_d, qm, &qm_regs_fops);
1821 
1822 	return 0;
1823 
1824 failed_to_create:
1825 	debugfs_remove_recursive(qm_d);
1826 	return ret;
1827 }
1828 EXPORT_SYMBOL_GPL(hisi_qm_debug_init);
1829 
1830 /**
1831  * hisi_qm_debug_regs_clear() - clear qm debug related registers.
1832  * @qm: The qm for which we want to clear its debug registers.
1833  */
1834 void hisi_qm_debug_regs_clear(struct hisi_qm *qm)
1835 {
1836 	struct qm_dfx_registers *regs;
1837 	int i;
1838 
1839 	/* clear current_q */
1840 	writel(0x0, qm->io_base + QM_DFX_SQE_CNT_VF_SQN);
1841 	writel(0x0, qm->io_base + QM_DFX_CQE_CNT_VF_CQN);
1842 
1843 	/*
1844 	 * these registers are reading and clearing, so clear them after
1845 	 * reading them.
1846 	 */
1847 	writel(0x1, qm->io_base + QM_DFX_CNT_CLR_CE);
1848 
1849 	regs = qm_dfx_regs;
1850 	for (i = 0; i < CNT_CYC_REGS_NUM; i++) {
1851 		readl(qm->io_base + regs->reg_offset);
1852 		regs++;
1853 	}
1854 
1855 	writel(0x0, qm->io_base + QM_DFX_CNT_CLR_CE);
1856 }
1857 EXPORT_SYMBOL_GPL(hisi_qm_debug_regs_clear);
1858 
1859 /**
1860  * hisi_qm_hw_error_init() - Configure qm hardware error report method.
1861  * @qm: The qm which we want to configure.
1862  * @ce: Bit mask of correctable error configure.
1863  * @nfe: Bit mask of non-fatal error configure.
1864  * @fe: Bit mask of fatal error configure.
1865  * @msi: Bit mask of error reported by message signal interrupt.
1866  *
1867  * Hardware errors of qm can be reported either by RAS interrupts which will
1868  * be handled by UEFI and then PCIe AER or by device MSI. User can configure
1869  * each error to use either of above two methods. For RAS interrupts, we can
1870  * configure an error as one of correctable error, non-fatal error or
1871  * fatal error.
1872  *
1873  * Bits indicating errors can be configured to ce, nfe, fe and msi to enable
1874  * related report methods. Error report will be masked if related error bit
1875  * does not configure.
1876  */
1877 void hisi_qm_hw_error_init(struct hisi_qm *qm, u32 ce, u32 nfe, u32 fe,
1878 			   u32 msi)
1879 {
1880 	if (!qm->ops->hw_error_init) {
1881 		dev_err(&qm->pdev->dev, "QM doesn't support hw error handling!\n");
1882 		return;
1883 	}
1884 
1885 	qm->ops->hw_error_init(qm, ce, nfe, fe, msi);
1886 }
1887 EXPORT_SYMBOL_GPL(hisi_qm_hw_error_init);
1888 
1889 /**
1890  * hisi_qm_hw_error_handle() - Handle qm non-fatal hardware errors.
1891  * @qm: The qm which has non-fatal hardware errors.
1892  *
1893  * Accelerators use this function to handle qm non-fatal hardware errors.
1894  */
1895 pci_ers_result_t hisi_qm_hw_error_handle(struct hisi_qm *qm)
1896 {
1897 	if (!qm->ops->hw_error_handle) {
1898 		dev_err(&qm->pdev->dev, "QM doesn't support hw error report!\n");
1899 		return PCI_ERS_RESULT_NONE;
1900 	}
1901 
1902 	return qm->ops->hw_error_handle(qm);
1903 }
1904 EXPORT_SYMBOL_GPL(hisi_qm_hw_error_handle);
1905 
1906 /**
1907  * hisi_qm_get_hw_version() - Get hardware version of a qm.
1908  * @pdev: The device which hardware version we want to get.
1909  *
1910  * This function gets the hardware version of a qm. Return QM_HW_UNKNOWN
1911  * if the hardware version is not supported.
1912  */
1913 enum qm_hw_ver hisi_qm_get_hw_version(struct pci_dev *pdev)
1914 {
1915 	switch (pdev->revision) {
1916 	case QM_HW_V1:
1917 	case QM_HW_V2:
1918 		return pdev->revision;
1919 	default:
1920 		return QM_HW_UNKNOWN;
1921 	}
1922 }
1923 EXPORT_SYMBOL_GPL(hisi_qm_get_hw_version);
1924 
1925 MODULE_LICENSE("GPL v2");
1926 MODULE_AUTHOR("Zhou Wang <wangzhou1@hisilicon.com>");
1927 MODULE_DESCRIPTION("HiSilicon Accelerator queue manager driver");
1928