1 /* SPDX-License-Identifier: GPL-2.0-only
2  * Copyright (C) 2020 Marvell.
3  */
4 #ifndef __OTX2_CPTLF_H
5 #define __OTX2_CPTLF_H
6 
7 #include <linux/soc/marvell/octeontx2/asm.h>
8 #include <mbox.h>
9 #include <rvu.h>
10 #include "otx2_cpt_common.h"
11 #include "otx2_cpt_reqmgr.h"
12 
13 /*
14  * CPT instruction and pending queues user requested length in CPT_INST_S msgs
15  */
16 #define OTX2_CPT_USER_REQUESTED_QLEN_MSGS 8200
17 
18 /*
19  * CPT instruction queue size passed to HW is in units of 40*CPT_INST_S
20  * messages.
21  */
22 #define OTX2_CPT_SIZE_DIV40 (OTX2_CPT_USER_REQUESTED_QLEN_MSGS/40)
23 
24 /*
25  * CPT instruction and pending queues length in CPT_INST_S messages
26  */
27 #define OTX2_CPT_INST_QLEN_MSGS	((OTX2_CPT_SIZE_DIV40 - 1) * 40)
28 
29 /* CPT instruction queue length in bytes */
30 #define OTX2_CPT_INST_QLEN_BYTES (OTX2_CPT_SIZE_DIV40 * 40 * \
31 				  OTX2_CPT_INST_SIZE)
32 
33 /* CPT instruction group queue length in bytes */
34 #define OTX2_CPT_INST_GRP_QLEN_BYTES (OTX2_CPT_SIZE_DIV40 * 16)
35 
36 /* CPT FC length in bytes */
37 #define OTX2_CPT_Q_FC_LEN 128
38 
39 /* CPT instruction queue alignment */
40 #define OTX2_CPT_INST_Q_ALIGNMENT  128
41 
42 /* Mask which selects all engine groups */
43 #define OTX2_CPT_ALL_ENG_GRPS_MASK 0xFF
44 
45 /* Maximum LFs supported in OcteonTX2 for CPT */
46 #define OTX2_CPT_MAX_LFS_NUM    64
47 
48 /* Queue priority */
49 #define OTX2_CPT_QUEUE_HI_PRIO  0x1
50 #define OTX2_CPT_QUEUE_LOW_PRIO 0x0
51 
52 enum otx2_cptlf_state {
53 	OTX2_CPTLF_IN_RESET,
54 	OTX2_CPTLF_STARTED,
55 };
56 
57 struct otx2_cpt_inst_queue {
58 	u8 *vaddr;
59 	u8 *real_vaddr;
60 	dma_addr_t dma_addr;
61 	dma_addr_t real_dma_addr;
62 	u32 size;
63 };
64 
65 struct otx2_cptlfs_info;
66 struct otx2_cptlf_wqe {
67 	struct tasklet_struct work;
68 	struct otx2_cptlfs_info *lfs;
69 	u8 lf_num;
70 };
71 
72 struct otx2_cptlf_info {
73 	struct otx2_cptlfs_info *lfs;           /* Ptr to cptlfs_info struct */
74 	void __iomem *lmtline;                  /* Address of LMTLINE */
75 	void __iomem *ioreg;                    /* LMTLINE send register */
76 	int msix_offset;                        /* MSI-X interrupts offset */
77 	cpumask_var_t affinity_mask;            /* IRQs affinity mask */
78 	u8 irq_name[OTX2_CPT_LF_MSIX_VECTORS][32];/* Interrupts name */
79 	u8 is_irq_reg[OTX2_CPT_LF_MSIX_VECTORS];  /* Is interrupt registered */
80 	u8 slot;                                /* Slot number of this LF */
81 
82 	struct otx2_cpt_inst_queue iqueue;/* Instruction queue */
83 	struct otx2_cpt_pending_queue pqueue; /* Pending queue */
84 	struct otx2_cptlf_wqe *wqe;       /* Tasklet work info */
85 };
86 
87 struct otx2_cptlfs_info {
88 	/* Registers start address of VF/PF LFs are attached to */
89 	void __iomem *reg_base;
90 #define LMTLINE_SIZE  128
91 	void __iomem *lmt_base;
92 	struct pci_dev *pdev;   /* Device LFs are attached to */
93 	struct otx2_cptlf_info lf[OTX2_CPT_MAX_LFS_NUM];
94 	struct otx2_mbox *mbox;
95 	u8 are_lfs_attached;	/* Whether CPT LFs are attached */
96 	u8 lfs_num;		/* Number of CPT LFs */
97 	u8 kcrypto_eng_grp_num;	/* Kernel crypto engine group number */
98 	u8 kvf_limits;          /* Kernel crypto limits */
99 	atomic_t state;         /* LF's state. started/reset */
100 	int blkaddr;            /* CPT blkaddr: BLKADDR_CPT0/BLKADDR_CPT1 */
101 };
102 
103 static inline void otx2_cpt_free_instruction_queues(
104 					struct otx2_cptlfs_info *lfs)
105 {
106 	struct otx2_cpt_inst_queue *iq;
107 	int i;
108 
109 	for (i = 0; i < lfs->lfs_num; i++) {
110 		iq = &lfs->lf[i].iqueue;
111 		if (iq->real_vaddr)
112 			dma_free_coherent(&lfs->pdev->dev,
113 					  iq->size,
114 					  iq->real_vaddr,
115 					  iq->real_dma_addr);
116 		iq->real_vaddr = NULL;
117 		iq->vaddr = NULL;
118 	}
119 }
120 
121 static inline int otx2_cpt_alloc_instruction_queues(
122 					struct otx2_cptlfs_info *lfs)
123 {
124 	struct otx2_cpt_inst_queue *iq;
125 	int ret = 0, i;
126 
127 	if (!lfs->lfs_num)
128 		return -EINVAL;
129 
130 	for (i = 0; i < lfs->lfs_num; i++) {
131 		iq = &lfs->lf[i].iqueue;
132 		iq->size = OTX2_CPT_INST_QLEN_BYTES +
133 			   OTX2_CPT_Q_FC_LEN +
134 			   OTX2_CPT_INST_GRP_QLEN_BYTES +
135 			   OTX2_CPT_INST_Q_ALIGNMENT;
136 		iq->real_vaddr = dma_alloc_coherent(&lfs->pdev->dev, iq->size,
137 					&iq->real_dma_addr, GFP_KERNEL);
138 		if (!iq->real_vaddr) {
139 			ret = -ENOMEM;
140 			goto error;
141 		}
142 		iq->vaddr = iq->real_vaddr + OTX2_CPT_INST_GRP_QLEN_BYTES;
143 		iq->dma_addr = iq->real_dma_addr + OTX2_CPT_INST_GRP_QLEN_BYTES;
144 
145 		/* Align pointers */
146 		iq->vaddr = PTR_ALIGN(iq->vaddr, OTX2_CPT_INST_Q_ALIGNMENT);
147 		iq->dma_addr = PTR_ALIGN(iq->dma_addr,
148 					 OTX2_CPT_INST_Q_ALIGNMENT);
149 	}
150 	return 0;
151 
152 error:
153 	otx2_cpt_free_instruction_queues(lfs);
154 	return ret;
155 }
156 
157 static inline void otx2_cptlf_set_iqueues_base_addr(
158 					struct otx2_cptlfs_info *lfs)
159 {
160 	union otx2_cptx_lf_q_base lf_q_base;
161 	int slot;
162 
163 	for (slot = 0; slot < lfs->lfs_num; slot++) {
164 		lf_q_base.u = lfs->lf[slot].iqueue.dma_addr;
165 		otx2_cpt_write64(lfs->reg_base, BLKADDR_CPT0, slot,
166 				 OTX2_CPT_LF_Q_BASE, lf_q_base.u);
167 	}
168 }
169 
170 static inline void otx2_cptlf_do_set_iqueue_size(struct otx2_cptlf_info *lf)
171 {
172 	union otx2_cptx_lf_q_size lf_q_size = { .u = 0x0 };
173 
174 	lf_q_size.s.size_div40 = OTX2_CPT_SIZE_DIV40;
175 	otx2_cpt_write64(lf->lfs->reg_base, BLKADDR_CPT0, lf->slot,
176 			 OTX2_CPT_LF_Q_SIZE, lf_q_size.u);
177 }
178 
179 static inline void otx2_cptlf_set_iqueues_size(struct otx2_cptlfs_info *lfs)
180 {
181 	int slot;
182 
183 	for (slot = 0; slot < lfs->lfs_num; slot++)
184 		otx2_cptlf_do_set_iqueue_size(&lfs->lf[slot]);
185 }
186 
187 static inline void otx2_cptlf_do_disable_iqueue(struct otx2_cptlf_info *lf)
188 {
189 	union otx2_cptx_lf_ctl lf_ctl = { .u = 0x0 };
190 	union otx2_cptx_lf_inprog lf_inprog;
191 	int timeout = 20;
192 
193 	/* Disable instructions enqueuing */
194 	otx2_cpt_write64(lf->lfs->reg_base, BLKADDR_CPT0, lf->slot,
195 			 OTX2_CPT_LF_CTL, lf_ctl.u);
196 
197 	/* Wait for instruction queue to become empty */
198 	do {
199 		lf_inprog.u = otx2_cpt_read64(lf->lfs->reg_base, BLKADDR_CPT0,
200 					      lf->slot, OTX2_CPT_LF_INPROG);
201 		if (!lf_inprog.s.inflight)
202 			break;
203 
204 		usleep_range(10000, 20000);
205 		if (timeout-- < 0) {
206 			dev_err(&lf->lfs->pdev->dev,
207 				"Error LF %d is still busy.\n", lf->slot);
208 			break;
209 		}
210 
211 	} while (1);
212 
213 	/*
214 	 * Disable executions in the LF's queue,
215 	 * the queue should be empty at this point
216 	 */
217 	lf_inprog.s.eena = 0x0;
218 	otx2_cpt_write64(lf->lfs->reg_base, BLKADDR_CPT0, lf->slot,
219 			 OTX2_CPT_LF_INPROG, lf_inprog.u);
220 }
221 
222 static inline void otx2_cptlf_disable_iqueues(struct otx2_cptlfs_info *lfs)
223 {
224 	int slot;
225 
226 	for (slot = 0; slot < lfs->lfs_num; slot++)
227 		otx2_cptlf_do_disable_iqueue(&lfs->lf[slot]);
228 }
229 
230 static inline void otx2_cptlf_set_iqueue_enq(struct otx2_cptlf_info *lf,
231 					     bool enable)
232 {
233 	union otx2_cptx_lf_ctl lf_ctl;
234 
235 	lf_ctl.u = otx2_cpt_read64(lf->lfs->reg_base, BLKADDR_CPT0, lf->slot,
236 				   OTX2_CPT_LF_CTL);
237 
238 	/* Set iqueue's enqueuing */
239 	lf_ctl.s.ena = enable ? 0x1 : 0x0;
240 	otx2_cpt_write64(lf->lfs->reg_base, BLKADDR_CPT0, lf->slot,
241 			 OTX2_CPT_LF_CTL, lf_ctl.u);
242 }
243 
244 static inline void otx2_cptlf_enable_iqueue_enq(struct otx2_cptlf_info *lf)
245 {
246 	otx2_cptlf_set_iqueue_enq(lf, true);
247 }
248 
249 static inline void otx2_cptlf_set_iqueue_exec(struct otx2_cptlf_info *lf,
250 					      bool enable)
251 {
252 	union otx2_cptx_lf_inprog lf_inprog;
253 
254 	lf_inprog.u = otx2_cpt_read64(lf->lfs->reg_base, BLKADDR_CPT0, lf->slot,
255 				      OTX2_CPT_LF_INPROG);
256 
257 	/* Set iqueue's execution */
258 	lf_inprog.s.eena = enable ? 0x1 : 0x0;
259 	otx2_cpt_write64(lf->lfs->reg_base, BLKADDR_CPT0, lf->slot,
260 			 OTX2_CPT_LF_INPROG, lf_inprog.u);
261 }
262 
263 static inline void otx2_cptlf_enable_iqueue_exec(struct otx2_cptlf_info *lf)
264 {
265 	otx2_cptlf_set_iqueue_exec(lf, true);
266 }
267 
268 static inline void otx2_cptlf_disable_iqueue_exec(struct otx2_cptlf_info *lf)
269 {
270 	otx2_cptlf_set_iqueue_exec(lf, false);
271 }
272 
273 static inline void otx2_cptlf_enable_iqueues(struct otx2_cptlfs_info *lfs)
274 {
275 	int slot;
276 
277 	for (slot = 0; slot < lfs->lfs_num; slot++) {
278 		otx2_cptlf_enable_iqueue_exec(&lfs->lf[slot]);
279 		otx2_cptlf_enable_iqueue_enq(&lfs->lf[slot]);
280 	}
281 }
282 
283 static inline void otx2_cpt_fill_inst(union otx2_cpt_inst_s *cptinst,
284 				      struct otx2_cpt_iq_command *iq_cmd,
285 				      u64 comp_baddr)
286 {
287 	cptinst->u[0] = 0x0;
288 	cptinst->s.doneint = true;
289 	cptinst->s.res_addr = comp_baddr;
290 	cptinst->u[2] = 0x0;
291 	cptinst->u[3] = 0x0;
292 	cptinst->s.ei0 = iq_cmd->cmd.u;
293 	cptinst->s.ei1 = iq_cmd->dptr;
294 	cptinst->s.ei2 = iq_cmd->rptr;
295 	cptinst->s.ei3 = iq_cmd->cptr.u;
296 }
297 
298 /*
299  * On OcteonTX2 platform the parameter insts_num is used as a count of
300  * instructions to be enqueued. The valid values for insts_num are:
301  * 1 - 1 CPT instruction will be enqueued during LMTST operation
302  * 2 - 2 CPT instructions will be enqueued during LMTST operation
303  */
304 static inline void otx2_cpt_send_cmd(union otx2_cpt_inst_s *cptinst,
305 				     u32 insts_num, struct otx2_cptlf_info *lf)
306 {
307 	void __iomem *lmtline = lf->lmtline;
308 	long ret;
309 
310 	/*
311 	 * Make sure memory areas pointed in CPT_INST_S
312 	 * are flushed before the instruction is sent to CPT
313 	 */
314 	dma_wmb();
315 
316 	do {
317 		/* Copy CPT command to LMTLINE */
318 		memcpy_toio(lmtline, cptinst, insts_num * OTX2_CPT_INST_SIZE);
319 
320 		/*
321 		 * LDEOR initiates atomic transfer to I/O device
322 		 * The following will cause the LMTST to fail (the LDEOR
323 		 * returns zero):
324 		 * - No stores have been performed to the LMTLINE since it was
325 		 * last invalidated.
326 		 * - The bytes which have been stored to LMTLINE since it was
327 		 * last invalidated form a pattern that is non-contiguous, does
328 		 * not start at byte 0, or does not end on a 8-byte boundary.
329 		 * (i.e.comprises a formation of other than 1–16 8-byte
330 		 * words.)
331 		 *
332 		 * These rules are designed such that an operating system
333 		 * context switch or hypervisor guest switch need have no
334 		 * knowledge of the LMTST operations; the switch code does not
335 		 * need to store to LMTCANCEL. Also note as LMTLINE data cannot
336 		 * be read, there is no information leakage between processes.
337 		 */
338 		ret = otx2_lmt_flush(lf->ioreg);
339 
340 	} while (!ret);
341 }
342 
343 static inline bool otx2_cptlf_started(struct otx2_cptlfs_info *lfs)
344 {
345 	return atomic_read(&lfs->state) == OTX2_CPTLF_STARTED;
346 }
347 
348 int otx2_cptlf_init(struct otx2_cptlfs_info *lfs, u8 eng_grp_msk, int pri,
349 		    int lfs_num);
350 void otx2_cptlf_shutdown(struct otx2_cptlfs_info *lfs);
351 int otx2_cptlf_register_interrupts(struct otx2_cptlfs_info *lfs);
352 void otx2_cptlf_unregister_interrupts(struct otx2_cptlfs_info *lfs);
353 void otx2_cptlf_free_irqs_affinity(struct otx2_cptlfs_info *lfs);
354 int otx2_cptlf_set_irqs_affinity(struct otx2_cptlfs_info *lfs);
355 
356 #endif /* __OTX2_CPTLF_H */
357