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 	struct pci_dev *pdev;   /* Device LFs are attached to */
91 	struct otx2_cptlf_info lf[OTX2_CPT_MAX_LFS_NUM];
92 	struct otx2_mbox *mbox;
93 	u8 are_lfs_attached;	/* Whether CPT LFs are attached */
94 	u8 lfs_num;		/* Number of CPT LFs */
95 	u8 kcrypto_eng_grp_num;	/* Kernel crypto engine group number */
96 	u8 kvf_limits;          /* Kernel crypto limits */
97 	atomic_t state;         /* LF's state. started/reset */
98 };
99 
100 static inline void otx2_cpt_free_instruction_queues(
101 					struct otx2_cptlfs_info *lfs)
102 {
103 	struct otx2_cpt_inst_queue *iq;
104 	int i;
105 
106 	for (i = 0; i < lfs->lfs_num; i++) {
107 		iq = &lfs->lf[i].iqueue;
108 		if (iq->real_vaddr)
109 			dma_free_coherent(&lfs->pdev->dev,
110 					  iq->size,
111 					  iq->real_vaddr,
112 					  iq->real_dma_addr);
113 		iq->real_vaddr = NULL;
114 		iq->vaddr = NULL;
115 	}
116 }
117 
118 static inline int otx2_cpt_alloc_instruction_queues(
119 					struct otx2_cptlfs_info *lfs)
120 {
121 	struct otx2_cpt_inst_queue *iq;
122 	int ret = 0, i;
123 
124 	if (!lfs->lfs_num)
125 		return -EINVAL;
126 
127 	for (i = 0; i < lfs->lfs_num; i++) {
128 		iq = &lfs->lf[i].iqueue;
129 		iq->size = OTX2_CPT_INST_QLEN_BYTES +
130 			   OTX2_CPT_Q_FC_LEN +
131 			   OTX2_CPT_INST_GRP_QLEN_BYTES +
132 			   OTX2_CPT_INST_Q_ALIGNMENT;
133 		iq->real_vaddr = dma_alloc_coherent(&lfs->pdev->dev, iq->size,
134 					&iq->real_dma_addr, GFP_KERNEL);
135 		if (!iq->real_vaddr) {
136 			ret = -ENOMEM;
137 			goto error;
138 		}
139 		iq->vaddr = iq->real_vaddr + OTX2_CPT_INST_GRP_QLEN_BYTES;
140 		iq->dma_addr = iq->real_dma_addr + OTX2_CPT_INST_GRP_QLEN_BYTES;
141 
142 		/* Align pointers */
143 		iq->vaddr = PTR_ALIGN(iq->vaddr, OTX2_CPT_INST_Q_ALIGNMENT);
144 		iq->dma_addr = PTR_ALIGN(iq->dma_addr,
145 					 OTX2_CPT_INST_Q_ALIGNMENT);
146 	}
147 	return 0;
148 
149 error:
150 	otx2_cpt_free_instruction_queues(lfs);
151 	return ret;
152 }
153 
154 static inline void otx2_cptlf_set_iqueues_base_addr(
155 					struct otx2_cptlfs_info *lfs)
156 {
157 	union otx2_cptx_lf_q_base lf_q_base;
158 	int slot;
159 
160 	for (slot = 0; slot < lfs->lfs_num; slot++) {
161 		lf_q_base.u = lfs->lf[slot].iqueue.dma_addr;
162 		otx2_cpt_write64(lfs->reg_base, BLKADDR_CPT0, slot,
163 				 OTX2_CPT_LF_Q_BASE, lf_q_base.u);
164 	}
165 }
166 
167 static inline void otx2_cptlf_do_set_iqueue_size(struct otx2_cptlf_info *lf)
168 {
169 	union otx2_cptx_lf_q_size lf_q_size = { .u = 0x0 };
170 
171 	lf_q_size.s.size_div40 = OTX2_CPT_SIZE_DIV40;
172 	otx2_cpt_write64(lf->lfs->reg_base, BLKADDR_CPT0, lf->slot,
173 			 OTX2_CPT_LF_Q_SIZE, lf_q_size.u);
174 }
175 
176 static inline void otx2_cptlf_set_iqueues_size(struct otx2_cptlfs_info *lfs)
177 {
178 	int slot;
179 
180 	for (slot = 0; slot < lfs->lfs_num; slot++)
181 		otx2_cptlf_do_set_iqueue_size(&lfs->lf[slot]);
182 }
183 
184 static inline void otx2_cptlf_do_disable_iqueue(struct otx2_cptlf_info *lf)
185 {
186 	union otx2_cptx_lf_ctl lf_ctl = { .u = 0x0 };
187 	union otx2_cptx_lf_inprog lf_inprog;
188 	int timeout = 20;
189 
190 	/* Disable instructions enqueuing */
191 	otx2_cpt_write64(lf->lfs->reg_base, BLKADDR_CPT0, lf->slot,
192 			 OTX2_CPT_LF_CTL, lf_ctl.u);
193 
194 	/* Wait for instruction queue to become empty */
195 	do {
196 		lf_inprog.u = otx2_cpt_read64(lf->lfs->reg_base, BLKADDR_CPT0,
197 					      lf->slot, OTX2_CPT_LF_INPROG);
198 		if (!lf_inprog.s.inflight)
199 			break;
200 
201 		usleep_range(10000, 20000);
202 		if (timeout-- < 0) {
203 			dev_err(&lf->lfs->pdev->dev,
204 				"Error LF %d is still busy.\n", lf->slot);
205 			break;
206 		}
207 
208 	} while (1);
209 
210 	/*
211 	 * Disable executions in the LF's queue,
212 	 * the queue should be empty at this point
213 	 */
214 	lf_inprog.s.eena = 0x0;
215 	otx2_cpt_write64(lf->lfs->reg_base, BLKADDR_CPT0, lf->slot,
216 			 OTX2_CPT_LF_INPROG, lf_inprog.u);
217 }
218 
219 static inline void otx2_cptlf_disable_iqueues(struct otx2_cptlfs_info *lfs)
220 {
221 	int slot;
222 
223 	for (slot = 0; slot < lfs->lfs_num; slot++)
224 		otx2_cptlf_do_disable_iqueue(&lfs->lf[slot]);
225 }
226 
227 static inline void otx2_cptlf_set_iqueue_enq(struct otx2_cptlf_info *lf,
228 					     bool enable)
229 {
230 	union otx2_cptx_lf_ctl lf_ctl;
231 
232 	lf_ctl.u = otx2_cpt_read64(lf->lfs->reg_base, BLKADDR_CPT0, lf->slot,
233 				   OTX2_CPT_LF_CTL);
234 
235 	/* Set iqueue's enqueuing */
236 	lf_ctl.s.ena = enable ? 0x1 : 0x0;
237 	otx2_cpt_write64(lf->lfs->reg_base, BLKADDR_CPT0, lf->slot,
238 			 OTX2_CPT_LF_CTL, lf_ctl.u);
239 }
240 
241 static inline void otx2_cptlf_enable_iqueue_enq(struct otx2_cptlf_info *lf)
242 {
243 	otx2_cptlf_set_iqueue_enq(lf, true);
244 }
245 
246 static inline void otx2_cptlf_set_iqueue_exec(struct otx2_cptlf_info *lf,
247 					      bool enable)
248 {
249 	union otx2_cptx_lf_inprog lf_inprog;
250 
251 	lf_inprog.u = otx2_cpt_read64(lf->lfs->reg_base, BLKADDR_CPT0, lf->slot,
252 				      OTX2_CPT_LF_INPROG);
253 
254 	/* Set iqueue's execution */
255 	lf_inprog.s.eena = enable ? 0x1 : 0x0;
256 	otx2_cpt_write64(lf->lfs->reg_base, BLKADDR_CPT0, lf->slot,
257 			 OTX2_CPT_LF_INPROG, lf_inprog.u);
258 }
259 
260 static inline void otx2_cptlf_enable_iqueue_exec(struct otx2_cptlf_info *lf)
261 {
262 	otx2_cptlf_set_iqueue_exec(lf, true);
263 }
264 
265 static inline void otx2_cptlf_disable_iqueue_exec(struct otx2_cptlf_info *lf)
266 {
267 	otx2_cptlf_set_iqueue_exec(lf, false);
268 }
269 
270 static inline void otx2_cptlf_enable_iqueues(struct otx2_cptlfs_info *lfs)
271 {
272 	int slot;
273 
274 	for (slot = 0; slot < lfs->lfs_num; slot++) {
275 		otx2_cptlf_enable_iqueue_exec(&lfs->lf[slot]);
276 		otx2_cptlf_enable_iqueue_enq(&lfs->lf[slot]);
277 	}
278 }
279 
280 static inline void otx2_cpt_fill_inst(union otx2_cpt_inst_s *cptinst,
281 				      struct otx2_cpt_iq_command *iq_cmd,
282 				      u64 comp_baddr)
283 {
284 	cptinst->u[0] = 0x0;
285 	cptinst->s.doneint = true;
286 	cptinst->s.res_addr = comp_baddr;
287 	cptinst->u[2] = 0x0;
288 	cptinst->u[3] = 0x0;
289 	cptinst->s.ei0 = iq_cmd->cmd.u;
290 	cptinst->s.ei1 = iq_cmd->dptr;
291 	cptinst->s.ei2 = iq_cmd->rptr;
292 	cptinst->s.ei3 = iq_cmd->cptr.u;
293 }
294 
295 /*
296  * On OcteonTX2 platform the parameter insts_num is used as a count of
297  * instructions to be enqueued. The valid values for insts_num are:
298  * 1 - 1 CPT instruction will be enqueued during LMTST operation
299  * 2 - 2 CPT instructions will be enqueued during LMTST operation
300  */
301 static inline void otx2_cpt_send_cmd(union otx2_cpt_inst_s *cptinst,
302 				     u32 insts_num, struct otx2_cptlf_info *lf)
303 {
304 	void __iomem *lmtline = lf->lmtline;
305 	long ret;
306 
307 	/*
308 	 * Make sure memory areas pointed in CPT_INST_S
309 	 * are flushed before the instruction is sent to CPT
310 	 */
311 	dma_wmb();
312 
313 	do {
314 		/* Copy CPT command to LMTLINE */
315 		memcpy_toio(lmtline, cptinst, insts_num * OTX2_CPT_INST_SIZE);
316 
317 		/*
318 		 * LDEOR initiates atomic transfer to I/O device
319 		 * The following will cause the LMTST to fail (the LDEOR
320 		 * returns zero):
321 		 * - No stores have been performed to the LMTLINE since it was
322 		 * last invalidated.
323 		 * - The bytes which have been stored to LMTLINE since it was
324 		 * last invalidated form a pattern that is non-contiguous, does
325 		 * not start at byte 0, or does not end on a 8-byte boundary.
326 		 * (i.e.comprises a formation of other than 1–16 8-byte
327 		 * words.)
328 		 *
329 		 * These rules are designed such that an operating system
330 		 * context switch or hypervisor guest switch need have no
331 		 * knowledge of the LMTST operations; the switch code does not
332 		 * need to store to LMTCANCEL. Also note as LMTLINE data cannot
333 		 * be read, there is no information leakage between processes.
334 		 */
335 		ret = otx2_lmt_flush(lf->ioreg);
336 
337 	} while (!ret);
338 }
339 
340 static inline bool otx2_cptlf_started(struct otx2_cptlfs_info *lfs)
341 {
342 	return atomic_read(&lfs->state) == OTX2_CPTLF_STARTED;
343 }
344 
345 int otx2_cptlf_init(struct otx2_cptlfs_info *lfs, u8 eng_grp_msk, int pri,
346 		    int lfs_num);
347 void otx2_cptlf_shutdown(struct otx2_cptlfs_info *lfs);
348 int otx2_cptlf_register_interrupts(struct otx2_cptlfs_info *lfs);
349 void otx2_cptlf_unregister_interrupts(struct otx2_cptlfs_info *lfs);
350 void otx2_cptlf_free_irqs_affinity(struct otx2_cptlfs_info *lfs);
351 int otx2_cptlf_set_irqs_affinity(struct otx2_cptlfs_info *lfs);
352 
353 #endif /* __OTX2_CPTLF_H */
354