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_cptlf_wqe *wqe;       /* Tasklet work info */
84 };
85 
86 struct otx2_cptlfs_info {
87 	/* Registers start address of VF/PF LFs are attached to */
88 	void __iomem *reg_base;
89 	struct pci_dev *pdev;   /* Device LFs are attached to */
90 	struct otx2_cptlf_info lf[OTX2_CPT_MAX_LFS_NUM];
91 	struct otx2_mbox *mbox;
92 	u8 are_lfs_attached;	/* Whether CPT LFs are attached */
93 	u8 lfs_num;		/* Number of CPT LFs */
94 	atomic_t state;         /* LF's state. started/reset */
95 };
96 
97 static inline void otx2_cpt_free_instruction_queues(
98 					struct otx2_cptlfs_info *lfs)
99 {
100 	struct otx2_cpt_inst_queue *iq;
101 	int i;
102 
103 	for (i = 0; i < lfs->lfs_num; i++) {
104 		iq = &lfs->lf[i].iqueue;
105 		if (iq->real_vaddr)
106 			dma_free_coherent(&lfs->pdev->dev,
107 					  iq->size,
108 					  iq->real_vaddr,
109 					  iq->real_dma_addr);
110 		iq->real_vaddr = NULL;
111 		iq->vaddr = NULL;
112 	}
113 }
114 
115 static inline int otx2_cpt_alloc_instruction_queues(
116 					struct otx2_cptlfs_info *lfs)
117 {
118 	struct otx2_cpt_inst_queue *iq;
119 	int ret = 0, i;
120 
121 	if (!lfs->lfs_num)
122 		return -EINVAL;
123 
124 	for (i = 0; i < lfs->lfs_num; i++) {
125 		iq = &lfs->lf[i].iqueue;
126 		iq->size = OTX2_CPT_INST_QLEN_BYTES +
127 			   OTX2_CPT_Q_FC_LEN +
128 			   OTX2_CPT_INST_GRP_QLEN_BYTES +
129 			   OTX2_CPT_INST_Q_ALIGNMENT;
130 		iq->real_vaddr = dma_alloc_coherent(&lfs->pdev->dev, iq->size,
131 					&iq->real_dma_addr, GFP_KERNEL);
132 		if (!iq->real_vaddr) {
133 			ret = -ENOMEM;
134 			goto error;
135 		}
136 		iq->vaddr = iq->real_vaddr + OTX2_CPT_INST_GRP_QLEN_BYTES;
137 		iq->dma_addr = iq->real_dma_addr + OTX2_CPT_INST_GRP_QLEN_BYTES;
138 
139 		/* Align pointers */
140 		iq->vaddr = PTR_ALIGN(iq->vaddr, OTX2_CPT_INST_Q_ALIGNMENT);
141 		iq->dma_addr = PTR_ALIGN(iq->dma_addr,
142 					 OTX2_CPT_INST_Q_ALIGNMENT);
143 	}
144 	return 0;
145 
146 error:
147 	otx2_cpt_free_instruction_queues(lfs);
148 	return ret;
149 }
150 
151 static inline void otx2_cptlf_set_iqueues_base_addr(
152 					struct otx2_cptlfs_info *lfs)
153 {
154 	union otx2_cptx_lf_q_base lf_q_base;
155 	int slot;
156 
157 	for (slot = 0; slot < lfs->lfs_num; slot++) {
158 		lf_q_base.u = lfs->lf[slot].iqueue.dma_addr;
159 		otx2_cpt_write64(lfs->reg_base, BLKADDR_CPT0, slot,
160 				 OTX2_CPT_LF_Q_BASE, lf_q_base.u);
161 	}
162 }
163 
164 static inline void otx2_cptlf_do_set_iqueue_size(struct otx2_cptlf_info *lf)
165 {
166 	union otx2_cptx_lf_q_size lf_q_size = { .u = 0x0 };
167 
168 	lf_q_size.s.size_div40 = OTX2_CPT_SIZE_DIV40;
169 	otx2_cpt_write64(lf->lfs->reg_base, BLKADDR_CPT0, lf->slot,
170 			 OTX2_CPT_LF_Q_SIZE, lf_q_size.u);
171 }
172 
173 static inline void otx2_cptlf_set_iqueues_size(struct otx2_cptlfs_info *lfs)
174 {
175 	int slot;
176 
177 	for (slot = 0; slot < lfs->lfs_num; slot++)
178 		otx2_cptlf_do_set_iqueue_size(&lfs->lf[slot]);
179 }
180 
181 static inline void otx2_cptlf_do_disable_iqueue(struct otx2_cptlf_info *lf)
182 {
183 	union otx2_cptx_lf_ctl lf_ctl = { .u = 0x0 };
184 	union otx2_cptx_lf_inprog lf_inprog;
185 	int timeout = 20;
186 
187 	/* Disable instructions enqueuing */
188 	otx2_cpt_write64(lf->lfs->reg_base, BLKADDR_CPT0, lf->slot,
189 			 OTX2_CPT_LF_CTL, lf_ctl.u);
190 
191 	/* Wait for instruction queue to become empty */
192 	do {
193 		lf_inprog.u = otx2_cpt_read64(lf->lfs->reg_base, BLKADDR_CPT0,
194 					      lf->slot, OTX2_CPT_LF_INPROG);
195 		if (!lf_inprog.s.inflight)
196 			break;
197 
198 		usleep_range(10000, 20000);
199 		if (timeout-- < 0) {
200 			dev_err(&lf->lfs->pdev->dev,
201 				"Error LF %d is still busy.\n", lf->slot);
202 			break;
203 		}
204 
205 	} while (1);
206 
207 	/*
208 	 * Disable executions in the LF's queue,
209 	 * the queue should be empty at this point
210 	 */
211 	lf_inprog.s.eena = 0x0;
212 	otx2_cpt_write64(lf->lfs->reg_base, BLKADDR_CPT0, lf->slot,
213 			 OTX2_CPT_LF_INPROG, lf_inprog.u);
214 }
215 
216 static inline void otx2_cptlf_disable_iqueues(struct otx2_cptlfs_info *lfs)
217 {
218 	int slot;
219 
220 	for (slot = 0; slot < lfs->lfs_num; slot++)
221 		otx2_cptlf_do_disable_iqueue(&lfs->lf[slot]);
222 }
223 
224 static inline void otx2_cptlf_set_iqueue_enq(struct otx2_cptlf_info *lf,
225 					     bool enable)
226 {
227 	union otx2_cptx_lf_ctl lf_ctl;
228 
229 	lf_ctl.u = otx2_cpt_read64(lf->lfs->reg_base, BLKADDR_CPT0, lf->slot,
230 				   OTX2_CPT_LF_CTL);
231 
232 	/* Set iqueue's enqueuing */
233 	lf_ctl.s.ena = enable ? 0x1 : 0x0;
234 	otx2_cpt_write64(lf->lfs->reg_base, BLKADDR_CPT0, lf->slot,
235 			 OTX2_CPT_LF_CTL, lf_ctl.u);
236 }
237 
238 static inline void otx2_cptlf_enable_iqueue_enq(struct otx2_cptlf_info *lf)
239 {
240 	otx2_cptlf_set_iqueue_enq(lf, true);
241 }
242 
243 static inline void otx2_cptlf_set_iqueue_exec(struct otx2_cptlf_info *lf,
244 					      bool enable)
245 {
246 	union otx2_cptx_lf_inprog lf_inprog;
247 
248 	lf_inprog.u = otx2_cpt_read64(lf->lfs->reg_base, BLKADDR_CPT0, lf->slot,
249 				      OTX2_CPT_LF_INPROG);
250 
251 	/* Set iqueue's execution */
252 	lf_inprog.s.eena = enable ? 0x1 : 0x0;
253 	otx2_cpt_write64(lf->lfs->reg_base, BLKADDR_CPT0, lf->slot,
254 			 OTX2_CPT_LF_INPROG, lf_inprog.u);
255 }
256 
257 static inline void otx2_cptlf_enable_iqueue_exec(struct otx2_cptlf_info *lf)
258 {
259 	otx2_cptlf_set_iqueue_exec(lf, true);
260 }
261 
262 static inline void otx2_cptlf_disable_iqueue_exec(struct otx2_cptlf_info *lf)
263 {
264 	otx2_cptlf_set_iqueue_exec(lf, false);
265 }
266 
267 static inline void otx2_cptlf_enable_iqueues(struct otx2_cptlfs_info *lfs)
268 {
269 	int slot;
270 
271 	for (slot = 0; slot < lfs->lfs_num; slot++) {
272 		otx2_cptlf_enable_iqueue_exec(&lfs->lf[slot]);
273 		otx2_cptlf_enable_iqueue_enq(&lfs->lf[slot]);
274 	}
275 }
276 
277 static inline void otx2_cpt_fill_inst(union otx2_cpt_inst_s *cptinst,
278 				      struct otx2_cpt_iq_command *iq_cmd,
279 				      u64 comp_baddr)
280 {
281 	cptinst->u[0] = 0x0;
282 	cptinst->s.doneint = true;
283 	cptinst->s.res_addr = comp_baddr;
284 	cptinst->u[2] = 0x0;
285 	cptinst->u[3] = 0x0;
286 	cptinst->s.ei0 = iq_cmd->cmd.u;
287 	cptinst->s.ei1 = iq_cmd->dptr;
288 	cptinst->s.ei2 = iq_cmd->rptr;
289 	cptinst->s.ei3 = iq_cmd->cptr.u;
290 }
291 
292 /*
293  * On OcteonTX2 platform the parameter insts_num is used as a count of
294  * instructions to be enqueued. The valid values for insts_num are:
295  * 1 - 1 CPT instruction will be enqueued during LMTST operation
296  * 2 - 2 CPT instructions will be enqueued during LMTST operation
297  */
298 static inline void otx2_cpt_send_cmd(union otx2_cpt_inst_s *cptinst,
299 				     u32 insts_num, struct otx2_cptlf_info *lf)
300 {
301 	void __iomem *lmtline = lf->lmtline;
302 	long ret;
303 
304 	/*
305 	 * Make sure memory areas pointed in CPT_INST_S
306 	 * are flushed before the instruction is sent to CPT
307 	 */
308 	dma_wmb();
309 
310 	do {
311 		/* Copy CPT command to LMTLINE */
312 		memcpy_toio(lmtline, cptinst, insts_num * OTX2_CPT_INST_SIZE);
313 
314 		/*
315 		 * LDEOR initiates atomic transfer to I/O device
316 		 * The following will cause the LMTST to fail (the LDEOR
317 		 * returns zero):
318 		 * - No stores have been performed to the LMTLINE since it was
319 		 * last invalidated.
320 		 * - The bytes which have been stored to LMTLINE since it was
321 		 * last invalidated form a pattern that is non-contiguous, does
322 		 * not start at byte 0, or does not end on a 8-byte boundary.
323 		 * (i.e.comprises a formation of other than 1–16 8-byte
324 		 * words.)
325 		 *
326 		 * These rules are designed such that an operating system
327 		 * context switch or hypervisor guest switch need have no
328 		 * knowledge of the LMTST operations; the switch code does not
329 		 * need to store to LMTCANCEL. Also note as LMTLINE data cannot
330 		 * be read, there is no information leakage between processes.
331 		 */
332 		ret = otx2_lmt_flush(lf->ioreg);
333 
334 	} while (!ret);
335 }
336 
337 int otx2_cptlf_init(struct otx2_cptlfs_info *lfs, u8 eng_grp_msk, int pri,
338 		    int lfs_num);
339 void otx2_cptlf_shutdown(struct otx2_cptlfs_info *lfs);
340 int otx2_cptlf_register_interrupts(struct otx2_cptlfs_info *lfs);
341 void otx2_cptlf_unregister_interrupts(struct otx2_cptlfs_info *lfs);
342 void otx2_cptlf_free_irqs_affinity(struct otx2_cptlfs_info *lfs);
343 int otx2_cptlf_set_irqs_affinity(struct otx2_cptlfs_info *lfs);
344 
345 #endif /* __OTX2_CPTLF_H */
346