xref: /linux/drivers/mailbox/bcm-pdc-mailbox.c (revision a1ff5a7d78a036d6c2178ee5acd6ba4946243800)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright 2016 Broadcom
4  */
5 
6 /*
7  * Broadcom PDC Mailbox Driver
8  * The PDC provides a ring based programming interface to one or more hardware
9  * offload engines. For example, the PDC driver works with both SPU-M and SPU2
10  * cryptographic offload hardware. In some chips the PDC is referred to as MDE,
11  * and in others the FA2/FA+ hardware is used with this PDC driver.
12  *
13  * The PDC driver registers with the Linux mailbox framework as a mailbox
14  * controller, once for each PDC instance. Ring 0 for each PDC is registered as
15  * a mailbox channel. The PDC driver uses interrupts to determine when data
16  * transfers to and from an offload engine are complete. The PDC driver uses
17  * threaded IRQs so that response messages are handled outside of interrupt
18  * context.
19  *
20  * The PDC driver allows multiple messages to be pending in the descriptor
21  * rings. The tx_msg_start descriptor index indicates where the last message
22  * starts. The txin_numd value at this index indicates how many descriptor
23  * indexes make up the message. Similar state is kept on the receive side. When
24  * an rx interrupt indicates a response is ready, the PDC driver processes numd
25  * descriptors from the tx and rx ring, thus processing one response at a time.
26  */
27 
28 #include <linux/errno.h>
29 #include <linux/module.h>
30 #include <linux/init.h>
31 #include <linux/slab.h>
32 #include <linux/debugfs.h>
33 #include <linux/interrupt.h>
34 #include <linux/wait.h>
35 #include <linux/platform_device.h>
36 #include <linux/property.h>
37 #include <linux/io.h>
38 #include <linux/of.h>
39 #include <linux/of_irq.h>
40 #include <linux/mailbox_controller.h>
41 #include <linux/mailbox/brcm-message.h>
42 #include <linux/scatterlist.h>
43 #include <linux/dma-direction.h>
44 #include <linux/dma-mapping.h>
45 #include <linux/dmapool.h>
46 #include <linux/workqueue.h>
47 
48 #define PDC_SUCCESS  0
49 
50 #define RING_ENTRY_SIZE   sizeof(struct dma64dd)
51 
52 /* # entries in PDC dma ring */
53 #define PDC_RING_ENTRIES  512
54 /*
55  * Minimum number of ring descriptor entries that must be free to tell mailbox
56  * framework that it can submit another request
57  */
58 #define PDC_RING_SPACE_MIN  15
59 
60 #define PDC_RING_SIZE    (PDC_RING_ENTRIES * RING_ENTRY_SIZE)
61 /* Rings are 8k aligned */
62 #define RING_ALIGN_ORDER  13
63 #define RING_ALIGN        BIT(RING_ALIGN_ORDER)
64 
65 #define RX_BUF_ALIGN_ORDER  5
66 #define RX_BUF_ALIGN	    BIT(RX_BUF_ALIGN_ORDER)
67 
68 /* descriptor bumping macros */
69 #define XXD(x, max_mask)              ((x) & (max_mask))
70 #define TXD(x, max_mask)              XXD((x), (max_mask))
71 #define RXD(x, max_mask)              XXD((x), (max_mask))
72 #define NEXTTXD(i, max_mask)          TXD((i) + 1, (max_mask))
73 #define PREVTXD(i, max_mask)          TXD((i) - 1, (max_mask))
74 #define NEXTRXD(i, max_mask)          RXD((i) + 1, (max_mask))
75 #define PREVRXD(i, max_mask)          RXD((i) - 1, (max_mask))
76 #define NTXDACTIVE(h, t, max_mask)    TXD((t) - (h), (max_mask))
77 #define NRXDACTIVE(h, t, max_mask)    RXD((t) - (h), (max_mask))
78 
79 /* Length of BCM header at start of SPU msg, in bytes */
80 #define BCM_HDR_LEN  8
81 
82 /*
83  * PDC driver reserves ringset 0 on each SPU for its own use. The driver does
84  * not currently support use of multiple ringsets on a single PDC engine.
85  */
86 #define PDC_RINGSET  0
87 
88 /*
89  * Interrupt mask and status definitions. Enable interrupts for tx and rx on
90  * ring 0
91  */
92 #define PDC_RCVINT_0         (16 + PDC_RINGSET)
93 #define PDC_RCVINTEN_0       BIT(PDC_RCVINT_0)
94 #define PDC_INTMASK	     (PDC_RCVINTEN_0)
95 #define PDC_LAZY_FRAMECOUNT  1
96 #define PDC_LAZY_TIMEOUT     10000
97 #define PDC_LAZY_INT  (PDC_LAZY_TIMEOUT | (PDC_LAZY_FRAMECOUNT << 24))
98 #define PDC_INTMASK_OFFSET   0x24
99 #define PDC_INTSTATUS_OFFSET 0x20
100 #define PDC_RCVLAZY0_OFFSET  (0x30 + 4 * PDC_RINGSET)
101 #define FA_RCVLAZY0_OFFSET   0x100
102 
103 /*
104  * For SPU2, configure MDE_CKSUM_CONTROL to write 17 bytes of metadata
105  * before frame
106  */
107 #define PDC_SPU2_RESP_HDR_LEN  17
108 #define PDC_CKSUM_CTRL         BIT(27)
109 #define PDC_CKSUM_CTRL_OFFSET  0x400
110 
111 #define PDC_SPUM_RESP_HDR_LEN  32
112 
113 /*
114  * Sets the following bits for write to transmit control reg:
115  * 11    - PtyChkDisable - parity check is disabled
116  * 20:18 - BurstLen = 3 -> 2^7 = 128 byte data reads from memory
117  */
118 #define PDC_TX_CTL		0x000C0800
119 
120 /* Bit in tx control reg to enable tx channel */
121 #define PDC_TX_ENABLE		0x1
122 
123 /*
124  * Sets the following bits for write to receive control reg:
125  * 7:1   - RcvOffset - size in bytes of status region at start of rx frame buf
126  * 9     - SepRxHdrDescEn - place start of new frames only in descriptors
127  *                          that have StartOfFrame set
128  * 10    - OflowContinue - on rx FIFO overflow, clear rx fifo, discard all
129  *                         remaining bytes in current frame, report error
130  *                         in rx frame status for current frame
131  * 11    - PtyChkDisable - parity check is disabled
132  * 20:18 - BurstLen = 3 -> 2^7 = 128 byte data reads from memory
133  */
134 #define PDC_RX_CTL		0x000C0E00
135 
136 /* Bit in rx control reg to enable rx channel */
137 #define PDC_RX_ENABLE		0x1
138 
139 #define CRYPTO_D64_RS0_CD_MASK   ((PDC_RING_ENTRIES * RING_ENTRY_SIZE) - 1)
140 
141 /* descriptor flags */
142 #define D64_CTRL1_EOT   BIT(28)	/* end of descriptor table */
143 #define D64_CTRL1_IOC   BIT(29)	/* interrupt on complete */
144 #define D64_CTRL1_EOF   BIT(30)	/* end of frame */
145 #define D64_CTRL1_SOF   BIT(31)	/* start of frame */
146 
147 #define RX_STATUS_OVERFLOW       0x00800000
148 #define RX_STATUS_LEN            0x0000FFFF
149 
150 #define PDC_TXREGS_OFFSET  0x200
151 #define PDC_RXREGS_OFFSET  0x220
152 
153 /* Maximum size buffer the DMA engine can handle */
154 #define PDC_DMA_BUF_MAX 16384
155 
156 enum pdc_hw {
157 	FA_HW,		/* FA2/FA+ hardware (i.e. Northstar Plus) */
158 	PDC_HW		/* PDC/MDE hardware (i.e. Northstar 2, Pegasus) */
159 };
160 
161 /* dma descriptor */
162 struct dma64dd {
163 	u32 ctrl1;      /* misc control bits */
164 	u32 ctrl2;      /* buffer count and address extension */
165 	u32 addrlow;    /* memory address of the date buffer, bits 31:0 */
166 	u32 addrhigh;   /* memory address of the date buffer, bits 63:32 */
167 };
168 
169 /* dma registers per channel(xmt or rcv) */
170 struct dma64_regs {
171 	u32  control;   /* enable, et al */
172 	u32  ptr;       /* last descriptor posted to chip */
173 	u32  addrlow;   /* descriptor ring base address low 32-bits */
174 	u32  addrhigh;  /* descriptor ring base address bits 63:32 */
175 	u32  status0;   /* last rx descriptor written by hw */
176 	u32  status1;   /* driver does not use */
177 };
178 
179 /* cpp contortions to concatenate w/arg prescan */
180 #ifndef PAD
181 #define _PADLINE(line)  pad ## line
182 #define _XSTR(line)     _PADLINE(line)
183 #define PAD             _XSTR(__LINE__)
184 #endif  /* PAD */
185 
186 /* dma registers. matches hw layout. */
187 struct dma64 {
188 	struct dma64_regs dmaxmt;  /* dma tx */
189 	u32          PAD[2];
190 	struct dma64_regs dmarcv;  /* dma rx */
191 	u32          PAD[2];
192 };
193 
194 /* PDC registers */
195 struct pdc_regs {
196 	u32  devcontrol;             /* 0x000 */
197 	u32  devstatus;              /* 0x004 */
198 	u32  PAD;
199 	u32  biststatus;             /* 0x00c */
200 	u32  PAD[4];
201 	u32  intstatus;              /* 0x020 */
202 	u32  intmask;                /* 0x024 */
203 	u32  gptimer;                /* 0x028 */
204 
205 	u32  PAD;
206 	u32  intrcvlazy_0;           /* 0x030 (Only in PDC, not FA2) */
207 	u32  intrcvlazy_1;           /* 0x034 (Only in PDC, not FA2) */
208 	u32  intrcvlazy_2;           /* 0x038 (Only in PDC, not FA2) */
209 	u32  intrcvlazy_3;           /* 0x03c (Only in PDC, not FA2) */
210 
211 	u32  PAD[48];
212 	u32  fa_intrecvlazy;         /* 0x100 (Only in FA2, not PDC) */
213 	u32  flowctlthresh;          /* 0x104 */
214 	u32  wrrthresh;              /* 0x108 */
215 	u32  gmac_idle_cnt_thresh;   /* 0x10c */
216 
217 	u32  PAD[4];
218 	u32  ifioaccessaddr;         /* 0x120 */
219 	u32  ifioaccessbyte;         /* 0x124 */
220 	u32  ifioaccessdata;         /* 0x128 */
221 
222 	u32  PAD[21];
223 	u32  phyaccess;              /* 0x180 */
224 	u32  PAD;
225 	u32  phycontrol;             /* 0x188 */
226 	u32  txqctl;                 /* 0x18c */
227 	u32  rxqctl;                 /* 0x190 */
228 	u32  gpioselect;             /* 0x194 */
229 	u32  gpio_output_en;         /* 0x198 */
230 	u32  PAD;                    /* 0x19c */
231 	u32  txq_rxq_mem_ctl;        /* 0x1a0 */
232 	u32  memory_ecc_status;      /* 0x1a4 */
233 	u32  serdes_ctl;             /* 0x1a8 */
234 	u32  serdes_status0;         /* 0x1ac */
235 	u32  serdes_status1;         /* 0x1b0 */
236 	u32  PAD[11];                /* 0x1b4-1dc */
237 	u32  clk_ctl_st;             /* 0x1e0 */
238 	u32  hw_war;                 /* 0x1e4 (Only in PDC, not FA2) */
239 	u32  pwrctl;                 /* 0x1e8 */
240 	u32  PAD[5];
241 
242 #define PDC_NUM_DMA_RINGS   4
243 	struct dma64 dmaregs[PDC_NUM_DMA_RINGS];  /* 0x0200 - 0x2fc */
244 
245 	/* more registers follow, but we don't use them */
246 };
247 
248 /* structure for allocating/freeing DMA rings */
249 struct pdc_ring_alloc {
250 	dma_addr_t  dmabase; /* DMA address of start of ring */
251 	void	   *vbase;   /* base kernel virtual address of ring */
252 	u32	    size;    /* ring allocation size in bytes */
253 };
254 
255 /*
256  * context associated with a receive descriptor.
257  * @rxp_ctx: opaque context associated with frame that starts at each
258  *           rx ring index.
259  * @dst_sg:  Scatterlist used to form reply frames beginning at a given ring
260  *           index. Retained in order to unmap each sg after reply is processed.
261  * @rxin_numd: Number of rx descriptors associated with the message that starts
262  *             at a descriptor index. Not set for every index. For example,
263  *             if descriptor index i points to a scatterlist with 4 entries,
264  *             then the next three descriptor indexes don't have a value set.
265  * @resp_hdr: Virtual address of buffer used to catch DMA rx status
266  * @resp_hdr_daddr: physical address of DMA rx status buffer
267  */
268 struct pdc_rx_ctx {
269 	void *rxp_ctx;
270 	struct scatterlist *dst_sg;
271 	u32  rxin_numd;
272 	void *resp_hdr;
273 	dma_addr_t resp_hdr_daddr;
274 };
275 
276 /* PDC state structure */
277 struct pdc_state {
278 	/* Index of the PDC whose state is in this structure instance */
279 	u8 pdc_idx;
280 
281 	/* Platform device for this PDC instance */
282 	struct platform_device *pdev;
283 
284 	/*
285 	 * Each PDC instance has a mailbox controller. PDC receives request
286 	 * messages through mailboxes, and sends response messages through the
287 	 * mailbox framework.
288 	 */
289 	struct mbox_controller mbc;
290 
291 	unsigned int pdc_irq;
292 
293 	/* work for deferred processing after DMA rx interrupt */
294 	struct work_struct rx_work;
295 
296 	/* Number of bytes of receive status prior to each rx frame */
297 	u32 rx_status_len;
298 	/* Whether a BCM header is prepended to each frame */
299 	bool use_bcm_hdr;
300 	/* Sum of length of BCM header and rx status header */
301 	u32 pdc_resp_hdr_len;
302 
303 	/* The base virtual address of DMA hw registers */
304 	void __iomem *pdc_reg_vbase;
305 
306 	/* Pool for allocation of DMA rings */
307 	struct dma_pool *ring_pool;
308 
309 	/* Pool for allocation of metadata buffers for response messages */
310 	struct dma_pool *rx_buf_pool;
311 
312 	/*
313 	 * The base virtual address of DMA tx/rx descriptor rings. Corresponding
314 	 * DMA address and size of ring allocation.
315 	 */
316 	struct pdc_ring_alloc tx_ring_alloc;
317 	struct pdc_ring_alloc rx_ring_alloc;
318 
319 	struct pdc_regs *regs;    /* start of PDC registers */
320 
321 	struct dma64_regs *txregs_64; /* dma tx engine registers */
322 	struct dma64_regs *rxregs_64; /* dma rx engine registers */
323 
324 	/*
325 	 * Arrays of PDC_RING_ENTRIES descriptors
326 	 * To use multiple ringsets, this needs to be extended
327 	 */
328 	struct dma64dd   *txd_64;  /* tx descriptor ring */
329 	struct dma64dd   *rxd_64;  /* rx descriptor ring */
330 
331 	/* descriptor ring sizes */
332 	u32      ntxd;       /* # tx descriptors */
333 	u32      nrxd;       /* # rx descriptors */
334 	u32      nrxpost;    /* # rx buffers to keep posted */
335 	u32      ntxpost;    /* max number of tx buffers that can be posted */
336 
337 	/*
338 	 * Index of next tx descriptor to reclaim. That is, the descriptor
339 	 * index of the oldest tx buffer for which the host has yet to process
340 	 * the corresponding response.
341 	 */
342 	u32  txin;
343 
344 	/*
345 	 * Index of the first receive descriptor for the sequence of
346 	 * message fragments currently under construction. Used to build up
347 	 * the rxin_numd count for a message. Updated to rxout when the host
348 	 * starts a new sequence of rx buffers for a new message.
349 	 */
350 	u32  tx_msg_start;
351 
352 	/* Index of next tx descriptor to post. */
353 	u32  txout;
354 
355 	/*
356 	 * Number of tx descriptors associated with the message that starts
357 	 * at this tx descriptor index.
358 	 */
359 	u32      txin_numd[PDC_RING_ENTRIES];
360 
361 	/*
362 	 * Index of next rx descriptor to reclaim. This is the index of
363 	 * the next descriptor whose data has yet to be processed by the host.
364 	 */
365 	u32  rxin;
366 
367 	/*
368 	 * Index of the first receive descriptor for the sequence of
369 	 * message fragments currently under construction. Used to build up
370 	 * the rxin_numd count for a message. Updated to rxout when the host
371 	 * starts a new sequence of rx buffers for a new message.
372 	 */
373 	u32  rx_msg_start;
374 
375 	/*
376 	 * Saved value of current hardware rx descriptor index.
377 	 * The last rx buffer written by the hw is the index previous to
378 	 * this one.
379 	 */
380 	u32  last_rx_curr;
381 
382 	/* Index of next rx descriptor to post. */
383 	u32  rxout;
384 
385 	struct pdc_rx_ctx rx_ctx[PDC_RING_ENTRIES];
386 
387 	/*
388 	 * Scatterlists used to form request and reply frames beginning at a
389 	 * given ring index. Retained in order to unmap each sg after reply
390 	 * is processed
391 	 */
392 	struct scatterlist *src_sg[PDC_RING_ENTRIES];
393 
394 	/* counters */
395 	u32  pdc_requests;     /* number of request messages submitted */
396 	u32  pdc_replies;      /* number of reply messages received */
397 	u32  last_tx_not_done; /* too few tx descriptors to indicate done */
398 	u32  tx_ring_full;     /* unable to accept msg because tx ring full */
399 	u32  rx_ring_full;     /* unable to accept msg because rx ring full */
400 	u32  txnobuf;          /* unable to create tx descriptor */
401 	u32  rxnobuf;          /* unable to create rx descriptor */
402 	u32  rx_oflow;         /* count of rx overflows */
403 
404 	/* hardware type - FA2 or PDC/MDE */
405 	enum pdc_hw hw_type;
406 };
407 
408 /* Global variables */
409 
410 struct pdc_globals {
411 	/* Actual number of SPUs in hardware, as reported by device tree */
412 	u32 num_spu;
413 };
414 
415 static struct pdc_globals pdcg;
416 
417 /* top level debug FS directory for PDC driver */
418 static struct dentry *debugfs_dir;
419 
pdc_debugfs_read(struct file * filp,char __user * ubuf,size_t count,loff_t * offp)420 static ssize_t pdc_debugfs_read(struct file *filp, char __user *ubuf,
421 				size_t count, loff_t *offp)
422 {
423 	struct pdc_state *pdcs;
424 	char *buf;
425 	ssize_t ret, out_offset, out_count;
426 
427 	out_count = 512;
428 
429 	buf = kmalloc(out_count, GFP_KERNEL);
430 	if (!buf)
431 		return -ENOMEM;
432 
433 	pdcs = filp->private_data;
434 	out_offset = 0;
435 	out_offset += scnprintf(buf + out_offset, out_count - out_offset,
436 			       "SPU %u stats:\n", pdcs->pdc_idx);
437 	out_offset += scnprintf(buf + out_offset, out_count - out_offset,
438 			       "PDC requests....................%u\n",
439 			       pdcs->pdc_requests);
440 	out_offset += scnprintf(buf + out_offset, out_count - out_offset,
441 			       "PDC responses...................%u\n",
442 			       pdcs->pdc_replies);
443 	out_offset += scnprintf(buf + out_offset, out_count - out_offset,
444 			       "Tx not done.....................%u\n",
445 			       pdcs->last_tx_not_done);
446 	out_offset += scnprintf(buf + out_offset, out_count - out_offset,
447 			       "Tx ring full....................%u\n",
448 			       pdcs->tx_ring_full);
449 	out_offset += scnprintf(buf + out_offset, out_count - out_offset,
450 			       "Rx ring full....................%u\n",
451 			       pdcs->rx_ring_full);
452 	out_offset += scnprintf(buf + out_offset, out_count - out_offset,
453 			       "Tx desc write fail. Ring full...%u\n",
454 			       pdcs->txnobuf);
455 	out_offset += scnprintf(buf + out_offset, out_count - out_offset,
456 			       "Rx desc write fail. Ring full...%u\n",
457 			       pdcs->rxnobuf);
458 	out_offset += scnprintf(buf + out_offset, out_count - out_offset,
459 			       "Receive overflow................%u\n",
460 			       pdcs->rx_oflow);
461 	out_offset += scnprintf(buf + out_offset, out_count - out_offset,
462 			       "Num frags in rx ring............%u\n",
463 			       NRXDACTIVE(pdcs->rxin, pdcs->last_rx_curr,
464 					  pdcs->nrxpost));
465 
466 	if (out_offset > out_count)
467 		out_offset = out_count;
468 
469 	ret = simple_read_from_buffer(ubuf, count, offp, buf, out_offset);
470 	kfree(buf);
471 	return ret;
472 }
473 
474 static const struct file_operations pdc_debugfs_stats = {
475 	.owner = THIS_MODULE,
476 	.open = simple_open,
477 	.read = pdc_debugfs_read,
478 };
479 
480 /**
481  * pdc_setup_debugfs() - Create the debug FS directories. If the top-level
482  * directory has not yet been created, create it now. Create a stats file in
483  * this directory for a SPU.
484  * @pdcs: PDC state structure
485  */
pdc_setup_debugfs(struct pdc_state * pdcs)486 static void pdc_setup_debugfs(struct pdc_state *pdcs)
487 {
488 	char spu_stats_name[16];
489 
490 	if (!debugfs_initialized())
491 		return;
492 
493 	snprintf(spu_stats_name, 16, "pdc%d_stats", pdcs->pdc_idx);
494 	if (!debugfs_dir)
495 		debugfs_dir = debugfs_create_dir(KBUILD_MODNAME, NULL);
496 
497 	/* S_IRUSR == 0400 */
498 	debugfs_create_file(spu_stats_name, 0400, debugfs_dir, pdcs,
499 			    &pdc_debugfs_stats);
500 }
501 
pdc_free_debugfs(void)502 static void pdc_free_debugfs(void)
503 {
504 	debugfs_remove_recursive(debugfs_dir);
505 	debugfs_dir = NULL;
506 }
507 
508 /**
509  * pdc_build_rxd() - Build DMA descriptor to receive SPU result.
510  * @pdcs:      PDC state for SPU that will generate result
511  * @dma_addr:  DMA address of buffer that descriptor is being built for
512  * @buf_len:   Length of the receive buffer, in bytes
513  * @flags:     Flags to be stored in descriptor
514  */
515 static inline void
pdc_build_rxd(struct pdc_state * pdcs,dma_addr_t dma_addr,u32 buf_len,u32 flags)516 pdc_build_rxd(struct pdc_state *pdcs, dma_addr_t dma_addr,
517 	      u32 buf_len, u32 flags)
518 {
519 	struct device *dev = &pdcs->pdev->dev;
520 	struct dma64dd *rxd = &pdcs->rxd_64[pdcs->rxout];
521 
522 	dev_dbg(dev,
523 		"Writing rx descriptor for PDC %u at index %u with length %u. flags %#x\n",
524 		pdcs->pdc_idx, pdcs->rxout, buf_len, flags);
525 
526 	rxd->addrlow = cpu_to_le32(lower_32_bits(dma_addr));
527 	rxd->addrhigh = cpu_to_le32(upper_32_bits(dma_addr));
528 	rxd->ctrl1 = cpu_to_le32(flags);
529 	rxd->ctrl2 = cpu_to_le32(buf_len);
530 
531 	/* bump ring index and return */
532 	pdcs->rxout = NEXTRXD(pdcs->rxout, pdcs->nrxpost);
533 }
534 
535 /**
536  * pdc_build_txd() - Build a DMA descriptor to transmit a SPU request to
537  * hardware.
538  * @pdcs:        PDC state for the SPU that will process this request
539  * @dma_addr:    DMA address of packet to be transmitted
540  * @buf_len:     Length of tx buffer, in bytes
541  * @flags:       Flags to be stored in descriptor
542  */
543 static inline void
pdc_build_txd(struct pdc_state * pdcs,dma_addr_t dma_addr,u32 buf_len,u32 flags)544 pdc_build_txd(struct pdc_state *pdcs, dma_addr_t dma_addr, u32 buf_len,
545 	      u32 flags)
546 {
547 	struct device *dev = &pdcs->pdev->dev;
548 	struct dma64dd *txd = &pdcs->txd_64[pdcs->txout];
549 
550 	dev_dbg(dev,
551 		"Writing tx descriptor for PDC %u at index %u with length %u, flags %#x\n",
552 		pdcs->pdc_idx, pdcs->txout, buf_len, flags);
553 
554 	txd->addrlow = cpu_to_le32(lower_32_bits(dma_addr));
555 	txd->addrhigh = cpu_to_le32(upper_32_bits(dma_addr));
556 	txd->ctrl1 = cpu_to_le32(flags);
557 	txd->ctrl2 = cpu_to_le32(buf_len);
558 
559 	/* bump ring index and return */
560 	pdcs->txout = NEXTTXD(pdcs->txout, pdcs->ntxpost);
561 }
562 
563 /**
564  * pdc_receive_one() - Receive a response message from a given SPU.
565  * @pdcs:    PDC state for the SPU to receive from
566  *
567  * When the return code indicates success, the response message is available in
568  * the receive buffers provided prior to submission of the request.
569  *
570  * Return:  PDC_SUCCESS if one or more receive descriptors was processed
571  *          -EAGAIN indicates that no response message is available
572  *          -EIO an error occurred
573  */
574 static int
pdc_receive_one(struct pdc_state * pdcs)575 pdc_receive_one(struct pdc_state *pdcs)
576 {
577 	struct device *dev = &pdcs->pdev->dev;
578 	struct mbox_controller *mbc;
579 	struct mbox_chan *chan;
580 	struct brcm_message mssg;
581 	u32 len, rx_status;
582 	u32 num_frags;
583 	u8 *resp_hdr;    /* virtual addr of start of resp message DMA header */
584 	u32 frags_rdy;   /* number of fragments ready to read */
585 	u32 rx_idx;      /* ring index of start of receive frame */
586 	dma_addr_t resp_hdr_daddr;
587 	struct pdc_rx_ctx *rx_ctx;
588 
589 	mbc = &pdcs->mbc;
590 	chan = &mbc->chans[0];
591 	mssg.type = BRCM_MESSAGE_SPU;
592 
593 	/*
594 	 * return if a complete response message is not yet ready.
595 	 * rxin_numd[rxin] is the number of fragments in the next msg
596 	 * to read.
597 	 */
598 	frags_rdy = NRXDACTIVE(pdcs->rxin, pdcs->last_rx_curr, pdcs->nrxpost);
599 	if ((frags_rdy == 0) ||
600 	    (frags_rdy < pdcs->rx_ctx[pdcs->rxin].rxin_numd))
601 		/* No response ready */
602 		return -EAGAIN;
603 
604 	num_frags = pdcs->txin_numd[pdcs->txin];
605 	WARN_ON(num_frags == 0);
606 
607 	dma_unmap_sg(dev, pdcs->src_sg[pdcs->txin],
608 		     sg_nents(pdcs->src_sg[pdcs->txin]), DMA_TO_DEVICE);
609 
610 	pdcs->txin = (pdcs->txin + num_frags) & pdcs->ntxpost;
611 
612 	dev_dbg(dev, "PDC %u reclaimed %d tx descriptors",
613 		pdcs->pdc_idx, num_frags);
614 
615 	rx_idx = pdcs->rxin;
616 	rx_ctx = &pdcs->rx_ctx[rx_idx];
617 	num_frags = rx_ctx->rxin_numd;
618 	/* Return opaque context with result */
619 	mssg.ctx = rx_ctx->rxp_ctx;
620 	rx_ctx->rxp_ctx = NULL;
621 	resp_hdr = rx_ctx->resp_hdr;
622 	resp_hdr_daddr = rx_ctx->resp_hdr_daddr;
623 	dma_unmap_sg(dev, rx_ctx->dst_sg, sg_nents(rx_ctx->dst_sg),
624 		     DMA_FROM_DEVICE);
625 
626 	pdcs->rxin = (pdcs->rxin + num_frags) & pdcs->nrxpost;
627 
628 	dev_dbg(dev, "PDC %u reclaimed %d rx descriptors",
629 		pdcs->pdc_idx, num_frags);
630 
631 	dev_dbg(dev,
632 		"PDC %u txin %u, txout %u, rxin %u, rxout %u, last_rx_curr %u\n",
633 		pdcs->pdc_idx, pdcs->txin, pdcs->txout, pdcs->rxin,
634 		pdcs->rxout, pdcs->last_rx_curr);
635 
636 	if (pdcs->pdc_resp_hdr_len == PDC_SPUM_RESP_HDR_LEN) {
637 		/*
638 		 * For SPU-M, get length of response msg and rx overflow status.
639 		 */
640 		rx_status = *((u32 *)resp_hdr);
641 		len = rx_status & RX_STATUS_LEN;
642 		dev_dbg(dev,
643 			"SPU response length %u bytes", len);
644 		if (unlikely(((rx_status & RX_STATUS_OVERFLOW) || (!len)))) {
645 			if (rx_status & RX_STATUS_OVERFLOW) {
646 				dev_err_ratelimited(dev,
647 						    "crypto receive overflow");
648 				pdcs->rx_oflow++;
649 			} else {
650 				dev_info_ratelimited(dev, "crypto rx len = 0");
651 			}
652 			return -EIO;
653 		}
654 	}
655 
656 	dma_pool_free(pdcs->rx_buf_pool, resp_hdr, resp_hdr_daddr);
657 
658 	mbox_chan_received_data(chan, &mssg);
659 
660 	pdcs->pdc_replies++;
661 	return PDC_SUCCESS;
662 }
663 
664 /**
665  * pdc_receive() - Process as many responses as are available in the rx ring.
666  * @pdcs:  PDC state
667  *
668  * Called within the hard IRQ.
669  * Return:
670  */
671 static int
pdc_receive(struct pdc_state * pdcs)672 pdc_receive(struct pdc_state *pdcs)
673 {
674 	int rx_status;
675 
676 	/* read last_rx_curr from register once */
677 	pdcs->last_rx_curr =
678 	    (ioread32((const void __iomem *)&pdcs->rxregs_64->status0) &
679 	     CRYPTO_D64_RS0_CD_MASK) / RING_ENTRY_SIZE;
680 
681 	do {
682 		/* Could be many frames ready */
683 		rx_status = pdc_receive_one(pdcs);
684 	} while (rx_status == PDC_SUCCESS);
685 
686 	return 0;
687 }
688 
689 /**
690  * pdc_tx_list_sg_add() - Add the buffers in a scatterlist to the transmit
691  * descriptors for a given SPU. The scatterlist buffers contain the data for a
692  * SPU request message.
693  * @pdcs:      PDC state for the SPU that will process this request
694  * @sg:        Scatterlist whose buffers contain part of the SPU request
695  *
696  * If a scatterlist buffer is larger than PDC_DMA_BUF_MAX, multiple descriptors
697  * are written for that buffer, each <= PDC_DMA_BUF_MAX byte in length.
698  *
699  * Return: PDC_SUCCESS if successful
700  *         < 0 otherwise
701  */
pdc_tx_list_sg_add(struct pdc_state * pdcs,struct scatterlist * sg)702 static int pdc_tx_list_sg_add(struct pdc_state *pdcs, struct scatterlist *sg)
703 {
704 	u32 flags = 0;
705 	u32 eot;
706 	u32 tx_avail;
707 
708 	/*
709 	 * Num descriptors needed. Conservatively assume we need a descriptor
710 	 * for every entry in sg.
711 	 */
712 	u32 num_desc;
713 	u32 desc_w = 0;	/* Number of tx descriptors written */
714 	u32 bufcnt;	/* Number of bytes of buffer pointed to by descriptor */
715 	dma_addr_t databufptr;	/* DMA address to put in descriptor */
716 
717 	num_desc = (u32)sg_nents(sg);
718 
719 	/* check whether enough tx descriptors are available */
720 	tx_avail = pdcs->ntxpost - NTXDACTIVE(pdcs->txin, pdcs->txout,
721 					      pdcs->ntxpost);
722 	if (unlikely(num_desc > tx_avail)) {
723 		pdcs->txnobuf++;
724 		return -ENOSPC;
725 	}
726 
727 	/* build tx descriptors */
728 	if (pdcs->tx_msg_start == pdcs->txout) {
729 		/* Start of frame */
730 		pdcs->txin_numd[pdcs->tx_msg_start] = 0;
731 		pdcs->src_sg[pdcs->txout] = sg;
732 		flags = D64_CTRL1_SOF;
733 	}
734 
735 	while (sg) {
736 		if (unlikely(pdcs->txout == (pdcs->ntxd - 1)))
737 			eot = D64_CTRL1_EOT;
738 		else
739 			eot = 0;
740 
741 		/*
742 		 * If sg buffer larger than PDC limit, split across
743 		 * multiple descriptors
744 		 */
745 		bufcnt = sg_dma_len(sg);
746 		databufptr = sg_dma_address(sg);
747 		while (bufcnt > PDC_DMA_BUF_MAX) {
748 			pdc_build_txd(pdcs, databufptr, PDC_DMA_BUF_MAX,
749 				      flags | eot);
750 			desc_w++;
751 			bufcnt -= PDC_DMA_BUF_MAX;
752 			databufptr += PDC_DMA_BUF_MAX;
753 			if (unlikely(pdcs->txout == (pdcs->ntxd - 1)))
754 				eot = D64_CTRL1_EOT;
755 			else
756 				eot = 0;
757 		}
758 		sg = sg_next(sg);
759 		if (!sg)
760 			/* Writing last descriptor for frame */
761 			flags |= (D64_CTRL1_EOF | D64_CTRL1_IOC);
762 		pdc_build_txd(pdcs, databufptr, bufcnt, flags | eot);
763 		desc_w++;
764 		/* Clear start of frame after first descriptor */
765 		flags &= ~D64_CTRL1_SOF;
766 	}
767 	pdcs->txin_numd[pdcs->tx_msg_start] += desc_w;
768 
769 	return PDC_SUCCESS;
770 }
771 
772 /**
773  * pdc_tx_list_final() - Initiate DMA transfer of last frame written to tx
774  * ring.
775  * @pdcs:  PDC state for SPU to process the request
776  *
777  * Sets the index of the last descriptor written in both the rx and tx ring.
778  *
779  * Return: PDC_SUCCESS
780  */
pdc_tx_list_final(struct pdc_state * pdcs)781 static int pdc_tx_list_final(struct pdc_state *pdcs)
782 {
783 	/*
784 	 * write barrier to ensure all register writes are complete
785 	 * before chip starts to process new request
786 	 */
787 	wmb();
788 	iowrite32(pdcs->rxout << 4, &pdcs->rxregs_64->ptr);
789 	iowrite32(pdcs->txout << 4, &pdcs->txregs_64->ptr);
790 	pdcs->pdc_requests++;
791 
792 	return PDC_SUCCESS;
793 }
794 
795 /**
796  * pdc_rx_list_init() - Start a new receive descriptor list for a given PDC.
797  * @pdcs:   PDC state for SPU handling request
798  * @dst_sg: scatterlist providing rx buffers for response to be returned to
799  *	    mailbox client
800  * @ctx:    Opaque context for this request
801  *
802  * Posts a single receive descriptor to hold the metadata that precedes a
803  * response. For example, with SPU-M, the metadata is a 32-byte DMA header and
804  * an 8-byte BCM header. Moves the msg_start descriptor indexes for both tx and
805  * rx to indicate the start of a new message.
806  *
807  * Return:  PDC_SUCCESS if successful
808  *          < 0 if an error (e.g., rx ring is full)
809  */
pdc_rx_list_init(struct pdc_state * pdcs,struct scatterlist * dst_sg,void * ctx)810 static int pdc_rx_list_init(struct pdc_state *pdcs, struct scatterlist *dst_sg,
811 			    void *ctx)
812 {
813 	u32 flags = 0;
814 	u32 rx_avail;
815 	u32 rx_pkt_cnt = 1;	/* Adding a single rx buffer */
816 	dma_addr_t daddr;
817 	void *vaddr;
818 	struct pdc_rx_ctx *rx_ctx;
819 
820 	rx_avail = pdcs->nrxpost - NRXDACTIVE(pdcs->rxin, pdcs->rxout,
821 					      pdcs->nrxpost);
822 	if (unlikely(rx_pkt_cnt > rx_avail)) {
823 		pdcs->rxnobuf++;
824 		return -ENOSPC;
825 	}
826 
827 	/* allocate a buffer for the dma rx status */
828 	vaddr = dma_pool_zalloc(pdcs->rx_buf_pool, GFP_ATOMIC, &daddr);
829 	if (unlikely(!vaddr))
830 		return -ENOMEM;
831 
832 	/*
833 	 * Update msg_start indexes for both tx and rx to indicate the start
834 	 * of a new sequence of descriptor indexes that contain the fragments
835 	 * of the same message.
836 	 */
837 	pdcs->rx_msg_start = pdcs->rxout;
838 	pdcs->tx_msg_start = pdcs->txout;
839 
840 	/* This is always the first descriptor in the receive sequence */
841 	flags = D64_CTRL1_SOF;
842 	pdcs->rx_ctx[pdcs->rx_msg_start].rxin_numd = 1;
843 
844 	if (unlikely(pdcs->rxout == (pdcs->nrxd - 1)))
845 		flags |= D64_CTRL1_EOT;
846 
847 	rx_ctx = &pdcs->rx_ctx[pdcs->rxout];
848 	rx_ctx->rxp_ctx = ctx;
849 	rx_ctx->dst_sg = dst_sg;
850 	rx_ctx->resp_hdr = vaddr;
851 	rx_ctx->resp_hdr_daddr = daddr;
852 	pdc_build_rxd(pdcs, daddr, pdcs->pdc_resp_hdr_len, flags);
853 	return PDC_SUCCESS;
854 }
855 
856 /**
857  * pdc_rx_list_sg_add() - Add the buffers in a scatterlist to the receive
858  * descriptors for a given SPU. The caller must have already DMA mapped the
859  * scatterlist.
860  * @pdcs:       PDC state for the SPU that will process this request
861  * @sg:         Scatterlist whose buffers are added to the receive ring
862  *
863  * If a receive buffer in the scatterlist is larger than PDC_DMA_BUF_MAX,
864  * multiple receive descriptors are written, each with a buffer <=
865  * PDC_DMA_BUF_MAX.
866  *
867  * Return: PDC_SUCCESS if successful
868  *         < 0 otherwise (e.g., receive ring is full)
869  */
pdc_rx_list_sg_add(struct pdc_state * pdcs,struct scatterlist * sg)870 static int pdc_rx_list_sg_add(struct pdc_state *pdcs, struct scatterlist *sg)
871 {
872 	u32 flags = 0;
873 	u32 rx_avail;
874 
875 	/*
876 	 * Num descriptors needed. Conservatively assume we need a descriptor
877 	 * for every entry from our starting point in the scatterlist.
878 	 */
879 	u32 num_desc;
880 	u32 desc_w = 0;	/* Number of tx descriptors written */
881 	u32 bufcnt;	/* Number of bytes of buffer pointed to by descriptor */
882 	dma_addr_t databufptr;	/* DMA address to put in descriptor */
883 
884 	num_desc = (u32)sg_nents(sg);
885 
886 	rx_avail = pdcs->nrxpost - NRXDACTIVE(pdcs->rxin, pdcs->rxout,
887 					      pdcs->nrxpost);
888 	if (unlikely(num_desc > rx_avail)) {
889 		pdcs->rxnobuf++;
890 		return -ENOSPC;
891 	}
892 
893 	while (sg) {
894 		if (unlikely(pdcs->rxout == (pdcs->nrxd - 1)))
895 			flags = D64_CTRL1_EOT;
896 		else
897 			flags = 0;
898 
899 		/*
900 		 * If sg buffer larger than PDC limit, split across
901 		 * multiple descriptors
902 		 */
903 		bufcnt = sg_dma_len(sg);
904 		databufptr = sg_dma_address(sg);
905 		while (bufcnt > PDC_DMA_BUF_MAX) {
906 			pdc_build_rxd(pdcs, databufptr, PDC_DMA_BUF_MAX, flags);
907 			desc_w++;
908 			bufcnt -= PDC_DMA_BUF_MAX;
909 			databufptr += PDC_DMA_BUF_MAX;
910 			if (unlikely(pdcs->rxout == (pdcs->nrxd - 1)))
911 				flags = D64_CTRL1_EOT;
912 			else
913 				flags = 0;
914 		}
915 		pdc_build_rxd(pdcs, databufptr, bufcnt, flags);
916 		desc_w++;
917 		sg = sg_next(sg);
918 	}
919 	pdcs->rx_ctx[pdcs->rx_msg_start].rxin_numd += desc_w;
920 
921 	return PDC_SUCCESS;
922 }
923 
924 /**
925  * pdc_irq_handler() - Interrupt handler called in interrupt context.
926  * @irq:      Interrupt number that has fired
927  * @data:     device struct for DMA engine that generated the interrupt
928  *
929  * We have to clear the device interrupt status flags here. So cache the
930  * status for later use in the thread function. Other than that, just return
931  * WAKE_THREAD to invoke the thread function.
932  *
933  * Return: IRQ_WAKE_THREAD if interrupt is ours
934  *         IRQ_NONE otherwise
935  */
pdc_irq_handler(int irq,void * data)936 static irqreturn_t pdc_irq_handler(int irq, void *data)
937 {
938 	struct device *dev = (struct device *)data;
939 	struct pdc_state *pdcs = dev_get_drvdata(dev);
940 	u32 intstatus = ioread32(pdcs->pdc_reg_vbase + PDC_INTSTATUS_OFFSET);
941 
942 	if (unlikely(intstatus == 0))
943 		return IRQ_NONE;
944 
945 	/* Disable interrupts until soft handler runs */
946 	iowrite32(0, pdcs->pdc_reg_vbase + PDC_INTMASK_OFFSET);
947 
948 	/* Clear interrupt flags in device */
949 	iowrite32(intstatus, pdcs->pdc_reg_vbase + PDC_INTSTATUS_OFFSET);
950 
951 	/* Wakeup IRQ thread */
952 	queue_work(system_bh_wq, &pdcs->rx_work);
953 	return IRQ_HANDLED;
954 }
955 
956 /**
957  * pdc_work_cb() - Work callback that runs the deferred processing after
958  * a DMA receive interrupt. Reenables the receive interrupt.
959  * @t: Pointer to the Altera sSGDMA channel structure
960  */
pdc_work_cb(struct work_struct * t)961 static void pdc_work_cb(struct work_struct *t)
962 {
963 	struct pdc_state *pdcs = from_work(pdcs, t, rx_work);
964 
965 	pdc_receive(pdcs);
966 
967 	/* reenable interrupts */
968 	iowrite32(PDC_INTMASK, pdcs->pdc_reg_vbase + PDC_INTMASK_OFFSET);
969 }
970 
971 /**
972  * pdc_ring_init() - Allocate DMA rings and initialize constant fields of
973  * descriptors in one ringset.
974  * @pdcs:    PDC instance state
975  * @ringset: index of ringset being used
976  *
977  * Return: PDC_SUCCESS if ring initialized
978  *         < 0 otherwise
979  */
pdc_ring_init(struct pdc_state * pdcs,int ringset)980 static int pdc_ring_init(struct pdc_state *pdcs, int ringset)
981 {
982 	int i;
983 	int err = PDC_SUCCESS;
984 	struct dma64 *dma_reg;
985 	struct device *dev = &pdcs->pdev->dev;
986 	struct pdc_ring_alloc tx;
987 	struct pdc_ring_alloc rx;
988 
989 	/* Allocate tx ring */
990 	tx.vbase = dma_pool_zalloc(pdcs->ring_pool, GFP_KERNEL, &tx.dmabase);
991 	if (unlikely(!tx.vbase)) {
992 		err = -ENOMEM;
993 		goto done;
994 	}
995 
996 	/* Allocate rx ring */
997 	rx.vbase = dma_pool_zalloc(pdcs->ring_pool, GFP_KERNEL, &rx.dmabase);
998 	if (unlikely(!rx.vbase)) {
999 		err = -ENOMEM;
1000 		goto fail_dealloc;
1001 	}
1002 
1003 	dev_dbg(dev, " - base DMA addr of tx ring      %pad", &tx.dmabase);
1004 	dev_dbg(dev, " - base virtual addr of tx ring  %p", tx.vbase);
1005 	dev_dbg(dev, " - base DMA addr of rx ring      %pad", &rx.dmabase);
1006 	dev_dbg(dev, " - base virtual addr of rx ring  %p", rx.vbase);
1007 
1008 	memcpy(&pdcs->tx_ring_alloc, &tx, sizeof(tx));
1009 	memcpy(&pdcs->rx_ring_alloc, &rx, sizeof(rx));
1010 
1011 	pdcs->rxin = 0;
1012 	pdcs->rx_msg_start = 0;
1013 	pdcs->last_rx_curr = 0;
1014 	pdcs->rxout = 0;
1015 	pdcs->txin = 0;
1016 	pdcs->tx_msg_start = 0;
1017 	pdcs->txout = 0;
1018 
1019 	/* Set descriptor array base addresses */
1020 	pdcs->txd_64 = (struct dma64dd *)pdcs->tx_ring_alloc.vbase;
1021 	pdcs->rxd_64 = (struct dma64dd *)pdcs->rx_ring_alloc.vbase;
1022 
1023 	/* Tell device the base DMA address of each ring */
1024 	dma_reg = &pdcs->regs->dmaregs[ringset];
1025 
1026 	/* But first disable DMA and set curptr to 0 for both TX & RX */
1027 	iowrite32(PDC_TX_CTL, &dma_reg->dmaxmt.control);
1028 	iowrite32((PDC_RX_CTL + (pdcs->rx_status_len << 1)),
1029 		  &dma_reg->dmarcv.control);
1030 	iowrite32(0, &dma_reg->dmaxmt.ptr);
1031 	iowrite32(0, &dma_reg->dmarcv.ptr);
1032 
1033 	/* Set base DMA addresses */
1034 	iowrite32(lower_32_bits(pdcs->tx_ring_alloc.dmabase),
1035 		  &dma_reg->dmaxmt.addrlow);
1036 	iowrite32(upper_32_bits(pdcs->tx_ring_alloc.dmabase),
1037 		  &dma_reg->dmaxmt.addrhigh);
1038 
1039 	iowrite32(lower_32_bits(pdcs->rx_ring_alloc.dmabase),
1040 		  &dma_reg->dmarcv.addrlow);
1041 	iowrite32(upper_32_bits(pdcs->rx_ring_alloc.dmabase),
1042 		  &dma_reg->dmarcv.addrhigh);
1043 
1044 	/* Re-enable DMA */
1045 	iowrite32(PDC_TX_CTL | PDC_TX_ENABLE, &dma_reg->dmaxmt.control);
1046 	iowrite32((PDC_RX_CTL | PDC_RX_ENABLE | (pdcs->rx_status_len << 1)),
1047 		  &dma_reg->dmarcv.control);
1048 
1049 	/* Initialize descriptors */
1050 	for (i = 0; i < PDC_RING_ENTRIES; i++) {
1051 		/* Every tx descriptor can be used for start of frame. */
1052 		if (i != pdcs->ntxpost) {
1053 			iowrite32(D64_CTRL1_SOF | D64_CTRL1_EOF,
1054 				  &pdcs->txd_64[i].ctrl1);
1055 		} else {
1056 			/* Last descriptor in ringset. Set End of Table. */
1057 			iowrite32(D64_CTRL1_SOF | D64_CTRL1_EOF |
1058 				  D64_CTRL1_EOT, &pdcs->txd_64[i].ctrl1);
1059 		}
1060 
1061 		/* Every rx descriptor can be used for start of frame */
1062 		if (i != pdcs->nrxpost) {
1063 			iowrite32(D64_CTRL1_SOF,
1064 				  &pdcs->rxd_64[i].ctrl1);
1065 		} else {
1066 			/* Last descriptor in ringset. Set End of Table. */
1067 			iowrite32(D64_CTRL1_SOF | D64_CTRL1_EOT,
1068 				  &pdcs->rxd_64[i].ctrl1);
1069 		}
1070 	}
1071 	return PDC_SUCCESS;
1072 
1073 fail_dealloc:
1074 	dma_pool_free(pdcs->ring_pool, tx.vbase, tx.dmabase);
1075 done:
1076 	return err;
1077 }
1078 
pdc_ring_free(struct pdc_state * pdcs)1079 static void pdc_ring_free(struct pdc_state *pdcs)
1080 {
1081 	if (pdcs->tx_ring_alloc.vbase) {
1082 		dma_pool_free(pdcs->ring_pool, pdcs->tx_ring_alloc.vbase,
1083 			      pdcs->tx_ring_alloc.dmabase);
1084 		pdcs->tx_ring_alloc.vbase = NULL;
1085 	}
1086 
1087 	if (pdcs->rx_ring_alloc.vbase) {
1088 		dma_pool_free(pdcs->ring_pool, pdcs->rx_ring_alloc.vbase,
1089 			      pdcs->rx_ring_alloc.dmabase);
1090 		pdcs->rx_ring_alloc.vbase = NULL;
1091 	}
1092 }
1093 
1094 /**
1095  * pdc_desc_count() - Count the number of DMA descriptors that will be required
1096  * for a given scatterlist. Account for the max length of a DMA buffer.
1097  * @sg:    Scatterlist to be DMA'd
1098  * Return: Number of descriptors required
1099  */
pdc_desc_count(struct scatterlist * sg)1100 static u32 pdc_desc_count(struct scatterlist *sg)
1101 {
1102 	u32 cnt = 0;
1103 
1104 	while (sg) {
1105 		cnt += ((sg->length / PDC_DMA_BUF_MAX) + 1);
1106 		sg = sg_next(sg);
1107 	}
1108 	return cnt;
1109 }
1110 
1111 /**
1112  * pdc_rings_full() - Check whether the tx ring has room for tx_cnt descriptors
1113  * and the rx ring has room for rx_cnt descriptors.
1114  * @pdcs:  PDC state
1115  * @tx_cnt: The number of descriptors required in the tx ring
1116  * @rx_cnt: The number of descriptors required i the rx ring
1117  *
1118  * Return: true if one of the rings does not have enough space
1119  *         false if sufficient space is available in both rings
1120  */
pdc_rings_full(struct pdc_state * pdcs,int tx_cnt,int rx_cnt)1121 static bool pdc_rings_full(struct pdc_state *pdcs, int tx_cnt, int rx_cnt)
1122 {
1123 	u32 rx_avail;
1124 	u32 tx_avail;
1125 	bool full = false;
1126 
1127 	/* Check if the tx and rx rings are likely to have enough space */
1128 	rx_avail = pdcs->nrxpost - NRXDACTIVE(pdcs->rxin, pdcs->rxout,
1129 					      pdcs->nrxpost);
1130 	if (unlikely(rx_cnt > rx_avail)) {
1131 		pdcs->rx_ring_full++;
1132 		full = true;
1133 	}
1134 
1135 	if (likely(!full)) {
1136 		tx_avail = pdcs->ntxpost - NTXDACTIVE(pdcs->txin, pdcs->txout,
1137 						      pdcs->ntxpost);
1138 		if (unlikely(tx_cnt > tx_avail)) {
1139 			pdcs->tx_ring_full++;
1140 			full = true;
1141 		}
1142 	}
1143 	return full;
1144 }
1145 
1146 /**
1147  * pdc_last_tx_done() - If both the tx and rx rings have at least
1148  * PDC_RING_SPACE_MIN descriptors available, then indicate that the mailbox
1149  * framework can submit another message.
1150  * @chan:  mailbox channel to check
1151  * Return: true if PDC can accept another message on this channel
1152  */
pdc_last_tx_done(struct mbox_chan * chan)1153 static bool pdc_last_tx_done(struct mbox_chan *chan)
1154 {
1155 	struct pdc_state *pdcs = chan->con_priv;
1156 	bool ret;
1157 
1158 	if (unlikely(pdc_rings_full(pdcs, PDC_RING_SPACE_MIN,
1159 				    PDC_RING_SPACE_MIN))) {
1160 		pdcs->last_tx_not_done++;
1161 		ret = false;
1162 	} else {
1163 		ret = true;
1164 	}
1165 	return ret;
1166 }
1167 
1168 /**
1169  * pdc_send_data() - mailbox send_data function
1170  * @chan:	The mailbox channel on which the data is sent. The channel
1171  *              corresponds to a DMA ringset.
1172  * @data:	The mailbox message to be sent. The message must be a
1173  *              brcm_message structure.
1174  *
1175  * This function is registered as the send_data function for the mailbox
1176  * controller. From the destination scatterlist in the mailbox message, it
1177  * creates a sequence of receive descriptors in the rx ring. From the source
1178  * scatterlist, it creates a sequence of transmit descriptors in the tx ring.
1179  * After creating the descriptors, it writes the rx ptr and tx ptr registers to
1180  * initiate the DMA transfer.
1181  *
1182  * This function does the DMA map and unmap of the src and dst scatterlists in
1183  * the mailbox message.
1184  *
1185  * Return: 0 if successful
1186  *	   -ENOTSUPP if the mailbox message is a type this driver does not
1187  *			support
1188  *         < 0 if an error
1189  */
pdc_send_data(struct mbox_chan * chan,void * data)1190 static int pdc_send_data(struct mbox_chan *chan, void *data)
1191 {
1192 	struct pdc_state *pdcs = chan->con_priv;
1193 	struct device *dev = &pdcs->pdev->dev;
1194 	struct brcm_message *mssg = data;
1195 	int err = PDC_SUCCESS;
1196 	int src_nent;
1197 	int dst_nent;
1198 	int nent;
1199 	u32 tx_desc_req;
1200 	u32 rx_desc_req;
1201 
1202 	if (unlikely(mssg->type != BRCM_MESSAGE_SPU))
1203 		return -ENOTSUPP;
1204 
1205 	src_nent = sg_nents(mssg->spu.src);
1206 	if (likely(src_nent)) {
1207 		nent = dma_map_sg(dev, mssg->spu.src, src_nent, DMA_TO_DEVICE);
1208 		if (unlikely(nent == 0))
1209 			return -EIO;
1210 	}
1211 
1212 	dst_nent = sg_nents(mssg->spu.dst);
1213 	if (likely(dst_nent)) {
1214 		nent = dma_map_sg(dev, mssg->spu.dst, dst_nent,
1215 				  DMA_FROM_DEVICE);
1216 		if (unlikely(nent == 0)) {
1217 			dma_unmap_sg(dev, mssg->spu.src, src_nent,
1218 				     DMA_TO_DEVICE);
1219 			return -EIO;
1220 		}
1221 	}
1222 
1223 	/*
1224 	 * Check if the tx and rx rings have enough space. Do this prior to
1225 	 * writing any tx or rx descriptors. Need to ensure that we do not write
1226 	 * a partial set of descriptors, or write just rx descriptors but
1227 	 * corresponding tx descriptors don't fit. Note that we want this check
1228 	 * and the entire sequence of descriptor to happen without another
1229 	 * thread getting in. The channel spin lock in the mailbox framework
1230 	 * ensures this.
1231 	 */
1232 	tx_desc_req = pdc_desc_count(mssg->spu.src);
1233 	rx_desc_req = pdc_desc_count(mssg->spu.dst);
1234 	if (unlikely(pdc_rings_full(pdcs, tx_desc_req, rx_desc_req + 1)))
1235 		return -ENOSPC;
1236 
1237 	/* Create rx descriptors to SPU catch response */
1238 	err = pdc_rx_list_init(pdcs, mssg->spu.dst, mssg->ctx);
1239 	err |= pdc_rx_list_sg_add(pdcs, mssg->spu.dst);
1240 
1241 	/* Create tx descriptors to submit SPU request */
1242 	err |= pdc_tx_list_sg_add(pdcs, mssg->spu.src);
1243 	err |= pdc_tx_list_final(pdcs);	/* initiate transfer */
1244 
1245 	if (unlikely(err))
1246 		dev_err(&pdcs->pdev->dev,
1247 			"%s failed with error %d", __func__, err);
1248 
1249 	return err;
1250 }
1251 
pdc_startup(struct mbox_chan * chan)1252 static int pdc_startup(struct mbox_chan *chan)
1253 {
1254 	return pdc_ring_init(chan->con_priv, PDC_RINGSET);
1255 }
1256 
pdc_shutdown(struct mbox_chan * chan)1257 static void pdc_shutdown(struct mbox_chan *chan)
1258 {
1259 	struct pdc_state *pdcs = chan->con_priv;
1260 
1261 	if (!pdcs)
1262 		return;
1263 
1264 	dev_dbg(&pdcs->pdev->dev,
1265 		"Shutdown mailbox channel for PDC %u", pdcs->pdc_idx);
1266 	pdc_ring_free(pdcs);
1267 }
1268 
1269 /**
1270  * pdc_hw_init() - Use the given initialization parameters to initialize the
1271  * state for one of the PDCs.
1272  * @pdcs:  state of the PDC
1273  */
1274 static
pdc_hw_init(struct pdc_state * pdcs)1275 void pdc_hw_init(struct pdc_state *pdcs)
1276 {
1277 	struct platform_device *pdev;
1278 	struct device *dev;
1279 	struct dma64 *dma_reg;
1280 	int ringset = PDC_RINGSET;
1281 
1282 	pdev = pdcs->pdev;
1283 	dev = &pdev->dev;
1284 
1285 	dev_dbg(dev, "PDC %u initial values:", pdcs->pdc_idx);
1286 	dev_dbg(dev, "state structure:                   %p",
1287 		pdcs);
1288 	dev_dbg(dev, " - base virtual addr of hw regs    %p",
1289 		pdcs->pdc_reg_vbase);
1290 
1291 	/* initialize data structures */
1292 	pdcs->regs = (struct pdc_regs *)pdcs->pdc_reg_vbase;
1293 	pdcs->txregs_64 = (struct dma64_regs *)
1294 	    (((u8 *)pdcs->pdc_reg_vbase) +
1295 		     PDC_TXREGS_OFFSET + (sizeof(struct dma64) * ringset));
1296 	pdcs->rxregs_64 = (struct dma64_regs *)
1297 	    (((u8 *)pdcs->pdc_reg_vbase) +
1298 		     PDC_RXREGS_OFFSET + (sizeof(struct dma64) * ringset));
1299 
1300 	pdcs->ntxd = PDC_RING_ENTRIES;
1301 	pdcs->nrxd = PDC_RING_ENTRIES;
1302 	pdcs->ntxpost = PDC_RING_ENTRIES - 1;
1303 	pdcs->nrxpost = PDC_RING_ENTRIES - 1;
1304 	iowrite32(0, &pdcs->regs->intmask);
1305 
1306 	dma_reg = &pdcs->regs->dmaregs[ringset];
1307 
1308 	/* Configure DMA but will enable later in pdc_ring_init() */
1309 	iowrite32(PDC_TX_CTL, &dma_reg->dmaxmt.control);
1310 
1311 	iowrite32(PDC_RX_CTL + (pdcs->rx_status_len << 1),
1312 		  &dma_reg->dmarcv.control);
1313 
1314 	/* Reset current index pointers after making sure DMA is disabled */
1315 	iowrite32(0, &dma_reg->dmaxmt.ptr);
1316 	iowrite32(0, &dma_reg->dmarcv.ptr);
1317 
1318 	if (pdcs->pdc_resp_hdr_len == PDC_SPU2_RESP_HDR_LEN)
1319 		iowrite32(PDC_CKSUM_CTRL,
1320 			  pdcs->pdc_reg_vbase + PDC_CKSUM_CTRL_OFFSET);
1321 }
1322 
1323 /**
1324  * pdc_hw_disable() - Disable the tx and rx control in the hw.
1325  * @pdcs: PDC state structure
1326  *
1327  */
pdc_hw_disable(struct pdc_state * pdcs)1328 static void pdc_hw_disable(struct pdc_state *pdcs)
1329 {
1330 	struct dma64 *dma_reg;
1331 
1332 	dma_reg = &pdcs->regs->dmaregs[PDC_RINGSET];
1333 	iowrite32(PDC_TX_CTL, &dma_reg->dmaxmt.control);
1334 	iowrite32(PDC_RX_CTL + (pdcs->rx_status_len << 1),
1335 		  &dma_reg->dmarcv.control);
1336 }
1337 
1338 /**
1339  * pdc_rx_buf_pool_create() - Pool of receive buffers used to catch the metadata
1340  * header returned with each response message.
1341  * @pdcs: PDC state structure
1342  *
1343  * The metadata is not returned to the mailbox client. So the PDC driver
1344  * manages these buffers.
1345  *
1346  * Return: PDC_SUCCESS
1347  *         -ENOMEM if pool creation fails
1348  */
pdc_rx_buf_pool_create(struct pdc_state * pdcs)1349 static int pdc_rx_buf_pool_create(struct pdc_state *pdcs)
1350 {
1351 	struct platform_device *pdev;
1352 	struct device *dev;
1353 
1354 	pdev = pdcs->pdev;
1355 	dev = &pdev->dev;
1356 
1357 	pdcs->pdc_resp_hdr_len = pdcs->rx_status_len;
1358 	if (pdcs->use_bcm_hdr)
1359 		pdcs->pdc_resp_hdr_len += BCM_HDR_LEN;
1360 
1361 	pdcs->rx_buf_pool = dma_pool_create("pdc rx bufs", dev,
1362 					    pdcs->pdc_resp_hdr_len,
1363 					    RX_BUF_ALIGN, 0);
1364 	if (!pdcs->rx_buf_pool)
1365 		return -ENOMEM;
1366 
1367 	return PDC_SUCCESS;
1368 }
1369 
1370 /**
1371  * pdc_interrupts_init() - Initialize the interrupt configuration for a PDC and
1372  * specify a threaded IRQ handler for deferred handling of interrupts outside of
1373  * interrupt context.
1374  * @pdcs:   PDC state
1375  *
1376  * Set the interrupt mask for transmit and receive done.
1377  * Set the lazy interrupt frame count to generate an interrupt for just one pkt.
1378  *
1379  * Return:  PDC_SUCCESS
1380  *          <0 if threaded irq request fails
1381  */
pdc_interrupts_init(struct pdc_state * pdcs)1382 static int pdc_interrupts_init(struct pdc_state *pdcs)
1383 {
1384 	struct platform_device *pdev = pdcs->pdev;
1385 	struct device *dev = &pdev->dev;
1386 	struct device_node *dn = pdev->dev.of_node;
1387 	int err;
1388 
1389 	/* interrupt configuration */
1390 	iowrite32(PDC_INTMASK, pdcs->pdc_reg_vbase + PDC_INTMASK_OFFSET);
1391 
1392 	if (pdcs->hw_type == FA_HW)
1393 		iowrite32(PDC_LAZY_INT, pdcs->pdc_reg_vbase +
1394 			  FA_RCVLAZY0_OFFSET);
1395 	else
1396 		iowrite32(PDC_LAZY_INT, pdcs->pdc_reg_vbase +
1397 			  PDC_RCVLAZY0_OFFSET);
1398 
1399 	/* read irq from device tree */
1400 	pdcs->pdc_irq = irq_of_parse_and_map(dn, 0);
1401 	dev_dbg(dev, "pdc device %s irq %u for pdcs %p",
1402 		dev_name(dev), pdcs->pdc_irq, pdcs);
1403 
1404 	err = devm_request_irq(dev, pdcs->pdc_irq, pdc_irq_handler, 0,
1405 			       dev_name(dev), dev);
1406 	if (err) {
1407 		dev_err(dev, "IRQ %u request failed with err %d\n",
1408 			pdcs->pdc_irq, err);
1409 		return err;
1410 	}
1411 	return PDC_SUCCESS;
1412 }
1413 
1414 static const struct mbox_chan_ops pdc_mbox_chan_ops = {
1415 	.send_data = pdc_send_data,
1416 	.last_tx_done = pdc_last_tx_done,
1417 	.startup = pdc_startup,
1418 	.shutdown = pdc_shutdown
1419 };
1420 
1421 /**
1422  * pdc_mb_init() - Initialize the mailbox controller.
1423  * @pdcs:  PDC state
1424  *
1425  * Each PDC is a mailbox controller. Each ringset is a mailbox channel. Kernel
1426  * driver only uses one ringset and thus one mb channel. PDC uses the transmit
1427  * complete interrupt to determine when a mailbox message has successfully been
1428  * transmitted.
1429  *
1430  * Return: 0 on success
1431  *         < 0 if there is an allocation or registration failure
1432  */
pdc_mb_init(struct pdc_state * pdcs)1433 static int pdc_mb_init(struct pdc_state *pdcs)
1434 {
1435 	struct device *dev = &pdcs->pdev->dev;
1436 	struct mbox_controller *mbc;
1437 	int chan_index;
1438 	int err;
1439 
1440 	mbc = &pdcs->mbc;
1441 	mbc->dev = dev;
1442 	mbc->ops = &pdc_mbox_chan_ops;
1443 	mbc->num_chans = 1;
1444 	mbc->chans = devm_kcalloc(dev, mbc->num_chans, sizeof(*mbc->chans),
1445 				  GFP_KERNEL);
1446 	if (!mbc->chans)
1447 		return -ENOMEM;
1448 
1449 	mbc->txdone_irq = false;
1450 	mbc->txdone_poll = true;
1451 	mbc->txpoll_period = 1;
1452 	for (chan_index = 0; chan_index < mbc->num_chans; chan_index++)
1453 		mbc->chans[chan_index].con_priv = pdcs;
1454 
1455 	/* Register mailbox controller */
1456 	err = devm_mbox_controller_register(dev, mbc);
1457 	if (err) {
1458 		dev_crit(dev,
1459 			 "Failed to register PDC mailbox controller. Error %d.",
1460 			 err);
1461 		return err;
1462 	}
1463 	return 0;
1464 }
1465 
1466 /* Device tree API */
1467 static const int pdc_hw = PDC_HW;
1468 static const int fa_hw = FA_HW;
1469 
1470 static const struct of_device_id pdc_mbox_of_match[] = {
1471 	{.compatible = "brcm,iproc-pdc-mbox", .data = &pdc_hw},
1472 	{.compatible = "brcm,iproc-fa2-mbox", .data = &fa_hw},
1473 	{ /* sentinel */ }
1474 };
1475 MODULE_DEVICE_TABLE(of, pdc_mbox_of_match);
1476 
1477 /**
1478  * pdc_dt_read() - Read application-specific data from device tree.
1479  * @pdev:  Platform device
1480  * @pdcs:  PDC state
1481  *
1482  * Reads the number of bytes of receive status that precede each received frame.
1483  * Reads whether transmit and received frames should be preceded by an 8-byte
1484  * BCM header.
1485  *
1486  * Return: 0 if successful
1487  *         -ENODEV if device not available
1488  */
pdc_dt_read(struct platform_device * pdev,struct pdc_state * pdcs)1489 static int pdc_dt_read(struct platform_device *pdev, struct pdc_state *pdcs)
1490 {
1491 	struct device *dev = &pdev->dev;
1492 	struct device_node *dn = pdev->dev.of_node;
1493 	const int *hw_type;
1494 	int err;
1495 
1496 	err = of_property_read_u32(dn, "brcm,rx-status-len",
1497 				   &pdcs->rx_status_len);
1498 	if (err < 0)
1499 		dev_err(dev,
1500 			"%s failed to get DMA receive status length from device tree",
1501 			__func__);
1502 
1503 	pdcs->use_bcm_hdr = of_property_read_bool(dn, "brcm,use-bcm-hdr");
1504 
1505 	pdcs->hw_type = PDC_HW;
1506 
1507 	hw_type = device_get_match_data(dev);
1508 	if (hw_type)
1509 		pdcs->hw_type = *hw_type;
1510 
1511 	return 0;
1512 }
1513 
1514 /**
1515  * pdc_probe() - Probe function for PDC driver.
1516  * @pdev:   PDC platform device
1517  *
1518  * Reserve and map register regions defined in device tree.
1519  * Allocate and initialize tx and rx DMA rings.
1520  * Initialize a mailbox controller for each PDC.
1521  *
1522  * Return: 0 if successful
1523  *         < 0 if an error
1524  */
pdc_probe(struct platform_device * pdev)1525 static int pdc_probe(struct platform_device *pdev)
1526 {
1527 	int err = 0;
1528 	struct device *dev = &pdev->dev;
1529 	struct resource *pdc_regs;
1530 	struct pdc_state *pdcs;
1531 
1532 	/* PDC state for one SPU */
1533 	pdcs = devm_kzalloc(dev, sizeof(*pdcs), GFP_KERNEL);
1534 	if (!pdcs) {
1535 		err = -ENOMEM;
1536 		goto cleanup;
1537 	}
1538 
1539 	pdcs->pdev = pdev;
1540 	platform_set_drvdata(pdev, pdcs);
1541 	pdcs->pdc_idx = pdcg.num_spu;
1542 	pdcg.num_spu++;
1543 
1544 	err = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(39));
1545 	if (err) {
1546 		dev_warn(dev, "PDC device cannot perform DMA. Error %d.", err);
1547 		goto cleanup;
1548 	}
1549 
1550 	/* Create DMA pool for tx ring */
1551 	pdcs->ring_pool = dma_pool_create("pdc rings", dev, PDC_RING_SIZE,
1552 					  RING_ALIGN, 0);
1553 	if (!pdcs->ring_pool) {
1554 		err = -ENOMEM;
1555 		goto cleanup;
1556 	}
1557 
1558 	err = pdc_dt_read(pdev, pdcs);
1559 	if (err)
1560 		goto cleanup_ring_pool;
1561 
1562 	pdcs->pdc_reg_vbase = devm_platform_get_and_ioremap_resource(pdev, 0, &pdc_regs);
1563 	if (IS_ERR(pdcs->pdc_reg_vbase)) {
1564 		err = PTR_ERR(pdcs->pdc_reg_vbase);
1565 		goto cleanup_ring_pool;
1566 	}
1567 	dev_dbg(dev, "PDC register region res.start = %pa, res.end = %pa",
1568 		&pdc_regs->start, &pdc_regs->end);
1569 
1570 	/* create rx buffer pool after dt read to know how big buffers are */
1571 	err = pdc_rx_buf_pool_create(pdcs);
1572 	if (err)
1573 		goto cleanup_ring_pool;
1574 
1575 	pdc_hw_init(pdcs);
1576 
1577 	/* Init work for deferred DMA rx processing */
1578 	INIT_WORK(&pdcs->rx_work, pdc_work_cb);
1579 
1580 	err = pdc_interrupts_init(pdcs);
1581 	if (err)
1582 		goto cleanup_buf_pool;
1583 
1584 	/* Initialize mailbox controller */
1585 	err = pdc_mb_init(pdcs);
1586 	if (err)
1587 		goto cleanup_buf_pool;
1588 
1589 	pdc_setup_debugfs(pdcs);
1590 
1591 	dev_dbg(dev, "pdc_probe() successful");
1592 	return PDC_SUCCESS;
1593 
1594 cleanup_buf_pool:
1595 	cancel_work_sync(&pdcs->rx_work);
1596 	dma_pool_destroy(pdcs->rx_buf_pool);
1597 
1598 cleanup_ring_pool:
1599 	dma_pool_destroy(pdcs->ring_pool);
1600 
1601 cleanup:
1602 	return err;
1603 }
1604 
pdc_remove(struct platform_device * pdev)1605 static void pdc_remove(struct platform_device *pdev)
1606 {
1607 	struct pdc_state *pdcs = platform_get_drvdata(pdev);
1608 
1609 	pdc_free_debugfs();
1610 
1611 	cancel_work_sync(&pdcs->rx_work);
1612 
1613 	pdc_hw_disable(pdcs);
1614 
1615 	dma_pool_destroy(pdcs->rx_buf_pool);
1616 	dma_pool_destroy(pdcs->ring_pool);
1617 }
1618 
1619 static struct platform_driver pdc_mbox_driver = {
1620 	.probe = pdc_probe,
1621 	.remove_new = pdc_remove,
1622 	.driver = {
1623 		   .name = "brcm-iproc-pdc-mbox",
1624 		   .of_match_table = pdc_mbox_of_match,
1625 		   },
1626 };
1627 module_platform_driver(pdc_mbox_driver);
1628 
1629 MODULE_AUTHOR("Rob Rice <rob.rice@broadcom.com>");
1630 MODULE_DESCRIPTION("Broadcom PDC mailbox driver");
1631 MODULE_LICENSE("GPL v2");
1632