1// SPDX-License-Identifier: GPL-2.0
2/*
3 * ZynqMP R5 Remote Processor driver
4 *
5 */
6
7#include <dt-bindings/power/xlnx-zynqmp-power.h>
8#include <linux/dma-mapping.h>
9#include <linux/firmware/xlnx-zynqmp.h>
10#include <linux/kernel.h>
11#include <linux/mailbox_client.h>
12#include <linux/mailbox/zynqmp-ipi-message.h>
13#include <linux/module.h>
14#include <linux/of_address.h>
15#include <linux/of_platform.h>
16#include <linux/of_reserved_mem.h>
17#include <linux/platform_device.h>
18#include <linux/remoteproc.h>
19
20#include "remoteproc_internal.h"
21
22/* IPI buffer MAX length */
23#define IPI_BUF_LEN_MAX	32U
24
25/* RX mailbox client buffer max length */
26#define MBOX_CLIENT_BUF_MAX	(IPI_BUF_LEN_MAX + \
27				 sizeof(struct zynqmp_ipi_message))
28/*
29 * settings for RPU cluster mode which
30 * reflects possible values of xlnx,cluster-mode dt-property
31 */
32enum zynqmp_r5_cluster_mode {
33	SPLIT_MODE = 0, /* When cores run as separate processor */
34	LOCKSTEP_MODE = 1, /* cores execute same code in lockstep,clk-for-clk */
35	SINGLE_CPU_MODE = 2, /* core0 is held in reset and only core1 runs */
36};
37
38/**
39 * struct mem_bank_data - Memory Bank description
40 *
41 * @addr: Start address of memory bank
42 * @da: device address
43 * @size: Size of Memory bank
44 * @pm_domain_id: Power-domains id of memory bank for firmware to turn on/off
45 * @bank_name: name of the bank for remoteproc framework
46 */
47struct mem_bank_data {
48	phys_addr_t addr;
49	u32 da;
50	size_t size;
51	u32 pm_domain_id;
52	char *bank_name;
53};
54
55/**
56 * struct mbox_info
57 *
58 * @rx_mc_buf: to copy data from mailbox rx channel
59 * @tx_mc_buf: to copy data to mailbox tx channel
60 * @r5_core: this mailbox's corresponding r5_core pointer
61 * @mbox_work: schedule work after receiving data from mailbox
62 * @mbox_cl: mailbox client
63 * @tx_chan: mailbox tx channel
64 * @rx_chan: mailbox rx channel
65 */
66struct mbox_info {
67	unsigned char rx_mc_buf[MBOX_CLIENT_BUF_MAX];
68	unsigned char tx_mc_buf[MBOX_CLIENT_BUF_MAX];
69	struct zynqmp_r5_core *r5_core;
70	struct work_struct mbox_work;
71	struct mbox_client mbox_cl;
72	struct mbox_chan *tx_chan;
73	struct mbox_chan *rx_chan;
74};
75
76/*
77 * Hardcoded TCM bank values. This will be removed once TCM bindings are
78 * accepted for system-dt specifications and upstreamed in linux kernel
79 */
80static const struct mem_bank_data zynqmp_tcm_banks_split[] = {
81	{0xffe00000UL, 0x0, 0x10000UL, PD_R5_0_ATCM, "atcm0"}, /* TCM 64KB each */
82	{0xffe20000UL, 0x20000, 0x10000UL, PD_R5_0_BTCM, "btcm0"},
83	{0xffe90000UL, 0x0, 0x10000UL, PD_R5_1_ATCM, "atcm1"},
84	{0xffeb0000UL, 0x20000, 0x10000UL, PD_R5_1_BTCM, "btcm1"},
85};
86
87/* In lockstep mode cluster combines each 64KB TCM and makes 128KB TCM */
88static const struct mem_bank_data zynqmp_tcm_banks_lockstep[] = {
89	{0xffe00000UL, 0x0, 0x20000UL, PD_R5_0_ATCM, "atcm0"}, /* TCM 128KB each */
90	{0xffe20000UL, 0x20000, 0x20000UL, PD_R5_0_BTCM, "btcm0"},
91	{0, 0, 0, PD_R5_1_ATCM, ""},
92	{0, 0, 0, PD_R5_1_BTCM, ""},
93};
94
95/**
96 * struct zynqmp_r5_core
97 *
98 * @dev: device of RPU instance
99 * @np: device node of RPU instance
100 * @tcm_bank_count: number TCM banks accessible to this RPU
101 * @tcm_banks: array of each TCM bank data
102 * @rproc: rproc handle
103 * @pm_domain_id: RPU CPU power domain id
104 * @ipi: pointer to mailbox information
105 */
106struct zynqmp_r5_core {
107	struct device *dev;
108	struct device_node *np;
109	int tcm_bank_count;
110	struct mem_bank_data **tcm_banks;
111	struct rproc *rproc;
112	u32 pm_domain_id;
113	struct mbox_info *ipi;
114};
115
116/**
117 * struct zynqmp_r5_cluster
118 *
119 * @dev: r5f subsystem cluster device node
120 * @mode: cluster mode of type zynqmp_r5_cluster_mode
121 * @core_count: number of r5 cores used for this cluster mode
122 * @r5_cores: Array of pointers pointing to r5 core
123 */
124struct zynqmp_r5_cluster {
125	struct device *dev;
126	enum  zynqmp_r5_cluster_mode mode;
127	int core_count;
128	struct zynqmp_r5_core **r5_cores;
129};
130
131/**
132 * event_notified_idr_cb() - callback for vq_interrupt per notifyid
133 * @id: rproc->notify id
134 * @ptr: pointer to idr private data
135 * @data: data passed to idr_for_each callback
136 *
137 * Pass notification to remoteproc virtio
138 *
139 * Return: 0. having return is to satisfy the idr_for_each() function
140 *          pointer input argument requirement.
141 **/
142static int event_notified_idr_cb(int id, void *ptr, void *data)
143{
144	struct rproc *rproc = data;
145
146	if (rproc_vq_interrupt(rproc, id) == IRQ_NONE)
147		dev_dbg(&rproc->dev, "data not found for vqid=%d\n", id);
148
149	return 0;
150}
151
152/**
153 * handle_event_notified() - remoteproc notification work function
154 * @work: pointer to the work structure
155 *
156 * It checks each registered remoteproc notify IDs.
157 */
158static void handle_event_notified(struct work_struct *work)
159{
160	struct mbox_info *ipi;
161	struct rproc *rproc;
162
163	ipi = container_of(work, struct mbox_info, mbox_work);
164	rproc = ipi->r5_core->rproc;
165
166	/*
167	 * We only use IPI for interrupt. The RPU firmware side may or may
168	 * not write the notifyid when it trigger IPI.
169	 * And thus, we scan through all the registered notifyids and
170	 * find which one is valid to get the message.
171	 * Even if message from firmware is NULL, we attempt to get vqid
172	 */
173	idr_for_each(&rproc->notifyids, event_notified_idr_cb, rproc);
174}
175
176/**
177 * zynqmp_r5_mb_rx_cb() - receive channel mailbox callback
178 * @cl: mailbox client
179 * @msg: message pointer
180 *
181 * Receive data from ipi buffer, ack interrupt and then
182 * it will schedule the R5 notification work.
183 */
184static void zynqmp_r5_mb_rx_cb(struct mbox_client *cl, void *msg)
185{
186	struct zynqmp_ipi_message *ipi_msg, *buf_msg;
187	struct mbox_info *ipi;
188	size_t len;
189
190	ipi = container_of(cl, struct mbox_info, mbox_cl);
191
192	/* copy data from ipi buffer to r5_core */
193	ipi_msg = (struct zynqmp_ipi_message *)msg;
194	buf_msg = (struct zynqmp_ipi_message *)ipi->rx_mc_buf;
195	len = ipi_msg->len;
196	if (len > IPI_BUF_LEN_MAX) {
197		dev_warn(cl->dev, "msg size exceeded than %d\n",
198			 IPI_BUF_LEN_MAX);
199		len = IPI_BUF_LEN_MAX;
200	}
201	buf_msg->len = len;
202	memcpy(buf_msg->data, ipi_msg->data, len);
203
204	/* received and processed interrupt ack */
205	if (mbox_send_message(ipi->rx_chan, NULL) < 0)
206		dev_err(cl->dev, "ack failed to mbox rx_chan\n");
207
208	schedule_work(&ipi->mbox_work);
209}
210
211/**
212 * zynqmp_r5_setup_mbox() - Setup mailboxes related properties
213 *			    this is used for each individual R5 core
214 *
215 * @cdev: child node device
216 *
217 * Function to setup mailboxes related properties
218 * return : NULL if failed else pointer to mbox_info
219 */
220static struct mbox_info *zynqmp_r5_setup_mbox(struct device *cdev)
221{
222	struct mbox_client *mbox_cl;
223	struct mbox_info *ipi;
224
225	ipi = kzalloc(sizeof(*ipi), GFP_KERNEL);
226	if (!ipi)
227		return NULL;
228
229	mbox_cl = &ipi->mbox_cl;
230	mbox_cl->rx_callback = zynqmp_r5_mb_rx_cb;
231	mbox_cl->tx_block = false;
232	mbox_cl->knows_txdone = false;
233	mbox_cl->tx_done = NULL;
234	mbox_cl->dev = cdev;
235
236	/* Request TX and RX channels */
237	ipi->tx_chan = mbox_request_channel_byname(mbox_cl, "tx");
238	if (IS_ERR(ipi->tx_chan)) {
239		ipi->tx_chan = NULL;
240		kfree(ipi);
241		dev_warn(cdev, "mbox tx channel request failed\n");
242		return NULL;
243	}
244
245	ipi->rx_chan = mbox_request_channel_byname(mbox_cl, "rx");
246	if (IS_ERR(ipi->rx_chan)) {
247		mbox_free_channel(ipi->tx_chan);
248		ipi->rx_chan = NULL;
249		ipi->tx_chan = NULL;
250		kfree(ipi);
251		dev_warn(cdev, "mbox rx channel request failed\n");
252		return NULL;
253	}
254
255	INIT_WORK(&ipi->mbox_work, handle_event_notified);
256
257	return ipi;
258}
259
260static void zynqmp_r5_free_mbox(struct mbox_info *ipi)
261{
262	if (!ipi)
263		return;
264
265	if (ipi->tx_chan) {
266		mbox_free_channel(ipi->tx_chan);
267		ipi->tx_chan = NULL;
268	}
269
270	if (ipi->rx_chan) {
271		mbox_free_channel(ipi->rx_chan);
272		ipi->rx_chan = NULL;
273	}
274
275	kfree(ipi);
276}
277
278/*
279 * zynqmp_r5_core_kick() - kick a firmware if mbox is provided
280 * @rproc: r5 core's corresponding rproc structure
281 * @vqid: virtqueue ID
282 */
283static void zynqmp_r5_rproc_kick(struct rproc *rproc, int vqid)
284{
285	struct zynqmp_r5_core *r5_core = rproc->priv;
286	struct device *dev = r5_core->dev;
287	struct zynqmp_ipi_message *mb_msg;
288	struct mbox_info *ipi;
289	int ret;
290
291	ipi = r5_core->ipi;
292	if (!ipi)
293		return;
294
295	mb_msg = (struct zynqmp_ipi_message *)ipi->tx_mc_buf;
296	memcpy(mb_msg->data, &vqid, sizeof(vqid));
297	mb_msg->len = sizeof(vqid);
298	ret = mbox_send_message(ipi->tx_chan, mb_msg);
299	if (ret < 0)
300		dev_warn(dev, "failed to send message\n");
301}
302
303/*
304 * zynqmp_r5_set_mode()
305 *
306 * set RPU cluster and TCM operation mode
307 *
308 * @r5_core: pointer to zynqmp_r5_core type object
309 * @fw_reg_val: value expected by firmware to configure RPU cluster mode
310 * @tcm_mode: value expected by fw to configure TCM mode (lockstep or split)
311 *
312 * Return: 0 for success and < 0 for failure
313 */
314static int zynqmp_r5_set_mode(struct zynqmp_r5_core *r5_core,
315			      enum rpu_oper_mode fw_reg_val,
316			      enum rpu_tcm_comb tcm_mode)
317{
318	int ret;
319
320	ret = zynqmp_pm_set_rpu_mode(r5_core->pm_domain_id, fw_reg_val);
321	if (ret < 0) {
322		dev_err(r5_core->dev, "failed to set RPU mode\n");
323		return ret;
324	}
325
326	ret = zynqmp_pm_set_tcm_config(r5_core->pm_domain_id, tcm_mode);
327	if (ret < 0)
328		dev_err(r5_core->dev, "failed to configure TCM\n");
329
330	return ret;
331}
332
333/*
334 * zynqmp_r5_rproc_start()
335 * @rproc: single R5 core's corresponding rproc instance
336 *
337 * Start R5 Core from designated boot address.
338 *
339 * return 0 on success, otherwise non-zero value on failure
340 */
341static int zynqmp_r5_rproc_start(struct rproc *rproc)
342{
343	struct zynqmp_r5_core *r5_core = rproc->priv;
344	enum rpu_boot_mem bootmem;
345	int ret;
346
347	/*
348	 * The exception vector pointers (EVP) refer to the base-address of
349	 * exception vectors (for reset, IRQ, FIQ, etc). The reset-vector
350	 * starts at the base-address and subsequent vectors are on 4-byte
351	 * boundaries.
352	 *
353	 * Exception vectors can start either from 0x0000_0000 (LOVEC) or
354	 * from 0xFFFF_0000 (HIVEC) which is mapped in the OCM (On-Chip Memory)
355	 *
356	 * Usually firmware will put Exception vectors at LOVEC.
357	 *
358	 * It is not recommend that you change the exception vector.
359	 * Changing the EVP to HIVEC will result in increased interrupt latency
360	 * and jitter. Also, if the OCM is secured and the Cortex-R5F processor
361	 * is non-secured, then the Cortex-R5F processor cannot access the
362	 * HIVEC exception vectors in the OCM.
363	 */
364	bootmem = (rproc->bootaddr >= 0xFFFC0000) ?
365		   PM_RPU_BOOTMEM_HIVEC : PM_RPU_BOOTMEM_LOVEC;
366
367	dev_dbg(r5_core->dev, "RPU boot addr 0x%llx from %s.", rproc->bootaddr,
368		bootmem == PM_RPU_BOOTMEM_HIVEC ? "OCM" : "TCM");
369
370	ret = zynqmp_pm_request_wake(r5_core->pm_domain_id, 1,
371				     bootmem, ZYNQMP_PM_REQUEST_ACK_NO);
372	if (ret)
373		dev_err(r5_core->dev,
374			"failed to start RPU = 0x%x\n", r5_core->pm_domain_id);
375	return ret;
376}
377
378/*
379 * zynqmp_r5_rproc_stop()
380 * @rproc: single R5 core's corresponding rproc instance
381 *
382 * Power down  R5 Core.
383 *
384 * return 0 on success, otherwise non-zero value on failure
385 */
386static int zynqmp_r5_rproc_stop(struct rproc *rproc)
387{
388	struct zynqmp_r5_core *r5_core = rproc->priv;
389	int ret;
390
391	ret = zynqmp_pm_force_pwrdwn(r5_core->pm_domain_id,
392				     ZYNQMP_PM_REQUEST_ACK_BLOCKING);
393	if (ret)
394		dev_err(r5_core->dev, "failed to stop remoteproc RPU %d\n", ret);
395
396	return ret;
397}
398
399/*
400 * zynqmp_r5_mem_region_map()
401 * @rproc: single R5 core's corresponding rproc instance
402 * @mem: mem descriptor to map reserved memory-regions
403 *
404 * Callback to map va for memory-region's carveout.
405 *
406 * return 0 on success, otherwise non-zero value on failure
407 */
408static int zynqmp_r5_mem_region_map(struct rproc *rproc,
409				    struct rproc_mem_entry *mem)
410{
411	void __iomem *va;
412
413	va = ioremap_wc(mem->dma, mem->len);
414	if (IS_ERR_OR_NULL(va))
415		return -ENOMEM;
416
417	mem->va = (void *)va;
418
419	return 0;
420}
421
422/*
423 * zynqmp_r5_rproc_mem_unmap
424 * @rproc: single R5 core's corresponding rproc instance
425 * @mem: mem entry to unmap
426 *
427 * Unmap memory-region carveout
428 *
429 * return: always returns 0
430 */
431static int zynqmp_r5_mem_region_unmap(struct rproc *rproc,
432				      struct rproc_mem_entry *mem)
433{
434	iounmap((void __iomem *)mem->va);
435	return 0;
436}
437
438/*
439 * add_mem_regions_carveout()
440 * @rproc: single R5 core's corresponding rproc instance
441 *
442 * Construct rproc mem carveouts from memory-region property nodes
443 *
444 * return 0 on success, otherwise non-zero value on failure
445 */
446static int add_mem_regions_carveout(struct rproc *rproc)
447{
448	struct rproc_mem_entry *rproc_mem;
449	struct zynqmp_r5_core *r5_core;
450	struct of_phandle_iterator it;
451	struct reserved_mem *rmem;
452	int i = 0;
453
454	r5_core = rproc->priv;
455
456	/* Register associated reserved memory regions */
457	of_phandle_iterator_init(&it, r5_core->np, "memory-region", NULL, 0);
458
459	while (of_phandle_iterator_next(&it) == 0) {
460		rmem = of_reserved_mem_lookup(it.node);
461		if (!rmem) {
462			of_node_put(it.node);
463			dev_err(&rproc->dev, "unable to acquire memory-region\n");
464			return -EINVAL;
465		}
466
467		if (!strcmp(it.node->name, "vdev0buffer")) {
468			/* Init reserved memory for vdev buffer */
469			rproc_mem = rproc_of_resm_mem_entry_init(&rproc->dev, i,
470								 rmem->size,
471								 rmem->base,
472								 it.node->name);
473		} else {
474			/* Register associated reserved memory regions */
475			rproc_mem = rproc_mem_entry_init(&rproc->dev, NULL,
476							 (dma_addr_t)rmem->base,
477							 rmem->size, rmem->base,
478							 zynqmp_r5_mem_region_map,
479							 zynqmp_r5_mem_region_unmap,
480							 it.node->name);
481		}
482
483		if (!rproc_mem) {
484			of_node_put(it.node);
485			return -ENOMEM;
486		}
487
488		rproc_add_carveout(rproc, rproc_mem);
489
490		dev_dbg(&rproc->dev, "reserved mem carveout %s addr=%llx, size=0x%llx",
491			it.node->name, rmem->base, rmem->size);
492		i++;
493	}
494
495	return 0;
496}
497
498/*
499 * tcm_mem_unmap()
500 * @rproc: single R5 core's corresponding rproc instance
501 * @mem: tcm mem entry to unmap
502 *
503 * Unmap TCM banks when powering down R5 core.
504 *
505 * return always 0
506 */
507static int tcm_mem_unmap(struct rproc *rproc, struct rproc_mem_entry *mem)
508{
509	iounmap((void __iomem *)mem->va);
510
511	return 0;
512}
513
514/*
515 * tcm_mem_map()
516 * @rproc: single R5 core's corresponding rproc instance
517 * @mem: tcm memory entry descriptor
518 *
519 * Given TCM bank entry, this func setup virtual address for TCM bank
520 * remoteproc carveout. It also takes care of va to da address translation
521 *
522 * return 0 on success, otherwise non-zero value on failure
523 */
524static int tcm_mem_map(struct rproc *rproc,
525		       struct rproc_mem_entry *mem)
526{
527	void __iomem *va;
528
529	va = ioremap_wc(mem->dma, mem->len);
530	if (IS_ERR_OR_NULL(va))
531		return -ENOMEM;
532
533	/* Update memory entry va */
534	mem->va = (void *)va;
535
536	/* clear TCMs */
537	memset_io(va, 0, mem->len);
538
539	return 0;
540}
541
542/*
543 * add_tcm_carveout_split_mode()
544 * @rproc: single R5 core's corresponding rproc instance
545 *
546 * allocate and add remoteproc carveout for TCM memory in split mode
547 *
548 * return 0 on success, otherwise non-zero value on failure
549 */
550static int add_tcm_carveout_split_mode(struct rproc *rproc)
551{
552	struct rproc_mem_entry *rproc_mem;
553	struct zynqmp_r5_core *r5_core;
554	int i, num_banks, ret;
555	phys_addr_t bank_addr;
556	struct device *dev;
557	u32 pm_domain_id;
558	size_t bank_size;
559	char *bank_name;
560	u32 da;
561
562	r5_core = rproc->priv;
563	dev = r5_core->dev;
564	num_banks = r5_core->tcm_bank_count;
565
566	/*
567	 * Power-on Each 64KB TCM,
568	 * register its address space, map and unmap functions
569	 * and add carveouts accordingly
570	 */
571	for (i = 0; i < num_banks; i++) {
572		bank_addr = r5_core->tcm_banks[i]->addr;
573		da = r5_core->tcm_banks[i]->da;
574		bank_name = r5_core->tcm_banks[i]->bank_name;
575		bank_size = r5_core->tcm_banks[i]->size;
576		pm_domain_id = r5_core->tcm_banks[i]->pm_domain_id;
577
578		ret = zynqmp_pm_request_node(pm_domain_id,
579					     ZYNQMP_PM_CAPABILITY_ACCESS, 0,
580					     ZYNQMP_PM_REQUEST_ACK_BLOCKING);
581		if (ret < 0) {
582			dev_err(dev, "failed to turn on TCM 0x%x", pm_domain_id);
583			goto release_tcm_split;
584		}
585
586		dev_dbg(dev, "TCM carveout split mode %s addr=%llx, da=0x%x, size=0x%lx",
587			bank_name, bank_addr, da, bank_size);
588
589		rproc_mem = rproc_mem_entry_init(dev, NULL, bank_addr,
590						 bank_size, da,
591						 tcm_mem_map, tcm_mem_unmap,
592						 bank_name);
593		if (!rproc_mem) {
594			ret = -ENOMEM;
595			zynqmp_pm_release_node(pm_domain_id);
596			goto release_tcm_split;
597		}
598
599		rproc_add_carveout(rproc, rproc_mem);
600	}
601
602	return 0;
603
604release_tcm_split:
605	/* If failed, Turn off all TCM banks turned on before */
606	for (i--; i >= 0; i--) {
607		pm_domain_id = r5_core->tcm_banks[i]->pm_domain_id;
608		zynqmp_pm_release_node(pm_domain_id);
609	}
610	return ret;
611}
612
613/*
614 * add_tcm_carveout_lockstep_mode()
615 * @rproc: single R5 core's corresponding rproc instance
616 *
617 * allocate and add remoteproc carveout for TCM memory in lockstep mode
618 *
619 * return 0 on success, otherwise non-zero value on failure
620 */
621static int add_tcm_carveout_lockstep_mode(struct rproc *rproc)
622{
623	struct rproc_mem_entry *rproc_mem;
624	struct zynqmp_r5_core *r5_core;
625	int i, num_banks, ret;
626	phys_addr_t bank_addr;
627	size_t bank_size = 0;
628	struct device *dev;
629	u32 pm_domain_id;
630	char *bank_name;
631	u32 da;
632
633	r5_core = rproc->priv;
634	dev = r5_core->dev;
635
636	/* Go through zynqmp banks for r5 node */
637	num_banks = r5_core->tcm_bank_count;
638
639	/*
640	 * In lockstep mode, TCM is contiguous memory block
641	 * However, each TCM block still needs to be enabled individually.
642	 * So, Enable each TCM block individually.
643	 * Although ATCM and BTCM is contiguous memory block, add two separate
644	 * carveouts for both.
645	 */
646	for (i = 0; i < num_banks; i++) {
647		pm_domain_id = r5_core->tcm_banks[i]->pm_domain_id;
648
649		/* Turn on each TCM bank individually */
650		ret = zynqmp_pm_request_node(pm_domain_id,
651					     ZYNQMP_PM_CAPABILITY_ACCESS, 0,
652					     ZYNQMP_PM_REQUEST_ACK_BLOCKING);
653		if (ret < 0) {
654			dev_err(dev, "failed to turn on TCM 0x%x", pm_domain_id);
655			goto release_tcm_lockstep;
656		}
657
658		bank_size = r5_core->tcm_banks[i]->size;
659		if (bank_size == 0)
660			continue;
661
662		bank_addr = r5_core->tcm_banks[i]->addr;
663		da = r5_core->tcm_banks[i]->da;
664		bank_name = r5_core->tcm_banks[i]->bank_name;
665
666		/* Register TCM address range, TCM map and unmap functions */
667		rproc_mem = rproc_mem_entry_init(dev, NULL, bank_addr,
668						 bank_size, da,
669						 tcm_mem_map, tcm_mem_unmap,
670						 bank_name);
671		if (!rproc_mem) {
672			ret = -ENOMEM;
673			zynqmp_pm_release_node(pm_domain_id);
674			goto release_tcm_lockstep;
675		}
676
677		/* If registration is success, add carveouts */
678		rproc_add_carveout(rproc, rproc_mem);
679
680		dev_dbg(dev, "TCM carveout lockstep mode %s addr=0x%llx, da=0x%x, size=0x%lx",
681			bank_name, bank_addr, da, bank_size);
682	}
683
684	return 0;
685
686release_tcm_lockstep:
687	/* If failed, Turn off all TCM banks turned on before */
688	for (i--; i >= 0; i--) {
689		pm_domain_id = r5_core->tcm_banks[i]->pm_domain_id;
690		zynqmp_pm_release_node(pm_domain_id);
691	}
692	return ret;
693}
694
695/*
696 * add_tcm_banks()
697 * @rproc: single R5 core's corresponding rproc instance
698 *
699 * allocate and add remoteproc carveouts for TCM memory based on cluster mode
700 *
701 * return 0 on success, otherwise non-zero value on failure
702 */
703static int add_tcm_banks(struct rproc *rproc)
704{
705	struct zynqmp_r5_cluster *cluster;
706	struct zynqmp_r5_core *r5_core;
707	struct device *dev;
708
709	r5_core = rproc->priv;
710	if (!r5_core)
711		return -EINVAL;
712
713	dev = r5_core->dev;
714
715	cluster = dev_get_drvdata(dev->parent);
716	if (!cluster) {
717		dev_err(dev->parent, "Invalid driver data\n");
718		return -EINVAL;
719	}
720
721	/*
722	 * In lockstep mode TCM banks are one contiguous memory region of 256Kb
723	 * In split mode, each TCM bank is 64Kb and not contiguous.
724	 * We add memory carveouts accordingly.
725	 */
726	if (cluster->mode == SPLIT_MODE)
727		return add_tcm_carveout_split_mode(rproc);
728	else if (cluster->mode == LOCKSTEP_MODE)
729		return add_tcm_carveout_lockstep_mode(rproc);
730
731	return -EINVAL;
732}
733
734/*
735 * zynqmp_r5_parse_fw()
736 * @rproc: single R5 core's corresponding rproc instance
737 * @fw: ptr to firmware to be loaded onto r5 core
738 *
739 * get resource table if available
740 *
741 * return 0 on success, otherwise non-zero value on failure
742 */
743static int zynqmp_r5_parse_fw(struct rproc *rproc, const struct firmware *fw)
744{
745	int ret;
746
747	ret = rproc_elf_load_rsc_table(rproc, fw);
748	if (ret == -EINVAL) {
749		/*
750		 * resource table only required for IPC.
751		 * if not present, this is not necessarily an error;
752		 * for example, loading r5 hello world application
753		 * so simply inform user and keep going.
754		 */
755		dev_info(&rproc->dev, "no resource table found.\n");
756		ret = 0;
757	}
758	return ret;
759}
760
761/**
762 * zynqmp_r5_rproc_prepare()
763 * adds carveouts for TCM bank and reserved memory regions
764 *
765 * @rproc: Device node of each rproc
766 *
767 * Return: 0 for success else < 0 error code
768 */
769static int zynqmp_r5_rproc_prepare(struct rproc *rproc)
770{
771	int ret;
772
773	ret = add_tcm_banks(rproc);
774	if (ret) {
775		dev_err(&rproc->dev, "failed to get TCM banks, err %d\n", ret);
776		return ret;
777	}
778
779	ret = add_mem_regions_carveout(rproc);
780	if (ret) {
781		dev_err(&rproc->dev, "failed to get reserve mem regions %d\n", ret);
782		return ret;
783	}
784
785	return 0;
786}
787
788/**
789 * zynqmp_r5_rproc_unprepare()
790 * Turns off TCM banks using power-domain id
791 *
792 * @rproc: Device node of each rproc
793 *
794 * Return: always 0
795 */
796static int zynqmp_r5_rproc_unprepare(struct rproc *rproc)
797{
798	struct zynqmp_r5_core *r5_core;
799	u32 pm_domain_id;
800	int i;
801
802	r5_core = rproc->priv;
803
804	for (i = 0; i < r5_core->tcm_bank_count; i++) {
805		pm_domain_id = r5_core->tcm_banks[i]->pm_domain_id;
806		if (zynqmp_pm_release_node(pm_domain_id))
807			dev_warn(r5_core->dev,
808				 "can't turn off TCM bank 0x%x", pm_domain_id);
809	}
810
811	return 0;
812}
813
814static const struct rproc_ops zynqmp_r5_rproc_ops = {
815	.prepare	= zynqmp_r5_rproc_prepare,
816	.unprepare	= zynqmp_r5_rproc_unprepare,
817	.start		= zynqmp_r5_rproc_start,
818	.stop		= zynqmp_r5_rproc_stop,
819	.load		= rproc_elf_load_segments,
820	.parse_fw	= zynqmp_r5_parse_fw,
821	.find_loaded_rsc_table = rproc_elf_find_loaded_rsc_table,
822	.sanity_check	= rproc_elf_sanity_check,
823	.get_boot_addr	= rproc_elf_get_boot_addr,
824	.kick		= zynqmp_r5_rproc_kick,
825};
826
827/**
828 * zynqmp_r5_add_rproc_core()
829 * Allocate and add struct rproc object for each r5f core
830 * This is called for each individual r5f core
831 *
832 * @cdev: Device node of each r5 core
833 *
834 * Return: zynqmp_r5_core object for success else error code pointer
835 */
836static struct zynqmp_r5_core *zynqmp_r5_add_rproc_core(struct device *cdev)
837{
838	struct zynqmp_r5_core *r5_core;
839	struct rproc *r5_rproc;
840	int ret;
841
842	/* Set up DMA mask */
843	ret = dma_set_coherent_mask(cdev, DMA_BIT_MASK(32));
844	if (ret)
845		return ERR_PTR(ret);
846
847	/* Allocate remoteproc instance */
848	r5_rproc = rproc_alloc(cdev, dev_name(cdev),
849			       &zynqmp_r5_rproc_ops,
850			       NULL, sizeof(struct zynqmp_r5_core));
851	if (!r5_rproc) {
852		dev_err(cdev, "failed to allocate memory for rproc instance\n");
853		return ERR_PTR(-ENOMEM);
854	}
855
856	r5_rproc->auto_boot = false;
857	r5_core = r5_rproc->priv;
858	r5_core->dev = cdev;
859	r5_core->np = dev_of_node(cdev);
860	if (!r5_core->np) {
861		dev_err(cdev, "can't get device node for r5 core\n");
862		ret = -EINVAL;
863		goto free_rproc;
864	}
865
866	/* Add R5 remoteproc core */
867	ret = rproc_add(r5_rproc);
868	if (ret) {
869		dev_err(cdev, "failed to add r5 remoteproc\n");
870		goto free_rproc;
871	}
872
873	r5_core->rproc = r5_rproc;
874	return r5_core;
875
876free_rproc:
877	rproc_free(r5_rproc);
878	return ERR_PTR(ret);
879}
880
881/**
882 * zynqmp_r5_get_tcm_node()
883 * Ideally this function should parse tcm node and store information
884 * in r5_core instance. For now, Hardcoded TCM information is used.
885 * This approach is used as TCM bindings for system-dt is being developed
886 *
887 * @cluster: pointer to zynqmp_r5_cluster type object
888 *
889 * Return: 0 for success and < 0 error code for failure.
890 */
891static int zynqmp_r5_get_tcm_node(struct zynqmp_r5_cluster *cluster)
892{
893	const struct mem_bank_data *zynqmp_tcm_banks;
894	struct device *dev = cluster->dev;
895	struct zynqmp_r5_core *r5_core;
896	int tcm_bank_count, tcm_node;
897	int i, j;
898
899	if (cluster->mode == SPLIT_MODE) {
900		zynqmp_tcm_banks = zynqmp_tcm_banks_split;
901		tcm_bank_count = ARRAY_SIZE(zynqmp_tcm_banks_split);
902	} else {
903		zynqmp_tcm_banks = zynqmp_tcm_banks_lockstep;
904		tcm_bank_count = ARRAY_SIZE(zynqmp_tcm_banks_lockstep);
905	}
906
907	/* count per core tcm banks */
908	tcm_bank_count = tcm_bank_count / cluster->core_count;
909
910	/*
911	 * r5 core 0 will use all of TCM banks in lockstep mode.
912	 * In split mode, r5 core0 will use 128k and r5 core1 will use another
913	 * 128k. Assign TCM banks to each core accordingly
914	 */
915	tcm_node = 0;
916	for (i = 0; i < cluster->core_count; i++) {
917		r5_core = cluster->r5_cores[i];
918		r5_core->tcm_banks = devm_kcalloc(dev, tcm_bank_count,
919						  sizeof(struct mem_bank_data *),
920						  GFP_KERNEL);
921		if (!r5_core->tcm_banks)
922			return -ENOMEM;
923
924		for (j = 0; j < tcm_bank_count; j++) {
925			/*
926			 * Use pre-defined TCM reg values.
927			 * Eventually this should be replaced by values
928			 * parsed from dts.
929			 */
930			r5_core->tcm_banks[j] =
931				(struct mem_bank_data *)&zynqmp_tcm_banks[tcm_node];
932			tcm_node++;
933		}
934
935		r5_core->tcm_bank_count = tcm_bank_count;
936	}
937
938	return 0;
939}
940
941/*
942 * zynqmp_r5_core_init()
943 * Create and initialize zynqmp_r5_core type object
944 *
945 * @cluster: pointer to zynqmp_r5_cluster type object
946 * @fw_reg_val: value expected by firmware to configure RPU cluster mode
947 * @tcm_mode: value expected by fw to configure TCM mode (lockstep or split)
948 *
949 * Return: 0 for success and error code for failure.
950 */
951static int zynqmp_r5_core_init(struct zynqmp_r5_cluster *cluster,
952			       enum rpu_oper_mode fw_reg_val,
953			       enum rpu_tcm_comb tcm_mode)
954{
955	struct device *dev = cluster->dev;
956	struct zynqmp_r5_core *r5_core;
957	int ret, i;
958
959	ret = zynqmp_r5_get_tcm_node(cluster);
960	if (ret < 0) {
961		dev_err(dev, "can't get tcm node, err %d\n", ret);
962		return ret;
963	}
964
965	for (i = 0; i < cluster->core_count; i++) {
966		r5_core = cluster->r5_cores[i];
967
968		/* Initialize r5 cores with power-domains parsed from dts */
969		ret = of_property_read_u32_index(r5_core->np, "power-domains",
970						 1, &r5_core->pm_domain_id);
971		if (ret) {
972			dev_err(dev, "failed to get power-domains property\n");
973			return ret;
974		}
975
976		ret = zynqmp_r5_set_mode(r5_core, fw_reg_val, tcm_mode);
977		if (ret) {
978			dev_err(dev, "failed to set r5 cluster mode %d, err %d\n",
979				cluster->mode, ret);
980			return ret;
981		}
982	}
983
984	return 0;
985}
986
987/*
988 * zynqmp_r5_cluster_init()
989 * Create and initialize zynqmp_r5_cluster type object
990 *
991 * @cluster: pointer to zynqmp_r5_cluster type object
992 *
993 * Return: 0 for success and error code for failure.
994 */
995static int zynqmp_r5_cluster_init(struct zynqmp_r5_cluster *cluster)
996{
997	enum zynqmp_r5_cluster_mode cluster_mode = LOCKSTEP_MODE;
998	struct device *dev = cluster->dev;
999	struct device_node *dev_node = dev_of_node(dev);
1000	struct platform_device *child_pdev;
1001	struct zynqmp_r5_core **r5_cores;
1002	enum rpu_oper_mode fw_reg_val;
1003	struct device **child_devs;
1004	struct device_node *child;
1005	enum rpu_tcm_comb tcm_mode;
1006	int core_count, ret, i;
1007	struct mbox_info *ipi;
1008
1009	ret = of_property_read_u32(dev_node, "xlnx,cluster-mode", &cluster_mode);
1010
1011	/*
1012	 * on success returns 0, if not defined then returns -EINVAL,
1013	 * In that case, default is LOCKSTEP mode. Other than that
1014	 * returns relative error code < 0.
1015	 */
1016	if (ret != -EINVAL && ret != 0) {
1017		dev_err(dev, "Invalid xlnx,cluster-mode property\n");
1018		return ret;
1019	}
1020
1021	/*
1022	 * For now driver only supports split mode and lockstep mode.
1023	 * fail driver probe if either of that is not set in dts.
1024	 */
1025	if (cluster_mode == LOCKSTEP_MODE) {
1026		tcm_mode = PM_RPU_TCM_COMB;
1027		fw_reg_val = PM_RPU_MODE_LOCKSTEP;
1028	} else if (cluster_mode == SPLIT_MODE) {
1029		tcm_mode = PM_RPU_TCM_SPLIT;
1030		fw_reg_val = PM_RPU_MODE_SPLIT;
1031	} else {
1032		dev_err(dev, "driver does not support cluster mode %d\n", cluster_mode);
1033		return -EINVAL;
1034	}
1035
1036	/*
1037	 * Number of cores is decided by number of child nodes of
1038	 * r5f subsystem node in dts. If Split mode is used in dts
1039	 * 2 child nodes are expected.
1040	 * In lockstep mode if two child nodes are available,
1041	 * only use first child node and consider it as core0
1042	 * and ignore core1 dt node.
1043	 */
1044	core_count = of_get_available_child_count(dev_node);
1045	if (core_count == 0) {
1046		dev_err(dev, "Invalid number of r5 cores %d", core_count);
1047		return -EINVAL;
1048	} else if (cluster_mode == SPLIT_MODE && core_count != 2) {
1049		dev_err(dev, "Invalid number of r5 cores for split mode\n");
1050		return -EINVAL;
1051	} else if (cluster_mode == LOCKSTEP_MODE && core_count == 2) {
1052		dev_warn(dev, "Only r5 core0 will be used\n");
1053		core_count = 1;
1054	}
1055
1056	child_devs = kcalloc(core_count, sizeof(struct device *), GFP_KERNEL);
1057	if (!child_devs)
1058		return -ENOMEM;
1059
1060	r5_cores = kcalloc(core_count,
1061			   sizeof(struct zynqmp_r5_core *), GFP_KERNEL);
1062	if (!r5_cores) {
1063		kfree(child_devs);
1064		return -ENOMEM;
1065	}
1066
1067	i = 0;
1068	for_each_available_child_of_node(dev_node, child) {
1069		child_pdev = of_find_device_by_node(child);
1070		if (!child_pdev) {
1071			of_node_put(child);
1072			ret = -ENODEV;
1073			goto release_r5_cores;
1074		}
1075
1076		child_devs[i] = &child_pdev->dev;
1077
1078		/* create and add remoteproc instance of type struct rproc */
1079		r5_cores[i] = zynqmp_r5_add_rproc_core(&child_pdev->dev);
1080		if (IS_ERR(r5_cores[i])) {
1081			of_node_put(child);
1082			ret = PTR_ERR(r5_cores[i]);
1083			r5_cores[i] = NULL;
1084			goto release_r5_cores;
1085		}
1086
1087		/*
1088		 * If mailbox nodes are disabled using "status" property then
1089		 * setting up mailbox channels will fail.
1090		 */
1091		ipi = zynqmp_r5_setup_mbox(&child_pdev->dev);
1092		if (ipi) {
1093			r5_cores[i]->ipi = ipi;
1094			ipi->r5_core = r5_cores[i];
1095		}
1096
1097		/*
1098		 * If two child nodes are available in dts in lockstep mode,
1099		 * then ignore second child node.
1100		 */
1101		if (cluster_mode == LOCKSTEP_MODE) {
1102			of_node_put(child);
1103			break;
1104		}
1105
1106		i++;
1107	}
1108
1109	cluster->mode = cluster_mode;
1110	cluster->core_count = core_count;
1111	cluster->r5_cores = r5_cores;
1112
1113	ret = zynqmp_r5_core_init(cluster, fw_reg_val, tcm_mode);
1114	if (ret < 0) {
1115		dev_err(dev, "failed to init r5 core err %d\n", ret);
1116		cluster->core_count = 0;
1117		cluster->r5_cores = NULL;
1118
1119		/*
1120		 * at this point rproc resources for each core are allocated.
1121		 * adjust index to free resources in reverse order
1122		 */
1123		i = core_count - 1;
1124		goto release_r5_cores;
1125	}
1126
1127	kfree(child_devs);
1128	return 0;
1129
1130release_r5_cores:
1131	while (i >= 0) {
1132		put_device(child_devs[i]);
1133		if (r5_cores[i]) {
1134			zynqmp_r5_free_mbox(r5_cores[i]->ipi);
1135			of_reserved_mem_device_release(r5_cores[i]->dev);
1136			rproc_del(r5_cores[i]->rproc);
1137			rproc_free(r5_cores[i]->rproc);
1138		}
1139		i--;
1140	}
1141	kfree(r5_cores);
1142	kfree(child_devs);
1143	return ret;
1144}
1145
1146static void zynqmp_r5_cluster_exit(void *data)
1147{
1148	struct platform_device *pdev = data;
1149	struct zynqmp_r5_cluster *cluster;
1150	struct zynqmp_r5_core *r5_core;
1151	int i;
1152
1153	cluster = platform_get_drvdata(pdev);
1154	if (!cluster)
1155		return;
1156
1157	for (i = 0; i < cluster->core_count; i++) {
1158		r5_core = cluster->r5_cores[i];
1159		zynqmp_r5_free_mbox(r5_core->ipi);
1160		of_reserved_mem_device_release(r5_core->dev);
1161		put_device(r5_core->dev);
1162		rproc_del(r5_core->rproc);
1163		rproc_free(r5_core->rproc);
1164	}
1165
1166	kfree(cluster->r5_cores);
1167	kfree(cluster);
1168	platform_set_drvdata(pdev, NULL);
1169}
1170
1171/*
1172 * zynqmp_r5_remoteproc_probe()
1173 * parse device-tree, initialize hardware and allocate required resources
1174 * and remoteproc ops
1175 *
1176 * @pdev: domain platform device for R5 cluster
1177 *
1178 * Return: 0 for success and < 0 for failure.
1179 */
1180static int zynqmp_r5_remoteproc_probe(struct platform_device *pdev)
1181{
1182	struct zynqmp_r5_cluster *cluster;
1183	struct device *dev = &pdev->dev;
1184	int ret;
1185
1186	cluster = kzalloc(sizeof(*cluster), GFP_KERNEL);
1187	if (!cluster)
1188		return -ENOMEM;
1189
1190	cluster->dev = dev;
1191
1192	ret = devm_of_platform_populate(dev);
1193	if (ret) {
1194		dev_err_probe(dev, ret, "failed to populate platform dev\n");
1195		kfree(cluster);
1196		return ret;
1197	}
1198
1199	/* wire in so each core can be cleaned up at driver remove */
1200	platform_set_drvdata(pdev, cluster);
1201
1202	ret = zynqmp_r5_cluster_init(cluster);
1203	if (ret) {
1204		kfree(cluster);
1205		platform_set_drvdata(pdev, NULL);
1206		dev_err_probe(dev, ret, "Invalid r5f subsystem device tree\n");
1207		return ret;
1208	}
1209
1210	ret = devm_add_action_or_reset(dev, zynqmp_r5_cluster_exit, pdev);
1211	if (ret)
1212		return ret;
1213
1214	return 0;
1215}
1216
1217/* Match table for OF platform binding */
1218static const struct of_device_id zynqmp_r5_remoteproc_match[] = {
1219	{ .compatible = "xlnx,zynqmp-r5fss", },
1220	{ /* end of list */ },
1221};
1222MODULE_DEVICE_TABLE(of, zynqmp_r5_remoteproc_match);
1223
1224static struct platform_driver zynqmp_r5_remoteproc_driver = {
1225	.probe = zynqmp_r5_remoteproc_probe,
1226	.driver = {
1227		.name = "zynqmp_r5_remoteproc",
1228		.of_match_table = zynqmp_r5_remoteproc_match,
1229	},
1230};
1231module_platform_driver(zynqmp_r5_remoteproc_driver);
1232
1233MODULE_DESCRIPTION("Xilinx R5F remote processor driver");
1234MODULE_AUTHOR("Xilinx Inc.");
1235MODULE_LICENSE("GPL");
1236