1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright(C) 2016 Linaro Limited. All rights reserved.
4  * Author: Mathieu Poirier <mathieu.poirier@linaro.org>
5  */
6 
7 #include <linux/atomic.h>
8 #include <linux/coresight.h>
9 #include <linux/dma-mapping.h>
10 #include <linux/iommu.h>
11 #include <linux/idr.h>
12 #include <linux/mutex.h>
13 #include <linux/refcount.h>
14 #include <linux/slab.h>
15 #include <linux/types.h>
16 #include <linux/vmalloc.h>
17 #include "coresight-catu.h"
18 #include "coresight-etm-perf.h"
19 #include "coresight-priv.h"
20 #include "coresight-tmc.h"
21 
22 struct etr_flat_buf {
23 	struct device	*dev;
24 	dma_addr_t	daddr;
25 	void		*vaddr;
26 	size_t		size;
27 };
28 
29 struct etr_buf_hw {
30 	bool	has_iommu;
31 	bool	has_etr_sg;
32 	bool	has_catu;
33 	bool	has_resrv;
34 };
35 
36 /*
37  * etr_perf_buffer - Perf buffer used for ETR
38  * @drvdata		- The ETR drvdaga this buffer has been allocated for.
39  * @etr_buf		- Actual buffer used by the ETR
40  * @pid			- The PID of the session owner that etr_perf_buffer
41  *			  belongs to.
42  * @snaphost		- Perf session mode
43  * @nr_pages		- Number of pages in the ring buffer.
44  * @pages		- Array of Pages in the ring buffer.
45  */
46 struct etr_perf_buffer {
47 	struct tmc_drvdata	*drvdata;
48 	struct etr_buf		*etr_buf;
49 	pid_t			pid;
50 	bool			snapshot;
51 	int			nr_pages;
52 	void			**pages;
53 };
54 
55 /* Convert the perf index to an offset within the ETR buffer */
56 #define PERF_IDX2OFF(idx, buf)		\
57 		((idx) % ((unsigned long)(buf)->nr_pages << PAGE_SHIFT))
58 
59 /* Lower limit for ETR hardware buffer */
60 #define TMC_ETR_PERF_MIN_BUF_SIZE	SZ_1M
61 
62 /*
63  * The TMC ETR SG has a page size of 4K. The SG table contains pointers
64  * to 4KB buffers. However, the OS may use a PAGE_SIZE different from
65  * 4K (i.e, 16KB or 64KB). This implies that a single OS page could
66  * contain more than one SG buffer and tables.
67  *
68  * A table entry has the following format:
69  *
70  * ---Bit31------------Bit4-------Bit1-----Bit0--
71  * |     Address[39:12]    | SBZ |  Entry Type  |
72  * ----------------------------------------------
73  *
74  * Address: Bits [39:12] of a physical page address. Bits [11:0] are
75  *	    always zero.
76  *
77  * Entry type:
78  *	b00 - Reserved.
79  *	b01 - Last entry in the tables, points to 4K page buffer.
80  *	b10 - Normal entry, points to 4K page buffer.
81  *	b11 - Link. The address points to the base of next table.
82  */
83 
84 typedef u32 sgte_t;
85 
86 #define ETR_SG_PAGE_SHIFT		12
87 #define ETR_SG_PAGE_SIZE		(1UL << ETR_SG_PAGE_SHIFT)
88 #define ETR_SG_PAGES_PER_SYSPAGE	(PAGE_SIZE / ETR_SG_PAGE_SIZE)
89 #define ETR_SG_PTRS_PER_PAGE		(ETR_SG_PAGE_SIZE / sizeof(sgte_t))
90 #define ETR_SG_PTRS_PER_SYSPAGE		(PAGE_SIZE / sizeof(sgte_t))
91 
92 #define ETR_SG_ET_MASK			0x3
93 #define ETR_SG_ET_LAST			0x1
94 #define ETR_SG_ET_NORMAL		0x2
95 #define ETR_SG_ET_LINK			0x3
96 
97 #define ETR_SG_ADDR_SHIFT		4
98 
99 #define ETR_SG_ENTRY(addr, type) \
100 	(sgte_t)((((addr) >> ETR_SG_PAGE_SHIFT) << ETR_SG_ADDR_SHIFT) | \
101 		 (type & ETR_SG_ET_MASK))
102 
103 #define ETR_SG_ADDR(entry) \
104 	(((dma_addr_t)(entry) >> ETR_SG_ADDR_SHIFT) << ETR_SG_PAGE_SHIFT)
105 #define ETR_SG_ET(entry)		((entry) & ETR_SG_ET_MASK)
106 
107 /*
108  * struct etr_sg_table : ETR SG Table
109  * @sg_table:		Generic SG Table holding the data/table pages.
110  * @hwaddr:		hwaddress used by the TMC, which is the base
111  *			address of the table.
112  */
113 struct etr_sg_table {
114 	struct tmc_sg_table	*sg_table;
115 	dma_addr_t		hwaddr;
116 };
117 
118 /*
119  * tmc_etr_sg_table_entries: Total number of table entries required to map
120  * @nr_pages system pages.
121  *
122  * We need to map @nr_pages * ETR_SG_PAGES_PER_SYSPAGE data pages.
123  * Each TMC page can map (ETR_SG_PTRS_PER_PAGE - 1) buffer pointers,
124  * with the last entry pointing to another page of table entries.
125  * If we spill over to a new page for mapping 1 entry, we could as
126  * well replace the link entry of the previous page with the last entry.
127  */
128 static inline unsigned long __attribute_const__
tmc_etr_sg_table_entries(int nr_pages)129 tmc_etr_sg_table_entries(int nr_pages)
130 {
131 	unsigned long nr_sgpages = nr_pages * ETR_SG_PAGES_PER_SYSPAGE;
132 	unsigned long nr_sglinks = nr_sgpages / (ETR_SG_PTRS_PER_PAGE - 1);
133 	/*
134 	 * If we spill over to a new page for 1 entry, we could as well
135 	 * make it the LAST entry in the previous page, skipping the Link
136 	 * address.
137 	 */
138 	if (nr_sglinks && (nr_sgpages % (ETR_SG_PTRS_PER_PAGE - 1) < 2))
139 		nr_sglinks--;
140 	return nr_sgpages + nr_sglinks;
141 }
142 
143 /*
144  * tmc_pages_get_offset:  Go through all the pages in the tmc_pages
145  * and map the device address @addr to an offset within the virtual
146  * contiguous buffer.
147  */
148 static long
tmc_pages_get_offset(struct tmc_pages * tmc_pages,dma_addr_t addr)149 tmc_pages_get_offset(struct tmc_pages *tmc_pages, dma_addr_t addr)
150 {
151 	int i;
152 	dma_addr_t page_start;
153 
154 	for (i = 0; i < tmc_pages->nr_pages; i++) {
155 		page_start = tmc_pages->daddrs[i];
156 		if (addr >= page_start && addr < (page_start + PAGE_SIZE))
157 			return i * PAGE_SIZE + (addr - page_start);
158 	}
159 
160 	return -EINVAL;
161 }
162 
163 /*
164  * tmc_pages_free : Unmap and free the pages used by tmc_pages.
165  * If the pages were not allocated in tmc_pages_alloc(), we would
166  * simply drop the refcount.
167  */
tmc_pages_free(struct tmc_pages * tmc_pages,struct device * dev,enum dma_data_direction dir)168 static void tmc_pages_free(struct tmc_pages *tmc_pages,
169 			   struct device *dev, enum dma_data_direction dir)
170 {
171 	int i;
172 	struct device *real_dev = dev->parent;
173 
174 	for (i = 0; i < tmc_pages->nr_pages; i++) {
175 		if (tmc_pages->daddrs && tmc_pages->daddrs[i])
176 			dma_unmap_page(real_dev, tmc_pages->daddrs[i],
177 					 PAGE_SIZE, dir);
178 		if (tmc_pages->pages && tmc_pages->pages[i])
179 			__free_page(tmc_pages->pages[i]);
180 	}
181 
182 	kfree(tmc_pages->pages);
183 	kfree(tmc_pages->daddrs);
184 	tmc_pages->pages = NULL;
185 	tmc_pages->daddrs = NULL;
186 	tmc_pages->nr_pages = 0;
187 }
188 
189 /*
190  * tmc_pages_alloc : Allocate and map pages for a given @tmc_pages.
191  * If @pages is not NULL, the list of page virtual addresses are
192  * used as the data pages. The pages are then dma_map'ed for @dev
193  * with dma_direction @dir.
194  *
195  * Returns 0 upon success, else the error number.
196  */
tmc_pages_alloc(struct tmc_pages * tmc_pages,struct device * dev,int node,enum dma_data_direction dir,void ** pages)197 static int tmc_pages_alloc(struct tmc_pages *tmc_pages,
198 			   struct device *dev, int node,
199 			   enum dma_data_direction dir, void **pages)
200 {
201 	int i, nr_pages;
202 	dma_addr_t paddr;
203 	struct page *page;
204 	struct device *real_dev = dev->parent;
205 
206 	nr_pages = tmc_pages->nr_pages;
207 	tmc_pages->daddrs = kcalloc(nr_pages, sizeof(*tmc_pages->daddrs),
208 					 GFP_KERNEL);
209 	if (!tmc_pages->daddrs)
210 		return -ENOMEM;
211 	tmc_pages->pages = kcalloc(nr_pages, sizeof(*tmc_pages->pages),
212 					 GFP_KERNEL);
213 	if (!tmc_pages->pages) {
214 		kfree(tmc_pages->daddrs);
215 		tmc_pages->daddrs = NULL;
216 		return -ENOMEM;
217 	}
218 
219 	for (i = 0; i < nr_pages; i++) {
220 		if (pages && pages[i]) {
221 			page = virt_to_page(pages[i]);
222 			/* Hold a refcount on the page */
223 			get_page(page);
224 		} else {
225 			page = alloc_pages_node(node,
226 						GFP_KERNEL | __GFP_ZERO, 0);
227 			if (!page)
228 				goto err;
229 		}
230 		paddr = dma_map_page(real_dev, page, 0, PAGE_SIZE, dir);
231 		if (dma_mapping_error(real_dev, paddr))
232 			goto err;
233 		tmc_pages->daddrs[i] = paddr;
234 		tmc_pages->pages[i] = page;
235 	}
236 	return 0;
237 err:
238 	tmc_pages_free(tmc_pages, dev, dir);
239 	return -ENOMEM;
240 }
241 
242 static inline long
tmc_sg_get_data_page_offset(struct tmc_sg_table * sg_table,dma_addr_t addr)243 tmc_sg_get_data_page_offset(struct tmc_sg_table *sg_table, dma_addr_t addr)
244 {
245 	return tmc_pages_get_offset(&sg_table->data_pages, addr);
246 }
247 
tmc_free_table_pages(struct tmc_sg_table * sg_table)248 static inline void tmc_free_table_pages(struct tmc_sg_table *sg_table)
249 {
250 	if (sg_table->table_vaddr)
251 		vunmap(sg_table->table_vaddr);
252 	tmc_pages_free(&sg_table->table_pages, sg_table->dev, DMA_TO_DEVICE);
253 }
254 
tmc_free_data_pages(struct tmc_sg_table * sg_table)255 static void tmc_free_data_pages(struct tmc_sg_table *sg_table)
256 {
257 	if (sg_table->data_vaddr)
258 		vunmap(sg_table->data_vaddr);
259 	tmc_pages_free(&sg_table->data_pages, sg_table->dev, DMA_FROM_DEVICE);
260 }
261 
tmc_free_sg_table(struct tmc_sg_table * sg_table)262 void tmc_free_sg_table(struct tmc_sg_table *sg_table)
263 {
264 	tmc_free_table_pages(sg_table);
265 	tmc_free_data_pages(sg_table);
266 	kfree(sg_table);
267 }
268 EXPORT_SYMBOL_GPL(tmc_free_sg_table);
269 
270 /*
271  * Alloc pages for the table. Since this will be used by the device,
272  * allocate the pages closer to the device (i.e, dev_to_node(dev)
273  * rather than the CPU node).
274  */
tmc_alloc_table_pages(struct tmc_sg_table * sg_table)275 static int tmc_alloc_table_pages(struct tmc_sg_table *sg_table)
276 {
277 	int rc;
278 	struct tmc_pages *table_pages = &sg_table->table_pages;
279 
280 	rc = tmc_pages_alloc(table_pages, sg_table->dev,
281 			     dev_to_node(sg_table->dev),
282 			     DMA_TO_DEVICE, NULL);
283 	if (rc)
284 		return rc;
285 	sg_table->table_vaddr = vmap(table_pages->pages,
286 				     table_pages->nr_pages,
287 				     VM_MAP,
288 				     PAGE_KERNEL);
289 	if (!sg_table->table_vaddr)
290 		rc = -ENOMEM;
291 	else
292 		sg_table->table_daddr = table_pages->daddrs[0];
293 	return rc;
294 }
295 
tmc_alloc_data_pages(struct tmc_sg_table * sg_table,void ** pages)296 static int tmc_alloc_data_pages(struct tmc_sg_table *sg_table, void **pages)
297 {
298 	int rc;
299 
300 	/* Allocate data pages on the node requested by the caller */
301 	rc = tmc_pages_alloc(&sg_table->data_pages,
302 			     sg_table->dev, sg_table->node,
303 			     DMA_FROM_DEVICE, pages);
304 	if (!rc) {
305 		sg_table->data_vaddr = vmap(sg_table->data_pages.pages,
306 					    sg_table->data_pages.nr_pages,
307 					    VM_MAP,
308 					    PAGE_KERNEL);
309 		if (!sg_table->data_vaddr)
310 			rc = -ENOMEM;
311 	}
312 	return rc;
313 }
314 
315 /*
316  * tmc_alloc_sg_table: Allocate and setup dma pages for the TMC SG table
317  * and data buffers. TMC writes to the data buffers and reads from the SG
318  * Table pages.
319  *
320  * @dev		- Coresight device to which page should be DMA mapped.
321  * @node	- Numa node for mem allocations
322  * @nr_tpages	- Number of pages for the table entries.
323  * @nr_dpages	- Number of pages for Data buffer.
324  * @pages	- Optional list of virtual address of pages.
325  */
tmc_alloc_sg_table(struct device * dev,int node,int nr_tpages,int nr_dpages,void ** pages)326 struct tmc_sg_table *tmc_alloc_sg_table(struct device *dev,
327 					int node,
328 					int nr_tpages,
329 					int nr_dpages,
330 					void **pages)
331 {
332 	long rc;
333 	struct tmc_sg_table *sg_table;
334 
335 	sg_table = kzalloc(sizeof(*sg_table), GFP_KERNEL);
336 	if (!sg_table)
337 		return ERR_PTR(-ENOMEM);
338 	sg_table->data_pages.nr_pages = nr_dpages;
339 	sg_table->table_pages.nr_pages = nr_tpages;
340 	sg_table->node = node;
341 	sg_table->dev = dev;
342 
343 	rc  = tmc_alloc_data_pages(sg_table, pages);
344 	if (!rc)
345 		rc = tmc_alloc_table_pages(sg_table);
346 	if (rc) {
347 		tmc_free_sg_table(sg_table);
348 		return ERR_PTR(rc);
349 	}
350 
351 	return sg_table;
352 }
353 EXPORT_SYMBOL_GPL(tmc_alloc_sg_table);
354 
355 /*
356  * tmc_sg_table_sync_data_range: Sync the data buffer written
357  * by the device from @offset upto a @size bytes.
358  */
tmc_sg_table_sync_data_range(struct tmc_sg_table * table,u64 offset,u64 size)359 void tmc_sg_table_sync_data_range(struct tmc_sg_table *table,
360 				  u64 offset, u64 size)
361 {
362 	int i, index, start;
363 	int npages = DIV_ROUND_UP(size, PAGE_SIZE);
364 	struct device *real_dev = table->dev->parent;
365 	struct tmc_pages *data = &table->data_pages;
366 
367 	start = offset >> PAGE_SHIFT;
368 	for (i = start; i < (start + npages); i++) {
369 		index = i % data->nr_pages;
370 		dma_sync_single_for_cpu(real_dev, data->daddrs[index],
371 					PAGE_SIZE, DMA_FROM_DEVICE);
372 	}
373 }
374 EXPORT_SYMBOL_GPL(tmc_sg_table_sync_data_range);
375 
376 /* tmc_sg_sync_table: Sync the page table */
tmc_sg_table_sync_table(struct tmc_sg_table * sg_table)377 void tmc_sg_table_sync_table(struct tmc_sg_table *sg_table)
378 {
379 	int i;
380 	struct device *real_dev = sg_table->dev->parent;
381 	struct tmc_pages *table_pages = &sg_table->table_pages;
382 
383 	for (i = 0; i < table_pages->nr_pages; i++)
384 		dma_sync_single_for_device(real_dev, table_pages->daddrs[i],
385 					   PAGE_SIZE, DMA_TO_DEVICE);
386 }
387 EXPORT_SYMBOL_GPL(tmc_sg_table_sync_table);
388 
389 /*
390  * tmc_sg_table_get_data: Get the buffer pointer for data @offset
391  * in the SG buffer. The @bufpp is updated to point to the buffer.
392  * Returns :
393  *	the length of linear data available at @offset.
394  *	or
395  *	<= 0 if no data is available.
396  */
tmc_sg_table_get_data(struct tmc_sg_table * sg_table,u64 offset,size_t len,char ** bufpp)397 ssize_t tmc_sg_table_get_data(struct tmc_sg_table *sg_table,
398 			      u64 offset, size_t len, char **bufpp)
399 {
400 	size_t size;
401 	int pg_idx = offset >> PAGE_SHIFT;
402 	int pg_offset = offset & (PAGE_SIZE - 1);
403 	struct tmc_pages *data_pages = &sg_table->data_pages;
404 
405 	size = tmc_sg_table_buf_size(sg_table);
406 	if (offset >= size)
407 		return -EINVAL;
408 
409 	/* Make sure we don't go beyond the end */
410 	len = (len < (size - offset)) ? len : size - offset;
411 	/* Respect the page boundaries */
412 	len = (len < (PAGE_SIZE - pg_offset)) ? len : (PAGE_SIZE - pg_offset);
413 	if (len > 0)
414 		*bufpp = page_address(data_pages->pages[pg_idx]) + pg_offset;
415 	return len;
416 }
417 EXPORT_SYMBOL_GPL(tmc_sg_table_get_data);
418 
419 #ifdef ETR_SG_DEBUG
420 /* Map a dma address to virtual address */
421 static unsigned long
tmc_sg_daddr_to_vaddr(struct tmc_sg_table * sg_table,dma_addr_t addr,bool table)422 tmc_sg_daddr_to_vaddr(struct tmc_sg_table *sg_table,
423 		      dma_addr_t addr, bool table)
424 {
425 	long offset;
426 	unsigned long base;
427 	struct tmc_pages *tmc_pages;
428 
429 	if (table) {
430 		tmc_pages = &sg_table->table_pages;
431 		base = (unsigned long)sg_table->table_vaddr;
432 	} else {
433 		tmc_pages = &sg_table->data_pages;
434 		base = (unsigned long)sg_table->data_vaddr;
435 	}
436 
437 	offset = tmc_pages_get_offset(tmc_pages, addr);
438 	if (offset < 0)
439 		return 0;
440 	return base + offset;
441 }
442 
443 /* Dump the given sg_table */
tmc_etr_sg_table_dump(struct etr_sg_table * etr_table)444 static void tmc_etr_sg_table_dump(struct etr_sg_table *etr_table)
445 {
446 	sgte_t *ptr;
447 	int i = 0;
448 	dma_addr_t addr;
449 	struct tmc_sg_table *sg_table = etr_table->sg_table;
450 
451 	ptr = (sgte_t *)tmc_sg_daddr_to_vaddr(sg_table,
452 					      etr_table->hwaddr, true);
453 	while (ptr) {
454 		addr = ETR_SG_ADDR(*ptr);
455 		switch (ETR_SG_ET(*ptr)) {
456 		case ETR_SG_ET_NORMAL:
457 			dev_dbg(sg_table->dev,
458 				"%05d: %p\t:[N] 0x%llx\n", i, ptr, addr);
459 			ptr++;
460 			break;
461 		case ETR_SG_ET_LINK:
462 			dev_dbg(sg_table->dev,
463 				"%05d: *** %p\t:{L} 0x%llx ***\n",
464 				 i, ptr, addr);
465 			ptr = (sgte_t *)tmc_sg_daddr_to_vaddr(sg_table,
466 							      addr, true);
467 			break;
468 		case ETR_SG_ET_LAST:
469 			dev_dbg(sg_table->dev,
470 				"%05d: ### %p\t:[L] 0x%llx ###\n",
471 				 i, ptr, addr);
472 			return;
473 		default:
474 			dev_dbg(sg_table->dev,
475 				"%05d: xxx %p\t:[INVALID] 0x%llx xxx\n",
476 				 i, ptr, addr);
477 			return;
478 		}
479 		i++;
480 	}
481 	dev_dbg(sg_table->dev, "******* End of Table *****\n");
482 }
483 #else
tmc_etr_sg_table_dump(struct etr_sg_table * etr_table)484 static inline void tmc_etr_sg_table_dump(struct etr_sg_table *etr_table) {}
485 #endif
486 
487 /*
488  * Populate the SG Table page table entries from table/data
489  * pages allocated. Each Data page has ETR_SG_PAGES_PER_SYSPAGE SG pages.
490  * So does a Table page. So we keep track of indices of the tables
491  * in each system page and move the pointers accordingly.
492  */
493 #define INC_IDX_ROUND(idx, size) ((idx) = ((idx) + 1) % (size))
tmc_etr_sg_table_populate(struct etr_sg_table * etr_table)494 static void tmc_etr_sg_table_populate(struct etr_sg_table *etr_table)
495 {
496 	dma_addr_t paddr;
497 	int i, type, nr_entries;
498 	int tpidx = 0; /* index to the current system table_page */
499 	int sgtidx = 0;	/* index to the sg_table within the current syspage */
500 	int sgtentry = 0; /* the entry within the sg_table */
501 	int dpidx = 0; /* index to the current system data_page */
502 	int spidx = 0; /* index to the SG page within the current data page */
503 	sgte_t *ptr; /* pointer to the table entry to fill */
504 	struct tmc_sg_table *sg_table = etr_table->sg_table;
505 	dma_addr_t *table_daddrs = sg_table->table_pages.daddrs;
506 	dma_addr_t *data_daddrs = sg_table->data_pages.daddrs;
507 
508 	nr_entries = tmc_etr_sg_table_entries(sg_table->data_pages.nr_pages);
509 	/*
510 	 * Use the contiguous virtual address of the table to update entries.
511 	 */
512 	ptr = sg_table->table_vaddr;
513 	/*
514 	 * Fill all the entries, except the last entry to avoid special
515 	 * checks within the loop.
516 	 */
517 	for (i = 0; i < nr_entries - 1; i++) {
518 		if (sgtentry == ETR_SG_PTRS_PER_PAGE - 1) {
519 			/*
520 			 * Last entry in a sg_table page is a link address to
521 			 * the next table page. If this sg_table is the last
522 			 * one in the system page, it links to the first
523 			 * sg_table in the next system page. Otherwise, it
524 			 * links to the next sg_table page within the system
525 			 * page.
526 			 */
527 			if (sgtidx == ETR_SG_PAGES_PER_SYSPAGE - 1) {
528 				paddr = table_daddrs[tpidx + 1];
529 			} else {
530 				paddr = table_daddrs[tpidx] +
531 					(ETR_SG_PAGE_SIZE * (sgtidx + 1));
532 			}
533 			type = ETR_SG_ET_LINK;
534 		} else {
535 			/*
536 			 * Update the indices to the data_pages to point to the
537 			 * next sg_page in the data buffer.
538 			 */
539 			type = ETR_SG_ET_NORMAL;
540 			paddr = data_daddrs[dpidx] + spidx * ETR_SG_PAGE_SIZE;
541 			if (!INC_IDX_ROUND(spidx, ETR_SG_PAGES_PER_SYSPAGE))
542 				dpidx++;
543 		}
544 		*ptr++ = ETR_SG_ENTRY(paddr, type);
545 		/*
546 		 * Move to the next table pointer, moving the table page index
547 		 * if necessary
548 		 */
549 		if (!INC_IDX_ROUND(sgtentry, ETR_SG_PTRS_PER_PAGE)) {
550 			if (!INC_IDX_ROUND(sgtidx, ETR_SG_PAGES_PER_SYSPAGE))
551 				tpidx++;
552 		}
553 	}
554 
555 	/* Set up the last entry, which is always a data pointer */
556 	paddr = data_daddrs[dpidx] + spidx * ETR_SG_PAGE_SIZE;
557 	*ptr++ = ETR_SG_ENTRY(paddr, ETR_SG_ET_LAST);
558 }
559 
560 /*
561  * tmc_init_etr_sg_table: Allocate a TMC ETR SG table, data buffer of @size and
562  * populate the table.
563  *
564  * @dev		- Device pointer for the TMC
565  * @node	- NUMA node where the memory should be allocated
566  * @size	- Total size of the data buffer
567  * @pages	- Optional list of page virtual address
568  */
569 static struct etr_sg_table *
tmc_init_etr_sg_table(struct device * dev,int node,unsigned long size,void ** pages)570 tmc_init_etr_sg_table(struct device *dev, int node,
571 		      unsigned long size, void **pages)
572 {
573 	int nr_entries, nr_tpages;
574 	int nr_dpages = size >> PAGE_SHIFT;
575 	struct tmc_sg_table *sg_table;
576 	struct etr_sg_table *etr_table;
577 
578 	etr_table = kzalloc(sizeof(*etr_table), GFP_KERNEL);
579 	if (!etr_table)
580 		return ERR_PTR(-ENOMEM);
581 	nr_entries = tmc_etr_sg_table_entries(nr_dpages);
582 	nr_tpages = DIV_ROUND_UP(nr_entries, ETR_SG_PTRS_PER_SYSPAGE);
583 
584 	sg_table = tmc_alloc_sg_table(dev, node, nr_tpages, nr_dpages, pages);
585 	if (IS_ERR(sg_table)) {
586 		kfree(etr_table);
587 		return ERR_CAST(sg_table);
588 	}
589 
590 	etr_table->sg_table = sg_table;
591 	/* TMC should use table base address for DBA */
592 	etr_table->hwaddr = sg_table->table_daddr;
593 	tmc_etr_sg_table_populate(etr_table);
594 	/* Sync the table pages for the HW */
595 	tmc_sg_table_sync_table(sg_table);
596 	tmc_etr_sg_table_dump(etr_table);
597 
598 	return etr_table;
599 }
600 
601 /*
602  * tmc_etr_alloc_flat_buf: Allocate a contiguous DMA buffer.
603  */
tmc_etr_alloc_flat_buf(struct tmc_drvdata * drvdata,struct etr_buf * etr_buf,int node,void ** pages)604 static int tmc_etr_alloc_flat_buf(struct tmc_drvdata *drvdata,
605 				  struct etr_buf *etr_buf, int node,
606 				  void **pages)
607 {
608 	struct etr_flat_buf *flat_buf;
609 	struct device *real_dev = drvdata->csdev->dev.parent;
610 
611 	/* We cannot reuse existing pages for flat buf */
612 	if (pages)
613 		return -EINVAL;
614 
615 	flat_buf = kzalloc(sizeof(*flat_buf), GFP_KERNEL);
616 	if (!flat_buf)
617 		return -ENOMEM;
618 
619 	flat_buf->vaddr = dma_alloc_noncoherent(real_dev, etr_buf->size,
620 						&flat_buf->daddr,
621 						DMA_FROM_DEVICE,
622 						GFP_KERNEL | __GFP_NOWARN);
623 	if (!flat_buf->vaddr) {
624 		kfree(flat_buf);
625 		return -ENOMEM;
626 	}
627 
628 	flat_buf->size = etr_buf->size;
629 	flat_buf->dev = &drvdata->csdev->dev;
630 	etr_buf->hwaddr = flat_buf->daddr;
631 	etr_buf->mode = ETR_MODE_FLAT;
632 	etr_buf->private = flat_buf;
633 	return 0;
634 }
635 
tmc_etr_free_flat_buf(struct etr_buf * etr_buf)636 static void tmc_etr_free_flat_buf(struct etr_buf *etr_buf)
637 {
638 	struct etr_flat_buf *flat_buf = etr_buf->private;
639 
640 	if (flat_buf && flat_buf->daddr) {
641 		struct device *real_dev = flat_buf->dev->parent;
642 
643 		dma_free_noncoherent(real_dev, etr_buf->size,
644 				     flat_buf->vaddr, flat_buf->daddr,
645 				     DMA_FROM_DEVICE);
646 	}
647 	kfree(flat_buf);
648 }
649 
tmc_etr_sync_flat_buf(struct etr_buf * etr_buf,u64 rrp,u64 rwp)650 static void tmc_etr_sync_flat_buf(struct etr_buf *etr_buf, u64 rrp, u64 rwp)
651 {
652 	struct etr_flat_buf *flat_buf = etr_buf->private;
653 	struct device *real_dev = flat_buf->dev->parent;
654 
655 	/*
656 	 * Adjust the buffer to point to the beginning of the trace data
657 	 * and update the available trace data.
658 	 */
659 	etr_buf->offset = rrp - etr_buf->hwaddr;
660 	if (etr_buf->full)
661 		etr_buf->len = etr_buf->size;
662 	else
663 		etr_buf->len = rwp - rrp;
664 
665 	/*
666 	 * The driver always starts tracing at the beginning of the buffer,
667 	 * the only reason why we would get a wrap around is when the buffer
668 	 * is full.  Sync the entire buffer in one go for this case.
669 	 */
670 	if (etr_buf->offset + etr_buf->len > etr_buf->size)
671 		dma_sync_single_for_cpu(real_dev, flat_buf->daddr,
672 					etr_buf->size, DMA_FROM_DEVICE);
673 	else
674 		dma_sync_single_for_cpu(real_dev,
675 					flat_buf->daddr + etr_buf->offset,
676 					etr_buf->len, DMA_FROM_DEVICE);
677 }
678 
tmc_etr_get_data_flat_buf(struct etr_buf * etr_buf,u64 offset,size_t len,char ** bufpp)679 static ssize_t tmc_etr_get_data_flat_buf(struct etr_buf *etr_buf,
680 					 u64 offset, size_t len, char **bufpp)
681 {
682 	struct etr_flat_buf *flat_buf = etr_buf->private;
683 
684 	*bufpp = (char *)flat_buf->vaddr + offset;
685 	/*
686 	 * tmc_etr_buf_get_data already adjusts the length to handle
687 	 * buffer wrapping around.
688 	 */
689 	return len;
690 }
691 
692 static const struct etr_buf_operations etr_flat_buf_ops = {
693 	.alloc = tmc_etr_alloc_flat_buf,
694 	.free = tmc_etr_free_flat_buf,
695 	.sync = tmc_etr_sync_flat_buf,
696 	.get_data = tmc_etr_get_data_flat_buf,
697 };
698 
699 /*
700  * tmc_etr_alloc_resrv_buf: Allocate a contiguous DMA buffer from reserved region.
701  */
tmc_etr_alloc_resrv_buf(struct tmc_drvdata * drvdata,struct etr_buf * etr_buf,int node,void ** pages)702 static int tmc_etr_alloc_resrv_buf(struct tmc_drvdata *drvdata,
703 				  struct etr_buf *etr_buf, int node,
704 				  void **pages)
705 {
706 	struct etr_flat_buf *resrv_buf;
707 	struct device *real_dev = drvdata->csdev->dev.parent;
708 
709 	/* We cannot reuse existing pages for resrv buf */
710 	if (pages)
711 		return -EINVAL;
712 
713 	resrv_buf = kzalloc(sizeof(*resrv_buf), GFP_KERNEL);
714 	if (!resrv_buf)
715 		return -ENOMEM;
716 
717 	resrv_buf->daddr = dma_map_resource(real_dev, drvdata->resrv_buf.paddr,
718 					   drvdata->resrv_buf.size,
719 					   DMA_FROM_DEVICE, 0);
720 	if (dma_mapping_error(real_dev, resrv_buf->daddr)) {
721 		dev_err(real_dev, "failed to map source buffer address\n");
722 		kfree(resrv_buf);
723 		return -ENOMEM;
724 	}
725 
726 	resrv_buf->vaddr = drvdata->resrv_buf.vaddr;
727 	resrv_buf->size = etr_buf->size = drvdata->resrv_buf.size;
728 	resrv_buf->dev = &drvdata->csdev->dev;
729 	etr_buf->hwaddr = resrv_buf->daddr;
730 	etr_buf->mode = ETR_MODE_RESRV;
731 	etr_buf->private = resrv_buf;
732 	return 0;
733 }
734 
tmc_etr_free_resrv_buf(struct etr_buf * etr_buf)735 static void tmc_etr_free_resrv_buf(struct etr_buf *etr_buf)
736 {
737 	struct etr_flat_buf *resrv_buf = etr_buf->private;
738 
739 	if (resrv_buf && resrv_buf->daddr) {
740 		struct device *real_dev = resrv_buf->dev->parent;
741 
742 		dma_unmap_resource(real_dev, resrv_buf->daddr,
743 				resrv_buf->size, DMA_FROM_DEVICE, 0);
744 	}
745 	kfree(resrv_buf);
746 }
747 
tmc_etr_sync_resrv_buf(struct etr_buf * etr_buf,u64 rrp,u64 rwp)748 static void tmc_etr_sync_resrv_buf(struct etr_buf *etr_buf, u64 rrp, u64 rwp)
749 {
750 	/*
751 	 * Adjust the buffer to point to the beginning of the trace data
752 	 * and update the available trace data.
753 	 */
754 	etr_buf->offset = rrp - etr_buf->hwaddr;
755 	if (etr_buf->full)
756 		etr_buf->len = etr_buf->size;
757 	else
758 		etr_buf->len = rwp - rrp;
759 }
760 
761 static const struct etr_buf_operations etr_resrv_buf_ops = {
762 	.alloc = tmc_etr_alloc_resrv_buf,
763 	.free = tmc_etr_free_resrv_buf,
764 	.sync = tmc_etr_sync_resrv_buf,
765 	.get_data = tmc_etr_get_data_flat_buf,
766 };
767 
768 /*
769  * tmc_etr_alloc_sg_buf: Allocate an SG buf @etr_buf. Setup the parameters
770  * appropriately.
771  */
tmc_etr_alloc_sg_buf(struct tmc_drvdata * drvdata,struct etr_buf * etr_buf,int node,void ** pages)772 static int tmc_etr_alloc_sg_buf(struct tmc_drvdata *drvdata,
773 				struct etr_buf *etr_buf, int node,
774 				void **pages)
775 {
776 	struct etr_sg_table *etr_table;
777 	struct device *dev = &drvdata->csdev->dev;
778 
779 	etr_table = tmc_init_etr_sg_table(dev, node,
780 					  etr_buf->size, pages);
781 	if (IS_ERR(etr_table))
782 		return -ENOMEM;
783 	etr_buf->hwaddr = etr_table->hwaddr;
784 	etr_buf->mode = ETR_MODE_ETR_SG;
785 	etr_buf->private = etr_table;
786 	return 0;
787 }
788 
tmc_etr_free_sg_buf(struct etr_buf * etr_buf)789 static void tmc_etr_free_sg_buf(struct etr_buf *etr_buf)
790 {
791 	struct etr_sg_table *etr_table = etr_buf->private;
792 
793 	if (etr_table) {
794 		tmc_free_sg_table(etr_table->sg_table);
795 		kfree(etr_table);
796 	}
797 }
798 
tmc_etr_get_data_sg_buf(struct etr_buf * etr_buf,u64 offset,size_t len,char ** bufpp)799 static ssize_t tmc_etr_get_data_sg_buf(struct etr_buf *etr_buf, u64 offset,
800 				       size_t len, char **bufpp)
801 {
802 	struct etr_sg_table *etr_table = etr_buf->private;
803 
804 	return tmc_sg_table_get_data(etr_table->sg_table, offset, len, bufpp);
805 }
806 
tmc_etr_sync_sg_buf(struct etr_buf * etr_buf,u64 rrp,u64 rwp)807 static void tmc_etr_sync_sg_buf(struct etr_buf *etr_buf, u64 rrp, u64 rwp)
808 {
809 	long r_offset, w_offset;
810 	struct etr_sg_table *etr_table = etr_buf->private;
811 	struct tmc_sg_table *table = etr_table->sg_table;
812 
813 	/* Convert hw address to offset in the buffer */
814 	r_offset = tmc_sg_get_data_page_offset(table, rrp);
815 	if (r_offset < 0) {
816 		dev_warn(table->dev,
817 			 "Unable to map RRP %llx to offset\n", rrp);
818 		etr_buf->len = 0;
819 		return;
820 	}
821 
822 	w_offset = tmc_sg_get_data_page_offset(table, rwp);
823 	if (w_offset < 0) {
824 		dev_warn(table->dev,
825 			 "Unable to map RWP %llx to offset\n", rwp);
826 		etr_buf->len = 0;
827 		return;
828 	}
829 
830 	etr_buf->offset = r_offset;
831 	if (etr_buf->full)
832 		etr_buf->len = etr_buf->size;
833 	else
834 		etr_buf->len = ((w_offset < r_offset) ? etr_buf->size : 0) +
835 				w_offset - r_offset;
836 	tmc_sg_table_sync_data_range(table, r_offset, etr_buf->len);
837 }
838 
839 static const struct etr_buf_operations etr_sg_buf_ops = {
840 	.alloc = tmc_etr_alloc_sg_buf,
841 	.free = tmc_etr_free_sg_buf,
842 	.sync = tmc_etr_sync_sg_buf,
843 	.get_data = tmc_etr_get_data_sg_buf,
844 };
845 
846 /*
847  * TMC ETR could be connected to a CATU device, which can provide address
848  * translation service. This is represented by the Output port of the TMC
849  * (ETR) connected to the input port of the CATU.
850  *
851  * Returns	: coresight_device ptr for the CATU device if a CATU is found.
852  *		: NULL otherwise.
853  */
854 struct coresight_device *
tmc_etr_get_catu_device(struct tmc_drvdata * drvdata)855 tmc_etr_get_catu_device(struct tmc_drvdata *drvdata)
856 {
857 	struct coresight_device *etr = drvdata->csdev;
858 	union coresight_dev_subtype catu_subtype = {
859 		.helper_subtype = CORESIGHT_DEV_SUBTYPE_HELPER_CATU
860 	};
861 
862 	if (!IS_ENABLED(CONFIG_CORESIGHT_CATU))
863 		return NULL;
864 
865 	return coresight_find_output_type(etr->pdata, CORESIGHT_DEV_TYPE_HELPER,
866 					  catu_subtype);
867 }
868 EXPORT_SYMBOL_GPL(tmc_etr_get_catu_device);
869 
870 static const struct etr_buf_operations *etr_buf_ops[] = {
871 	[ETR_MODE_FLAT] = &etr_flat_buf_ops,
872 	[ETR_MODE_ETR_SG] = &etr_sg_buf_ops,
873 	[ETR_MODE_CATU] = NULL,
874 	[ETR_MODE_RESRV] = &etr_resrv_buf_ops
875 };
876 
tmc_etr_set_catu_ops(const struct etr_buf_operations * catu)877 void tmc_etr_set_catu_ops(const struct etr_buf_operations *catu)
878 {
879 	etr_buf_ops[ETR_MODE_CATU] = catu;
880 }
881 EXPORT_SYMBOL_GPL(tmc_etr_set_catu_ops);
882 
tmc_etr_remove_catu_ops(void)883 void tmc_etr_remove_catu_ops(void)
884 {
885 	etr_buf_ops[ETR_MODE_CATU] = NULL;
886 }
887 EXPORT_SYMBOL_GPL(tmc_etr_remove_catu_ops);
888 
tmc_etr_mode_alloc_buf(int mode,struct tmc_drvdata * drvdata,struct etr_buf * etr_buf,int node,void ** pages)889 static inline int tmc_etr_mode_alloc_buf(int mode,
890 					 struct tmc_drvdata *drvdata,
891 					 struct etr_buf *etr_buf, int node,
892 					 void **pages)
893 {
894 	int rc = -EINVAL;
895 
896 	switch (mode) {
897 	case ETR_MODE_FLAT:
898 	case ETR_MODE_ETR_SG:
899 	case ETR_MODE_CATU:
900 	case ETR_MODE_RESRV:
901 		if (etr_buf_ops[mode] && etr_buf_ops[mode]->alloc)
902 			rc = etr_buf_ops[mode]->alloc(drvdata, etr_buf,
903 						      node, pages);
904 		if (!rc)
905 			etr_buf->ops = etr_buf_ops[mode];
906 		return rc;
907 	default:
908 		return -EINVAL;
909 	}
910 }
911 
get_etr_buf_hw(struct device * dev,struct etr_buf_hw * buf_hw)912 static void get_etr_buf_hw(struct device *dev, struct etr_buf_hw *buf_hw)
913 {
914 	struct tmc_drvdata *drvdata = dev_get_drvdata(dev->parent);
915 
916 	buf_hw->has_iommu = iommu_get_domain_for_dev(dev->parent);
917 	buf_hw->has_etr_sg = tmc_etr_has_cap(drvdata, TMC_ETR_SG);
918 	buf_hw->has_catu = !!tmc_etr_get_catu_device(drvdata);
919 	buf_hw->has_resrv = tmc_has_reserved_buffer(drvdata);
920 }
921 
etr_can_use_flat_mode(struct etr_buf_hw * buf_hw,ssize_t etr_buf_size)922 static bool etr_can_use_flat_mode(struct etr_buf_hw *buf_hw, ssize_t etr_buf_size)
923 {
924 	bool has_sg = buf_hw->has_catu || buf_hw->has_etr_sg;
925 
926 	return !has_sg || buf_hw->has_iommu || etr_buf_size < SZ_1M;
927 }
928 
929 /*
930  * tmc_alloc_etr_buf: Allocate a buffer use by ETR.
931  * @drvdata	: ETR device details.
932  * @size	: size of the requested buffer.
933  * @flags	: Required properties for the buffer.
934  * @node	: Node for memory allocations.
935  * @pages	: An optional list of pages.
936  */
tmc_alloc_etr_buf(struct tmc_drvdata * drvdata,ssize_t size,int flags,int node,void ** pages)937 static struct etr_buf *tmc_alloc_etr_buf(struct tmc_drvdata *drvdata,
938 					 ssize_t size, int flags,
939 					 int node, void **pages)
940 {
941 	int rc = -ENOMEM;
942 	struct etr_buf *etr_buf;
943 	struct etr_buf_hw buf_hw;
944 	struct device *dev = &drvdata->csdev->dev;
945 
946 	get_etr_buf_hw(dev, &buf_hw);
947 	etr_buf = kzalloc(sizeof(*etr_buf), GFP_KERNEL);
948 	if (!etr_buf)
949 		return ERR_PTR(-ENOMEM);
950 
951 	etr_buf->size = size;
952 
953 	/* If there is user directive for buffer mode, try that first */
954 	if (drvdata->etr_mode != ETR_MODE_AUTO)
955 		rc = tmc_etr_mode_alloc_buf(drvdata->etr_mode, drvdata,
956 					    etr_buf, node, pages);
957 
958 	/*
959 	 * If we have to use an existing list of pages, we cannot reliably
960 	 * use a contiguous DMA memory (even if we have an IOMMU). Otherwise,
961 	 * we use the contiguous DMA memory if at least one of the following
962 	 * conditions is true:
963 	 *  a) The ETR cannot use Scatter-Gather.
964 	 *  b) we have a backing IOMMU
965 	 *  c) The requested memory size is smaller (< 1M).
966 	 *
967 	 * Fallback to available mechanisms.
968 	 *
969 	 */
970 	if (rc && !pages && etr_can_use_flat_mode(&buf_hw, size))
971 		rc = tmc_etr_mode_alloc_buf(ETR_MODE_FLAT, drvdata,
972 					    etr_buf, node, pages);
973 	if (rc && buf_hw.has_etr_sg)
974 		rc = tmc_etr_mode_alloc_buf(ETR_MODE_ETR_SG, drvdata,
975 					    etr_buf, node, pages);
976 	if (rc && buf_hw.has_catu)
977 		rc = tmc_etr_mode_alloc_buf(ETR_MODE_CATU, drvdata,
978 					    etr_buf, node, pages);
979 	if (rc) {
980 		kfree(etr_buf);
981 		return ERR_PTR(rc);
982 	}
983 
984 	refcount_set(&etr_buf->refcount, 1);
985 	dev_dbg(dev, "allocated buffer of size %ldKB in mode %d\n",
986 		(unsigned long)size >> 10, etr_buf->mode);
987 	return etr_buf;
988 }
989 
tmc_free_etr_buf(struct etr_buf * etr_buf)990 static void tmc_free_etr_buf(struct etr_buf *etr_buf)
991 {
992 	WARN_ON(!etr_buf->ops || !etr_buf->ops->free);
993 	etr_buf->ops->free(etr_buf);
994 	kfree(etr_buf);
995 }
996 
997 /*
998  * tmc_etr_buf_get_data: Get the pointer the trace data at @offset
999  * with a maximum of @len bytes.
1000  * Returns: The size of the linear data available @pos, with *bufpp
1001  * updated to point to the buffer.
1002  */
tmc_etr_buf_get_data(struct etr_buf * etr_buf,u64 offset,size_t len,char ** bufpp)1003 static ssize_t tmc_etr_buf_get_data(struct etr_buf *etr_buf,
1004 				    u64 offset, size_t len, char **bufpp)
1005 {
1006 	/* Adjust the length to limit this transaction to end of buffer */
1007 	len = (len < (etr_buf->size - offset)) ? len : etr_buf->size - offset;
1008 
1009 	return etr_buf->ops->get_data(etr_buf, (u64)offset, len, bufpp);
1010 }
1011 
1012 static inline s64
tmc_etr_buf_insert_barrier_packet(struct etr_buf * etr_buf,u64 offset)1013 tmc_etr_buf_insert_barrier_packet(struct etr_buf *etr_buf, u64 offset)
1014 {
1015 	ssize_t len;
1016 	char *bufp;
1017 
1018 	len = tmc_etr_buf_get_data(etr_buf, offset,
1019 				   CORESIGHT_BARRIER_PKT_SIZE, &bufp);
1020 	if (WARN_ON(len < 0 || len < CORESIGHT_BARRIER_PKT_SIZE))
1021 		return -EINVAL;
1022 	coresight_insert_barrier_packet(bufp);
1023 	return offset + CORESIGHT_BARRIER_PKT_SIZE;
1024 }
1025 
1026 /*
1027  * tmc_sync_etr_buf: Sync the trace buffer availability with drvdata.
1028  * Makes sure the trace data is synced to the memory for consumption.
1029  * @etr_buf->offset will hold the offset to the beginning of the trace data
1030  * within the buffer, with @etr_buf->len bytes to consume.
1031  */
tmc_sync_etr_buf(struct tmc_drvdata * drvdata)1032 static void tmc_sync_etr_buf(struct tmc_drvdata *drvdata)
1033 {
1034 	struct etr_buf *etr_buf = drvdata->etr_buf;
1035 	u64 rrp, rwp;
1036 	u32 status;
1037 
1038 	rrp = tmc_read_rrp(drvdata);
1039 	rwp = tmc_read_rwp(drvdata);
1040 	status = readl_relaxed(drvdata->base + TMC_STS);
1041 
1042 	/*
1043 	 * If there were memory errors in the session, truncate the
1044 	 * buffer.
1045 	 */
1046 	if (WARN_ON_ONCE(status & TMC_STS_MEMERR)) {
1047 		dev_dbg(&drvdata->csdev->dev,
1048 			"tmc memory error detected, truncating buffer\n");
1049 		etr_buf->len = 0;
1050 		etr_buf->full = false;
1051 		return;
1052 	}
1053 
1054 	etr_buf->full = !!(status & TMC_STS_FULL);
1055 
1056 	WARN_ON(!etr_buf->ops || !etr_buf->ops->sync);
1057 
1058 	etr_buf->ops->sync(etr_buf, rrp, rwp);
1059 }
1060 
__tmc_etr_enable_hw(struct tmc_drvdata * drvdata)1061 static int __tmc_etr_enable_hw(struct tmc_drvdata *drvdata)
1062 {
1063 	u32 axictl, sts, ffcr;
1064 	struct etr_buf *etr_buf = drvdata->etr_buf;
1065 	int rc = 0;
1066 
1067 	CS_UNLOCK(drvdata->base);
1068 
1069 	/* Wait for TMCSReady bit to be set */
1070 	rc = tmc_wait_for_tmcready(drvdata);
1071 	if (rc) {
1072 		dev_err(&drvdata->csdev->dev,
1073 			"Failed to enable : TMC not ready\n");
1074 		CS_LOCK(drvdata->base);
1075 		return rc;
1076 	}
1077 
1078 	writel_relaxed(etr_buf->size / 4, drvdata->base + TMC_RSZ);
1079 	writel_relaxed(TMC_MODE_CIRCULAR_BUFFER, drvdata->base + TMC_MODE);
1080 
1081 	axictl = readl_relaxed(drvdata->base + TMC_AXICTL);
1082 	axictl &= ~TMC_AXICTL_CLEAR_MASK;
1083 	axictl |= TMC_AXICTL_PROT_CTL_B1;
1084 	axictl |= TMC_AXICTL_WR_BURST(drvdata->max_burst_size);
1085 	axictl |= TMC_AXICTL_AXCACHE_OS;
1086 
1087 	if (tmc_etr_has_cap(drvdata, TMC_ETR_AXI_ARCACHE)) {
1088 		axictl &= ~TMC_AXICTL_ARCACHE_MASK;
1089 		axictl |= TMC_AXICTL_ARCACHE_OS;
1090 	}
1091 
1092 	if (etr_buf->mode == ETR_MODE_ETR_SG)
1093 		axictl |= TMC_AXICTL_SCT_GAT_MODE;
1094 
1095 	writel_relaxed(axictl, drvdata->base + TMC_AXICTL);
1096 	tmc_write_dba(drvdata, etr_buf->hwaddr);
1097 	/*
1098 	 * If the TMC pointers must be programmed before the session,
1099 	 * we have to set it properly (i.e, RRP/RWP to base address and
1100 	 * STS to "not full").
1101 	 */
1102 	if (tmc_etr_has_cap(drvdata, TMC_ETR_SAVE_RESTORE)) {
1103 		tmc_write_rrp(drvdata, etr_buf->hwaddr);
1104 		tmc_write_rwp(drvdata, etr_buf->hwaddr);
1105 		sts = readl_relaxed(drvdata->base + TMC_STS) & ~TMC_STS_FULL;
1106 		writel_relaxed(sts, drvdata->base + TMC_STS);
1107 	}
1108 
1109 	ffcr = TMC_FFCR_EN_FMT | TMC_FFCR_EN_TI | TMC_FFCR_FON_FLIN |
1110 		TMC_FFCR_FON_TRIG_EVT | TMC_FFCR_TRIGON_TRIGIN;
1111 	if (drvdata->stop_on_flush)
1112 		ffcr |= TMC_FFCR_STOP_ON_FLUSH;
1113 	writel_relaxed(ffcr, drvdata->base + TMC_FFCR);
1114 
1115 	writel_relaxed(drvdata->trigger_cntr, drvdata->base + TMC_TRG);
1116 	tmc_enable_hw(drvdata);
1117 
1118 	CS_LOCK(drvdata->base);
1119 	return rc;
1120 }
1121 
tmc_etr_enable_hw(struct tmc_drvdata * drvdata,struct etr_buf * etr_buf)1122 static int tmc_etr_enable_hw(struct tmc_drvdata *drvdata,
1123 			     struct etr_buf *etr_buf)
1124 {
1125 	int rc;
1126 
1127 	/* Callers should provide an appropriate buffer for use */
1128 	if (WARN_ON(!etr_buf))
1129 		return -EINVAL;
1130 
1131 	if ((etr_buf->mode == ETR_MODE_ETR_SG) &&
1132 	    WARN_ON(!tmc_etr_has_cap(drvdata, TMC_ETR_SG)))
1133 		return -EINVAL;
1134 
1135 	if (WARN_ON(drvdata->etr_buf))
1136 		return -EBUSY;
1137 
1138 	rc = coresight_claim_device(drvdata->csdev);
1139 	if (!rc) {
1140 		drvdata->etr_buf = etr_buf;
1141 		rc = __tmc_etr_enable_hw(drvdata);
1142 		if (rc) {
1143 			drvdata->etr_buf = NULL;
1144 			coresight_disclaim_device(drvdata->csdev);
1145 		}
1146 	}
1147 
1148 	return rc;
1149 }
1150 
1151 /*
1152  * Return the available trace data in the buffer (starts at etr_buf->offset,
1153  * limited by etr_buf->len) from @pos, with a maximum limit of @len,
1154  * also updating the @bufpp on where to find it. Since the trace data
1155  * starts at anywhere in the buffer, depending on the RRP, we adjust the
1156  * @len returned to handle buffer wrapping around.
1157  *
1158  * We are protected here by drvdata->reading != 0, which ensures the
1159  * sysfs_buf stays alive.
1160  */
tmc_etr_get_sysfs_trace(struct tmc_drvdata * drvdata,loff_t pos,size_t len,char ** bufpp)1161 ssize_t tmc_etr_get_sysfs_trace(struct tmc_drvdata *drvdata,
1162 				loff_t pos, size_t len, char **bufpp)
1163 {
1164 	s64 offset;
1165 	ssize_t actual = len;
1166 	struct etr_buf *etr_buf = drvdata->sysfs_buf;
1167 
1168 	if (pos + actual > etr_buf->len)
1169 		actual = etr_buf->len - pos;
1170 	if (actual <= 0)
1171 		return actual;
1172 
1173 	/* Compute the offset from which we read the data */
1174 	offset = etr_buf->offset + pos;
1175 	if (offset >= etr_buf->size)
1176 		offset -= etr_buf->size;
1177 	return tmc_etr_buf_get_data(etr_buf, offset, actual, bufpp);
1178 }
1179 
1180 static struct etr_buf *
tmc_etr_setup_sysfs_buf(struct tmc_drvdata * drvdata)1181 tmc_etr_setup_sysfs_buf(struct tmc_drvdata *drvdata)
1182 {
1183 	return tmc_alloc_etr_buf(drvdata, drvdata->size,
1184 				 0, cpu_to_node(0), NULL);
1185 }
1186 
1187 static void
tmc_etr_free_sysfs_buf(struct etr_buf * buf)1188 tmc_etr_free_sysfs_buf(struct etr_buf *buf)
1189 {
1190 	if (buf)
1191 		tmc_free_etr_buf(buf);
1192 }
1193 
tmc_etr_sync_sysfs_buf(struct tmc_drvdata * drvdata)1194 static void tmc_etr_sync_sysfs_buf(struct tmc_drvdata *drvdata)
1195 {
1196 	struct etr_buf *etr_buf = drvdata->etr_buf;
1197 
1198 	if (WARN_ON(drvdata->sysfs_buf != etr_buf)) {
1199 		tmc_etr_free_sysfs_buf(drvdata->sysfs_buf);
1200 		drvdata->sysfs_buf = NULL;
1201 	} else {
1202 		tmc_sync_etr_buf(drvdata);
1203 		/*
1204 		 * Insert barrier packets at the beginning, if there was
1205 		 * an overflow.
1206 		 */
1207 		if (etr_buf->full)
1208 			tmc_etr_buf_insert_barrier_packet(etr_buf,
1209 							  etr_buf->offset);
1210 	}
1211 }
1212 
__tmc_etr_disable_hw(struct tmc_drvdata * drvdata)1213 static void __tmc_etr_disable_hw(struct tmc_drvdata *drvdata)
1214 {
1215 	CS_UNLOCK(drvdata->base);
1216 
1217 	tmc_flush_and_stop(drvdata);
1218 	/*
1219 	 * When operating in sysFS mode the content of the buffer needs to be
1220 	 * read before the TMC is disabled.
1221 	 */
1222 	if (coresight_get_mode(drvdata->csdev) == CS_MODE_SYSFS)
1223 		tmc_etr_sync_sysfs_buf(drvdata);
1224 
1225 	tmc_disable_hw(drvdata);
1226 
1227 	CS_LOCK(drvdata->base);
1228 
1229 }
1230 
tmc_etr_disable_hw(struct tmc_drvdata * drvdata)1231 void tmc_etr_disable_hw(struct tmc_drvdata *drvdata)
1232 {
1233 	__tmc_etr_disable_hw(drvdata);
1234 	coresight_disclaim_device(drvdata->csdev);
1235 	/* Reset the ETR buf used by hardware */
1236 	drvdata->etr_buf = NULL;
1237 }
1238 
tmc_etr_get_sysfs_buffer(struct coresight_device * csdev)1239 static struct etr_buf *tmc_etr_get_sysfs_buffer(struct coresight_device *csdev)
1240 {
1241 	int ret = 0;
1242 	unsigned long flags;
1243 	struct tmc_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent);
1244 	struct etr_buf *sysfs_buf = NULL, *new_buf = NULL, *free_buf = NULL;
1245 
1246 	/*
1247 	 * If we are enabling the ETR from disabled state, we need to make
1248 	 * sure we have a buffer with the right size. The etr_buf is not reset
1249 	 * immediately after we stop the tracing in SYSFS mode as we wait for
1250 	 * the user to collect the data. We may be able to reuse the existing
1251 	 * buffer, provided the size matches. Any allocation has to be done
1252 	 * with the lock released.
1253 	 */
1254 	raw_spin_lock_irqsave(&drvdata->spinlock, flags);
1255 	sysfs_buf = READ_ONCE(drvdata->sysfs_buf);
1256 	if (!sysfs_buf || (sysfs_buf->size != drvdata->size)) {
1257 		raw_spin_unlock_irqrestore(&drvdata->spinlock, flags);
1258 
1259 		/* Allocate memory with the locks released */
1260 		free_buf = new_buf = tmc_etr_setup_sysfs_buf(drvdata);
1261 		if (IS_ERR(new_buf))
1262 			return new_buf;
1263 
1264 		/* Let's try again */
1265 		raw_spin_lock_irqsave(&drvdata->spinlock, flags);
1266 	}
1267 
1268 	if (drvdata->reading || coresight_get_mode(csdev) == CS_MODE_PERF) {
1269 		ret = -EBUSY;
1270 		goto out;
1271 	}
1272 
1273 	/*
1274 	 * If we don't have a buffer or it doesn't match the requested size,
1275 	 * use the buffer allocated above. Otherwise reuse the existing buffer.
1276 	 */
1277 	sysfs_buf = READ_ONCE(drvdata->sysfs_buf);
1278 	if (!sysfs_buf || (new_buf && sysfs_buf->size != new_buf->size)) {
1279 		free_buf = sysfs_buf;
1280 		drvdata->sysfs_buf = new_buf;
1281 	}
1282 
1283 out:
1284 	raw_spin_unlock_irqrestore(&drvdata->spinlock, flags);
1285 
1286 	/* Free memory outside the spinlock if need be */
1287 	if (free_buf)
1288 		tmc_etr_free_sysfs_buf(free_buf);
1289 	return ret ? ERR_PTR(ret) : drvdata->sysfs_buf;
1290 }
1291 
tmc_enable_etr_sink_sysfs(struct coresight_device * csdev)1292 static int tmc_enable_etr_sink_sysfs(struct coresight_device *csdev)
1293 {
1294 	int ret = 0;
1295 	unsigned long flags;
1296 	struct tmc_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent);
1297 	struct etr_buf *sysfs_buf = tmc_etr_get_sysfs_buffer(csdev);
1298 
1299 	if (IS_ERR(sysfs_buf))
1300 		return PTR_ERR(sysfs_buf);
1301 
1302 	raw_spin_lock_irqsave(&drvdata->spinlock, flags);
1303 
1304 	/*
1305 	 * In sysFS mode we can have multiple writers per sink.  Since this
1306 	 * sink is already enabled no memory is needed and the HW need not be
1307 	 * touched, even if the buffer size has changed.
1308 	 */
1309 	if (coresight_get_mode(csdev) == CS_MODE_SYSFS) {
1310 		csdev->refcnt++;
1311 		goto out;
1312 	}
1313 
1314 	ret = tmc_etr_enable_hw(drvdata, sysfs_buf);
1315 	if (!ret) {
1316 		coresight_set_mode(csdev, CS_MODE_SYSFS);
1317 		csdev->refcnt++;
1318 	}
1319 
1320 out:
1321 	raw_spin_unlock_irqrestore(&drvdata->spinlock, flags);
1322 
1323 	if (!ret)
1324 		dev_dbg(&csdev->dev, "TMC-ETR enabled\n");
1325 
1326 	return ret;
1327 }
1328 
tmc_etr_get_buffer(struct coresight_device * csdev,enum cs_mode mode,void * data)1329 struct etr_buf *tmc_etr_get_buffer(struct coresight_device *csdev,
1330 				   enum cs_mode mode, void *data)
1331 {
1332 	struct perf_output_handle *handle = data;
1333 	struct etr_perf_buffer *etr_perf;
1334 
1335 	switch (mode) {
1336 	case CS_MODE_SYSFS:
1337 		return tmc_etr_get_sysfs_buffer(csdev);
1338 	case CS_MODE_PERF:
1339 		etr_perf = etm_perf_sink_config(handle);
1340 		if (WARN_ON(!etr_perf || !etr_perf->etr_buf))
1341 			return ERR_PTR(-EINVAL);
1342 		return etr_perf->etr_buf;
1343 	default:
1344 		return ERR_PTR(-EINVAL);
1345 	}
1346 }
1347 EXPORT_SYMBOL_GPL(tmc_etr_get_buffer);
1348 
1349 /*
1350  * alloc_etr_buf: Allocate ETR buffer for use by perf.
1351  * The size of the hardware buffer is dependent on the size configured
1352  * via sysfs and the perf ring buffer size. We prefer to allocate the
1353  * largest possible size, scaling down the size by half until it
1354  * reaches a minimum limit (1M), beyond which we give up.
1355  */
1356 static struct etr_buf *
alloc_etr_buf(struct tmc_drvdata * drvdata,struct perf_event * event,int nr_pages,void ** pages,bool snapshot)1357 alloc_etr_buf(struct tmc_drvdata *drvdata, struct perf_event *event,
1358 	      int nr_pages, void **pages, bool snapshot)
1359 {
1360 	int node;
1361 	struct etr_buf *etr_buf;
1362 	unsigned long size;
1363 
1364 	node = (event->cpu == -1) ? NUMA_NO_NODE : cpu_to_node(event->cpu);
1365 	/*
1366 	 * Try to match the perf ring buffer size if it is larger
1367 	 * than the size requested via sysfs.
1368 	 */
1369 	if ((nr_pages << PAGE_SHIFT) > drvdata->size) {
1370 		etr_buf = tmc_alloc_etr_buf(drvdata, ((ssize_t)nr_pages << PAGE_SHIFT),
1371 					    0, node, NULL);
1372 		if (!IS_ERR(etr_buf))
1373 			goto done;
1374 	}
1375 
1376 	/*
1377 	 * Else switch to configured size for this ETR
1378 	 * and scale down until we hit the minimum limit.
1379 	 */
1380 	size = drvdata->size;
1381 	do {
1382 		etr_buf = tmc_alloc_etr_buf(drvdata, size, 0, node, NULL);
1383 		if (!IS_ERR(etr_buf))
1384 			goto done;
1385 		size /= 2;
1386 	} while (size >= TMC_ETR_PERF_MIN_BUF_SIZE);
1387 
1388 	return ERR_PTR(-ENOMEM);
1389 
1390 done:
1391 	return etr_buf;
1392 }
1393 
1394 static struct etr_buf *
get_perf_etr_buf_cpu_wide(struct tmc_drvdata * drvdata,struct perf_event * event,int nr_pages,void ** pages,bool snapshot)1395 get_perf_etr_buf_cpu_wide(struct tmc_drvdata *drvdata,
1396 			  struct perf_event *event, int nr_pages,
1397 			  void **pages, bool snapshot)
1398 {
1399 	int ret;
1400 	pid_t pid = task_pid_nr(event->owner);
1401 	struct etr_buf *etr_buf;
1402 
1403 retry:
1404 	/*
1405 	 * An etr_perf_buffer is associated with an event and holds a reference
1406 	 * to the AUX ring buffer that was created for that event.  In CPU-wide
1407 	 * N:1 mode multiple events (one per CPU), each with its own AUX ring
1408 	 * buffer, share a sink.  As such an etr_perf_buffer is created for each
1409 	 * event but a single etr_buf associated with the ETR is shared between
1410 	 * them.  The last event in a trace session will copy the content of the
1411 	 * etr_buf to its AUX ring buffer.  Ring buffer associated to other
1412 	 * events are simply not used an freed as events are destoyed.  We still
1413 	 * need to allocate a ring buffer for each event since we don't know
1414 	 * which event will be last.
1415 	 */
1416 
1417 	/*
1418 	 * The first thing to do here is check if an etr_buf has already been
1419 	 * allocated for this session.  If so it is shared with this event,
1420 	 * otherwise it is created.
1421 	 */
1422 	mutex_lock(&drvdata->idr_mutex);
1423 	etr_buf = idr_find(&drvdata->idr, pid);
1424 	if (etr_buf) {
1425 		refcount_inc(&etr_buf->refcount);
1426 		mutex_unlock(&drvdata->idr_mutex);
1427 		return etr_buf;
1428 	}
1429 
1430 	/* If we made it here no buffer has been allocated, do so now. */
1431 	mutex_unlock(&drvdata->idr_mutex);
1432 
1433 	etr_buf = alloc_etr_buf(drvdata, event, nr_pages, pages, snapshot);
1434 	if (IS_ERR(etr_buf))
1435 		return etr_buf;
1436 
1437 	/* Now that we have a buffer, add it to the IDR. */
1438 	mutex_lock(&drvdata->idr_mutex);
1439 	ret = idr_alloc(&drvdata->idr, etr_buf, pid, pid + 1, GFP_KERNEL);
1440 	mutex_unlock(&drvdata->idr_mutex);
1441 
1442 	/* Another event with this session ID has allocated this buffer. */
1443 	if (ret == -ENOSPC) {
1444 		tmc_free_etr_buf(etr_buf);
1445 		goto retry;
1446 	}
1447 
1448 	/* The IDR can't allocate room for a new session, abandon ship. */
1449 	if (ret == -ENOMEM) {
1450 		tmc_free_etr_buf(etr_buf);
1451 		return ERR_PTR(ret);
1452 	}
1453 
1454 
1455 	return etr_buf;
1456 }
1457 
1458 static struct etr_buf *
get_perf_etr_buf_per_thread(struct tmc_drvdata * drvdata,struct perf_event * event,int nr_pages,void ** pages,bool snapshot)1459 get_perf_etr_buf_per_thread(struct tmc_drvdata *drvdata,
1460 			    struct perf_event *event, int nr_pages,
1461 			    void **pages, bool snapshot)
1462 {
1463 	/*
1464 	 * In per-thread mode the etr_buf isn't shared, so just go ahead
1465 	 * with memory allocation.
1466 	 */
1467 	return alloc_etr_buf(drvdata, event, nr_pages, pages, snapshot);
1468 }
1469 
1470 static struct etr_buf *
get_perf_etr_buf(struct tmc_drvdata * drvdata,struct perf_event * event,int nr_pages,void ** pages,bool snapshot)1471 get_perf_etr_buf(struct tmc_drvdata *drvdata, struct perf_event *event,
1472 		 int nr_pages, void **pages, bool snapshot)
1473 {
1474 	if (event->cpu == -1)
1475 		return get_perf_etr_buf_per_thread(drvdata, event, nr_pages,
1476 						   pages, snapshot);
1477 
1478 	return get_perf_etr_buf_cpu_wide(drvdata, event, nr_pages,
1479 					 pages, snapshot);
1480 }
1481 
1482 static struct etr_perf_buffer *
tmc_etr_setup_perf_buf(struct tmc_drvdata * drvdata,struct perf_event * event,int nr_pages,void ** pages,bool snapshot)1483 tmc_etr_setup_perf_buf(struct tmc_drvdata *drvdata, struct perf_event *event,
1484 		       int nr_pages, void **pages, bool snapshot)
1485 {
1486 	int node;
1487 	struct etr_buf *etr_buf;
1488 	struct etr_perf_buffer *etr_perf;
1489 
1490 	node = (event->cpu == -1) ? NUMA_NO_NODE : cpu_to_node(event->cpu);
1491 
1492 	etr_perf = kzalloc_node(sizeof(*etr_perf), GFP_KERNEL, node);
1493 	if (!etr_perf)
1494 		return ERR_PTR(-ENOMEM);
1495 
1496 	etr_buf = get_perf_etr_buf(drvdata, event, nr_pages, pages, snapshot);
1497 	if (!IS_ERR(etr_buf))
1498 		goto done;
1499 
1500 	kfree(etr_perf);
1501 	return ERR_PTR(-ENOMEM);
1502 
1503 done:
1504 	/*
1505 	 * Keep a reference to the ETR this buffer has been allocated for
1506 	 * in order to have access to the IDR in tmc_free_etr_buffer().
1507 	 */
1508 	etr_perf->drvdata = drvdata;
1509 	etr_perf->etr_buf = etr_buf;
1510 
1511 	return etr_perf;
1512 }
1513 
1514 
tmc_alloc_etr_buffer(struct coresight_device * csdev,struct perf_event * event,void ** pages,int nr_pages,bool snapshot)1515 static void *tmc_alloc_etr_buffer(struct coresight_device *csdev,
1516 				  struct perf_event *event, void **pages,
1517 				  int nr_pages, bool snapshot)
1518 {
1519 	struct etr_perf_buffer *etr_perf;
1520 	struct tmc_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent);
1521 
1522 	etr_perf = tmc_etr_setup_perf_buf(drvdata, event,
1523 					  nr_pages, pages, snapshot);
1524 	if (IS_ERR(etr_perf)) {
1525 		dev_dbg(&csdev->dev, "Unable to allocate ETR buffer\n");
1526 		return NULL;
1527 	}
1528 
1529 	etr_perf->pid = task_pid_nr(event->owner);
1530 	etr_perf->snapshot = snapshot;
1531 	etr_perf->nr_pages = nr_pages;
1532 	etr_perf->pages = pages;
1533 
1534 	return etr_perf;
1535 }
1536 
tmc_free_etr_buffer(void * config)1537 static void tmc_free_etr_buffer(void *config)
1538 {
1539 	struct etr_perf_buffer *etr_perf = config;
1540 	struct tmc_drvdata *drvdata = etr_perf->drvdata;
1541 	struct etr_buf *buf, *etr_buf = etr_perf->etr_buf;
1542 
1543 	if (!etr_buf)
1544 		goto free_etr_perf_buffer;
1545 
1546 	mutex_lock(&drvdata->idr_mutex);
1547 	/* If we are not the last one to use the buffer, don't touch it. */
1548 	if (!refcount_dec_and_test(&etr_buf->refcount)) {
1549 		mutex_unlock(&drvdata->idr_mutex);
1550 		goto free_etr_perf_buffer;
1551 	}
1552 
1553 	/* We are the last one, remove from the IDR and free the buffer. */
1554 	buf = idr_remove(&drvdata->idr, etr_perf->pid);
1555 	mutex_unlock(&drvdata->idr_mutex);
1556 
1557 	/*
1558 	 * Something went very wrong if the buffer associated with this ID
1559 	 * is not the same in the IDR.  Leak to avoid use after free.
1560 	 */
1561 	if (buf && WARN_ON(buf != etr_buf))
1562 		goto free_etr_perf_buffer;
1563 
1564 	tmc_free_etr_buf(etr_perf->etr_buf);
1565 
1566 free_etr_perf_buffer:
1567 	kfree(etr_perf);
1568 }
1569 
1570 /*
1571  * tmc_etr_sync_perf_buffer: Copy the actual trace data from the hardware
1572  * buffer to the perf ring buffer.
1573  */
tmc_etr_sync_perf_buffer(struct etr_perf_buffer * etr_perf,unsigned long head,unsigned long src_offset,unsigned long to_copy)1574 static void tmc_etr_sync_perf_buffer(struct etr_perf_buffer *etr_perf,
1575 				     unsigned long head,
1576 				     unsigned long src_offset,
1577 				     unsigned long to_copy)
1578 {
1579 	long bytes;
1580 	long pg_idx, pg_offset;
1581 	char **dst_pages, *src_buf;
1582 	struct etr_buf *etr_buf = etr_perf->etr_buf;
1583 
1584 	head = PERF_IDX2OFF(head, etr_perf);
1585 	pg_idx = head >> PAGE_SHIFT;
1586 	pg_offset = head & (PAGE_SIZE - 1);
1587 	dst_pages = (char **)etr_perf->pages;
1588 
1589 	while (to_copy > 0) {
1590 		/*
1591 		 * In one iteration, we can copy minimum of :
1592 		 *  1) what is available in the source buffer,
1593 		 *  2) what is available in the source buffer, before it
1594 		 *     wraps around.
1595 		 *  3) what is available in the destination page.
1596 		 * in one iteration.
1597 		 */
1598 		if (src_offset >= etr_buf->size)
1599 			src_offset -= etr_buf->size;
1600 		bytes = tmc_etr_buf_get_data(etr_buf, src_offset, to_copy,
1601 					     &src_buf);
1602 		if (WARN_ON_ONCE(bytes <= 0))
1603 			break;
1604 		bytes = min(bytes, (long)(PAGE_SIZE - pg_offset));
1605 
1606 		memcpy(dst_pages[pg_idx] + pg_offset, src_buf, bytes);
1607 
1608 		to_copy -= bytes;
1609 
1610 		/* Move destination pointers */
1611 		pg_offset += bytes;
1612 		if (pg_offset == PAGE_SIZE) {
1613 			pg_offset = 0;
1614 			if (++pg_idx == etr_perf->nr_pages)
1615 				pg_idx = 0;
1616 		}
1617 
1618 		/* Move source pointers */
1619 		src_offset += bytes;
1620 	}
1621 }
1622 
1623 /*
1624  * tmc_update_etr_buffer : Update the perf ring buffer with the
1625  * available trace data. We use software double buffering at the moment.
1626  *
1627  * TODO: Add support for reusing the perf ring buffer.
1628  */
1629 static unsigned long
tmc_update_etr_buffer(struct coresight_device * csdev,struct perf_output_handle * handle,void * config)1630 tmc_update_etr_buffer(struct coresight_device *csdev,
1631 		      struct perf_output_handle *handle,
1632 		      void *config)
1633 {
1634 	bool lost = false;
1635 	unsigned long flags, offset, size = 0;
1636 	struct tmc_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent);
1637 	struct etr_perf_buffer *etr_perf = config;
1638 	struct etr_buf *etr_buf = etr_perf->etr_buf;
1639 
1640 	raw_spin_lock_irqsave(&drvdata->spinlock, flags);
1641 
1642 	/* Don't do anything if another tracer is using this sink */
1643 	if (csdev->refcnt != 1) {
1644 		raw_spin_unlock_irqrestore(&drvdata->spinlock, flags);
1645 		goto out;
1646 	}
1647 
1648 	if (WARN_ON(drvdata->perf_buf != etr_buf)) {
1649 		lost = true;
1650 		raw_spin_unlock_irqrestore(&drvdata->spinlock, flags);
1651 		goto out;
1652 	}
1653 
1654 	CS_UNLOCK(drvdata->base);
1655 
1656 	tmc_flush_and_stop(drvdata);
1657 	tmc_sync_etr_buf(drvdata);
1658 
1659 	CS_LOCK(drvdata->base);
1660 	raw_spin_unlock_irqrestore(&drvdata->spinlock, flags);
1661 
1662 	lost = etr_buf->full;
1663 	offset = etr_buf->offset;
1664 	size = etr_buf->len;
1665 
1666 	/*
1667 	 * The ETR buffer may be bigger than the space available in the
1668 	 * perf ring buffer (handle->size).  If so advance the offset so that we
1669 	 * get the latest trace data.  In snapshot mode none of that matters
1670 	 * since we are expected to clobber stale data in favour of the latest
1671 	 * traces.
1672 	 */
1673 	if (!etr_perf->snapshot && size > handle->size) {
1674 		u32 mask = tmc_get_memwidth_mask(drvdata);
1675 
1676 		/*
1677 		 * Make sure the new size is aligned in accordance with the
1678 		 * requirement explained in function tmc_get_memwidth_mask().
1679 		 */
1680 		size = handle->size & mask;
1681 		offset = etr_buf->offset + etr_buf->len - size;
1682 
1683 		if (offset >= etr_buf->size)
1684 			offset -= etr_buf->size;
1685 		lost = true;
1686 	}
1687 
1688 	/* Insert barrier packets at the beginning, if there was an overflow */
1689 	if (lost)
1690 		tmc_etr_buf_insert_barrier_packet(etr_buf, offset);
1691 	tmc_etr_sync_perf_buffer(etr_perf, handle->head, offset, size);
1692 
1693 	/*
1694 	 * In snapshot mode we simply increment the head by the number of byte
1695 	 * that were written.  User space will figure out how many bytes to get
1696 	 * from the AUX buffer based on the position of the head.
1697 	 */
1698 	if (etr_perf->snapshot)
1699 		handle->head += size;
1700 
1701 	/*
1702 	 * Ensure that the AUX trace data is visible before the aux_head
1703 	 * is updated via perf_aux_output_end(), as expected by the
1704 	 * perf ring buffer.
1705 	 */
1706 	smp_wmb();
1707 
1708 out:
1709 	/*
1710 	 * Don't set the TRUNCATED flag in snapshot mode because 1) the
1711 	 * captured buffer is expected to be truncated and 2) a full buffer
1712 	 * prevents the event from being re-enabled by the perf core,
1713 	 * resulting in stale data being send to user space.
1714 	 */
1715 	if (!etr_perf->snapshot && lost)
1716 		perf_aux_output_flag(handle, PERF_AUX_FLAG_TRUNCATED);
1717 	return size;
1718 }
1719 
tmc_enable_etr_sink_perf(struct coresight_device * csdev,void * data)1720 static int tmc_enable_etr_sink_perf(struct coresight_device *csdev, void *data)
1721 {
1722 	int rc = 0;
1723 	pid_t pid;
1724 	unsigned long flags;
1725 	struct tmc_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent);
1726 	struct perf_output_handle *handle = data;
1727 	struct etr_perf_buffer *etr_perf = etm_perf_sink_config(handle);
1728 
1729 	raw_spin_lock_irqsave(&drvdata->spinlock, flags);
1730 	 /* Don't use this sink if it is already claimed by sysFS */
1731 	if (coresight_get_mode(csdev) == CS_MODE_SYSFS) {
1732 		rc = -EBUSY;
1733 		goto unlock_out;
1734 	}
1735 
1736 	if (WARN_ON(!etr_perf || !etr_perf->etr_buf)) {
1737 		rc = -EINVAL;
1738 		goto unlock_out;
1739 	}
1740 
1741 	/* Get a handle on the pid of the session owner */
1742 	pid = etr_perf->pid;
1743 
1744 	/* Do not proceed if this device is associated with another session */
1745 	if (drvdata->pid != -1 && drvdata->pid != pid) {
1746 		rc = -EBUSY;
1747 		goto unlock_out;
1748 	}
1749 
1750 	/*
1751 	 * No HW configuration is needed if the sink is already in
1752 	 * use for this session.
1753 	 */
1754 	if (drvdata->pid == pid) {
1755 		csdev->refcnt++;
1756 		goto unlock_out;
1757 	}
1758 
1759 	rc = tmc_etr_enable_hw(drvdata, etr_perf->etr_buf);
1760 	if (!rc) {
1761 		/* Associate with monitored process. */
1762 		drvdata->pid = pid;
1763 		coresight_set_mode(csdev, CS_MODE_PERF);
1764 		drvdata->perf_buf = etr_perf->etr_buf;
1765 		csdev->refcnt++;
1766 	}
1767 
1768 unlock_out:
1769 	raw_spin_unlock_irqrestore(&drvdata->spinlock, flags);
1770 	return rc;
1771 }
1772 
tmc_enable_etr_sink(struct coresight_device * csdev,enum cs_mode mode,void * data)1773 static int tmc_enable_etr_sink(struct coresight_device *csdev,
1774 			       enum cs_mode mode, void *data)
1775 {
1776 	switch (mode) {
1777 	case CS_MODE_SYSFS:
1778 		return tmc_enable_etr_sink_sysfs(csdev);
1779 	case CS_MODE_PERF:
1780 		return tmc_enable_etr_sink_perf(csdev, data);
1781 	default:
1782 		return -EINVAL;
1783 	}
1784 }
1785 
tmc_disable_etr_sink(struct coresight_device * csdev)1786 static int tmc_disable_etr_sink(struct coresight_device *csdev)
1787 {
1788 	unsigned long flags;
1789 	struct tmc_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent);
1790 
1791 	raw_spin_lock_irqsave(&drvdata->spinlock, flags);
1792 
1793 	if (drvdata->reading) {
1794 		raw_spin_unlock_irqrestore(&drvdata->spinlock, flags);
1795 		return -EBUSY;
1796 	}
1797 
1798 	csdev->refcnt--;
1799 	if (csdev->refcnt) {
1800 		raw_spin_unlock_irqrestore(&drvdata->spinlock, flags);
1801 		return -EBUSY;
1802 	}
1803 
1804 	/* Complain if we (somehow) got out of sync */
1805 	WARN_ON_ONCE(coresight_get_mode(csdev) == CS_MODE_DISABLED);
1806 	tmc_etr_disable_hw(drvdata);
1807 	/* Dissociate from monitored process. */
1808 	drvdata->pid = -1;
1809 	coresight_set_mode(csdev, CS_MODE_DISABLED);
1810 	/* Reset perf specific data */
1811 	drvdata->perf_buf = NULL;
1812 
1813 	raw_spin_unlock_irqrestore(&drvdata->spinlock, flags);
1814 
1815 	dev_dbg(&csdev->dev, "TMC-ETR disabled\n");
1816 	return 0;
1817 }
1818 
tmc_panic_sync_etr(struct coresight_device * csdev)1819 static int tmc_panic_sync_etr(struct coresight_device *csdev)
1820 {
1821 	u32 val;
1822 	struct tmc_crash_metadata *mdata;
1823 	struct tmc_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent);
1824 
1825 	mdata = (struct tmc_crash_metadata *)drvdata->crash_mdata.vaddr;
1826 
1827 	if (!drvdata->etr_buf)
1828 		return 0;
1829 
1830 	/* Being in RESRV mode implies valid reserved memory as well */
1831 	if (drvdata->etr_buf->mode != ETR_MODE_RESRV)
1832 		return 0;
1833 
1834 	if (!tmc_has_crash_mdata_buffer(drvdata))
1835 		return 0;
1836 
1837 	CS_UNLOCK(drvdata->base);
1838 
1839 	/* Proceed only if ETR is enabled */
1840 	val = readl(drvdata->base + TMC_CTL);
1841 	if (!(val & TMC_CTL_CAPT_EN))
1842 		goto out;
1843 
1844 	val = readl(drvdata->base + TMC_FFSR);
1845 	/* Do manual flush and stop only if its not auto-stopped */
1846 	if (!(val & TMC_FFSR_FT_STOPPED)) {
1847 		dev_dbg(&csdev->dev,
1848 			 "%s: Triggering manual flush\n", __func__);
1849 		tmc_flush_and_stop(drvdata);
1850 	} else
1851 		tmc_wait_for_tmcready(drvdata);
1852 
1853 	/* Sync registers from hardware to metadata region */
1854 	mdata->tmc_ram_size = readl(drvdata->base + TMC_RSZ);
1855 	mdata->tmc_sts = readl(drvdata->base + TMC_STS);
1856 	mdata->tmc_mode = readl(drvdata->base + TMC_MODE);
1857 	mdata->tmc_ffcr = readl(drvdata->base + TMC_FFCR);
1858 	mdata->tmc_ffsr = readl(drvdata->base + TMC_FFSR);
1859 	mdata->tmc_rrp = tmc_read_rrp(drvdata);
1860 	mdata->tmc_rwp = tmc_read_rwp(drvdata);
1861 	mdata->tmc_dba = tmc_read_dba(drvdata);
1862 	mdata->trace_paddr = drvdata->resrv_buf.paddr;
1863 	mdata->version = CS_CRASHDATA_VERSION;
1864 
1865 	/*
1866 	 * Make sure all previous writes are ordered,
1867 	 * before we mark valid
1868 	 */
1869 	dmb(sy);
1870 	mdata->valid = true;
1871 	/*
1872 	 * Below order need to maintained, since crc of metadata
1873 	 * is dependent on first
1874 	 */
1875 	mdata->crc32_tdata = find_crash_tracedata_crc(drvdata, mdata);
1876 	mdata->crc32_mdata = find_crash_metadata_crc(mdata);
1877 
1878 	tmc_disable_hw(drvdata);
1879 
1880 	dev_dbg(&csdev->dev, "%s: success\n", __func__);
1881 out:
1882 	CS_UNLOCK(drvdata->base);
1883 
1884 	return 0;
1885 }
1886 
1887 static const struct coresight_ops_sink tmc_etr_sink_ops = {
1888 	.enable		= tmc_enable_etr_sink,
1889 	.disable	= tmc_disable_etr_sink,
1890 	.alloc_buffer	= tmc_alloc_etr_buffer,
1891 	.update_buffer	= tmc_update_etr_buffer,
1892 	.free_buffer	= tmc_free_etr_buffer,
1893 };
1894 
1895 static const struct coresight_ops_panic tmc_etr_sync_ops = {
1896 	.sync		= tmc_panic_sync_etr,
1897 };
1898 
1899 const struct coresight_ops tmc_etr_cs_ops = {
1900 	.sink_ops	= &tmc_etr_sink_ops,
1901 	.panic_ops	= &tmc_etr_sync_ops,
1902 };
1903 
tmc_read_prepare_etr(struct tmc_drvdata * drvdata)1904 int tmc_read_prepare_etr(struct tmc_drvdata *drvdata)
1905 {
1906 	int ret = 0;
1907 	unsigned long flags;
1908 
1909 	/* config types are set a boot time and never change */
1910 	if (WARN_ON_ONCE(drvdata->config_type != TMC_CONFIG_TYPE_ETR))
1911 		return -EINVAL;
1912 
1913 	raw_spin_lock_irqsave(&drvdata->spinlock, flags);
1914 	if (drvdata->reading) {
1915 		ret = -EBUSY;
1916 		goto out;
1917 	}
1918 
1919 	/*
1920 	 * We can safely allow reads even if the ETR is operating in PERF mode,
1921 	 * since the sysfs session is captured in mode specific data.
1922 	 * If drvdata::sysfs_data is NULL the trace data has been read already.
1923 	 */
1924 	if (!drvdata->sysfs_buf) {
1925 		ret = -EINVAL;
1926 		goto out;
1927 	}
1928 
1929 	/* Disable the TMC if we are trying to read from a running session. */
1930 	if (coresight_get_mode(drvdata->csdev) == CS_MODE_SYSFS)
1931 		__tmc_etr_disable_hw(drvdata);
1932 
1933 	drvdata->reading = true;
1934 out:
1935 	raw_spin_unlock_irqrestore(&drvdata->spinlock, flags);
1936 
1937 	return ret;
1938 }
1939 
tmc_read_unprepare_etr(struct tmc_drvdata * drvdata)1940 int tmc_read_unprepare_etr(struct tmc_drvdata *drvdata)
1941 {
1942 	unsigned long flags;
1943 	struct etr_buf *sysfs_buf = NULL;
1944 
1945 	/* config types are set a boot time and never change */
1946 	if (WARN_ON_ONCE(drvdata->config_type != TMC_CONFIG_TYPE_ETR))
1947 		return -EINVAL;
1948 
1949 	raw_spin_lock_irqsave(&drvdata->spinlock, flags);
1950 
1951 	/* RE-enable the TMC if need be */
1952 	if (coresight_get_mode(drvdata->csdev) == CS_MODE_SYSFS) {
1953 		/*
1954 		 * The trace run will continue with the same allocated trace
1955 		 * buffer. Since the tracer is still enabled drvdata::buf can't
1956 		 * be NULL.
1957 		 */
1958 		__tmc_etr_enable_hw(drvdata);
1959 	} else {
1960 		/*
1961 		 * The ETR is not tracing and the buffer was just read.
1962 		 * As such prepare to free the trace buffer.
1963 		 */
1964 		sysfs_buf = drvdata->sysfs_buf;
1965 		drvdata->sysfs_buf = NULL;
1966 	}
1967 
1968 	drvdata->reading = false;
1969 	raw_spin_unlock_irqrestore(&drvdata->spinlock, flags);
1970 
1971 	/* Free allocated memory out side of the spinlock */
1972 	if (sysfs_buf)
1973 		tmc_etr_free_sysfs_buf(sysfs_buf);
1974 
1975 	return 0;
1976 }
1977 
1978 static const char *const buf_modes_str[] = {
1979 	[ETR_MODE_FLAT]		= "flat",
1980 	[ETR_MODE_ETR_SG]	= "tmc-sg",
1981 	[ETR_MODE_CATU]		= "catu",
1982 	[ETR_MODE_RESRV]	= "resrv",
1983 	[ETR_MODE_AUTO]		= "auto",
1984 };
1985 
buf_modes_available_show(struct device * dev,struct device_attribute * attr,char * buf)1986 static ssize_t buf_modes_available_show(struct device *dev,
1987 					    struct device_attribute *attr, char *buf)
1988 {
1989 	struct etr_buf_hw buf_hw;
1990 	ssize_t size = 0;
1991 
1992 	get_etr_buf_hw(dev, &buf_hw);
1993 	size += sysfs_emit(buf, "%s ", buf_modes_str[ETR_MODE_AUTO]);
1994 	size += sysfs_emit_at(buf, size, "%s ", buf_modes_str[ETR_MODE_FLAT]);
1995 	if (buf_hw.has_etr_sg)
1996 		size += sysfs_emit_at(buf, size, "%s ", buf_modes_str[ETR_MODE_ETR_SG]);
1997 
1998 	if (buf_hw.has_catu)
1999 		size += sysfs_emit_at(buf, size, "%s ", buf_modes_str[ETR_MODE_CATU]);
2000 
2001 	if (buf_hw.has_resrv)
2002 		size += sysfs_emit_at(buf, size, "%s ", buf_modes_str[ETR_MODE_RESRV]);
2003 
2004 	size += sysfs_emit_at(buf, size, "\n");
2005 	return size;
2006 }
2007 static DEVICE_ATTR_RO(buf_modes_available);
2008 
buf_mode_preferred_show(struct device * dev,struct device_attribute * attr,char * buf)2009 static ssize_t buf_mode_preferred_show(struct device *dev,
2010 					 struct device_attribute *attr, char *buf)
2011 {
2012 	struct tmc_drvdata *drvdata = dev_get_drvdata(dev->parent);
2013 
2014 	return sysfs_emit(buf, "%s\n", buf_modes_str[drvdata->etr_mode]);
2015 }
2016 
buf_mode_set_resrv(struct tmc_drvdata * drvdata)2017 static int buf_mode_set_resrv(struct tmc_drvdata *drvdata)
2018 {
2019 	int err = -EBUSY;
2020 	unsigned long flags;
2021 	struct tmc_resrv_buf *rbuf;
2022 
2023 	rbuf = &drvdata->resrv_buf;
2024 
2025 	/* Ensure there are no active crashdata read sessions */
2026 	raw_spin_lock_irqsave(&drvdata->spinlock, flags);
2027 	if (!rbuf->reading) {
2028 		tmc_crashdata_set_invalid(drvdata);
2029 		rbuf->len = 0;
2030 		drvdata->etr_mode = ETR_MODE_RESRV;
2031 		err = 0;
2032 	}
2033 	raw_spin_unlock_irqrestore(&drvdata->spinlock, flags);
2034 	return err;
2035 }
2036 
buf_mode_preferred_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t size)2037 static ssize_t buf_mode_preferred_store(struct device *dev,
2038 					  struct device_attribute *attr,
2039 					  const char *buf, size_t size)
2040 {
2041 	struct tmc_drvdata *drvdata = dev_get_drvdata(dev->parent);
2042 	struct etr_buf_hw buf_hw;
2043 
2044 	get_etr_buf_hw(dev, &buf_hw);
2045 	if (sysfs_streq(buf, buf_modes_str[ETR_MODE_FLAT]))
2046 		drvdata->etr_mode = ETR_MODE_FLAT;
2047 	else if (sysfs_streq(buf, buf_modes_str[ETR_MODE_ETR_SG]) && buf_hw.has_etr_sg)
2048 		drvdata->etr_mode = ETR_MODE_ETR_SG;
2049 	else if (sysfs_streq(buf, buf_modes_str[ETR_MODE_CATU]) && buf_hw.has_catu)
2050 		drvdata->etr_mode = ETR_MODE_CATU;
2051 	else if (sysfs_streq(buf, buf_modes_str[ETR_MODE_RESRV]) && buf_hw.has_resrv)
2052 		return buf_mode_set_resrv(drvdata) ? : size;
2053 	else if (sysfs_streq(buf, buf_modes_str[ETR_MODE_AUTO]))
2054 		drvdata->etr_mode = ETR_MODE_AUTO;
2055 	else
2056 		return -EINVAL;
2057 	return size;
2058 }
2059 static DEVICE_ATTR_RW(buf_mode_preferred);
2060 
2061 static struct attribute *coresight_etr_attrs[] = {
2062 	&dev_attr_buf_modes_available.attr,
2063 	&dev_attr_buf_mode_preferred.attr,
2064 	NULL,
2065 };
2066 
2067 const struct attribute_group coresight_etr_group = {
2068 	.attrs = coresight_etr_attrs,
2069 };
2070