xref: /openbmc/linux/arch/parisc/kernel/cache.c (revision 22c9bb1e)
1 /*
2  * This file is subject to the terms and conditions of the GNU General Public
3  * License.  See the file "COPYING" in the main directory of this archive
4  * for more details.
5  *
6  * Copyright (C) 1999-2006 Helge Deller <deller@gmx.de> (07-13-1999)
7  * Copyright (C) 1999 SuSE GmbH Nuernberg
8  * Copyright (C) 2000 Philipp Rumpf (prumpf@tux.org)
9  *
10  * Cache and TLB management
11  *
12  */
13 
14 #include <linux/init.h>
15 #include <linux/kernel.h>
16 #include <linux/mm.h>
17 #include <linux/module.h>
18 #include <linux/seq_file.h>
19 #include <linux/pagemap.h>
20 #include <linux/sched.h>
21 #include <linux/sched/mm.h>
22 #include <linux/syscalls.h>
23 #include <asm/pdc.h>
24 #include <asm/cache.h>
25 #include <asm/cacheflush.h>
26 #include <asm/tlbflush.h>
27 #include <asm/page.h>
28 #include <asm/processor.h>
29 #include <asm/sections.h>
30 #include <asm/shmparam.h>
31 #include <asm/mmu_context.h>
32 #include <asm/cachectl.h>
33 
34 int split_tlb __ro_after_init;
35 int dcache_stride __ro_after_init;
36 int icache_stride __ro_after_init;
37 EXPORT_SYMBOL(dcache_stride);
38 
39 void flush_dcache_page_asm(unsigned long phys_addr, unsigned long vaddr);
40 EXPORT_SYMBOL(flush_dcache_page_asm);
41 void purge_dcache_page_asm(unsigned long phys_addr, unsigned long vaddr);
42 void flush_icache_page_asm(unsigned long phys_addr, unsigned long vaddr);
43 
44 /* Internal implementation in arch/parisc/kernel/pacache.S */
45 void flush_data_cache_local(void *);  /* flushes local data-cache only */
46 void flush_instruction_cache_local(void); /* flushes local code-cache only */
47 
48 /* On some machines (i.e., ones with the Merced bus), there can be
49  * only a single PxTLB broadcast at a time; this must be guaranteed
50  * by software. We need a spinlock around all TLB flushes to ensure
51  * this.
52  */
53 DEFINE_SPINLOCK(pa_tlb_flush_lock);
54 
55 #if defined(CONFIG_64BIT) && defined(CONFIG_SMP)
56 int pa_serialize_tlb_flushes __ro_after_init;
57 #endif
58 
59 struct pdc_cache_info cache_info __ro_after_init;
60 #ifndef CONFIG_PA20
61 struct pdc_btlb_info btlb_info __ro_after_init;
62 #endif
63 
64 DEFINE_STATIC_KEY_TRUE(parisc_has_cache);
65 DEFINE_STATIC_KEY_TRUE(parisc_has_dcache);
66 DEFINE_STATIC_KEY_TRUE(parisc_has_icache);
67 
68 static void cache_flush_local_cpu(void *dummy)
69 {
70 	if (static_branch_likely(&parisc_has_icache))
71 		flush_instruction_cache_local();
72 	if (static_branch_likely(&parisc_has_dcache))
73 		flush_data_cache_local(NULL);
74 }
75 
76 void flush_cache_all_local(void)
77 {
78 	cache_flush_local_cpu(NULL);
79 }
80 
81 void flush_cache_all(void)
82 {
83 	if (static_branch_likely(&parisc_has_cache))
84 		on_each_cpu(cache_flush_local_cpu, NULL, 1);
85 }
86 
87 static inline void flush_data_cache(void)
88 {
89 	if (static_branch_likely(&parisc_has_dcache))
90 		on_each_cpu(flush_data_cache_local, NULL, 1);
91 }
92 
93 
94 /* Kernel virtual address of pfn.  */
95 #define pfn_va(pfn)	__va(PFN_PHYS(pfn))
96 
97 void __update_cache(pte_t pte)
98 {
99 	unsigned long pfn = pte_pfn(pte);
100 	struct folio *folio;
101 	unsigned int nr;
102 
103 	/* We don't have pte special.  As a result, we can be called with
104 	   an invalid pfn and we don't need to flush the kernel dcache page.
105 	   This occurs with FireGL card in C8000.  */
106 	if (!pfn_valid(pfn))
107 		return;
108 
109 	folio = page_folio(pfn_to_page(pfn));
110 	pfn = folio_pfn(folio);
111 	nr = folio_nr_pages(folio);
112 	if (folio_flush_mapping(folio) &&
113 	    test_bit(PG_dcache_dirty, &folio->flags)) {
114 		while (nr--)
115 			flush_kernel_dcache_page_addr(pfn_va(pfn + nr));
116 		clear_bit(PG_dcache_dirty, &folio->flags);
117 	} else if (parisc_requires_coherency())
118 		while (nr--)
119 			flush_kernel_dcache_page_addr(pfn_va(pfn + nr));
120 }
121 
122 void
123 show_cache_info(struct seq_file *m)
124 {
125 	char buf[32];
126 
127 	seq_printf(m, "I-cache\t\t: %ld KB\n",
128 		cache_info.ic_size/1024 );
129 	if (cache_info.dc_loop != 1)
130 		snprintf(buf, 32, "%lu-way associative", cache_info.dc_loop);
131 	seq_printf(m, "D-cache\t\t: %ld KB (%s%s, %s, alias=%d)\n",
132 		cache_info.dc_size/1024,
133 		(cache_info.dc_conf.cc_wt ? "WT":"WB"),
134 		(cache_info.dc_conf.cc_sh ? ", shared I/D":""),
135 		((cache_info.dc_loop == 1) ? "direct mapped" : buf),
136 		cache_info.dc_conf.cc_alias
137 	);
138 	seq_printf(m, "ITLB entries\t: %ld\n" "DTLB entries\t: %ld%s\n",
139 		cache_info.it_size,
140 		cache_info.dt_size,
141 		cache_info.dt_conf.tc_sh ? " - shared with ITLB":""
142 	);
143 
144 #ifndef CONFIG_PA20
145 	/* BTLB - Block TLB */
146 	if (btlb_info.max_size==0) {
147 		seq_printf(m, "BTLB\t\t: not supported\n" );
148 	} else {
149 		seq_printf(m,
150 		"BTLB fixed\t: max. %d pages, pagesize=%d (%dMB)\n"
151 		"BTLB fix-entr.\t: %d instruction, %d data (%d combined)\n"
152 		"BTLB var-entr.\t: %d instruction, %d data (%d combined)\n",
153 		btlb_info.max_size, (int)4096,
154 		btlb_info.max_size>>8,
155 		btlb_info.fixed_range_info.num_i,
156 		btlb_info.fixed_range_info.num_d,
157 		btlb_info.fixed_range_info.num_comb,
158 		btlb_info.variable_range_info.num_i,
159 		btlb_info.variable_range_info.num_d,
160 		btlb_info.variable_range_info.num_comb
161 		);
162 	}
163 #endif
164 }
165 
166 void __init
167 parisc_cache_init(void)
168 {
169 	if (pdc_cache_info(&cache_info) < 0)
170 		panic("parisc_cache_init: pdc_cache_info failed");
171 
172 #if 0
173 	printk("ic_size %lx dc_size %lx it_size %lx\n",
174 		cache_info.ic_size,
175 		cache_info.dc_size,
176 		cache_info.it_size);
177 
178 	printk("DC  base 0x%lx stride 0x%lx count 0x%lx loop 0x%lx\n",
179 		cache_info.dc_base,
180 		cache_info.dc_stride,
181 		cache_info.dc_count,
182 		cache_info.dc_loop);
183 
184 	printk("dc_conf = 0x%lx  alias %d blk %d line %d shift %d\n",
185 		*(unsigned long *) (&cache_info.dc_conf),
186 		cache_info.dc_conf.cc_alias,
187 		cache_info.dc_conf.cc_block,
188 		cache_info.dc_conf.cc_line,
189 		cache_info.dc_conf.cc_shift);
190 	printk("	wt %d sh %d cst %d hv %d\n",
191 		cache_info.dc_conf.cc_wt,
192 		cache_info.dc_conf.cc_sh,
193 		cache_info.dc_conf.cc_cst,
194 		cache_info.dc_conf.cc_hv);
195 
196 	printk("IC  base 0x%lx stride 0x%lx count 0x%lx loop 0x%lx\n",
197 		cache_info.ic_base,
198 		cache_info.ic_stride,
199 		cache_info.ic_count,
200 		cache_info.ic_loop);
201 
202 	printk("IT  base 0x%lx stride 0x%lx count 0x%lx loop 0x%lx off_base 0x%lx off_stride 0x%lx off_count 0x%lx\n",
203 		cache_info.it_sp_base,
204 		cache_info.it_sp_stride,
205 		cache_info.it_sp_count,
206 		cache_info.it_loop,
207 		cache_info.it_off_base,
208 		cache_info.it_off_stride,
209 		cache_info.it_off_count);
210 
211 	printk("DT  base 0x%lx stride 0x%lx count 0x%lx loop 0x%lx off_base 0x%lx off_stride 0x%lx off_count 0x%lx\n",
212 		cache_info.dt_sp_base,
213 		cache_info.dt_sp_stride,
214 		cache_info.dt_sp_count,
215 		cache_info.dt_loop,
216 		cache_info.dt_off_base,
217 		cache_info.dt_off_stride,
218 		cache_info.dt_off_count);
219 
220 	printk("ic_conf = 0x%lx  alias %d blk %d line %d shift %d\n",
221 		*(unsigned long *) (&cache_info.ic_conf),
222 		cache_info.ic_conf.cc_alias,
223 		cache_info.ic_conf.cc_block,
224 		cache_info.ic_conf.cc_line,
225 		cache_info.ic_conf.cc_shift);
226 	printk("	wt %d sh %d cst %d hv %d\n",
227 		cache_info.ic_conf.cc_wt,
228 		cache_info.ic_conf.cc_sh,
229 		cache_info.ic_conf.cc_cst,
230 		cache_info.ic_conf.cc_hv);
231 
232 	printk("D-TLB conf: sh %d page %d cst %d aid %d sr %d\n",
233 		cache_info.dt_conf.tc_sh,
234 		cache_info.dt_conf.tc_page,
235 		cache_info.dt_conf.tc_cst,
236 		cache_info.dt_conf.tc_aid,
237 		cache_info.dt_conf.tc_sr);
238 
239 	printk("I-TLB conf: sh %d page %d cst %d aid %d sr %d\n",
240 		cache_info.it_conf.tc_sh,
241 		cache_info.it_conf.tc_page,
242 		cache_info.it_conf.tc_cst,
243 		cache_info.it_conf.tc_aid,
244 		cache_info.it_conf.tc_sr);
245 #endif
246 
247 	split_tlb = 0;
248 	if (cache_info.dt_conf.tc_sh == 0 || cache_info.dt_conf.tc_sh == 2) {
249 		if (cache_info.dt_conf.tc_sh == 2)
250 			printk(KERN_WARNING "Unexpected TLB configuration. "
251 			"Will flush I/D separately (could be optimized).\n");
252 
253 		split_tlb = 1;
254 	}
255 
256 	/* "New and Improved" version from Jim Hull
257 	 *	(1 << (cc_block-1)) * (cc_line << (4 + cnf.cc_shift))
258 	 * The following CAFL_STRIDE is an optimized version, see
259 	 * http://lists.parisc-linux.org/pipermail/parisc-linux/2004-June/023625.html
260 	 * http://lists.parisc-linux.org/pipermail/parisc-linux/2004-June/023671.html
261 	 */
262 #define CAFL_STRIDE(cnf) (cnf.cc_line << (3 + cnf.cc_block + cnf.cc_shift))
263 	dcache_stride = CAFL_STRIDE(cache_info.dc_conf);
264 	icache_stride = CAFL_STRIDE(cache_info.ic_conf);
265 #undef CAFL_STRIDE
266 
267 	if ((boot_cpu_data.pdc.capabilities & PDC_MODEL_NVA_MASK) ==
268 						PDC_MODEL_NVA_UNSUPPORTED) {
269 		printk(KERN_WARNING "parisc_cache_init: Only equivalent aliasing supported!\n");
270 #if 0
271 		panic("SMP kernel required to avoid non-equivalent aliasing");
272 #endif
273 	}
274 }
275 
276 void disable_sr_hashing(void)
277 {
278 	int srhash_type, retval;
279 	unsigned long space_bits;
280 
281 	switch (boot_cpu_data.cpu_type) {
282 	case pcx: /* We shouldn't get this far.  setup.c should prevent it. */
283 		BUG();
284 		return;
285 
286 	case pcxs:
287 	case pcxt:
288 	case pcxt_:
289 		srhash_type = SRHASH_PCXST;
290 		break;
291 
292 	case pcxl:
293 		srhash_type = SRHASH_PCXL;
294 		break;
295 
296 	case pcxl2: /* pcxl2 doesn't support space register hashing */
297 		return;
298 
299 	default: /* Currently all PA2.0 machines use the same ins. sequence */
300 		srhash_type = SRHASH_PA20;
301 		break;
302 	}
303 
304 	disable_sr_hashing_asm(srhash_type);
305 
306 	retval = pdc_spaceid_bits(&space_bits);
307 	/* If this procedure isn't implemented, don't panic. */
308 	if (retval < 0 && retval != PDC_BAD_OPTION)
309 		panic("pdc_spaceid_bits call failed.\n");
310 	if (space_bits != 0)
311 		panic("SpaceID hashing is still on!\n");
312 }
313 
314 static inline void
315 __flush_cache_page(struct vm_area_struct *vma, unsigned long vmaddr,
316 		   unsigned long physaddr)
317 {
318 	if (!static_branch_likely(&parisc_has_cache))
319 		return;
320 	preempt_disable();
321 	flush_dcache_page_asm(physaddr, vmaddr);
322 	if (vma->vm_flags & VM_EXEC)
323 		flush_icache_page_asm(physaddr, vmaddr);
324 	preempt_enable();
325 }
326 
327 static void flush_user_cache_page(struct vm_area_struct *vma, unsigned long vmaddr)
328 {
329 	unsigned long flags, space, pgd, prot;
330 #ifdef CONFIG_TLB_PTLOCK
331 	unsigned long pgd_lock;
332 #endif
333 
334 	vmaddr &= PAGE_MASK;
335 
336 	preempt_disable();
337 
338 	/* Set context for flush */
339 	local_irq_save(flags);
340 	prot = mfctl(8);
341 	space = mfsp(SR_USER);
342 	pgd = mfctl(25);
343 #ifdef CONFIG_TLB_PTLOCK
344 	pgd_lock = mfctl(28);
345 #endif
346 	switch_mm_irqs_off(NULL, vma->vm_mm, NULL);
347 	local_irq_restore(flags);
348 
349 	flush_user_dcache_range_asm(vmaddr, vmaddr + PAGE_SIZE);
350 	if (vma->vm_flags & VM_EXEC)
351 		flush_user_icache_range_asm(vmaddr, vmaddr + PAGE_SIZE);
352 	flush_tlb_page(vma, vmaddr);
353 
354 	/* Restore previous context */
355 	local_irq_save(flags);
356 #ifdef CONFIG_TLB_PTLOCK
357 	mtctl(pgd_lock, 28);
358 #endif
359 	mtctl(pgd, 25);
360 	mtsp(space, SR_USER);
361 	mtctl(prot, 8);
362 	local_irq_restore(flags);
363 
364 	preempt_enable();
365 }
366 
367 void flush_icache_pages(struct vm_area_struct *vma, struct page *page,
368 		unsigned int nr)
369 {
370 	void *kaddr = page_address(page);
371 
372 	for (;;) {
373 		flush_kernel_dcache_page_addr(kaddr);
374 		flush_kernel_icache_page(kaddr);
375 		if (--nr == 0)
376 			break;
377 		kaddr += PAGE_SIZE;
378 	}
379 }
380 
381 static inline pte_t *get_ptep(struct mm_struct *mm, unsigned long addr)
382 {
383 	pte_t *ptep = NULL;
384 	pgd_t *pgd = mm->pgd;
385 	p4d_t *p4d;
386 	pud_t *pud;
387 	pmd_t *pmd;
388 
389 	if (!pgd_none(*pgd)) {
390 		p4d = p4d_offset(pgd, addr);
391 		if (!p4d_none(*p4d)) {
392 			pud = pud_offset(p4d, addr);
393 			if (!pud_none(*pud)) {
394 				pmd = pmd_offset(pud, addr);
395 				if (!pmd_none(*pmd))
396 					ptep = pte_offset_map(pmd, addr);
397 			}
398 		}
399 	}
400 	return ptep;
401 }
402 
403 static inline bool pte_needs_flush(pte_t pte)
404 {
405 	return (pte_val(pte) & (_PAGE_PRESENT | _PAGE_ACCESSED | _PAGE_NO_CACHE))
406 		== (_PAGE_PRESENT | _PAGE_ACCESSED);
407 }
408 
409 void flush_dcache_folio(struct folio *folio)
410 {
411 	struct address_space *mapping = folio_flush_mapping(folio);
412 	struct vm_area_struct *vma;
413 	unsigned long addr, old_addr = 0;
414 	void *kaddr;
415 	unsigned long count = 0;
416 	unsigned long i, nr, flags;
417 	pgoff_t pgoff;
418 
419 	if (mapping && !mapping_mapped(mapping)) {
420 		set_bit(PG_dcache_dirty, &folio->flags);
421 		return;
422 	}
423 
424 	nr = folio_nr_pages(folio);
425 	kaddr = folio_address(folio);
426 	for (i = 0; i < nr; i++)
427 		flush_kernel_dcache_page_addr(kaddr + i * PAGE_SIZE);
428 
429 	if (!mapping)
430 		return;
431 
432 	pgoff = folio->index;
433 
434 	/*
435 	 * We have carefully arranged in arch_get_unmapped_area() that
436 	 * *any* mappings of a file are always congruently mapped (whether
437 	 * declared as MAP_PRIVATE or MAP_SHARED), so we only need
438 	 * to flush one address here for them all to become coherent
439 	 * on machines that support equivalent aliasing
440 	 */
441 	flush_dcache_mmap_lock_irqsave(mapping, flags);
442 	vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff + nr - 1) {
443 		unsigned long offset = pgoff - vma->vm_pgoff;
444 		unsigned long pfn = folio_pfn(folio);
445 
446 		addr = vma->vm_start;
447 		nr = folio_nr_pages(folio);
448 		if (offset > -nr) {
449 			pfn -= offset;
450 			nr += offset;
451 		} else {
452 			addr += offset * PAGE_SIZE;
453 		}
454 		if (addr + nr * PAGE_SIZE > vma->vm_end)
455 			nr = (vma->vm_end - addr) / PAGE_SIZE;
456 
457 		if (parisc_requires_coherency()) {
458 			for (i = 0; i < nr; i++) {
459 				pte_t *ptep = get_ptep(vma->vm_mm,
460 							addr + i * PAGE_SIZE);
461 				if (!ptep)
462 					continue;
463 				if (pte_needs_flush(*ptep))
464 					flush_user_cache_page(vma,
465 							addr + i * PAGE_SIZE);
466 				/* Optimise accesses to the same table? */
467 				pte_unmap(ptep);
468 			}
469 		} else {
470 			/*
471 			 * The TLB is the engine of coherence on parisc:
472 			 * The CPU is entitled to speculate any page
473 			 * with a TLB mapping, so here we kill the
474 			 * mapping then flush the page along a special
475 			 * flush only alias mapping. This guarantees that
476 			 * the page is no-longer in the cache for any
477 			 * process and nor may it be speculatively read
478 			 * in (until the user or kernel specifically
479 			 * accesses it, of course)
480 			 */
481 			for (i = 0; i < nr; i++)
482 				flush_tlb_page(vma, addr + i * PAGE_SIZE);
483 			if (old_addr == 0 || (old_addr & (SHM_COLOUR - 1))
484 					!= (addr & (SHM_COLOUR - 1))) {
485 				for (i = 0; i < nr; i++)
486 					__flush_cache_page(vma,
487 						addr + i * PAGE_SIZE,
488 						(pfn + i) * PAGE_SIZE);
489 				/*
490 				 * Software is allowed to have any number
491 				 * of private mappings to a page.
492 				 */
493 				if (!(vma->vm_flags & VM_SHARED))
494 					continue;
495 				if (old_addr)
496 					pr_err("INEQUIVALENT ALIASES 0x%lx and 0x%lx in file %pD\n",
497 						old_addr, addr, vma->vm_file);
498 				if (nr == folio_nr_pages(folio))
499 					old_addr = addr;
500 			}
501 		}
502 		WARN_ON(++count == 4096);
503 	}
504 	flush_dcache_mmap_unlock_irqrestore(mapping, flags);
505 }
506 EXPORT_SYMBOL(flush_dcache_folio);
507 
508 /* Defined in arch/parisc/kernel/pacache.S */
509 EXPORT_SYMBOL(flush_kernel_dcache_range_asm);
510 EXPORT_SYMBOL(flush_kernel_icache_range_asm);
511 
512 #define FLUSH_THRESHOLD 0x80000 /* 0.5MB */
513 static unsigned long parisc_cache_flush_threshold __ro_after_init = FLUSH_THRESHOLD;
514 
515 #define FLUSH_TLB_THRESHOLD (16*1024) /* 16 KiB minimum TLB threshold */
516 static unsigned long parisc_tlb_flush_threshold __ro_after_init = ~0UL;
517 
518 void __init parisc_setup_cache_timing(void)
519 {
520 	unsigned long rangetime, alltime;
521 	unsigned long size;
522 	unsigned long threshold, threshold2;
523 
524 	alltime = mfctl(16);
525 	flush_data_cache();
526 	alltime = mfctl(16) - alltime;
527 
528 	size = (unsigned long)(_end - _text);
529 	rangetime = mfctl(16);
530 	flush_kernel_dcache_range((unsigned long)_text, size);
531 	rangetime = mfctl(16) - rangetime;
532 
533 	printk(KERN_DEBUG "Whole cache flush %lu cycles, flushing %lu bytes %lu cycles\n",
534 		alltime, size, rangetime);
535 
536 	threshold = L1_CACHE_ALIGN((unsigned long)((uint64_t)size * alltime / rangetime));
537 	pr_info("Calculated flush threshold is %lu KiB\n",
538 		threshold/1024);
539 
540 	/*
541 	 * The threshold computed above isn't very reliable. The following
542 	 * heuristic works reasonably well on c8000/rp3440.
543 	 */
544 	threshold2 = cache_info.dc_size * num_online_cpus();
545 	parisc_cache_flush_threshold = threshold2;
546 	printk(KERN_INFO "Cache flush threshold set to %lu KiB\n",
547 		parisc_cache_flush_threshold/1024);
548 
549 	/* calculate TLB flush threshold */
550 
551 	/* On SMP machines, skip the TLB measure of kernel text which
552 	 * has been mapped as huge pages. */
553 	if (num_online_cpus() > 1 && !parisc_requires_coherency()) {
554 		threshold = max(cache_info.it_size, cache_info.dt_size);
555 		threshold *= PAGE_SIZE;
556 		threshold /= num_online_cpus();
557 		goto set_tlb_threshold;
558 	}
559 
560 	size = (unsigned long)_end - (unsigned long)_text;
561 	rangetime = mfctl(16);
562 	flush_tlb_kernel_range((unsigned long)_text, (unsigned long)_end);
563 	rangetime = mfctl(16) - rangetime;
564 
565 	alltime = mfctl(16);
566 	flush_tlb_all();
567 	alltime = mfctl(16) - alltime;
568 
569 	printk(KERN_INFO "Whole TLB flush %lu cycles, Range flush %lu bytes %lu cycles\n",
570 		alltime, size, rangetime);
571 
572 	threshold = PAGE_ALIGN((num_online_cpus() * size * alltime) / rangetime);
573 	printk(KERN_INFO "Calculated TLB flush threshold %lu KiB\n",
574 		threshold/1024);
575 
576 set_tlb_threshold:
577 	if (threshold > FLUSH_TLB_THRESHOLD)
578 		parisc_tlb_flush_threshold = threshold;
579 	else
580 		parisc_tlb_flush_threshold = FLUSH_TLB_THRESHOLD;
581 
582 	printk(KERN_INFO "TLB flush threshold set to %lu KiB\n",
583 		parisc_tlb_flush_threshold/1024);
584 }
585 
586 extern void purge_kernel_dcache_page_asm(unsigned long);
587 extern void clear_user_page_asm(void *, unsigned long);
588 extern void copy_user_page_asm(void *, void *, unsigned long);
589 
590 void flush_kernel_dcache_page_addr(const void *addr)
591 {
592 	unsigned long flags;
593 
594 	flush_kernel_dcache_page_asm(addr);
595 	purge_tlb_start(flags);
596 	pdtlb(SR_KERNEL, addr);
597 	purge_tlb_end(flags);
598 }
599 EXPORT_SYMBOL(flush_kernel_dcache_page_addr);
600 
601 static void flush_cache_page_if_present(struct vm_area_struct *vma,
602 	unsigned long vmaddr, unsigned long pfn)
603 {
604 	bool needs_flush = false;
605 	pte_t *ptep;
606 
607 	/*
608 	 * The pte check is racy and sometimes the flush will trigger
609 	 * a non-access TLB miss. Hopefully, the page has already been
610 	 * flushed.
611 	 */
612 	ptep = get_ptep(vma->vm_mm, vmaddr);
613 	if (ptep) {
614 		needs_flush = pte_needs_flush(*ptep);
615 		pte_unmap(ptep);
616 	}
617 	if (needs_flush)
618 		flush_cache_page(vma, vmaddr, pfn);
619 }
620 
621 void copy_user_highpage(struct page *to, struct page *from,
622 	unsigned long vaddr, struct vm_area_struct *vma)
623 {
624 	void *kto, *kfrom;
625 
626 	kfrom = kmap_local_page(from);
627 	kto = kmap_local_page(to);
628 	flush_cache_page_if_present(vma, vaddr, page_to_pfn(from));
629 	copy_page_asm(kto, kfrom);
630 	kunmap_local(kto);
631 	kunmap_local(kfrom);
632 }
633 
634 void copy_to_user_page(struct vm_area_struct *vma, struct page *page,
635 		unsigned long user_vaddr, void *dst, void *src, int len)
636 {
637 	flush_cache_page_if_present(vma, user_vaddr, page_to_pfn(page));
638 	memcpy(dst, src, len);
639 	flush_kernel_dcache_range_asm((unsigned long)dst, (unsigned long)dst + len);
640 }
641 
642 void copy_from_user_page(struct vm_area_struct *vma, struct page *page,
643 		unsigned long user_vaddr, void *dst, void *src, int len)
644 {
645 	flush_cache_page_if_present(vma, user_vaddr, page_to_pfn(page));
646 	memcpy(dst, src, len);
647 }
648 
649 /* __flush_tlb_range()
650  *
651  * returns 1 if all TLBs were flushed.
652  */
653 int __flush_tlb_range(unsigned long sid, unsigned long start,
654 		      unsigned long end)
655 {
656 	unsigned long flags;
657 
658 	if ((!IS_ENABLED(CONFIG_SMP) || !arch_irqs_disabled()) &&
659 	    end - start >= parisc_tlb_flush_threshold) {
660 		flush_tlb_all();
661 		return 1;
662 	}
663 
664 	/* Purge TLB entries for small ranges using the pdtlb and
665 	   pitlb instructions.  These instructions execute locally
666 	   but cause a purge request to be broadcast to other TLBs.  */
667 	while (start < end) {
668 		purge_tlb_start(flags);
669 		mtsp(sid, SR_TEMP1);
670 		pdtlb(SR_TEMP1, start);
671 		pitlb(SR_TEMP1, start);
672 		purge_tlb_end(flags);
673 		start += PAGE_SIZE;
674 	}
675 	return 0;
676 }
677 
678 static void flush_cache_pages(struct vm_area_struct *vma, unsigned long start, unsigned long end)
679 {
680 	unsigned long addr, pfn;
681 	pte_t *ptep;
682 
683 	for (addr = start; addr < end; addr += PAGE_SIZE) {
684 		bool needs_flush = false;
685 		/*
686 		 * The vma can contain pages that aren't present. Although
687 		 * the pte search is expensive, we need the pte to find the
688 		 * page pfn and to check whether the page should be flushed.
689 		 */
690 		ptep = get_ptep(vma->vm_mm, addr);
691 		if (ptep) {
692 			needs_flush = pte_needs_flush(*ptep);
693 			pfn = pte_pfn(*ptep);
694 			pte_unmap(ptep);
695 		}
696 		if (needs_flush) {
697 			if (parisc_requires_coherency()) {
698 				flush_user_cache_page(vma, addr);
699 			} else {
700 				if (WARN_ON(!pfn_valid(pfn)))
701 					return;
702 				__flush_cache_page(vma, addr, PFN_PHYS(pfn));
703 			}
704 		}
705 	}
706 }
707 
708 static inline unsigned long mm_total_size(struct mm_struct *mm)
709 {
710 	struct vm_area_struct *vma;
711 	unsigned long usize = 0;
712 	VMA_ITERATOR(vmi, mm, 0);
713 
714 	for_each_vma(vmi, vma) {
715 		if (usize >= parisc_cache_flush_threshold)
716 			break;
717 		usize += vma->vm_end - vma->vm_start;
718 	}
719 	return usize;
720 }
721 
722 void flush_cache_mm(struct mm_struct *mm)
723 {
724 	struct vm_area_struct *vma;
725 	VMA_ITERATOR(vmi, mm, 0);
726 
727 	/*
728 	 * Flushing the whole cache on each cpu takes forever on
729 	 * rp3440, etc. So, avoid it if the mm isn't too big.
730 	 *
731 	 * Note that we must flush the entire cache on machines
732 	 * with aliasing caches to prevent random segmentation
733 	 * faults.
734 	 */
735 	if (!parisc_requires_coherency()
736 	    ||  mm_total_size(mm) >= parisc_cache_flush_threshold) {
737 		if (WARN_ON(IS_ENABLED(CONFIG_SMP) && arch_irqs_disabled()))
738 			return;
739 		flush_tlb_all();
740 		flush_cache_all();
741 		return;
742 	}
743 
744 	/* Flush mm */
745 	for_each_vma(vmi, vma)
746 		flush_cache_pages(vma, vma->vm_start, vma->vm_end);
747 }
748 
749 void flush_cache_range(struct vm_area_struct *vma, unsigned long start, unsigned long end)
750 {
751 	if (!parisc_requires_coherency()
752 	    || end - start >= parisc_cache_flush_threshold) {
753 		if (WARN_ON(IS_ENABLED(CONFIG_SMP) && arch_irqs_disabled()))
754 			return;
755 		flush_tlb_range(vma, start, end);
756 		flush_cache_all();
757 		return;
758 	}
759 
760 	flush_cache_pages(vma, start, end);
761 }
762 
763 void flush_cache_page(struct vm_area_struct *vma, unsigned long vmaddr, unsigned long pfn)
764 {
765 	if (WARN_ON(!pfn_valid(pfn)))
766 		return;
767 	if (parisc_requires_coherency())
768 		flush_user_cache_page(vma, vmaddr);
769 	else
770 		__flush_cache_page(vma, vmaddr, PFN_PHYS(pfn));
771 }
772 
773 void flush_anon_page(struct vm_area_struct *vma, struct page *page, unsigned long vmaddr)
774 {
775 	if (!PageAnon(page))
776 		return;
777 
778 	if (parisc_requires_coherency()) {
779 		if (vma->vm_flags & VM_SHARED)
780 			flush_data_cache();
781 		else
782 			flush_user_cache_page(vma, vmaddr);
783 		return;
784 	}
785 
786 	flush_tlb_page(vma, vmaddr);
787 	preempt_disable();
788 	flush_dcache_page_asm(page_to_phys(page), vmaddr);
789 	preempt_enable();
790 }
791 
792 void flush_kernel_vmap_range(void *vaddr, int size)
793 {
794 	unsigned long start = (unsigned long)vaddr;
795 	unsigned long end = start + size;
796 
797 	if ((!IS_ENABLED(CONFIG_SMP) || !arch_irqs_disabled()) &&
798 	    (unsigned long)size >= parisc_cache_flush_threshold) {
799 		flush_tlb_kernel_range(start, end);
800 		flush_data_cache();
801 		return;
802 	}
803 
804 	flush_kernel_dcache_range_asm(start, end);
805 	flush_tlb_kernel_range(start, end);
806 }
807 EXPORT_SYMBOL(flush_kernel_vmap_range);
808 
809 void invalidate_kernel_vmap_range(void *vaddr, int size)
810 {
811 	unsigned long start = (unsigned long)vaddr;
812 	unsigned long end = start + size;
813 
814 	/* Ensure DMA is complete */
815 	asm_syncdma();
816 
817 	if ((!IS_ENABLED(CONFIG_SMP) || !arch_irqs_disabled()) &&
818 	    (unsigned long)size >= parisc_cache_flush_threshold) {
819 		flush_tlb_kernel_range(start, end);
820 		flush_data_cache();
821 		return;
822 	}
823 
824 	purge_kernel_dcache_range_asm(start, end);
825 	flush_tlb_kernel_range(start, end);
826 }
827 EXPORT_SYMBOL(invalidate_kernel_vmap_range);
828 
829 
830 SYSCALL_DEFINE3(cacheflush, unsigned long, addr, unsigned long, bytes,
831 	unsigned int, cache)
832 {
833 	unsigned long start, end;
834 	ASM_EXCEPTIONTABLE_VAR(error);
835 
836 	if (bytes == 0)
837 		return 0;
838 	if (!access_ok((void __user *) addr, bytes))
839 		return -EFAULT;
840 
841 	end = addr + bytes;
842 
843 	if (cache & DCACHE) {
844 		start = addr;
845 		__asm__ __volatile__ (
846 #ifdef CONFIG_64BIT
847 			"1: cmpb,*<<,n	%0,%2,1b\n"
848 #else
849 			"1: cmpb,<<,n	%0,%2,1b\n"
850 #endif
851 			"   fic,m	%3(%4,%0)\n"
852 			"2: sync\n"
853 			ASM_EXCEPTIONTABLE_ENTRY_EFAULT(1b, 2b)
854 			: "+r" (start), "+r" (error)
855 			: "r" (end), "r" (dcache_stride), "i" (SR_USER));
856 	}
857 
858 	if (cache & ICACHE && error == 0) {
859 		start = addr;
860 		__asm__ __volatile__ (
861 #ifdef CONFIG_64BIT
862 			"1: cmpb,*<<,n	%0,%2,1b\n"
863 #else
864 			"1: cmpb,<<,n	%0,%2,1b\n"
865 #endif
866 			"   fdc,m	%3(%4,%0)\n"
867 			"2: sync\n"
868 			ASM_EXCEPTIONTABLE_ENTRY_EFAULT(1b, 2b)
869 			: "+r" (start), "+r" (error)
870 			: "r" (end), "r" (icache_stride), "i" (SR_USER));
871 	}
872 
873 	return error;
874 }
875