xref: /openbmc/linux/arch/mips/mm/tlbex.c (revision e15a5365)
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  * Synthesize TLB refill handlers at runtime.
7  *
8  * Copyright (C) 2004, 2005, 2006, 2008	 Thiemo Seufer
9  * Copyright (C) 2005, 2007, 2008, 2009	 Maciej W. Rozycki
10  * Copyright (C) 2006  Ralf Baechle (ralf@linux-mips.org)
11  * Copyright (C) 2008, 2009 Cavium Networks, Inc.
12  * Copyright (C) 2011  MIPS Technologies, Inc.
13  *
14  * ... and the days got worse and worse and now you see
15  * I've gone completely out of my mind.
16  *
17  * They're coming to take me a away haha
18  * they're coming to take me a away hoho hihi haha
19  * to the funny farm where code is beautiful all the time ...
20  *
21  * (Condolences to Napoleon XIV)
22  */
23 
24 #include <linux/bug.h>
25 #include <linux/export.h>
26 #include <linux/kernel.h>
27 #include <linux/types.h>
28 #include <linux/smp.h>
29 #include <linux/string.h>
30 #include <linux/cache.h>
31 #include <linux/pgtable.h>
32 
33 #include <asm/cacheflush.h>
34 #include <asm/cpu-type.h>
35 #include <asm/mmu_context.h>
36 #include <asm/war.h>
37 #include <asm/uasm.h>
38 #include <asm/setup.h>
39 #include <asm/tlbex.h>
40 
41 static int mips_xpa_disabled;
42 
43 static int __init xpa_disable(char *s)
44 {
45 	mips_xpa_disabled = 1;
46 
47 	return 1;
48 }
49 
50 __setup("noxpa", xpa_disable);
51 
52 /*
53  * TLB load/store/modify handlers.
54  *
55  * Only the fastpath gets synthesized at runtime, the slowpath for
56  * do_page_fault remains normal asm.
57  */
58 extern void tlb_do_page_fault_0(void);
59 extern void tlb_do_page_fault_1(void);
60 
61 struct work_registers {
62 	int r1;
63 	int r2;
64 	int r3;
65 };
66 
67 struct tlb_reg_save {
68 	unsigned long a;
69 	unsigned long b;
70 } ____cacheline_aligned_in_smp;
71 
72 static struct tlb_reg_save handler_reg_save[NR_CPUS];
73 
74 static inline int r45k_bvahwbug(void)
75 {
76 	/* XXX: We should probe for the presence of this bug, but we don't. */
77 	return 0;
78 }
79 
80 static inline int r4k_250MHZhwbug(void)
81 {
82 	/* XXX: We should probe for the presence of this bug, but we don't. */
83 	return 0;
84 }
85 
86 extern int sb1250_m3_workaround_needed(void);
87 
88 static inline int __maybe_unused bcm1250_m3_war(void)
89 {
90 	if (IS_ENABLED(CONFIG_SB1_PASS_2_WORKAROUNDS))
91 		return sb1250_m3_workaround_needed();
92 	return 0;
93 }
94 
95 static inline int __maybe_unused r10000_llsc_war(void)
96 {
97 	return IS_ENABLED(CONFIG_WAR_R10000_LLSC);
98 }
99 
100 static int use_bbit_insns(void)
101 {
102 	switch (current_cpu_type()) {
103 	case CPU_CAVIUM_OCTEON:
104 	case CPU_CAVIUM_OCTEON_PLUS:
105 	case CPU_CAVIUM_OCTEON2:
106 	case CPU_CAVIUM_OCTEON3:
107 		return 1;
108 	default:
109 		return 0;
110 	}
111 }
112 
113 static int use_lwx_insns(void)
114 {
115 	switch (current_cpu_type()) {
116 	case CPU_CAVIUM_OCTEON2:
117 	case CPU_CAVIUM_OCTEON3:
118 		return 1;
119 	default:
120 		return 0;
121 	}
122 }
123 #if defined(CONFIG_CAVIUM_OCTEON_CVMSEG_SIZE) && \
124     CONFIG_CAVIUM_OCTEON_CVMSEG_SIZE > 0
125 static bool scratchpad_available(void)
126 {
127 	return true;
128 }
129 static int scratchpad_offset(int i)
130 {
131 	/*
132 	 * CVMSEG starts at address -32768 and extends for
133 	 * CAVIUM_OCTEON_CVMSEG_SIZE 128 byte cache lines.
134 	 */
135 	i += 1; /* Kernel use starts at the top and works down. */
136 	return CONFIG_CAVIUM_OCTEON_CVMSEG_SIZE * 128 - (8 * i) - 32768;
137 }
138 #else
139 static bool scratchpad_available(void)
140 {
141 	return false;
142 }
143 static int scratchpad_offset(int i)
144 {
145 	BUG();
146 	/* Really unreachable, but evidently some GCC want this. */
147 	return 0;
148 }
149 #endif
150 /*
151  * Found by experiment: At least some revisions of the 4kc throw under
152  * some circumstances a machine check exception, triggered by invalid
153  * values in the index register.  Delaying the tlbp instruction until
154  * after the next branch,  plus adding an additional nop in front of
155  * tlbwi/tlbwr avoids the invalid index register values. Nobody knows
156  * why; it's not an issue caused by the core RTL.
157  *
158  */
159 static int m4kc_tlbp_war(void)
160 {
161 	return current_cpu_type() == CPU_4KC;
162 }
163 
164 /* Handle labels (which must be positive integers). */
165 enum label_id {
166 	label_second_part = 1,
167 	label_leave,
168 	label_vmalloc,
169 	label_vmalloc_done,
170 	label_tlbw_hazard_0,
171 	label_split = label_tlbw_hazard_0 + 8,
172 	label_tlbl_goaround1,
173 	label_tlbl_goaround2,
174 	label_nopage_tlbl,
175 	label_nopage_tlbs,
176 	label_nopage_tlbm,
177 	label_smp_pgtable_change,
178 	label_r3000_write_probe_fail,
179 	label_large_segbits_fault,
180 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
181 	label_tlb_huge_update,
182 #endif
183 };
184 
185 UASM_L_LA(_second_part)
186 UASM_L_LA(_leave)
187 UASM_L_LA(_vmalloc)
188 UASM_L_LA(_vmalloc_done)
189 /* _tlbw_hazard_x is handled differently.  */
190 UASM_L_LA(_split)
191 UASM_L_LA(_tlbl_goaround1)
192 UASM_L_LA(_tlbl_goaround2)
193 UASM_L_LA(_nopage_tlbl)
194 UASM_L_LA(_nopage_tlbs)
195 UASM_L_LA(_nopage_tlbm)
196 UASM_L_LA(_smp_pgtable_change)
197 UASM_L_LA(_r3000_write_probe_fail)
198 UASM_L_LA(_large_segbits_fault)
199 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
200 UASM_L_LA(_tlb_huge_update)
201 #endif
202 
203 static int hazard_instance;
204 
205 static void uasm_bgezl_hazard(u32 **p, struct uasm_reloc **r, int instance)
206 {
207 	switch (instance) {
208 	case 0 ... 7:
209 		uasm_il_bgezl(p, r, 0, label_tlbw_hazard_0 + instance);
210 		return;
211 	default:
212 		BUG();
213 	}
214 }
215 
216 static void uasm_bgezl_label(struct uasm_label **l, u32 **p, int instance)
217 {
218 	switch (instance) {
219 	case 0 ... 7:
220 		uasm_build_label(l, *p, label_tlbw_hazard_0 + instance);
221 		break;
222 	default:
223 		BUG();
224 	}
225 }
226 
227 /*
228  * pgtable bits are assigned dynamically depending on processor feature
229  * and statically based on kernel configuration.  This spits out the actual
230  * values the kernel is using.	Required to make sense from disassembled
231  * TLB exception handlers.
232  */
233 static void output_pgtable_bits_defines(void)
234 {
235 #define pr_define(fmt, ...)					\
236 	pr_debug("#define " fmt, ##__VA_ARGS__)
237 
238 	pr_debug("#include <asm/asm.h>\n");
239 	pr_debug("#include <asm/regdef.h>\n");
240 	pr_debug("\n");
241 
242 	pr_define("_PAGE_PRESENT_SHIFT %d\n", _PAGE_PRESENT_SHIFT);
243 	pr_define("_PAGE_NO_READ_SHIFT %d\n", _PAGE_NO_READ_SHIFT);
244 	pr_define("_PAGE_WRITE_SHIFT %d\n", _PAGE_WRITE_SHIFT);
245 	pr_define("_PAGE_ACCESSED_SHIFT %d\n", _PAGE_ACCESSED_SHIFT);
246 	pr_define("_PAGE_MODIFIED_SHIFT %d\n", _PAGE_MODIFIED_SHIFT);
247 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
248 	pr_define("_PAGE_HUGE_SHIFT %d\n", _PAGE_HUGE_SHIFT);
249 #endif
250 #ifdef _PAGE_NO_EXEC_SHIFT
251 	if (cpu_has_rixi)
252 		pr_define("_PAGE_NO_EXEC_SHIFT %d\n", _PAGE_NO_EXEC_SHIFT);
253 #endif
254 	pr_define("_PAGE_GLOBAL_SHIFT %d\n", _PAGE_GLOBAL_SHIFT);
255 	pr_define("_PAGE_VALID_SHIFT %d\n", _PAGE_VALID_SHIFT);
256 	pr_define("_PAGE_DIRTY_SHIFT %d\n", _PAGE_DIRTY_SHIFT);
257 	pr_define("_PFN_SHIFT %d\n", _PFN_SHIFT);
258 	pr_debug("\n");
259 }
260 
261 static inline void dump_handler(const char *symbol, const void *start, const void *end)
262 {
263 	unsigned int count = (end - start) / sizeof(u32);
264 	const u32 *handler = start;
265 	int i;
266 
267 	pr_debug("LEAF(%s)\n", symbol);
268 
269 	pr_debug("\t.set push\n");
270 	pr_debug("\t.set noreorder\n");
271 
272 	for (i = 0; i < count; i++)
273 		pr_debug("\t.word\t0x%08x\t\t# %p\n", handler[i], &handler[i]);
274 
275 	pr_debug("\t.set\tpop\n");
276 
277 	pr_debug("\tEND(%s)\n", symbol);
278 }
279 
280 /* The only general purpose registers allowed in TLB handlers. */
281 #define K0		26
282 #define K1		27
283 
284 /* Some CP0 registers */
285 #define C0_INDEX	0, 0
286 #define C0_ENTRYLO0	2, 0
287 #define C0_TCBIND	2, 2
288 #define C0_ENTRYLO1	3, 0
289 #define C0_CONTEXT	4, 0
290 #define C0_PAGEMASK	5, 0
291 #define C0_PWBASE	5, 5
292 #define C0_PWFIELD	5, 6
293 #define C0_PWSIZE	5, 7
294 #define C0_PWCTL	6, 6
295 #define C0_BADVADDR	8, 0
296 #define C0_PGD		9, 7
297 #define C0_ENTRYHI	10, 0
298 #define C0_EPC		14, 0
299 #define C0_XCONTEXT	20, 0
300 
301 #ifdef CONFIG_64BIT
302 # define GET_CONTEXT(buf, reg) UASM_i_MFC0(buf, reg, C0_XCONTEXT)
303 #else
304 # define GET_CONTEXT(buf, reg) UASM_i_MFC0(buf, reg, C0_CONTEXT)
305 #endif
306 
307 /* The worst case length of the handler is around 18 instructions for
308  * R3000-style TLBs and up to 63 instructions for R4000-style TLBs.
309  * Maximum space available is 32 instructions for R3000 and 64
310  * instructions for R4000.
311  *
312  * We deliberately chose a buffer size of 128, so we won't scribble
313  * over anything important on overflow before we panic.
314  */
315 static u32 tlb_handler[128];
316 
317 /* simply assume worst case size for labels and relocs */
318 static struct uasm_label labels[128];
319 static struct uasm_reloc relocs[128];
320 
321 static int check_for_high_segbits;
322 static bool fill_includes_sw_bits;
323 
324 static unsigned int kscratch_used_mask;
325 
326 static inline int __maybe_unused c0_kscratch(void)
327 {
328 	switch (current_cpu_type()) {
329 	case CPU_XLP:
330 	case CPU_XLR:
331 		return 22;
332 	default:
333 		return 31;
334 	}
335 }
336 
337 static int allocate_kscratch(void)
338 {
339 	int r;
340 	unsigned int a = cpu_data[0].kscratch_mask & ~kscratch_used_mask;
341 
342 	r = ffs(a);
343 
344 	if (r == 0)
345 		return -1;
346 
347 	r--; /* make it zero based */
348 
349 	kscratch_used_mask |= (1 << r);
350 
351 	return r;
352 }
353 
354 static int scratch_reg;
355 int pgd_reg;
356 EXPORT_SYMBOL_GPL(pgd_reg);
357 enum vmalloc64_mode {not_refill, refill_scratch, refill_noscratch};
358 
359 static struct work_registers build_get_work_registers(u32 **p)
360 {
361 	struct work_registers r;
362 
363 	if (scratch_reg >= 0) {
364 		/* Save in CPU local C0_KScratch? */
365 		UASM_i_MTC0(p, 1, c0_kscratch(), scratch_reg);
366 		r.r1 = K0;
367 		r.r2 = K1;
368 		r.r3 = 1;
369 		return r;
370 	}
371 
372 	if (num_possible_cpus() > 1) {
373 		/* Get smp_processor_id */
374 		UASM_i_CPUID_MFC0(p, K0, SMP_CPUID_REG);
375 		UASM_i_SRL_SAFE(p, K0, K0, SMP_CPUID_REGSHIFT);
376 
377 		/* handler_reg_save index in K0 */
378 		UASM_i_SLL(p, K0, K0, ilog2(sizeof(struct tlb_reg_save)));
379 
380 		UASM_i_LA(p, K1, (long)&handler_reg_save);
381 		UASM_i_ADDU(p, K0, K0, K1);
382 	} else {
383 		UASM_i_LA(p, K0, (long)&handler_reg_save);
384 	}
385 	/* K0 now points to save area, save $1 and $2  */
386 	UASM_i_SW(p, 1, offsetof(struct tlb_reg_save, a), K0);
387 	UASM_i_SW(p, 2, offsetof(struct tlb_reg_save, b), K0);
388 
389 	r.r1 = K1;
390 	r.r2 = 1;
391 	r.r3 = 2;
392 	return r;
393 }
394 
395 static void build_restore_work_registers(u32 **p)
396 {
397 	if (scratch_reg >= 0) {
398 		uasm_i_ehb(p);
399 		UASM_i_MFC0(p, 1, c0_kscratch(), scratch_reg);
400 		return;
401 	}
402 	/* K0 already points to save area, restore $1 and $2  */
403 	UASM_i_LW(p, 1, offsetof(struct tlb_reg_save, a), K0);
404 	UASM_i_LW(p, 2, offsetof(struct tlb_reg_save, b), K0);
405 }
406 
407 #ifndef CONFIG_MIPS_PGD_C0_CONTEXT
408 
409 /*
410  * CONFIG_MIPS_PGD_C0_CONTEXT implies 64 bit and lack of pgd_current,
411  * we cannot do r3000 under these circumstances.
412  *
413  * The R3000 TLB handler is simple.
414  */
415 static void build_r3000_tlb_refill_handler(void)
416 {
417 	long pgdc = (long)pgd_current;
418 	u32 *p;
419 
420 	memset(tlb_handler, 0, sizeof(tlb_handler));
421 	p = tlb_handler;
422 
423 	uasm_i_mfc0(&p, K0, C0_BADVADDR);
424 	uasm_i_lui(&p, K1, uasm_rel_hi(pgdc)); /* cp0 delay */
425 	uasm_i_lw(&p, K1, uasm_rel_lo(pgdc), K1);
426 	uasm_i_srl(&p, K0, K0, 22); /* load delay */
427 	uasm_i_sll(&p, K0, K0, 2);
428 	uasm_i_addu(&p, K1, K1, K0);
429 	uasm_i_mfc0(&p, K0, C0_CONTEXT);
430 	uasm_i_lw(&p, K1, 0, K1); /* cp0 delay */
431 	uasm_i_andi(&p, K0, K0, 0xffc); /* load delay */
432 	uasm_i_addu(&p, K1, K1, K0);
433 	uasm_i_lw(&p, K0, 0, K1);
434 	uasm_i_nop(&p); /* load delay */
435 	uasm_i_mtc0(&p, K0, C0_ENTRYLO0);
436 	uasm_i_mfc0(&p, K1, C0_EPC); /* cp0 delay */
437 	uasm_i_tlbwr(&p); /* cp0 delay */
438 	uasm_i_jr(&p, K1);
439 	uasm_i_rfe(&p); /* branch delay */
440 
441 	if (p > tlb_handler + 32)
442 		panic("TLB refill handler space exceeded");
443 
444 	pr_debug("Wrote TLB refill handler (%u instructions).\n",
445 		 (unsigned int)(p - tlb_handler));
446 
447 	memcpy((void *)ebase, tlb_handler, 0x80);
448 	local_flush_icache_range(ebase, ebase + 0x80);
449 	dump_handler("r3000_tlb_refill", (u32 *)ebase, (u32 *)(ebase + 0x80));
450 }
451 #endif /* CONFIG_MIPS_PGD_C0_CONTEXT */
452 
453 /*
454  * The R4000 TLB handler is much more complicated. We have two
455  * consecutive handler areas with 32 instructions space each.
456  * Since they aren't used at the same time, we can overflow in the
457  * other one.To keep things simple, we first assume linear space,
458  * then we relocate it to the final handler layout as needed.
459  */
460 static u32 final_handler[64];
461 
462 /*
463  * Hazards
464  *
465  * From the IDT errata for the QED RM5230 (Nevada), processor revision 1.0:
466  * 2. A timing hazard exists for the TLBP instruction.
467  *
468  *	stalling_instruction
469  *	TLBP
470  *
471  * The JTLB is being read for the TLBP throughout the stall generated by the
472  * previous instruction. This is not really correct as the stalling instruction
473  * can modify the address used to access the JTLB.  The failure symptom is that
474  * the TLBP instruction will use an address created for the stalling instruction
475  * and not the address held in C0_ENHI and thus report the wrong results.
476  *
477  * The software work-around is to not allow the instruction preceding the TLBP
478  * to stall - make it an NOP or some other instruction guaranteed not to stall.
479  *
480  * Errata 2 will not be fixed.	This errata is also on the R5000.
481  *
482  * As if we MIPS hackers wouldn't know how to nop pipelines happy ...
483  */
484 static void __maybe_unused build_tlb_probe_entry(u32 **p)
485 {
486 	switch (current_cpu_type()) {
487 	/* Found by experiment: R4600 v2.0/R4700 needs this, too.  */
488 	case CPU_R4600:
489 	case CPU_R4700:
490 	case CPU_R5000:
491 	case CPU_NEVADA:
492 		uasm_i_nop(p);
493 		uasm_i_tlbp(p);
494 		break;
495 
496 	default:
497 		uasm_i_tlbp(p);
498 		break;
499 	}
500 }
501 
502 void build_tlb_write_entry(u32 **p, struct uasm_label **l,
503 			   struct uasm_reloc **r,
504 			   enum tlb_write_entry wmode)
505 {
506 	void(*tlbw)(u32 **) = NULL;
507 
508 	switch (wmode) {
509 	case tlb_random: tlbw = uasm_i_tlbwr; break;
510 	case tlb_indexed: tlbw = uasm_i_tlbwi; break;
511 	}
512 
513 	if (cpu_has_mips_r2_r6) {
514 		if (cpu_has_mips_r2_exec_hazard)
515 			uasm_i_ehb(p);
516 		tlbw(p);
517 		return;
518 	}
519 
520 	switch (current_cpu_type()) {
521 	case CPU_R4000PC:
522 	case CPU_R4000SC:
523 	case CPU_R4000MC:
524 	case CPU_R4400PC:
525 	case CPU_R4400SC:
526 	case CPU_R4400MC:
527 		/*
528 		 * This branch uses up a mtc0 hazard nop slot and saves
529 		 * two nops after the tlbw instruction.
530 		 */
531 		uasm_bgezl_hazard(p, r, hazard_instance);
532 		tlbw(p);
533 		uasm_bgezl_label(l, p, hazard_instance);
534 		hazard_instance++;
535 		uasm_i_nop(p);
536 		break;
537 
538 	case CPU_R4600:
539 	case CPU_R4700:
540 		uasm_i_nop(p);
541 		tlbw(p);
542 		uasm_i_nop(p);
543 		break;
544 
545 	case CPU_R5000:
546 	case CPU_NEVADA:
547 		uasm_i_nop(p); /* QED specifies 2 nops hazard */
548 		uasm_i_nop(p); /* QED specifies 2 nops hazard */
549 		tlbw(p);
550 		break;
551 
552 	case CPU_5KC:
553 	case CPU_TX49XX:
554 	case CPU_PR4450:
555 	case CPU_XLR:
556 		uasm_i_nop(p);
557 		tlbw(p);
558 		break;
559 
560 	case CPU_R10000:
561 	case CPU_R12000:
562 	case CPU_R14000:
563 	case CPU_R16000:
564 	case CPU_4KC:
565 	case CPU_4KEC:
566 	case CPU_M14KC:
567 	case CPU_M14KEC:
568 	case CPU_SB1:
569 	case CPU_SB1A:
570 	case CPU_4KSC:
571 	case CPU_20KC:
572 	case CPU_25KF:
573 	case CPU_BMIPS32:
574 	case CPU_BMIPS3300:
575 	case CPU_BMIPS4350:
576 	case CPU_BMIPS4380:
577 	case CPU_BMIPS5000:
578 	case CPU_LOONGSON2EF:
579 	case CPU_LOONGSON64:
580 	case CPU_R5500:
581 		if (m4kc_tlbp_war())
582 			uasm_i_nop(p);
583 		fallthrough;
584 	case CPU_ALCHEMY:
585 		tlbw(p);
586 		break;
587 
588 	case CPU_RM7000:
589 		uasm_i_nop(p);
590 		uasm_i_nop(p);
591 		uasm_i_nop(p);
592 		uasm_i_nop(p);
593 		tlbw(p);
594 		break;
595 
596 	case CPU_VR4111:
597 	case CPU_VR4121:
598 	case CPU_VR4122:
599 	case CPU_VR4181:
600 	case CPU_VR4181A:
601 		uasm_i_nop(p);
602 		uasm_i_nop(p);
603 		tlbw(p);
604 		uasm_i_nop(p);
605 		uasm_i_nop(p);
606 		break;
607 
608 	case CPU_VR4131:
609 	case CPU_VR4133:
610 		uasm_i_nop(p);
611 		uasm_i_nop(p);
612 		tlbw(p);
613 		break;
614 
615 	case CPU_XBURST:
616 		tlbw(p);
617 		uasm_i_nop(p);
618 		break;
619 
620 	default:
621 		panic("No TLB refill handler yet (CPU type: %d)",
622 		      current_cpu_type());
623 		break;
624 	}
625 }
626 EXPORT_SYMBOL_GPL(build_tlb_write_entry);
627 
628 static __maybe_unused void build_convert_pte_to_entrylo(u32 **p,
629 							unsigned int reg)
630 {
631 	if (_PAGE_GLOBAL_SHIFT == 0) {
632 		/* pte_t is already in EntryLo format */
633 		return;
634 	}
635 
636 	if (cpu_has_rixi && !!_PAGE_NO_EXEC) {
637 		if (fill_includes_sw_bits) {
638 			UASM_i_ROTR(p, reg, reg, ilog2(_PAGE_GLOBAL));
639 		} else {
640 			UASM_i_SRL(p, reg, reg, ilog2(_PAGE_NO_EXEC));
641 			UASM_i_ROTR(p, reg, reg,
642 				    ilog2(_PAGE_GLOBAL) - ilog2(_PAGE_NO_EXEC));
643 		}
644 	} else {
645 #ifdef CONFIG_PHYS_ADDR_T_64BIT
646 		uasm_i_dsrl_safe(p, reg, reg, ilog2(_PAGE_GLOBAL));
647 #else
648 		UASM_i_SRL(p, reg, reg, ilog2(_PAGE_GLOBAL));
649 #endif
650 	}
651 }
652 
653 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
654 
655 static void build_restore_pagemask(u32 **p, struct uasm_reloc **r,
656 				   unsigned int tmp, enum label_id lid,
657 				   int restore_scratch)
658 {
659 	if (restore_scratch) {
660 		/*
661 		 * Ensure the MFC0 below observes the value written to the
662 		 * KScratch register by the prior MTC0.
663 		 */
664 		if (scratch_reg >= 0)
665 			uasm_i_ehb(p);
666 
667 		/* Reset default page size */
668 		if (PM_DEFAULT_MASK >> 16) {
669 			uasm_i_lui(p, tmp, PM_DEFAULT_MASK >> 16);
670 			uasm_i_ori(p, tmp, tmp, PM_DEFAULT_MASK & 0xffff);
671 			uasm_i_mtc0(p, tmp, C0_PAGEMASK);
672 			uasm_il_b(p, r, lid);
673 		} else if (PM_DEFAULT_MASK) {
674 			uasm_i_ori(p, tmp, 0, PM_DEFAULT_MASK);
675 			uasm_i_mtc0(p, tmp, C0_PAGEMASK);
676 			uasm_il_b(p, r, lid);
677 		} else {
678 			uasm_i_mtc0(p, 0, C0_PAGEMASK);
679 			uasm_il_b(p, r, lid);
680 		}
681 		if (scratch_reg >= 0)
682 			UASM_i_MFC0(p, 1, c0_kscratch(), scratch_reg);
683 		else
684 			UASM_i_LW(p, 1, scratchpad_offset(0), 0);
685 	} else {
686 		/* Reset default page size */
687 		if (PM_DEFAULT_MASK >> 16) {
688 			uasm_i_lui(p, tmp, PM_DEFAULT_MASK >> 16);
689 			uasm_i_ori(p, tmp, tmp, PM_DEFAULT_MASK & 0xffff);
690 			uasm_il_b(p, r, lid);
691 			uasm_i_mtc0(p, tmp, C0_PAGEMASK);
692 		} else if (PM_DEFAULT_MASK) {
693 			uasm_i_ori(p, tmp, 0, PM_DEFAULT_MASK);
694 			uasm_il_b(p, r, lid);
695 			uasm_i_mtc0(p, tmp, C0_PAGEMASK);
696 		} else {
697 			uasm_il_b(p, r, lid);
698 			uasm_i_mtc0(p, 0, C0_PAGEMASK);
699 		}
700 	}
701 }
702 
703 static void build_huge_tlb_write_entry(u32 **p, struct uasm_label **l,
704 				       struct uasm_reloc **r,
705 				       unsigned int tmp,
706 				       enum tlb_write_entry wmode,
707 				       int restore_scratch)
708 {
709 	/* Set huge page tlb entry size */
710 	uasm_i_lui(p, tmp, PM_HUGE_MASK >> 16);
711 	uasm_i_ori(p, tmp, tmp, PM_HUGE_MASK & 0xffff);
712 	uasm_i_mtc0(p, tmp, C0_PAGEMASK);
713 
714 	build_tlb_write_entry(p, l, r, wmode);
715 
716 	build_restore_pagemask(p, r, tmp, label_leave, restore_scratch);
717 }
718 
719 /*
720  * Check if Huge PTE is present, if so then jump to LABEL.
721  */
722 static void
723 build_is_huge_pte(u32 **p, struct uasm_reloc **r, unsigned int tmp,
724 		  unsigned int pmd, int lid)
725 {
726 	UASM_i_LW(p, tmp, 0, pmd);
727 	if (use_bbit_insns()) {
728 		uasm_il_bbit1(p, r, tmp, ilog2(_PAGE_HUGE), lid);
729 	} else {
730 		uasm_i_andi(p, tmp, tmp, _PAGE_HUGE);
731 		uasm_il_bnez(p, r, tmp, lid);
732 	}
733 }
734 
735 static void build_huge_update_entries(u32 **p, unsigned int pte,
736 				      unsigned int tmp)
737 {
738 	int small_sequence;
739 
740 	/*
741 	 * A huge PTE describes an area the size of the
742 	 * configured huge page size. This is twice the
743 	 * of the large TLB entry size we intend to use.
744 	 * A TLB entry half the size of the configured
745 	 * huge page size is configured into entrylo0
746 	 * and entrylo1 to cover the contiguous huge PTE
747 	 * address space.
748 	 */
749 	small_sequence = (HPAGE_SIZE >> 7) < 0x10000;
750 
751 	/* We can clobber tmp.	It isn't used after this.*/
752 	if (!small_sequence)
753 		uasm_i_lui(p, tmp, HPAGE_SIZE >> (7 + 16));
754 
755 	build_convert_pte_to_entrylo(p, pte);
756 	UASM_i_MTC0(p, pte, C0_ENTRYLO0); /* load it */
757 	/* convert to entrylo1 */
758 	if (small_sequence)
759 		UASM_i_ADDIU(p, pte, pte, HPAGE_SIZE >> 7);
760 	else
761 		UASM_i_ADDU(p, pte, pte, tmp);
762 
763 	UASM_i_MTC0(p, pte, C0_ENTRYLO1); /* load it */
764 }
765 
766 static void build_huge_handler_tail(u32 **p, struct uasm_reloc **r,
767 				    struct uasm_label **l,
768 				    unsigned int pte,
769 				    unsigned int ptr,
770 				    unsigned int flush)
771 {
772 #ifdef CONFIG_SMP
773 	UASM_i_SC(p, pte, 0, ptr);
774 	uasm_il_beqz(p, r, pte, label_tlb_huge_update);
775 	UASM_i_LW(p, pte, 0, ptr); /* Needed because SC killed our PTE */
776 #else
777 	UASM_i_SW(p, pte, 0, ptr);
778 #endif
779 	if (cpu_has_ftlb && flush) {
780 		BUG_ON(!cpu_has_tlbinv);
781 
782 		UASM_i_MFC0(p, ptr, C0_ENTRYHI);
783 		uasm_i_ori(p, ptr, ptr, MIPS_ENTRYHI_EHINV);
784 		UASM_i_MTC0(p, ptr, C0_ENTRYHI);
785 		build_tlb_write_entry(p, l, r, tlb_indexed);
786 
787 		uasm_i_xori(p, ptr, ptr, MIPS_ENTRYHI_EHINV);
788 		UASM_i_MTC0(p, ptr, C0_ENTRYHI);
789 		build_huge_update_entries(p, pte, ptr);
790 		build_huge_tlb_write_entry(p, l, r, pte, tlb_random, 0);
791 
792 		return;
793 	}
794 
795 	build_huge_update_entries(p, pte, ptr);
796 	build_huge_tlb_write_entry(p, l, r, pte, tlb_indexed, 0);
797 }
798 #endif /* CONFIG_MIPS_HUGE_TLB_SUPPORT */
799 
800 #ifdef CONFIG_64BIT
801 /*
802  * TMP and PTR are scratch.
803  * TMP will be clobbered, PTR will hold the pmd entry.
804  */
805 void build_get_pmde64(u32 **p, struct uasm_label **l, struct uasm_reloc **r,
806 		      unsigned int tmp, unsigned int ptr)
807 {
808 #ifndef CONFIG_MIPS_PGD_C0_CONTEXT
809 	long pgdc = (long)pgd_current;
810 #endif
811 	/*
812 	 * The vmalloc handling is not in the hotpath.
813 	 */
814 	uasm_i_dmfc0(p, tmp, C0_BADVADDR);
815 
816 	if (check_for_high_segbits) {
817 		/*
818 		 * The kernel currently implicitely assumes that the
819 		 * MIPS SEGBITS parameter for the processor is
820 		 * (PGDIR_SHIFT+PGDIR_BITS) or less, and will never
821 		 * allocate virtual addresses outside the maximum
822 		 * range for SEGBITS = (PGDIR_SHIFT+PGDIR_BITS). But
823 		 * that doesn't prevent user code from accessing the
824 		 * higher xuseg addresses.  Here, we make sure that
825 		 * everything but the lower xuseg addresses goes down
826 		 * the module_alloc/vmalloc path.
827 		 */
828 		uasm_i_dsrl_safe(p, ptr, tmp, PGDIR_SHIFT + PGD_ORDER + PAGE_SHIFT - 3);
829 		uasm_il_bnez(p, r, ptr, label_vmalloc);
830 	} else {
831 		uasm_il_bltz(p, r, tmp, label_vmalloc);
832 	}
833 	/* No uasm_i_nop needed here, since the next insn doesn't touch TMP. */
834 
835 	if (pgd_reg != -1) {
836 		/* pgd is in pgd_reg */
837 		if (cpu_has_ldpte)
838 			UASM_i_MFC0(p, ptr, C0_PWBASE);
839 		else
840 			UASM_i_MFC0(p, ptr, c0_kscratch(), pgd_reg);
841 	} else {
842 #if defined(CONFIG_MIPS_PGD_C0_CONTEXT)
843 		/*
844 		 * &pgd << 11 stored in CONTEXT [23..63].
845 		 */
846 		UASM_i_MFC0(p, ptr, C0_CONTEXT);
847 
848 		/* Clear lower 23 bits of context. */
849 		uasm_i_dins(p, ptr, 0, 0, 23);
850 
851 		/* 1 0	1 0 1  << 6  xkphys cached */
852 		uasm_i_ori(p, ptr, ptr, 0x540);
853 		uasm_i_drotr(p, ptr, ptr, 11);
854 #elif defined(CONFIG_SMP)
855 		UASM_i_CPUID_MFC0(p, ptr, SMP_CPUID_REG);
856 		uasm_i_dsrl_safe(p, ptr, ptr, SMP_CPUID_PTRSHIFT);
857 		UASM_i_LA_mostly(p, tmp, pgdc);
858 		uasm_i_daddu(p, ptr, ptr, tmp);
859 		uasm_i_dmfc0(p, tmp, C0_BADVADDR);
860 		uasm_i_ld(p, ptr, uasm_rel_lo(pgdc), ptr);
861 #else
862 		UASM_i_LA_mostly(p, ptr, pgdc);
863 		uasm_i_ld(p, ptr, uasm_rel_lo(pgdc), ptr);
864 #endif
865 	}
866 
867 	uasm_l_vmalloc_done(l, *p);
868 
869 	/* get pgd offset in bytes */
870 	uasm_i_dsrl_safe(p, tmp, tmp, PGDIR_SHIFT - 3);
871 
872 	uasm_i_andi(p, tmp, tmp, (PTRS_PER_PGD - 1)<<3);
873 	uasm_i_daddu(p, ptr, ptr, tmp); /* add in pgd offset */
874 #ifndef __PAGETABLE_PUD_FOLDED
875 	uasm_i_dmfc0(p, tmp, C0_BADVADDR); /* get faulting address */
876 	uasm_i_ld(p, ptr, 0, ptr); /* get pud pointer */
877 	uasm_i_dsrl_safe(p, tmp, tmp, PUD_SHIFT - 3); /* get pud offset in bytes */
878 	uasm_i_andi(p, tmp, tmp, (PTRS_PER_PUD - 1) << 3);
879 	uasm_i_daddu(p, ptr, ptr, tmp); /* add in pud offset */
880 #endif
881 #ifndef __PAGETABLE_PMD_FOLDED
882 	uasm_i_dmfc0(p, tmp, C0_BADVADDR); /* get faulting address */
883 	uasm_i_ld(p, ptr, 0, ptr); /* get pmd pointer */
884 	uasm_i_dsrl_safe(p, tmp, tmp, PMD_SHIFT-3); /* get pmd offset in bytes */
885 	uasm_i_andi(p, tmp, tmp, (PTRS_PER_PMD - 1)<<3);
886 	uasm_i_daddu(p, ptr, ptr, tmp); /* add in pmd offset */
887 #endif
888 }
889 EXPORT_SYMBOL_GPL(build_get_pmde64);
890 
891 /*
892  * BVADDR is the faulting address, PTR is scratch.
893  * PTR will hold the pgd for vmalloc.
894  */
895 static void
896 build_get_pgd_vmalloc64(u32 **p, struct uasm_label **l, struct uasm_reloc **r,
897 			unsigned int bvaddr, unsigned int ptr,
898 			enum vmalloc64_mode mode)
899 {
900 	long swpd = (long)swapper_pg_dir;
901 	int single_insn_swpd;
902 	int did_vmalloc_branch = 0;
903 
904 	single_insn_swpd = uasm_in_compat_space_p(swpd) && !uasm_rel_lo(swpd);
905 
906 	uasm_l_vmalloc(l, *p);
907 
908 	if (mode != not_refill && check_for_high_segbits) {
909 		if (single_insn_swpd) {
910 			uasm_il_bltz(p, r, bvaddr, label_vmalloc_done);
911 			uasm_i_lui(p, ptr, uasm_rel_hi(swpd));
912 			did_vmalloc_branch = 1;
913 			/* fall through */
914 		} else {
915 			uasm_il_bgez(p, r, bvaddr, label_large_segbits_fault);
916 		}
917 	}
918 	if (!did_vmalloc_branch) {
919 		if (single_insn_swpd) {
920 			uasm_il_b(p, r, label_vmalloc_done);
921 			uasm_i_lui(p, ptr, uasm_rel_hi(swpd));
922 		} else {
923 			UASM_i_LA_mostly(p, ptr, swpd);
924 			uasm_il_b(p, r, label_vmalloc_done);
925 			if (uasm_in_compat_space_p(swpd))
926 				uasm_i_addiu(p, ptr, ptr, uasm_rel_lo(swpd));
927 			else
928 				uasm_i_daddiu(p, ptr, ptr, uasm_rel_lo(swpd));
929 		}
930 	}
931 	if (mode != not_refill && check_for_high_segbits) {
932 		uasm_l_large_segbits_fault(l, *p);
933 
934 		if (mode == refill_scratch && scratch_reg >= 0)
935 			uasm_i_ehb(p);
936 
937 		/*
938 		 * We get here if we are an xsseg address, or if we are
939 		 * an xuseg address above (PGDIR_SHIFT+PGDIR_BITS) boundary.
940 		 *
941 		 * Ignoring xsseg (assume disabled so would generate
942 		 * (address errors?), the only remaining possibility
943 		 * is the upper xuseg addresses.  On processors with
944 		 * TLB_SEGBITS <= PGDIR_SHIFT+PGDIR_BITS, these
945 		 * addresses would have taken an address error. We try
946 		 * to mimic that here by taking a load/istream page
947 		 * fault.
948 		 */
949 		if (IS_ENABLED(CONFIG_CPU_LOONGSON3_WORKAROUNDS))
950 			uasm_i_sync(p, 0);
951 		UASM_i_LA(p, ptr, (unsigned long)tlb_do_page_fault_0);
952 		uasm_i_jr(p, ptr);
953 
954 		if (mode == refill_scratch) {
955 			if (scratch_reg >= 0)
956 				UASM_i_MFC0(p, 1, c0_kscratch(), scratch_reg);
957 			else
958 				UASM_i_LW(p, 1, scratchpad_offset(0), 0);
959 		} else {
960 			uasm_i_nop(p);
961 		}
962 	}
963 }
964 
965 #else /* !CONFIG_64BIT */
966 
967 /*
968  * TMP and PTR are scratch.
969  * TMP will be clobbered, PTR will hold the pgd entry.
970  */
971 void build_get_pgde32(u32 **p, unsigned int tmp, unsigned int ptr)
972 {
973 	if (pgd_reg != -1) {
974 		/* pgd is in pgd_reg */
975 		uasm_i_mfc0(p, ptr, c0_kscratch(), pgd_reg);
976 		uasm_i_mfc0(p, tmp, C0_BADVADDR); /* get faulting address */
977 	} else {
978 		long pgdc = (long)pgd_current;
979 
980 		/* 32 bit SMP has smp_processor_id() stored in CONTEXT. */
981 #ifdef CONFIG_SMP
982 		uasm_i_mfc0(p, ptr, SMP_CPUID_REG);
983 		UASM_i_LA_mostly(p, tmp, pgdc);
984 		uasm_i_srl(p, ptr, ptr, SMP_CPUID_PTRSHIFT);
985 		uasm_i_addu(p, ptr, tmp, ptr);
986 #else
987 		UASM_i_LA_mostly(p, ptr, pgdc);
988 #endif
989 		uasm_i_mfc0(p, tmp, C0_BADVADDR); /* get faulting address */
990 		uasm_i_lw(p, ptr, uasm_rel_lo(pgdc), ptr);
991 	}
992 	uasm_i_srl(p, tmp, tmp, PGDIR_SHIFT); /* get pgd only bits */
993 	uasm_i_sll(p, tmp, tmp, PGD_T_LOG2);
994 	uasm_i_addu(p, ptr, ptr, tmp); /* add in pgd offset */
995 }
996 EXPORT_SYMBOL_GPL(build_get_pgde32);
997 
998 #endif /* !CONFIG_64BIT */
999 
1000 static void build_adjust_context(u32 **p, unsigned int ctx)
1001 {
1002 	unsigned int shift = 4 - (PTE_T_LOG2 + 1) + PAGE_SHIFT - 12;
1003 	unsigned int mask = (PTRS_PER_PTE / 2 - 1) << (PTE_T_LOG2 + 1);
1004 
1005 	switch (current_cpu_type()) {
1006 	case CPU_VR41XX:
1007 	case CPU_VR4111:
1008 	case CPU_VR4121:
1009 	case CPU_VR4122:
1010 	case CPU_VR4131:
1011 	case CPU_VR4181:
1012 	case CPU_VR4181A:
1013 	case CPU_VR4133:
1014 		shift += 2;
1015 		break;
1016 
1017 	default:
1018 		break;
1019 	}
1020 
1021 	if (shift)
1022 		UASM_i_SRL(p, ctx, ctx, shift);
1023 	uasm_i_andi(p, ctx, ctx, mask);
1024 }
1025 
1026 void build_get_ptep(u32 **p, unsigned int tmp, unsigned int ptr)
1027 {
1028 	/*
1029 	 * Bug workaround for the Nevada. It seems as if under certain
1030 	 * circumstances the move from cp0_context might produce a
1031 	 * bogus result when the mfc0 instruction and its consumer are
1032 	 * in a different cacheline or a load instruction, probably any
1033 	 * memory reference, is between them.
1034 	 */
1035 	switch (current_cpu_type()) {
1036 	case CPU_NEVADA:
1037 		UASM_i_LW(p, ptr, 0, ptr);
1038 		GET_CONTEXT(p, tmp); /* get context reg */
1039 		break;
1040 
1041 	default:
1042 		GET_CONTEXT(p, tmp); /* get context reg */
1043 		UASM_i_LW(p, ptr, 0, ptr);
1044 		break;
1045 	}
1046 
1047 	build_adjust_context(p, tmp);
1048 	UASM_i_ADDU(p, ptr, ptr, tmp); /* add in offset */
1049 }
1050 EXPORT_SYMBOL_GPL(build_get_ptep);
1051 
1052 void build_update_entries(u32 **p, unsigned int tmp, unsigned int ptep)
1053 {
1054 	int pte_off_even = 0;
1055 	int pte_off_odd = sizeof(pte_t);
1056 
1057 #if defined(CONFIG_CPU_MIPS32) && defined(CONFIG_PHYS_ADDR_T_64BIT)
1058 	/* The low 32 bits of EntryLo is stored in pte_high */
1059 	pte_off_even += offsetof(pte_t, pte_high);
1060 	pte_off_odd += offsetof(pte_t, pte_high);
1061 #endif
1062 
1063 	if (IS_ENABLED(CONFIG_XPA)) {
1064 		uasm_i_lw(p, tmp, pte_off_even, ptep); /* even pte */
1065 		UASM_i_ROTR(p, tmp, tmp, ilog2(_PAGE_GLOBAL));
1066 		UASM_i_MTC0(p, tmp, C0_ENTRYLO0);
1067 
1068 		if (cpu_has_xpa && !mips_xpa_disabled) {
1069 			uasm_i_lw(p, tmp, 0, ptep);
1070 			uasm_i_ext(p, tmp, tmp, 0, 24);
1071 			uasm_i_mthc0(p, tmp, C0_ENTRYLO0);
1072 		}
1073 
1074 		uasm_i_lw(p, tmp, pte_off_odd, ptep); /* odd pte */
1075 		UASM_i_ROTR(p, tmp, tmp, ilog2(_PAGE_GLOBAL));
1076 		UASM_i_MTC0(p, tmp, C0_ENTRYLO1);
1077 
1078 		if (cpu_has_xpa && !mips_xpa_disabled) {
1079 			uasm_i_lw(p, tmp, sizeof(pte_t), ptep);
1080 			uasm_i_ext(p, tmp, tmp, 0, 24);
1081 			uasm_i_mthc0(p, tmp, C0_ENTRYLO1);
1082 		}
1083 		return;
1084 	}
1085 
1086 	UASM_i_LW(p, tmp, pte_off_even, ptep); /* get even pte */
1087 	UASM_i_LW(p, ptep, pte_off_odd, ptep); /* get odd pte */
1088 	if (r45k_bvahwbug())
1089 		build_tlb_probe_entry(p);
1090 	build_convert_pte_to_entrylo(p, tmp);
1091 	if (r4k_250MHZhwbug())
1092 		UASM_i_MTC0(p, 0, C0_ENTRYLO0);
1093 	UASM_i_MTC0(p, tmp, C0_ENTRYLO0); /* load it */
1094 	build_convert_pte_to_entrylo(p, ptep);
1095 	if (r45k_bvahwbug())
1096 		uasm_i_mfc0(p, tmp, C0_INDEX);
1097 	if (r4k_250MHZhwbug())
1098 		UASM_i_MTC0(p, 0, C0_ENTRYLO1);
1099 	UASM_i_MTC0(p, ptep, C0_ENTRYLO1); /* load it */
1100 }
1101 EXPORT_SYMBOL_GPL(build_update_entries);
1102 
1103 struct mips_huge_tlb_info {
1104 	int huge_pte;
1105 	int restore_scratch;
1106 	bool need_reload_pte;
1107 };
1108 
1109 static struct mips_huge_tlb_info
1110 build_fast_tlb_refill_handler (u32 **p, struct uasm_label **l,
1111 			       struct uasm_reloc **r, unsigned int tmp,
1112 			       unsigned int ptr, int c0_scratch_reg)
1113 {
1114 	struct mips_huge_tlb_info rv;
1115 	unsigned int even, odd;
1116 	int vmalloc_branch_delay_filled = 0;
1117 	const int scratch = 1; /* Our extra working register */
1118 
1119 	rv.huge_pte = scratch;
1120 	rv.restore_scratch = 0;
1121 	rv.need_reload_pte = false;
1122 
1123 	if (check_for_high_segbits) {
1124 		UASM_i_MFC0(p, tmp, C0_BADVADDR);
1125 
1126 		if (pgd_reg != -1)
1127 			UASM_i_MFC0(p, ptr, c0_kscratch(), pgd_reg);
1128 		else
1129 			UASM_i_MFC0(p, ptr, C0_CONTEXT);
1130 
1131 		if (c0_scratch_reg >= 0)
1132 			UASM_i_MTC0(p, scratch, c0_kscratch(), c0_scratch_reg);
1133 		else
1134 			UASM_i_SW(p, scratch, scratchpad_offset(0), 0);
1135 
1136 		uasm_i_dsrl_safe(p, scratch, tmp,
1137 				 PGDIR_SHIFT + PGD_ORDER + PAGE_SHIFT - 3);
1138 		uasm_il_bnez(p, r, scratch, label_vmalloc);
1139 
1140 		if (pgd_reg == -1) {
1141 			vmalloc_branch_delay_filled = 1;
1142 			/* Clear lower 23 bits of context. */
1143 			uasm_i_dins(p, ptr, 0, 0, 23);
1144 		}
1145 	} else {
1146 		if (pgd_reg != -1)
1147 			UASM_i_MFC0(p, ptr, c0_kscratch(), pgd_reg);
1148 		else
1149 			UASM_i_MFC0(p, ptr, C0_CONTEXT);
1150 
1151 		UASM_i_MFC0(p, tmp, C0_BADVADDR);
1152 
1153 		if (c0_scratch_reg >= 0)
1154 			UASM_i_MTC0(p, scratch, c0_kscratch(), c0_scratch_reg);
1155 		else
1156 			UASM_i_SW(p, scratch, scratchpad_offset(0), 0);
1157 
1158 		if (pgd_reg == -1)
1159 			/* Clear lower 23 bits of context. */
1160 			uasm_i_dins(p, ptr, 0, 0, 23);
1161 
1162 		uasm_il_bltz(p, r, tmp, label_vmalloc);
1163 	}
1164 
1165 	if (pgd_reg == -1) {
1166 		vmalloc_branch_delay_filled = 1;
1167 		/* 1 0	1 0 1  << 6  xkphys cached */
1168 		uasm_i_ori(p, ptr, ptr, 0x540);
1169 		uasm_i_drotr(p, ptr, ptr, 11);
1170 	}
1171 
1172 #ifdef __PAGETABLE_PMD_FOLDED
1173 #define LOC_PTEP scratch
1174 #else
1175 #define LOC_PTEP ptr
1176 #endif
1177 
1178 	if (!vmalloc_branch_delay_filled)
1179 		/* get pgd offset in bytes */
1180 		uasm_i_dsrl_safe(p, scratch, tmp, PGDIR_SHIFT - 3);
1181 
1182 	uasm_l_vmalloc_done(l, *p);
1183 
1184 	/*
1185 	 *			   tmp		ptr
1186 	 * fall-through case =	 badvaddr  *pgd_current
1187 	 * vmalloc case	     =	 badvaddr  swapper_pg_dir
1188 	 */
1189 
1190 	if (vmalloc_branch_delay_filled)
1191 		/* get pgd offset in bytes */
1192 		uasm_i_dsrl_safe(p, scratch, tmp, PGDIR_SHIFT - 3);
1193 
1194 #ifdef __PAGETABLE_PMD_FOLDED
1195 	GET_CONTEXT(p, tmp); /* get context reg */
1196 #endif
1197 	uasm_i_andi(p, scratch, scratch, (PTRS_PER_PGD - 1) << 3);
1198 
1199 	if (use_lwx_insns()) {
1200 		UASM_i_LWX(p, LOC_PTEP, scratch, ptr);
1201 	} else {
1202 		uasm_i_daddu(p, ptr, ptr, scratch); /* add in pgd offset */
1203 		uasm_i_ld(p, LOC_PTEP, 0, ptr); /* get pmd pointer */
1204 	}
1205 
1206 #ifndef __PAGETABLE_PUD_FOLDED
1207 	/* get pud offset in bytes */
1208 	uasm_i_dsrl_safe(p, scratch, tmp, PUD_SHIFT - 3);
1209 	uasm_i_andi(p, scratch, scratch, (PTRS_PER_PUD - 1) << 3);
1210 
1211 	if (use_lwx_insns()) {
1212 		UASM_i_LWX(p, ptr, scratch, ptr);
1213 	} else {
1214 		uasm_i_daddu(p, ptr, ptr, scratch); /* add in pmd offset */
1215 		UASM_i_LW(p, ptr, 0, ptr);
1216 	}
1217 	/* ptr contains a pointer to PMD entry */
1218 	/* tmp contains the address */
1219 #endif
1220 
1221 #ifndef __PAGETABLE_PMD_FOLDED
1222 	/* get pmd offset in bytes */
1223 	uasm_i_dsrl_safe(p, scratch, tmp, PMD_SHIFT - 3);
1224 	uasm_i_andi(p, scratch, scratch, (PTRS_PER_PMD - 1) << 3);
1225 	GET_CONTEXT(p, tmp); /* get context reg */
1226 
1227 	if (use_lwx_insns()) {
1228 		UASM_i_LWX(p, scratch, scratch, ptr);
1229 	} else {
1230 		uasm_i_daddu(p, ptr, ptr, scratch); /* add in pmd offset */
1231 		UASM_i_LW(p, scratch, 0, ptr);
1232 	}
1233 #endif
1234 	/* Adjust the context during the load latency. */
1235 	build_adjust_context(p, tmp);
1236 
1237 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
1238 	uasm_il_bbit1(p, r, scratch, ilog2(_PAGE_HUGE), label_tlb_huge_update);
1239 	/*
1240 	 * The in the LWX case we don't want to do the load in the
1241 	 * delay slot.	It cannot issue in the same cycle and may be
1242 	 * speculative and unneeded.
1243 	 */
1244 	if (use_lwx_insns())
1245 		uasm_i_nop(p);
1246 #endif /* CONFIG_MIPS_HUGE_TLB_SUPPORT */
1247 
1248 
1249 	/* build_update_entries */
1250 	if (use_lwx_insns()) {
1251 		even = ptr;
1252 		odd = tmp;
1253 		UASM_i_LWX(p, even, scratch, tmp);
1254 		UASM_i_ADDIU(p, tmp, tmp, sizeof(pte_t));
1255 		UASM_i_LWX(p, odd, scratch, tmp);
1256 	} else {
1257 		UASM_i_ADDU(p, ptr, scratch, tmp); /* add in offset */
1258 		even = tmp;
1259 		odd = ptr;
1260 		UASM_i_LW(p, even, 0, ptr); /* get even pte */
1261 		UASM_i_LW(p, odd, sizeof(pte_t), ptr); /* get odd pte */
1262 	}
1263 	if (cpu_has_rixi) {
1264 		uasm_i_drotr(p, even, even, ilog2(_PAGE_GLOBAL));
1265 		UASM_i_MTC0(p, even, C0_ENTRYLO0); /* load it */
1266 		uasm_i_drotr(p, odd, odd, ilog2(_PAGE_GLOBAL));
1267 	} else {
1268 		uasm_i_dsrl_safe(p, even, even, ilog2(_PAGE_GLOBAL));
1269 		UASM_i_MTC0(p, even, C0_ENTRYLO0); /* load it */
1270 		uasm_i_dsrl_safe(p, odd, odd, ilog2(_PAGE_GLOBAL));
1271 	}
1272 	UASM_i_MTC0(p, odd, C0_ENTRYLO1); /* load it */
1273 
1274 	if (c0_scratch_reg >= 0) {
1275 		uasm_i_ehb(p);
1276 		UASM_i_MFC0(p, scratch, c0_kscratch(), c0_scratch_reg);
1277 		build_tlb_write_entry(p, l, r, tlb_random);
1278 		uasm_l_leave(l, *p);
1279 		rv.restore_scratch = 1;
1280 	} else if (PAGE_SHIFT == 14 || PAGE_SHIFT == 13)  {
1281 		build_tlb_write_entry(p, l, r, tlb_random);
1282 		uasm_l_leave(l, *p);
1283 		UASM_i_LW(p, scratch, scratchpad_offset(0), 0);
1284 	} else {
1285 		UASM_i_LW(p, scratch, scratchpad_offset(0), 0);
1286 		build_tlb_write_entry(p, l, r, tlb_random);
1287 		uasm_l_leave(l, *p);
1288 		rv.restore_scratch = 1;
1289 	}
1290 
1291 	uasm_i_eret(p); /* return from trap */
1292 
1293 	return rv;
1294 }
1295 
1296 /*
1297  * For a 64-bit kernel, we are using the 64-bit XTLB refill exception
1298  * because EXL == 0.  If we wrap, we can also use the 32 instruction
1299  * slots before the XTLB refill exception handler which belong to the
1300  * unused TLB refill exception.
1301  */
1302 #define MIPS64_REFILL_INSNS 32
1303 
1304 static void build_r4000_tlb_refill_handler(void)
1305 {
1306 	u32 *p = tlb_handler;
1307 	struct uasm_label *l = labels;
1308 	struct uasm_reloc *r = relocs;
1309 	u32 *f;
1310 	unsigned int final_len;
1311 	struct mips_huge_tlb_info htlb_info __maybe_unused;
1312 	enum vmalloc64_mode vmalloc_mode __maybe_unused;
1313 
1314 	memset(tlb_handler, 0, sizeof(tlb_handler));
1315 	memset(labels, 0, sizeof(labels));
1316 	memset(relocs, 0, sizeof(relocs));
1317 	memset(final_handler, 0, sizeof(final_handler));
1318 
1319 	if (IS_ENABLED(CONFIG_64BIT) && (scratch_reg >= 0 || scratchpad_available()) && use_bbit_insns()) {
1320 		htlb_info = build_fast_tlb_refill_handler(&p, &l, &r, K0, K1,
1321 							  scratch_reg);
1322 		vmalloc_mode = refill_scratch;
1323 	} else {
1324 		htlb_info.huge_pte = K0;
1325 		htlb_info.restore_scratch = 0;
1326 		htlb_info.need_reload_pte = true;
1327 		vmalloc_mode = refill_noscratch;
1328 		/*
1329 		 * create the plain linear handler
1330 		 */
1331 		if (bcm1250_m3_war()) {
1332 			unsigned int segbits = 44;
1333 
1334 			uasm_i_dmfc0(&p, K0, C0_BADVADDR);
1335 			uasm_i_dmfc0(&p, K1, C0_ENTRYHI);
1336 			uasm_i_xor(&p, K0, K0, K1);
1337 			uasm_i_dsrl_safe(&p, K1, K0, 62);
1338 			uasm_i_dsrl_safe(&p, K0, K0, 12 + 1);
1339 			uasm_i_dsll_safe(&p, K0, K0, 64 + 12 + 1 - segbits);
1340 			uasm_i_or(&p, K0, K0, K1);
1341 			uasm_il_bnez(&p, &r, K0, label_leave);
1342 			/* No need for uasm_i_nop */
1343 		}
1344 
1345 #ifdef CONFIG_64BIT
1346 		build_get_pmde64(&p, &l, &r, K0, K1); /* get pmd in K1 */
1347 #else
1348 		build_get_pgde32(&p, K0, K1); /* get pgd in K1 */
1349 #endif
1350 
1351 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
1352 		build_is_huge_pte(&p, &r, K0, K1, label_tlb_huge_update);
1353 #endif
1354 
1355 		build_get_ptep(&p, K0, K1);
1356 		build_update_entries(&p, K0, K1);
1357 		build_tlb_write_entry(&p, &l, &r, tlb_random);
1358 		uasm_l_leave(&l, p);
1359 		uasm_i_eret(&p); /* return from trap */
1360 	}
1361 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
1362 	uasm_l_tlb_huge_update(&l, p);
1363 	if (htlb_info.need_reload_pte)
1364 		UASM_i_LW(&p, htlb_info.huge_pte, 0, K1);
1365 	build_huge_update_entries(&p, htlb_info.huge_pte, K1);
1366 	build_huge_tlb_write_entry(&p, &l, &r, K0, tlb_random,
1367 				   htlb_info.restore_scratch);
1368 #endif
1369 
1370 #ifdef CONFIG_64BIT
1371 	build_get_pgd_vmalloc64(&p, &l, &r, K0, K1, vmalloc_mode);
1372 #endif
1373 
1374 	/*
1375 	 * Overflow check: For the 64bit handler, we need at least one
1376 	 * free instruction slot for the wrap-around branch. In worst
1377 	 * case, if the intended insertion point is a delay slot, we
1378 	 * need three, with the second nop'ed and the third being
1379 	 * unused.
1380 	 */
1381 	switch (boot_cpu_type()) {
1382 	default:
1383 		if (sizeof(long) == 4) {
1384 	case CPU_LOONGSON2EF:
1385 		/* Loongson2 ebase is different than r4k, we have more space */
1386 			if ((p - tlb_handler) > 64)
1387 				panic("TLB refill handler space exceeded");
1388 			/*
1389 			 * Now fold the handler in the TLB refill handler space.
1390 			 */
1391 			f = final_handler;
1392 			/* Simplest case, just copy the handler. */
1393 			uasm_copy_handler(relocs, labels, tlb_handler, p, f);
1394 			final_len = p - tlb_handler;
1395 			break;
1396 		} else {
1397 			if (((p - tlb_handler) > (MIPS64_REFILL_INSNS * 2) - 1)
1398 			    || (((p - tlb_handler) > (MIPS64_REFILL_INSNS * 2) - 3)
1399 				&& uasm_insn_has_bdelay(relocs,
1400 							tlb_handler + MIPS64_REFILL_INSNS - 3)))
1401 				panic("TLB refill handler space exceeded");
1402 			/*
1403 			 * Now fold the handler in the TLB refill handler space.
1404 			 */
1405 			f = final_handler + MIPS64_REFILL_INSNS;
1406 			if ((p - tlb_handler) <= MIPS64_REFILL_INSNS) {
1407 				/* Just copy the handler. */
1408 				uasm_copy_handler(relocs, labels, tlb_handler, p, f);
1409 				final_len = p - tlb_handler;
1410 			} else {
1411 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
1412 				const enum label_id ls = label_tlb_huge_update;
1413 #else
1414 				const enum label_id ls = label_vmalloc;
1415 #endif
1416 				u32 *split;
1417 				int ov = 0;
1418 				int i;
1419 
1420 				for (i = 0; i < ARRAY_SIZE(labels) && labels[i].lab != ls; i++)
1421 					;
1422 				BUG_ON(i == ARRAY_SIZE(labels));
1423 				split = labels[i].addr;
1424 
1425 				/*
1426 				 * See if we have overflown one way or the other.
1427 				 */
1428 				if (split > tlb_handler + MIPS64_REFILL_INSNS ||
1429 				    split < p - MIPS64_REFILL_INSNS)
1430 					ov = 1;
1431 
1432 				if (ov) {
1433 					/*
1434 					 * Split two instructions before the end.  One
1435 					 * for the branch and one for the instruction
1436 					 * in the delay slot.
1437 					 */
1438 					split = tlb_handler + MIPS64_REFILL_INSNS - 2;
1439 
1440 					/*
1441 					 * If the branch would fall in a delay slot,
1442 					 * we must back up an additional instruction
1443 					 * so that it is no longer in a delay slot.
1444 					 */
1445 					if (uasm_insn_has_bdelay(relocs, split - 1))
1446 						split--;
1447 				}
1448 				/* Copy first part of the handler. */
1449 				uasm_copy_handler(relocs, labels, tlb_handler, split, f);
1450 				f += split - tlb_handler;
1451 
1452 				if (ov) {
1453 					/* Insert branch. */
1454 					uasm_l_split(&l, final_handler);
1455 					uasm_il_b(&f, &r, label_split);
1456 					if (uasm_insn_has_bdelay(relocs, split))
1457 						uasm_i_nop(&f);
1458 					else {
1459 						uasm_copy_handler(relocs, labels,
1460 								  split, split + 1, f);
1461 						uasm_move_labels(labels, f, f + 1, -1);
1462 						f++;
1463 						split++;
1464 					}
1465 				}
1466 
1467 				/* Copy the rest of the handler. */
1468 				uasm_copy_handler(relocs, labels, split, p, final_handler);
1469 				final_len = (f - (final_handler + MIPS64_REFILL_INSNS)) +
1470 					    (p - split);
1471 			}
1472 		}
1473 		break;
1474 	}
1475 
1476 	uasm_resolve_relocs(relocs, labels);
1477 	pr_debug("Wrote TLB refill handler (%u instructions).\n",
1478 		 final_len);
1479 
1480 	memcpy((void *)ebase, final_handler, 0x100);
1481 	local_flush_icache_range(ebase, ebase + 0x100);
1482 	dump_handler("r4000_tlb_refill", (u32 *)ebase, (u32 *)(ebase + 0x100));
1483 }
1484 
1485 static void setup_pw(void)
1486 {
1487 	unsigned int pwctl;
1488 	unsigned long pgd_i, pgd_w;
1489 #ifndef __PAGETABLE_PMD_FOLDED
1490 	unsigned long pmd_i, pmd_w;
1491 #endif
1492 	unsigned long pt_i, pt_w;
1493 	unsigned long pte_i, pte_w;
1494 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
1495 	unsigned long psn;
1496 
1497 	psn = ilog2(_PAGE_HUGE);     /* bit used to indicate huge page */
1498 #endif
1499 	pgd_i = PGDIR_SHIFT;  /* 1st level PGD */
1500 #ifndef __PAGETABLE_PMD_FOLDED
1501 	pgd_w = PGDIR_SHIFT - PMD_SHIFT + PGD_ORDER;
1502 
1503 	pmd_i = PMD_SHIFT;    /* 2nd level PMD */
1504 	pmd_w = PMD_SHIFT - PAGE_SHIFT;
1505 #else
1506 	pgd_w = PGDIR_SHIFT - PAGE_SHIFT + PGD_ORDER;
1507 #endif
1508 
1509 	pt_i  = PAGE_SHIFT;    /* 3rd level PTE */
1510 	pt_w  = PAGE_SHIFT - 3;
1511 
1512 	pte_i = ilog2(_PAGE_GLOBAL);
1513 	pte_w = 0;
1514 	pwctl = 1 << 30; /* Set PWDirExt */
1515 
1516 #ifndef __PAGETABLE_PMD_FOLDED
1517 	write_c0_pwfield(pgd_i << 24 | pmd_i << 12 | pt_i << 6 | pte_i);
1518 	write_c0_pwsize(1 << 30 | pgd_w << 24 | pmd_w << 12 | pt_w << 6 | pte_w);
1519 #else
1520 	write_c0_pwfield(pgd_i << 24 | pt_i << 6 | pte_i);
1521 	write_c0_pwsize(1 << 30 | pgd_w << 24 | pt_w << 6 | pte_w);
1522 #endif
1523 
1524 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
1525 	pwctl |= (1 << 6 | psn);
1526 #endif
1527 	write_c0_pwctl(pwctl);
1528 	write_c0_kpgd((long)swapper_pg_dir);
1529 	kscratch_used_mask |= (1 << 7); /* KScratch6 is used for KPGD */
1530 }
1531 
1532 static void build_loongson3_tlb_refill_handler(void)
1533 {
1534 	u32 *p = tlb_handler;
1535 	struct uasm_label *l = labels;
1536 	struct uasm_reloc *r = relocs;
1537 
1538 	memset(labels, 0, sizeof(labels));
1539 	memset(relocs, 0, sizeof(relocs));
1540 	memset(tlb_handler, 0, sizeof(tlb_handler));
1541 
1542 	if (check_for_high_segbits) {
1543 		uasm_i_dmfc0(&p, K0, C0_BADVADDR);
1544 		uasm_i_dsrl_safe(&p, K1, K0, PGDIR_SHIFT + PGD_ORDER + PAGE_SHIFT - 3);
1545 		uasm_il_beqz(&p, &r, K1, label_vmalloc);
1546 		uasm_i_nop(&p);
1547 
1548 		uasm_il_bgez(&p, &r, K0, label_large_segbits_fault);
1549 		uasm_i_nop(&p);
1550 		uasm_l_vmalloc(&l, p);
1551 	}
1552 
1553 	uasm_i_dmfc0(&p, K1, C0_PGD);
1554 
1555 	uasm_i_lddir(&p, K0, K1, 3);  /* global page dir */
1556 #ifndef __PAGETABLE_PMD_FOLDED
1557 	uasm_i_lddir(&p, K1, K0, 1);  /* middle page dir */
1558 #endif
1559 	uasm_i_ldpte(&p, K1, 0);      /* even */
1560 	uasm_i_ldpte(&p, K1, 1);      /* odd */
1561 	uasm_i_tlbwr(&p);
1562 
1563 	/* restore page mask */
1564 	if (PM_DEFAULT_MASK >> 16) {
1565 		uasm_i_lui(&p, K0, PM_DEFAULT_MASK >> 16);
1566 		uasm_i_ori(&p, K0, K0, PM_DEFAULT_MASK & 0xffff);
1567 		uasm_i_mtc0(&p, K0, C0_PAGEMASK);
1568 	} else if (PM_DEFAULT_MASK) {
1569 		uasm_i_ori(&p, K0, 0, PM_DEFAULT_MASK);
1570 		uasm_i_mtc0(&p, K0, C0_PAGEMASK);
1571 	} else {
1572 		uasm_i_mtc0(&p, 0, C0_PAGEMASK);
1573 	}
1574 
1575 	uasm_i_eret(&p);
1576 
1577 	if (check_for_high_segbits) {
1578 		uasm_l_large_segbits_fault(&l, p);
1579 		UASM_i_LA(&p, K1, (unsigned long)tlb_do_page_fault_0);
1580 		uasm_i_jr(&p, K1);
1581 		uasm_i_nop(&p);
1582 	}
1583 
1584 	uasm_resolve_relocs(relocs, labels);
1585 	memcpy((void *)(ebase + 0x80), tlb_handler, 0x80);
1586 	local_flush_icache_range(ebase + 0x80, ebase + 0x100);
1587 	dump_handler("loongson3_tlb_refill",
1588 		     (u32 *)(ebase + 0x80), (u32 *)(ebase + 0x100));
1589 }
1590 
1591 static void build_setup_pgd(void)
1592 {
1593 	const int a0 = 4;
1594 	const int __maybe_unused a1 = 5;
1595 	const int __maybe_unused a2 = 6;
1596 	u32 *p = (u32 *)msk_isa16_mode((ulong)tlbmiss_handler_setup_pgd);
1597 #ifndef CONFIG_MIPS_PGD_C0_CONTEXT
1598 	long pgdc = (long)pgd_current;
1599 #endif
1600 
1601 	memset(p, 0, tlbmiss_handler_setup_pgd_end - (char *)p);
1602 	memset(labels, 0, sizeof(labels));
1603 	memset(relocs, 0, sizeof(relocs));
1604 	pgd_reg = allocate_kscratch();
1605 #ifdef CONFIG_MIPS_PGD_C0_CONTEXT
1606 	if (pgd_reg == -1) {
1607 		struct uasm_label *l = labels;
1608 		struct uasm_reloc *r = relocs;
1609 
1610 		/* PGD << 11 in c0_Context */
1611 		/*
1612 		 * If it is a ckseg0 address, convert to a physical
1613 		 * address.  Shifting right by 29 and adding 4 will
1614 		 * result in zero for these addresses.
1615 		 *
1616 		 */
1617 		UASM_i_SRA(&p, a1, a0, 29);
1618 		UASM_i_ADDIU(&p, a1, a1, 4);
1619 		uasm_il_bnez(&p, &r, a1, label_tlbl_goaround1);
1620 		uasm_i_nop(&p);
1621 		uasm_i_dinsm(&p, a0, 0, 29, 64 - 29);
1622 		uasm_l_tlbl_goaround1(&l, p);
1623 		UASM_i_SLL(&p, a0, a0, 11);
1624 		UASM_i_MTC0(&p, a0, C0_CONTEXT);
1625 		uasm_i_jr(&p, 31);
1626 		uasm_i_ehb(&p);
1627 	} else {
1628 		/* PGD in c0_KScratch */
1629 		if (cpu_has_ldpte)
1630 			UASM_i_MTC0(&p, a0, C0_PWBASE);
1631 		else
1632 			UASM_i_MTC0(&p, a0, c0_kscratch(), pgd_reg);
1633 		uasm_i_jr(&p, 31);
1634 		uasm_i_ehb(&p);
1635 	}
1636 #else
1637 #ifdef CONFIG_SMP
1638 	/* Save PGD to pgd_current[smp_processor_id()] */
1639 	UASM_i_CPUID_MFC0(&p, a1, SMP_CPUID_REG);
1640 	UASM_i_SRL_SAFE(&p, a1, a1, SMP_CPUID_PTRSHIFT);
1641 	UASM_i_LA_mostly(&p, a2, pgdc);
1642 	UASM_i_ADDU(&p, a2, a2, a1);
1643 	UASM_i_SW(&p, a0, uasm_rel_lo(pgdc), a2);
1644 #else
1645 	UASM_i_LA_mostly(&p, a2, pgdc);
1646 	UASM_i_SW(&p, a0, uasm_rel_lo(pgdc), a2);
1647 #endif /* SMP */
1648 
1649 	/* if pgd_reg is allocated, save PGD also to scratch register */
1650 	if (pgd_reg != -1) {
1651 		UASM_i_MTC0(&p, a0, c0_kscratch(), pgd_reg);
1652 		uasm_i_jr(&p, 31);
1653 		uasm_i_ehb(&p);
1654 	} else {
1655 		uasm_i_jr(&p, 31);
1656 		uasm_i_nop(&p);
1657 	}
1658 #endif
1659 	if (p >= (u32 *)tlbmiss_handler_setup_pgd_end)
1660 		panic("tlbmiss_handler_setup_pgd space exceeded");
1661 
1662 	uasm_resolve_relocs(relocs, labels);
1663 	pr_debug("Wrote tlbmiss_handler_setup_pgd (%u instructions).\n",
1664 		 (unsigned int)(p - (u32 *)tlbmiss_handler_setup_pgd));
1665 
1666 	dump_handler("tlbmiss_handler", tlbmiss_handler_setup_pgd,
1667 					tlbmiss_handler_setup_pgd_end);
1668 }
1669 
1670 static void
1671 iPTE_LW(u32 **p, unsigned int pte, unsigned int ptr)
1672 {
1673 #ifdef CONFIG_SMP
1674 	if (IS_ENABLED(CONFIG_CPU_LOONGSON3_WORKAROUNDS))
1675 		uasm_i_sync(p, 0);
1676 # ifdef CONFIG_PHYS_ADDR_T_64BIT
1677 	if (cpu_has_64bits)
1678 		uasm_i_lld(p, pte, 0, ptr);
1679 	else
1680 # endif
1681 		UASM_i_LL(p, pte, 0, ptr);
1682 #else
1683 # ifdef CONFIG_PHYS_ADDR_T_64BIT
1684 	if (cpu_has_64bits)
1685 		uasm_i_ld(p, pte, 0, ptr);
1686 	else
1687 # endif
1688 		UASM_i_LW(p, pte, 0, ptr);
1689 #endif
1690 }
1691 
1692 static void
1693 iPTE_SW(u32 **p, struct uasm_reloc **r, unsigned int pte, unsigned int ptr,
1694 	unsigned int mode, unsigned int scratch)
1695 {
1696 	unsigned int hwmode = mode & (_PAGE_VALID | _PAGE_DIRTY);
1697 	unsigned int swmode = mode & ~hwmode;
1698 
1699 	if (IS_ENABLED(CONFIG_XPA) && !cpu_has_64bits) {
1700 		uasm_i_lui(p, scratch, swmode >> 16);
1701 		uasm_i_or(p, pte, pte, scratch);
1702 		BUG_ON(swmode & 0xffff);
1703 	} else {
1704 		uasm_i_ori(p, pte, pte, mode);
1705 	}
1706 
1707 #ifdef CONFIG_SMP
1708 # ifdef CONFIG_PHYS_ADDR_T_64BIT
1709 	if (cpu_has_64bits)
1710 		uasm_i_scd(p, pte, 0, ptr);
1711 	else
1712 # endif
1713 		UASM_i_SC(p, pte, 0, ptr);
1714 
1715 	if (r10000_llsc_war())
1716 		uasm_il_beqzl(p, r, pte, label_smp_pgtable_change);
1717 	else
1718 		uasm_il_beqz(p, r, pte, label_smp_pgtable_change);
1719 
1720 # ifdef CONFIG_PHYS_ADDR_T_64BIT
1721 	if (!cpu_has_64bits) {
1722 		/* no uasm_i_nop needed */
1723 		uasm_i_ll(p, pte, sizeof(pte_t) / 2, ptr);
1724 		uasm_i_ori(p, pte, pte, hwmode);
1725 		BUG_ON(hwmode & ~0xffff);
1726 		uasm_i_sc(p, pte, sizeof(pte_t) / 2, ptr);
1727 		uasm_il_beqz(p, r, pte, label_smp_pgtable_change);
1728 		/* no uasm_i_nop needed */
1729 		uasm_i_lw(p, pte, 0, ptr);
1730 	} else
1731 		uasm_i_nop(p);
1732 # else
1733 	uasm_i_nop(p);
1734 # endif
1735 #else
1736 # ifdef CONFIG_PHYS_ADDR_T_64BIT
1737 	if (cpu_has_64bits)
1738 		uasm_i_sd(p, pte, 0, ptr);
1739 	else
1740 # endif
1741 		UASM_i_SW(p, pte, 0, ptr);
1742 
1743 # ifdef CONFIG_PHYS_ADDR_T_64BIT
1744 	if (!cpu_has_64bits) {
1745 		uasm_i_lw(p, pte, sizeof(pte_t) / 2, ptr);
1746 		uasm_i_ori(p, pte, pte, hwmode);
1747 		BUG_ON(hwmode & ~0xffff);
1748 		uasm_i_sw(p, pte, sizeof(pte_t) / 2, ptr);
1749 		uasm_i_lw(p, pte, 0, ptr);
1750 	}
1751 # endif
1752 #endif
1753 }
1754 
1755 /*
1756  * Check if PTE is present, if not then jump to LABEL. PTR points to
1757  * the page table where this PTE is located, PTE will be re-loaded
1758  * with it's original value.
1759  */
1760 static void
1761 build_pte_present(u32 **p, struct uasm_reloc **r,
1762 		  int pte, int ptr, int scratch, enum label_id lid)
1763 {
1764 	int t = scratch >= 0 ? scratch : pte;
1765 	int cur = pte;
1766 
1767 	if (cpu_has_rixi) {
1768 		if (use_bbit_insns()) {
1769 			uasm_il_bbit0(p, r, pte, ilog2(_PAGE_PRESENT), lid);
1770 			uasm_i_nop(p);
1771 		} else {
1772 			if (_PAGE_PRESENT_SHIFT) {
1773 				uasm_i_srl(p, t, cur, _PAGE_PRESENT_SHIFT);
1774 				cur = t;
1775 			}
1776 			uasm_i_andi(p, t, cur, 1);
1777 			uasm_il_beqz(p, r, t, lid);
1778 			if (pte == t)
1779 				/* You lose the SMP race :-(*/
1780 				iPTE_LW(p, pte, ptr);
1781 		}
1782 	} else {
1783 		if (_PAGE_PRESENT_SHIFT) {
1784 			uasm_i_srl(p, t, cur, _PAGE_PRESENT_SHIFT);
1785 			cur = t;
1786 		}
1787 		uasm_i_andi(p, t, cur,
1788 			(_PAGE_PRESENT | _PAGE_NO_READ) >> _PAGE_PRESENT_SHIFT);
1789 		uasm_i_xori(p, t, t, _PAGE_PRESENT >> _PAGE_PRESENT_SHIFT);
1790 		uasm_il_bnez(p, r, t, lid);
1791 		if (pte == t)
1792 			/* You lose the SMP race :-(*/
1793 			iPTE_LW(p, pte, ptr);
1794 	}
1795 }
1796 
1797 /* Make PTE valid, store result in PTR. */
1798 static void
1799 build_make_valid(u32 **p, struct uasm_reloc **r, unsigned int pte,
1800 		 unsigned int ptr, unsigned int scratch)
1801 {
1802 	unsigned int mode = _PAGE_VALID | _PAGE_ACCESSED;
1803 
1804 	iPTE_SW(p, r, pte, ptr, mode, scratch);
1805 }
1806 
1807 /*
1808  * Check if PTE can be written to, if not branch to LABEL. Regardless
1809  * restore PTE with value from PTR when done.
1810  */
1811 static void
1812 build_pte_writable(u32 **p, struct uasm_reloc **r,
1813 		   unsigned int pte, unsigned int ptr, int scratch,
1814 		   enum label_id lid)
1815 {
1816 	int t = scratch >= 0 ? scratch : pte;
1817 	int cur = pte;
1818 
1819 	if (_PAGE_PRESENT_SHIFT) {
1820 		uasm_i_srl(p, t, cur, _PAGE_PRESENT_SHIFT);
1821 		cur = t;
1822 	}
1823 	uasm_i_andi(p, t, cur,
1824 		    (_PAGE_PRESENT | _PAGE_WRITE) >> _PAGE_PRESENT_SHIFT);
1825 	uasm_i_xori(p, t, t,
1826 		    (_PAGE_PRESENT | _PAGE_WRITE) >> _PAGE_PRESENT_SHIFT);
1827 	uasm_il_bnez(p, r, t, lid);
1828 	if (pte == t)
1829 		/* You lose the SMP race :-(*/
1830 		iPTE_LW(p, pte, ptr);
1831 	else
1832 		uasm_i_nop(p);
1833 }
1834 
1835 /* Make PTE writable, update software status bits as well, then store
1836  * at PTR.
1837  */
1838 static void
1839 build_make_write(u32 **p, struct uasm_reloc **r, unsigned int pte,
1840 		 unsigned int ptr, unsigned int scratch)
1841 {
1842 	unsigned int mode = (_PAGE_ACCESSED | _PAGE_MODIFIED | _PAGE_VALID
1843 			     | _PAGE_DIRTY);
1844 
1845 	iPTE_SW(p, r, pte, ptr, mode, scratch);
1846 }
1847 
1848 /*
1849  * Check if PTE can be modified, if not branch to LABEL. Regardless
1850  * restore PTE with value from PTR when done.
1851  */
1852 static void
1853 build_pte_modifiable(u32 **p, struct uasm_reloc **r,
1854 		     unsigned int pte, unsigned int ptr, int scratch,
1855 		     enum label_id lid)
1856 {
1857 	if (use_bbit_insns()) {
1858 		uasm_il_bbit0(p, r, pte, ilog2(_PAGE_WRITE), lid);
1859 		uasm_i_nop(p);
1860 	} else {
1861 		int t = scratch >= 0 ? scratch : pte;
1862 		uasm_i_srl(p, t, pte, _PAGE_WRITE_SHIFT);
1863 		uasm_i_andi(p, t, t, 1);
1864 		uasm_il_beqz(p, r, t, lid);
1865 		if (pte == t)
1866 			/* You lose the SMP race :-(*/
1867 			iPTE_LW(p, pte, ptr);
1868 	}
1869 }
1870 
1871 #ifndef CONFIG_MIPS_PGD_C0_CONTEXT
1872 
1873 
1874 /*
1875  * R3000 style TLB load/store/modify handlers.
1876  */
1877 
1878 /*
1879  * This places the pte into ENTRYLO0 and writes it with tlbwi.
1880  * Then it returns.
1881  */
1882 static void
1883 build_r3000_pte_reload_tlbwi(u32 **p, unsigned int pte, unsigned int tmp)
1884 {
1885 	uasm_i_mtc0(p, pte, C0_ENTRYLO0); /* cp0 delay */
1886 	uasm_i_mfc0(p, tmp, C0_EPC); /* cp0 delay */
1887 	uasm_i_tlbwi(p);
1888 	uasm_i_jr(p, tmp);
1889 	uasm_i_rfe(p); /* branch delay */
1890 }
1891 
1892 /*
1893  * This places the pte into ENTRYLO0 and writes it with tlbwi
1894  * or tlbwr as appropriate.  This is because the index register
1895  * may have the probe fail bit set as a result of a trap on a
1896  * kseg2 access, i.e. without refill.  Then it returns.
1897  */
1898 static void
1899 build_r3000_tlb_reload_write(u32 **p, struct uasm_label **l,
1900 			     struct uasm_reloc **r, unsigned int pte,
1901 			     unsigned int tmp)
1902 {
1903 	uasm_i_mfc0(p, tmp, C0_INDEX);
1904 	uasm_i_mtc0(p, pte, C0_ENTRYLO0); /* cp0 delay */
1905 	uasm_il_bltz(p, r, tmp, label_r3000_write_probe_fail); /* cp0 delay */
1906 	uasm_i_mfc0(p, tmp, C0_EPC); /* branch delay */
1907 	uasm_i_tlbwi(p); /* cp0 delay */
1908 	uasm_i_jr(p, tmp);
1909 	uasm_i_rfe(p); /* branch delay */
1910 	uasm_l_r3000_write_probe_fail(l, *p);
1911 	uasm_i_tlbwr(p); /* cp0 delay */
1912 	uasm_i_jr(p, tmp);
1913 	uasm_i_rfe(p); /* branch delay */
1914 }
1915 
1916 static void
1917 build_r3000_tlbchange_handler_head(u32 **p, unsigned int pte,
1918 				   unsigned int ptr)
1919 {
1920 	long pgdc = (long)pgd_current;
1921 
1922 	uasm_i_mfc0(p, pte, C0_BADVADDR);
1923 	uasm_i_lui(p, ptr, uasm_rel_hi(pgdc)); /* cp0 delay */
1924 	uasm_i_lw(p, ptr, uasm_rel_lo(pgdc), ptr);
1925 	uasm_i_srl(p, pte, pte, 22); /* load delay */
1926 	uasm_i_sll(p, pte, pte, 2);
1927 	uasm_i_addu(p, ptr, ptr, pte);
1928 	uasm_i_mfc0(p, pte, C0_CONTEXT);
1929 	uasm_i_lw(p, ptr, 0, ptr); /* cp0 delay */
1930 	uasm_i_andi(p, pte, pte, 0xffc); /* load delay */
1931 	uasm_i_addu(p, ptr, ptr, pte);
1932 	uasm_i_lw(p, pte, 0, ptr);
1933 	uasm_i_tlbp(p); /* load delay */
1934 }
1935 
1936 static void build_r3000_tlb_load_handler(void)
1937 {
1938 	u32 *p = (u32 *)handle_tlbl;
1939 	struct uasm_label *l = labels;
1940 	struct uasm_reloc *r = relocs;
1941 
1942 	memset(p, 0, handle_tlbl_end - (char *)p);
1943 	memset(labels, 0, sizeof(labels));
1944 	memset(relocs, 0, sizeof(relocs));
1945 
1946 	build_r3000_tlbchange_handler_head(&p, K0, K1);
1947 	build_pte_present(&p, &r, K0, K1, -1, label_nopage_tlbl);
1948 	uasm_i_nop(&p); /* load delay */
1949 	build_make_valid(&p, &r, K0, K1, -1);
1950 	build_r3000_tlb_reload_write(&p, &l, &r, K0, K1);
1951 
1952 	uasm_l_nopage_tlbl(&l, p);
1953 	uasm_i_j(&p, (unsigned long)tlb_do_page_fault_0 & 0x0fffffff);
1954 	uasm_i_nop(&p);
1955 
1956 	if (p >= (u32 *)handle_tlbl_end)
1957 		panic("TLB load handler fastpath space exceeded");
1958 
1959 	uasm_resolve_relocs(relocs, labels);
1960 	pr_debug("Wrote TLB load handler fastpath (%u instructions).\n",
1961 		 (unsigned int)(p - (u32 *)handle_tlbl));
1962 
1963 	dump_handler("r3000_tlb_load", handle_tlbl, handle_tlbl_end);
1964 }
1965 
1966 static void build_r3000_tlb_store_handler(void)
1967 {
1968 	u32 *p = (u32 *)handle_tlbs;
1969 	struct uasm_label *l = labels;
1970 	struct uasm_reloc *r = relocs;
1971 
1972 	memset(p, 0, handle_tlbs_end - (char *)p);
1973 	memset(labels, 0, sizeof(labels));
1974 	memset(relocs, 0, sizeof(relocs));
1975 
1976 	build_r3000_tlbchange_handler_head(&p, K0, K1);
1977 	build_pte_writable(&p, &r, K0, K1, -1, label_nopage_tlbs);
1978 	uasm_i_nop(&p); /* load delay */
1979 	build_make_write(&p, &r, K0, K1, -1);
1980 	build_r3000_tlb_reload_write(&p, &l, &r, K0, K1);
1981 
1982 	uasm_l_nopage_tlbs(&l, p);
1983 	uasm_i_j(&p, (unsigned long)tlb_do_page_fault_1 & 0x0fffffff);
1984 	uasm_i_nop(&p);
1985 
1986 	if (p >= (u32 *)handle_tlbs_end)
1987 		panic("TLB store handler fastpath space exceeded");
1988 
1989 	uasm_resolve_relocs(relocs, labels);
1990 	pr_debug("Wrote TLB store handler fastpath (%u instructions).\n",
1991 		 (unsigned int)(p - (u32 *)handle_tlbs));
1992 
1993 	dump_handler("r3000_tlb_store", handle_tlbs, handle_tlbs_end);
1994 }
1995 
1996 static void build_r3000_tlb_modify_handler(void)
1997 {
1998 	u32 *p = (u32 *)handle_tlbm;
1999 	struct uasm_label *l = labels;
2000 	struct uasm_reloc *r = relocs;
2001 
2002 	memset(p, 0, handle_tlbm_end - (char *)p);
2003 	memset(labels, 0, sizeof(labels));
2004 	memset(relocs, 0, sizeof(relocs));
2005 
2006 	build_r3000_tlbchange_handler_head(&p, K0, K1);
2007 	build_pte_modifiable(&p, &r, K0, K1,  -1, label_nopage_tlbm);
2008 	uasm_i_nop(&p); /* load delay */
2009 	build_make_write(&p, &r, K0, K1, -1);
2010 	build_r3000_pte_reload_tlbwi(&p, K0, K1);
2011 
2012 	uasm_l_nopage_tlbm(&l, p);
2013 	uasm_i_j(&p, (unsigned long)tlb_do_page_fault_1 & 0x0fffffff);
2014 	uasm_i_nop(&p);
2015 
2016 	if (p >= (u32 *)handle_tlbm_end)
2017 		panic("TLB modify handler fastpath space exceeded");
2018 
2019 	uasm_resolve_relocs(relocs, labels);
2020 	pr_debug("Wrote TLB modify handler fastpath (%u instructions).\n",
2021 		 (unsigned int)(p - (u32 *)handle_tlbm));
2022 
2023 	dump_handler("r3000_tlb_modify", handle_tlbm, handle_tlbm_end);
2024 }
2025 #endif /* CONFIG_MIPS_PGD_C0_CONTEXT */
2026 
2027 static bool cpu_has_tlbex_tlbp_race(void)
2028 {
2029 	/*
2030 	 * When a Hardware Table Walker is running it can replace TLB entries
2031 	 * at any time, leading to a race between it & the CPU.
2032 	 */
2033 	if (cpu_has_htw)
2034 		return true;
2035 
2036 	/*
2037 	 * If the CPU shares FTLB RAM with its siblings then our entry may be
2038 	 * replaced at any time by a sibling performing a write to the FTLB.
2039 	 */
2040 	if (cpu_has_shared_ftlb_ram)
2041 		return true;
2042 
2043 	/* In all other cases there ought to be no race condition to handle */
2044 	return false;
2045 }
2046 
2047 /*
2048  * R4000 style TLB load/store/modify handlers.
2049  */
2050 static struct work_registers
2051 build_r4000_tlbchange_handler_head(u32 **p, struct uasm_label **l,
2052 				   struct uasm_reloc **r)
2053 {
2054 	struct work_registers wr = build_get_work_registers(p);
2055 
2056 #ifdef CONFIG_64BIT
2057 	build_get_pmde64(p, l, r, wr.r1, wr.r2); /* get pmd in ptr */
2058 #else
2059 	build_get_pgde32(p, wr.r1, wr.r2); /* get pgd in ptr */
2060 #endif
2061 
2062 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
2063 	/*
2064 	 * For huge tlb entries, pmd doesn't contain an address but
2065 	 * instead contains the tlb pte. Check the PAGE_HUGE bit and
2066 	 * see if we need to jump to huge tlb processing.
2067 	 */
2068 	build_is_huge_pte(p, r, wr.r1, wr.r2, label_tlb_huge_update);
2069 #endif
2070 
2071 	UASM_i_MFC0(p, wr.r1, C0_BADVADDR);
2072 	UASM_i_LW(p, wr.r2, 0, wr.r2);
2073 	UASM_i_SRL(p, wr.r1, wr.r1, PAGE_SHIFT + PTE_ORDER - PTE_T_LOG2);
2074 	uasm_i_andi(p, wr.r1, wr.r1, (PTRS_PER_PTE - 1) << PTE_T_LOG2);
2075 	UASM_i_ADDU(p, wr.r2, wr.r2, wr.r1);
2076 
2077 #ifdef CONFIG_SMP
2078 	uasm_l_smp_pgtable_change(l, *p);
2079 #endif
2080 	iPTE_LW(p, wr.r1, wr.r2); /* get even pte */
2081 	if (!m4kc_tlbp_war()) {
2082 		build_tlb_probe_entry(p);
2083 		if (cpu_has_tlbex_tlbp_race()) {
2084 			/* race condition happens, leaving */
2085 			uasm_i_ehb(p);
2086 			uasm_i_mfc0(p, wr.r3, C0_INDEX);
2087 			uasm_il_bltz(p, r, wr.r3, label_leave);
2088 			uasm_i_nop(p);
2089 		}
2090 	}
2091 	return wr;
2092 }
2093 
2094 static void
2095 build_r4000_tlbchange_handler_tail(u32 **p, struct uasm_label **l,
2096 				   struct uasm_reloc **r, unsigned int tmp,
2097 				   unsigned int ptr)
2098 {
2099 	uasm_i_ori(p, ptr, ptr, sizeof(pte_t));
2100 	uasm_i_xori(p, ptr, ptr, sizeof(pte_t));
2101 	build_update_entries(p, tmp, ptr);
2102 	build_tlb_write_entry(p, l, r, tlb_indexed);
2103 	uasm_l_leave(l, *p);
2104 	build_restore_work_registers(p);
2105 	uasm_i_eret(p); /* return from trap */
2106 
2107 #ifdef CONFIG_64BIT
2108 	build_get_pgd_vmalloc64(p, l, r, tmp, ptr, not_refill);
2109 #endif
2110 }
2111 
2112 static void build_r4000_tlb_load_handler(void)
2113 {
2114 	u32 *p = (u32 *)msk_isa16_mode((ulong)handle_tlbl);
2115 	struct uasm_label *l = labels;
2116 	struct uasm_reloc *r = relocs;
2117 	struct work_registers wr;
2118 
2119 	memset(p, 0, handle_tlbl_end - (char *)p);
2120 	memset(labels, 0, sizeof(labels));
2121 	memset(relocs, 0, sizeof(relocs));
2122 
2123 	if (bcm1250_m3_war()) {
2124 		unsigned int segbits = 44;
2125 
2126 		uasm_i_dmfc0(&p, K0, C0_BADVADDR);
2127 		uasm_i_dmfc0(&p, K1, C0_ENTRYHI);
2128 		uasm_i_xor(&p, K0, K0, K1);
2129 		uasm_i_dsrl_safe(&p, K1, K0, 62);
2130 		uasm_i_dsrl_safe(&p, K0, K0, 12 + 1);
2131 		uasm_i_dsll_safe(&p, K0, K0, 64 + 12 + 1 - segbits);
2132 		uasm_i_or(&p, K0, K0, K1);
2133 		uasm_il_bnez(&p, &r, K0, label_leave);
2134 		/* No need for uasm_i_nop */
2135 	}
2136 
2137 	wr = build_r4000_tlbchange_handler_head(&p, &l, &r);
2138 	build_pte_present(&p, &r, wr.r1, wr.r2, wr.r3, label_nopage_tlbl);
2139 	if (m4kc_tlbp_war())
2140 		build_tlb_probe_entry(&p);
2141 
2142 	if (cpu_has_rixi && !cpu_has_rixiex) {
2143 		/*
2144 		 * If the page is not _PAGE_VALID, RI or XI could not
2145 		 * have triggered it.  Skip the expensive test..
2146 		 */
2147 		if (use_bbit_insns()) {
2148 			uasm_il_bbit0(&p, &r, wr.r1, ilog2(_PAGE_VALID),
2149 				      label_tlbl_goaround1);
2150 		} else {
2151 			uasm_i_andi(&p, wr.r3, wr.r1, _PAGE_VALID);
2152 			uasm_il_beqz(&p, &r, wr.r3, label_tlbl_goaround1);
2153 		}
2154 		uasm_i_nop(&p);
2155 
2156 		/*
2157 		 * Warn if something may race with us & replace the TLB entry
2158 		 * before we read it here. Everything with such races should
2159 		 * also have dedicated RiXi exception handlers, so this
2160 		 * shouldn't be hit.
2161 		 */
2162 		WARN(cpu_has_tlbex_tlbp_race(), "Unhandled race in RiXi path");
2163 
2164 		uasm_i_tlbr(&p);
2165 
2166 		switch (current_cpu_type()) {
2167 		default:
2168 			if (cpu_has_mips_r2_exec_hazard) {
2169 				uasm_i_ehb(&p);
2170 
2171 		case CPU_CAVIUM_OCTEON:
2172 		case CPU_CAVIUM_OCTEON_PLUS:
2173 		case CPU_CAVIUM_OCTEON2:
2174 				break;
2175 			}
2176 		}
2177 
2178 		/* Examine  entrylo 0 or 1 based on ptr. */
2179 		if (use_bbit_insns()) {
2180 			uasm_i_bbit0(&p, wr.r2, ilog2(sizeof(pte_t)), 8);
2181 		} else {
2182 			uasm_i_andi(&p, wr.r3, wr.r2, sizeof(pte_t));
2183 			uasm_i_beqz(&p, wr.r3, 8);
2184 		}
2185 		/* load it in the delay slot*/
2186 		UASM_i_MFC0(&p, wr.r3, C0_ENTRYLO0);
2187 		/* load it if ptr is odd */
2188 		UASM_i_MFC0(&p, wr.r3, C0_ENTRYLO1);
2189 		/*
2190 		 * If the entryLo (now in wr.r3) is valid (bit 1), RI or
2191 		 * XI must have triggered it.
2192 		 */
2193 		if (use_bbit_insns()) {
2194 			uasm_il_bbit1(&p, &r, wr.r3, 1, label_nopage_tlbl);
2195 			uasm_i_nop(&p);
2196 			uasm_l_tlbl_goaround1(&l, p);
2197 		} else {
2198 			uasm_i_andi(&p, wr.r3, wr.r3, 2);
2199 			uasm_il_bnez(&p, &r, wr.r3, label_nopage_tlbl);
2200 			uasm_i_nop(&p);
2201 		}
2202 		uasm_l_tlbl_goaround1(&l, p);
2203 	}
2204 	build_make_valid(&p, &r, wr.r1, wr.r2, wr.r3);
2205 	build_r4000_tlbchange_handler_tail(&p, &l, &r, wr.r1, wr.r2);
2206 
2207 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
2208 	/*
2209 	 * This is the entry point when build_r4000_tlbchange_handler_head
2210 	 * spots a huge page.
2211 	 */
2212 	uasm_l_tlb_huge_update(&l, p);
2213 	iPTE_LW(&p, wr.r1, wr.r2);
2214 	build_pte_present(&p, &r, wr.r1, wr.r2, wr.r3, label_nopage_tlbl);
2215 	build_tlb_probe_entry(&p);
2216 
2217 	if (cpu_has_rixi && !cpu_has_rixiex) {
2218 		/*
2219 		 * If the page is not _PAGE_VALID, RI or XI could not
2220 		 * have triggered it.  Skip the expensive test..
2221 		 */
2222 		if (use_bbit_insns()) {
2223 			uasm_il_bbit0(&p, &r, wr.r1, ilog2(_PAGE_VALID),
2224 				      label_tlbl_goaround2);
2225 		} else {
2226 			uasm_i_andi(&p, wr.r3, wr.r1, _PAGE_VALID);
2227 			uasm_il_beqz(&p, &r, wr.r3, label_tlbl_goaround2);
2228 		}
2229 		uasm_i_nop(&p);
2230 
2231 		/*
2232 		 * Warn if something may race with us & replace the TLB entry
2233 		 * before we read it here. Everything with such races should
2234 		 * also have dedicated RiXi exception handlers, so this
2235 		 * shouldn't be hit.
2236 		 */
2237 		WARN(cpu_has_tlbex_tlbp_race(), "Unhandled race in RiXi path");
2238 
2239 		uasm_i_tlbr(&p);
2240 
2241 		switch (current_cpu_type()) {
2242 		default:
2243 			if (cpu_has_mips_r2_exec_hazard) {
2244 				uasm_i_ehb(&p);
2245 
2246 		case CPU_CAVIUM_OCTEON:
2247 		case CPU_CAVIUM_OCTEON_PLUS:
2248 		case CPU_CAVIUM_OCTEON2:
2249 				break;
2250 			}
2251 		}
2252 
2253 		/* Examine  entrylo 0 or 1 based on ptr. */
2254 		if (use_bbit_insns()) {
2255 			uasm_i_bbit0(&p, wr.r2, ilog2(sizeof(pte_t)), 8);
2256 		} else {
2257 			uasm_i_andi(&p, wr.r3, wr.r2, sizeof(pte_t));
2258 			uasm_i_beqz(&p, wr.r3, 8);
2259 		}
2260 		/* load it in the delay slot*/
2261 		UASM_i_MFC0(&p, wr.r3, C0_ENTRYLO0);
2262 		/* load it if ptr is odd */
2263 		UASM_i_MFC0(&p, wr.r3, C0_ENTRYLO1);
2264 		/*
2265 		 * If the entryLo (now in wr.r3) is valid (bit 1), RI or
2266 		 * XI must have triggered it.
2267 		 */
2268 		if (use_bbit_insns()) {
2269 			uasm_il_bbit0(&p, &r, wr.r3, 1, label_tlbl_goaround2);
2270 		} else {
2271 			uasm_i_andi(&p, wr.r3, wr.r3, 2);
2272 			uasm_il_beqz(&p, &r, wr.r3, label_tlbl_goaround2);
2273 		}
2274 		if (PM_DEFAULT_MASK == 0)
2275 			uasm_i_nop(&p);
2276 		/*
2277 		 * We clobbered C0_PAGEMASK, restore it.  On the other branch
2278 		 * it is restored in build_huge_tlb_write_entry.
2279 		 */
2280 		build_restore_pagemask(&p, &r, wr.r3, label_nopage_tlbl, 0);
2281 
2282 		uasm_l_tlbl_goaround2(&l, p);
2283 	}
2284 	uasm_i_ori(&p, wr.r1, wr.r1, (_PAGE_ACCESSED | _PAGE_VALID));
2285 	build_huge_handler_tail(&p, &r, &l, wr.r1, wr.r2, 1);
2286 #endif
2287 
2288 	uasm_l_nopage_tlbl(&l, p);
2289 	if (IS_ENABLED(CONFIG_CPU_LOONGSON3_WORKAROUNDS))
2290 		uasm_i_sync(&p, 0);
2291 	build_restore_work_registers(&p);
2292 #ifdef CONFIG_CPU_MICROMIPS
2293 	if ((unsigned long)tlb_do_page_fault_0 & 1) {
2294 		uasm_i_lui(&p, K0, uasm_rel_hi((long)tlb_do_page_fault_0));
2295 		uasm_i_addiu(&p, K0, K0, uasm_rel_lo((long)tlb_do_page_fault_0));
2296 		uasm_i_jr(&p, K0);
2297 	} else
2298 #endif
2299 	uasm_i_j(&p, (unsigned long)tlb_do_page_fault_0 & 0x0fffffff);
2300 	uasm_i_nop(&p);
2301 
2302 	if (p >= (u32 *)handle_tlbl_end)
2303 		panic("TLB load handler fastpath space exceeded");
2304 
2305 	uasm_resolve_relocs(relocs, labels);
2306 	pr_debug("Wrote TLB load handler fastpath (%u instructions).\n",
2307 		 (unsigned int)(p - (u32 *)handle_tlbl));
2308 
2309 	dump_handler("r4000_tlb_load", handle_tlbl, handle_tlbl_end);
2310 }
2311 
2312 static void build_r4000_tlb_store_handler(void)
2313 {
2314 	u32 *p = (u32 *)msk_isa16_mode((ulong)handle_tlbs);
2315 	struct uasm_label *l = labels;
2316 	struct uasm_reloc *r = relocs;
2317 	struct work_registers wr;
2318 
2319 	memset(p, 0, handle_tlbs_end - (char *)p);
2320 	memset(labels, 0, sizeof(labels));
2321 	memset(relocs, 0, sizeof(relocs));
2322 
2323 	wr = build_r4000_tlbchange_handler_head(&p, &l, &r);
2324 	build_pte_writable(&p, &r, wr.r1, wr.r2, wr.r3, label_nopage_tlbs);
2325 	if (m4kc_tlbp_war())
2326 		build_tlb_probe_entry(&p);
2327 	build_make_write(&p, &r, wr.r1, wr.r2, wr.r3);
2328 	build_r4000_tlbchange_handler_tail(&p, &l, &r, wr.r1, wr.r2);
2329 
2330 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
2331 	/*
2332 	 * This is the entry point when
2333 	 * build_r4000_tlbchange_handler_head spots a huge page.
2334 	 */
2335 	uasm_l_tlb_huge_update(&l, p);
2336 	iPTE_LW(&p, wr.r1, wr.r2);
2337 	build_pte_writable(&p, &r, wr.r1, wr.r2, wr.r3, label_nopage_tlbs);
2338 	build_tlb_probe_entry(&p);
2339 	uasm_i_ori(&p, wr.r1, wr.r1,
2340 		   _PAGE_ACCESSED | _PAGE_MODIFIED | _PAGE_VALID | _PAGE_DIRTY);
2341 	build_huge_handler_tail(&p, &r, &l, wr.r1, wr.r2, 1);
2342 #endif
2343 
2344 	uasm_l_nopage_tlbs(&l, p);
2345 	if (IS_ENABLED(CONFIG_CPU_LOONGSON3_WORKAROUNDS))
2346 		uasm_i_sync(&p, 0);
2347 	build_restore_work_registers(&p);
2348 #ifdef CONFIG_CPU_MICROMIPS
2349 	if ((unsigned long)tlb_do_page_fault_1 & 1) {
2350 		uasm_i_lui(&p, K0, uasm_rel_hi((long)tlb_do_page_fault_1));
2351 		uasm_i_addiu(&p, K0, K0, uasm_rel_lo((long)tlb_do_page_fault_1));
2352 		uasm_i_jr(&p, K0);
2353 	} else
2354 #endif
2355 	uasm_i_j(&p, (unsigned long)tlb_do_page_fault_1 & 0x0fffffff);
2356 	uasm_i_nop(&p);
2357 
2358 	if (p >= (u32 *)handle_tlbs_end)
2359 		panic("TLB store handler fastpath space exceeded");
2360 
2361 	uasm_resolve_relocs(relocs, labels);
2362 	pr_debug("Wrote TLB store handler fastpath (%u instructions).\n",
2363 		 (unsigned int)(p - (u32 *)handle_tlbs));
2364 
2365 	dump_handler("r4000_tlb_store", handle_tlbs, handle_tlbs_end);
2366 }
2367 
2368 static void build_r4000_tlb_modify_handler(void)
2369 {
2370 	u32 *p = (u32 *)msk_isa16_mode((ulong)handle_tlbm);
2371 	struct uasm_label *l = labels;
2372 	struct uasm_reloc *r = relocs;
2373 	struct work_registers wr;
2374 
2375 	memset(p, 0, handle_tlbm_end - (char *)p);
2376 	memset(labels, 0, sizeof(labels));
2377 	memset(relocs, 0, sizeof(relocs));
2378 
2379 	wr = build_r4000_tlbchange_handler_head(&p, &l, &r);
2380 	build_pte_modifiable(&p, &r, wr.r1, wr.r2, wr.r3, label_nopage_tlbm);
2381 	if (m4kc_tlbp_war())
2382 		build_tlb_probe_entry(&p);
2383 	/* Present and writable bits set, set accessed and dirty bits. */
2384 	build_make_write(&p, &r, wr.r1, wr.r2, wr.r3);
2385 	build_r4000_tlbchange_handler_tail(&p, &l, &r, wr.r1, wr.r2);
2386 
2387 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
2388 	/*
2389 	 * This is the entry point when
2390 	 * build_r4000_tlbchange_handler_head spots a huge page.
2391 	 */
2392 	uasm_l_tlb_huge_update(&l, p);
2393 	iPTE_LW(&p, wr.r1, wr.r2);
2394 	build_pte_modifiable(&p, &r, wr.r1, wr.r2,  wr.r3, label_nopage_tlbm);
2395 	build_tlb_probe_entry(&p);
2396 	uasm_i_ori(&p, wr.r1, wr.r1,
2397 		   _PAGE_ACCESSED | _PAGE_MODIFIED | _PAGE_VALID | _PAGE_DIRTY);
2398 	build_huge_handler_tail(&p, &r, &l, wr.r1, wr.r2, 0);
2399 #endif
2400 
2401 	uasm_l_nopage_tlbm(&l, p);
2402 	if (IS_ENABLED(CONFIG_CPU_LOONGSON3_WORKAROUNDS))
2403 		uasm_i_sync(&p, 0);
2404 	build_restore_work_registers(&p);
2405 #ifdef CONFIG_CPU_MICROMIPS
2406 	if ((unsigned long)tlb_do_page_fault_1 & 1) {
2407 		uasm_i_lui(&p, K0, uasm_rel_hi((long)tlb_do_page_fault_1));
2408 		uasm_i_addiu(&p, K0, K0, uasm_rel_lo((long)tlb_do_page_fault_1));
2409 		uasm_i_jr(&p, K0);
2410 	} else
2411 #endif
2412 	uasm_i_j(&p, (unsigned long)tlb_do_page_fault_1 & 0x0fffffff);
2413 	uasm_i_nop(&p);
2414 
2415 	if (p >= (u32 *)handle_tlbm_end)
2416 		panic("TLB modify handler fastpath space exceeded");
2417 
2418 	uasm_resolve_relocs(relocs, labels);
2419 	pr_debug("Wrote TLB modify handler fastpath (%u instructions).\n",
2420 		 (unsigned int)(p - (u32 *)handle_tlbm));
2421 
2422 	dump_handler("r4000_tlb_modify", handle_tlbm, handle_tlbm_end);
2423 }
2424 
2425 static void flush_tlb_handlers(void)
2426 {
2427 	local_flush_icache_range((unsigned long)handle_tlbl,
2428 			   (unsigned long)handle_tlbl_end);
2429 	local_flush_icache_range((unsigned long)handle_tlbs,
2430 			   (unsigned long)handle_tlbs_end);
2431 	local_flush_icache_range((unsigned long)handle_tlbm,
2432 			   (unsigned long)handle_tlbm_end);
2433 	local_flush_icache_range((unsigned long)tlbmiss_handler_setup_pgd,
2434 			   (unsigned long)tlbmiss_handler_setup_pgd_end);
2435 }
2436 
2437 static void print_htw_config(void)
2438 {
2439 	unsigned long config;
2440 	unsigned int pwctl;
2441 	const int field = 2 * sizeof(unsigned long);
2442 
2443 	config = read_c0_pwfield();
2444 	pr_debug("PWField (0x%0*lx): GDI: 0x%02lx  UDI: 0x%02lx  MDI: 0x%02lx  PTI: 0x%02lx  PTEI: 0x%02lx\n",
2445 		field, config,
2446 		(config & MIPS_PWFIELD_GDI_MASK) >> MIPS_PWFIELD_GDI_SHIFT,
2447 		(config & MIPS_PWFIELD_UDI_MASK) >> MIPS_PWFIELD_UDI_SHIFT,
2448 		(config & MIPS_PWFIELD_MDI_MASK) >> MIPS_PWFIELD_MDI_SHIFT,
2449 		(config & MIPS_PWFIELD_PTI_MASK) >> MIPS_PWFIELD_PTI_SHIFT,
2450 		(config & MIPS_PWFIELD_PTEI_MASK) >> MIPS_PWFIELD_PTEI_SHIFT);
2451 
2452 	config = read_c0_pwsize();
2453 	pr_debug("PWSize  (0x%0*lx): PS: 0x%lx  GDW: 0x%02lx  UDW: 0x%02lx  MDW: 0x%02lx  PTW: 0x%02lx  PTEW: 0x%02lx\n",
2454 		field, config,
2455 		(config & MIPS_PWSIZE_PS_MASK) >> MIPS_PWSIZE_PS_SHIFT,
2456 		(config & MIPS_PWSIZE_GDW_MASK) >> MIPS_PWSIZE_GDW_SHIFT,
2457 		(config & MIPS_PWSIZE_UDW_MASK) >> MIPS_PWSIZE_UDW_SHIFT,
2458 		(config & MIPS_PWSIZE_MDW_MASK) >> MIPS_PWSIZE_MDW_SHIFT,
2459 		(config & MIPS_PWSIZE_PTW_MASK) >> MIPS_PWSIZE_PTW_SHIFT,
2460 		(config & MIPS_PWSIZE_PTEW_MASK) >> MIPS_PWSIZE_PTEW_SHIFT);
2461 
2462 	pwctl = read_c0_pwctl();
2463 	pr_debug("PWCtl   (0x%x): PWEn: 0x%x  XK: 0x%x  XS: 0x%x  XU: 0x%x  DPH: 0x%x  HugePg: 0x%x  Psn: 0x%x\n",
2464 		pwctl,
2465 		(pwctl & MIPS_PWCTL_PWEN_MASK) >> MIPS_PWCTL_PWEN_SHIFT,
2466 		(pwctl & MIPS_PWCTL_XK_MASK) >> MIPS_PWCTL_XK_SHIFT,
2467 		(pwctl & MIPS_PWCTL_XS_MASK) >> MIPS_PWCTL_XS_SHIFT,
2468 		(pwctl & MIPS_PWCTL_XU_MASK) >> MIPS_PWCTL_XU_SHIFT,
2469 		(pwctl & MIPS_PWCTL_DPH_MASK) >> MIPS_PWCTL_DPH_SHIFT,
2470 		(pwctl & MIPS_PWCTL_HUGEPG_MASK) >> MIPS_PWCTL_HUGEPG_SHIFT,
2471 		(pwctl & MIPS_PWCTL_PSN_MASK) >> MIPS_PWCTL_PSN_SHIFT);
2472 }
2473 
2474 static void config_htw_params(void)
2475 {
2476 	unsigned long pwfield, pwsize, ptei;
2477 	unsigned int config;
2478 
2479 	/*
2480 	 * We are using 2-level page tables, so we only need to
2481 	 * setup GDW and PTW appropriately. UDW and MDW will remain 0.
2482 	 * The default value of GDI/UDI/MDI/PTI is 0xc. It is illegal to
2483 	 * write values less than 0xc in these fields because the entire
2484 	 * write will be dropped. As a result of which, we must preserve
2485 	 * the original reset values and overwrite only what we really want.
2486 	 */
2487 
2488 	pwfield = read_c0_pwfield();
2489 	/* re-initialize the GDI field */
2490 	pwfield &= ~MIPS_PWFIELD_GDI_MASK;
2491 	pwfield |= PGDIR_SHIFT << MIPS_PWFIELD_GDI_SHIFT;
2492 	/* re-initialize the PTI field including the even/odd bit */
2493 	pwfield &= ~MIPS_PWFIELD_PTI_MASK;
2494 	pwfield |= PAGE_SHIFT << MIPS_PWFIELD_PTI_SHIFT;
2495 	if (CONFIG_PGTABLE_LEVELS >= 3) {
2496 		pwfield &= ~MIPS_PWFIELD_MDI_MASK;
2497 		pwfield |= PMD_SHIFT << MIPS_PWFIELD_MDI_SHIFT;
2498 	}
2499 	/* Set the PTEI right shift */
2500 	ptei = _PAGE_GLOBAL_SHIFT << MIPS_PWFIELD_PTEI_SHIFT;
2501 	pwfield |= ptei;
2502 	write_c0_pwfield(pwfield);
2503 	/* Check whether the PTEI value is supported */
2504 	back_to_back_c0_hazard();
2505 	pwfield = read_c0_pwfield();
2506 	if (((pwfield & MIPS_PWFIELD_PTEI_MASK) << MIPS_PWFIELD_PTEI_SHIFT)
2507 		!= ptei) {
2508 		pr_warn("Unsupported PTEI field value: 0x%lx. HTW will not be enabled",
2509 			ptei);
2510 		/*
2511 		 * Drop option to avoid HTW being enabled via another path
2512 		 * (eg htw_reset())
2513 		 */
2514 		current_cpu_data.options &= ~MIPS_CPU_HTW;
2515 		return;
2516 	}
2517 
2518 	pwsize = ilog2(PTRS_PER_PGD) << MIPS_PWSIZE_GDW_SHIFT;
2519 	pwsize |= ilog2(PTRS_PER_PTE) << MIPS_PWSIZE_PTW_SHIFT;
2520 	if (CONFIG_PGTABLE_LEVELS >= 3)
2521 		pwsize |= ilog2(PTRS_PER_PMD) << MIPS_PWSIZE_MDW_SHIFT;
2522 
2523 	/* Set pointer size to size of directory pointers */
2524 	if (IS_ENABLED(CONFIG_64BIT))
2525 		pwsize |= MIPS_PWSIZE_PS_MASK;
2526 	/* PTEs may be multiple pointers long (e.g. with XPA) */
2527 	pwsize |= ((PTE_T_LOG2 - PGD_T_LOG2) << MIPS_PWSIZE_PTEW_SHIFT)
2528 			& MIPS_PWSIZE_PTEW_MASK;
2529 
2530 	write_c0_pwsize(pwsize);
2531 
2532 	/* Make sure everything is set before we enable the HTW */
2533 	back_to_back_c0_hazard();
2534 
2535 	/*
2536 	 * Enable HTW (and only for XUSeg on 64-bit), and disable the rest of
2537 	 * the pwctl fields.
2538 	 */
2539 	config = 1 << MIPS_PWCTL_PWEN_SHIFT;
2540 	if (IS_ENABLED(CONFIG_64BIT))
2541 		config |= MIPS_PWCTL_XU_MASK;
2542 	write_c0_pwctl(config);
2543 	pr_info("Hardware Page Table Walker enabled\n");
2544 
2545 	print_htw_config();
2546 }
2547 
2548 static void config_xpa_params(void)
2549 {
2550 #ifdef CONFIG_XPA
2551 	unsigned int pagegrain;
2552 
2553 	if (mips_xpa_disabled) {
2554 		pr_info("Extended Physical Addressing (XPA) disabled\n");
2555 		return;
2556 	}
2557 
2558 	pagegrain = read_c0_pagegrain();
2559 	write_c0_pagegrain(pagegrain | PG_ELPA);
2560 	back_to_back_c0_hazard();
2561 	pagegrain = read_c0_pagegrain();
2562 
2563 	if (pagegrain & PG_ELPA)
2564 		pr_info("Extended Physical Addressing (XPA) enabled\n");
2565 	else
2566 		panic("Extended Physical Addressing (XPA) disabled");
2567 #endif
2568 }
2569 
2570 static void check_pabits(void)
2571 {
2572 	unsigned long entry;
2573 	unsigned pabits, fillbits;
2574 
2575 	if (!cpu_has_rixi || !_PAGE_NO_EXEC) {
2576 		/*
2577 		 * We'll only be making use of the fact that we can rotate bits
2578 		 * into the fill if the CPU supports RIXI, so don't bother
2579 		 * probing this for CPUs which don't.
2580 		 */
2581 		return;
2582 	}
2583 
2584 	write_c0_entrylo0(~0ul);
2585 	back_to_back_c0_hazard();
2586 	entry = read_c0_entrylo0();
2587 
2588 	/* clear all non-PFN bits */
2589 	entry &= ~((1 << MIPS_ENTRYLO_PFN_SHIFT) - 1);
2590 	entry &= ~(MIPS_ENTRYLO_RI | MIPS_ENTRYLO_XI);
2591 
2592 	/* find a lower bound on PABITS, and upper bound on fill bits */
2593 	pabits = fls_long(entry) + 6;
2594 	fillbits = max_t(int, (int)BITS_PER_LONG - pabits, 0);
2595 
2596 	/* minus the RI & XI bits */
2597 	fillbits -= min_t(unsigned, fillbits, 2);
2598 
2599 	if (fillbits >= ilog2(_PAGE_NO_EXEC))
2600 		fill_includes_sw_bits = true;
2601 
2602 	pr_debug("Entry* registers contain %u fill bits\n", fillbits);
2603 }
2604 
2605 void build_tlb_refill_handler(void)
2606 {
2607 	/*
2608 	 * The refill handler is generated per-CPU, multi-node systems
2609 	 * may have local storage for it. The other handlers are only
2610 	 * needed once.
2611 	 */
2612 	static int run_once = 0;
2613 
2614 	if (IS_ENABLED(CONFIG_XPA) && !cpu_has_rixi)
2615 		panic("Kernels supporting XPA currently require CPUs with RIXI");
2616 
2617 	output_pgtable_bits_defines();
2618 	check_pabits();
2619 
2620 #ifdef CONFIG_64BIT
2621 	check_for_high_segbits = current_cpu_data.vmbits > (PGDIR_SHIFT + PGD_ORDER + PAGE_SHIFT - 3);
2622 #endif
2623 
2624 	if (cpu_has_3kex) {
2625 #ifndef CONFIG_MIPS_PGD_C0_CONTEXT
2626 		if (!run_once) {
2627 			build_setup_pgd();
2628 			build_r3000_tlb_refill_handler();
2629 			build_r3000_tlb_load_handler();
2630 			build_r3000_tlb_store_handler();
2631 			build_r3000_tlb_modify_handler();
2632 			flush_tlb_handlers();
2633 			run_once++;
2634 		}
2635 #else
2636 		panic("No R3000 TLB refill handler");
2637 #endif
2638 		return;
2639 	}
2640 
2641 	if (cpu_has_ldpte)
2642 		setup_pw();
2643 
2644 	if (!run_once) {
2645 		scratch_reg = allocate_kscratch();
2646 		build_setup_pgd();
2647 		build_r4000_tlb_load_handler();
2648 		build_r4000_tlb_store_handler();
2649 		build_r4000_tlb_modify_handler();
2650 		if (cpu_has_ldpte)
2651 			build_loongson3_tlb_refill_handler();
2652 		else
2653 			build_r4000_tlb_refill_handler();
2654 		flush_tlb_handlers();
2655 		run_once++;
2656 	}
2657 	if (cpu_has_xpa)
2658 		config_xpa_params();
2659 	if (cpu_has_htw)
2660 		config_htw_params();
2661 }
2662