xref: /openbmc/linux/arch/mips/mm/tlbex.c (revision d236d361)
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 
32 #include <asm/cacheflush.h>
33 #include <asm/cpu-type.h>
34 #include <asm/pgtable.h>
35 #include <asm/war.h>
36 #include <asm/uasm.h>
37 #include <asm/setup.h>
38 #include <asm/tlbex.h>
39 
40 static int mips_xpa_disabled;
41 
42 static int __init xpa_disable(char *s)
43 {
44 	mips_xpa_disabled = 1;
45 
46 	return 1;
47 }
48 
49 __setup("noxpa", xpa_disable);
50 
51 /*
52  * TLB load/store/modify handlers.
53  *
54  * Only the fastpath gets synthesized at runtime, the slowpath for
55  * do_page_fault remains normal asm.
56  */
57 extern void tlb_do_page_fault_0(void);
58 extern void tlb_do_page_fault_1(void);
59 
60 struct work_registers {
61 	int r1;
62 	int r2;
63 	int r3;
64 };
65 
66 struct tlb_reg_save {
67 	unsigned long a;
68 	unsigned long b;
69 } ____cacheline_aligned_in_smp;
70 
71 static struct tlb_reg_save handler_reg_save[NR_CPUS];
72 
73 static inline int r45k_bvahwbug(void)
74 {
75 	/* XXX: We should probe for the presence of this bug, but we don't. */
76 	return 0;
77 }
78 
79 static inline int r4k_250MHZhwbug(void)
80 {
81 	/* XXX: We should probe for the presence of this bug, but we don't. */
82 	return 0;
83 }
84 
85 static inline int __maybe_unused bcm1250_m3_war(void)
86 {
87 	return BCM1250_M3_WAR;
88 }
89 
90 static inline int __maybe_unused r10000_llsc_war(void)
91 {
92 	return R10000_LLSC_WAR;
93 }
94 
95 static int use_bbit_insns(void)
96 {
97 	switch (current_cpu_type()) {
98 	case CPU_CAVIUM_OCTEON:
99 	case CPU_CAVIUM_OCTEON_PLUS:
100 	case CPU_CAVIUM_OCTEON2:
101 	case CPU_CAVIUM_OCTEON3:
102 		return 1;
103 	default:
104 		return 0;
105 	}
106 }
107 
108 static int use_lwx_insns(void)
109 {
110 	switch (current_cpu_type()) {
111 	case CPU_CAVIUM_OCTEON2:
112 	case CPU_CAVIUM_OCTEON3:
113 		return 1;
114 	default:
115 		return 0;
116 	}
117 }
118 #if defined(CONFIG_CAVIUM_OCTEON_CVMSEG_SIZE) && \
119     CONFIG_CAVIUM_OCTEON_CVMSEG_SIZE > 0
120 static bool scratchpad_available(void)
121 {
122 	return true;
123 }
124 static int scratchpad_offset(int i)
125 {
126 	/*
127 	 * CVMSEG starts at address -32768 and extends for
128 	 * CAVIUM_OCTEON_CVMSEG_SIZE 128 byte cache lines.
129 	 */
130 	i += 1; /* Kernel use starts at the top and works down. */
131 	return CONFIG_CAVIUM_OCTEON_CVMSEG_SIZE * 128 - (8 * i) - 32768;
132 }
133 #else
134 static bool scratchpad_available(void)
135 {
136 	return false;
137 }
138 static int scratchpad_offset(int i)
139 {
140 	BUG();
141 	/* Really unreachable, but evidently some GCC want this. */
142 	return 0;
143 }
144 #endif
145 /*
146  * Found by experiment: At least some revisions of the 4kc throw under
147  * some circumstances a machine check exception, triggered by invalid
148  * values in the index register.  Delaying the tlbp instruction until
149  * after the next branch,  plus adding an additional nop in front of
150  * tlbwi/tlbwr avoids the invalid index register values. Nobody knows
151  * why; it's not an issue caused by the core RTL.
152  *
153  */
154 static int m4kc_tlbp_war(void)
155 {
156 	return (current_cpu_data.processor_id & 0xffff00) ==
157 	       (PRID_COMP_MIPS | PRID_IMP_4KC);
158 }
159 
160 /* Handle labels (which must be positive integers). */
161 enum label_id {
162 	label_second_part = 1,
163 	label_leave,
164 	label_vmalloc,
165 	label_vmalloc_done,
166 	label_tlbw_hazard_0,
167 	label_split = label_tlbw_hazard_0 + 8,
168 	label_tlbl_goaround1,
169 	label_tlbl_goaround2,
170 	label_nopage_tlbl,
171 	label_nopage_tlbs,
172 	label_nopage_tlbm,
173 	label_smp_pgtable_change,
174 	label_r3000_write_probe_fail,
175 	label_large_segbits_fault,
176 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
177 	label_tlb_huge_update,
178 #endif
179 };
180 
181 UASM_L_LA(_second_part)
182 UASM_L_LA(_leave)
183 UASM_L_LA(_vmalloc)
184 UASM_L_LA(_vmalloc_done)
185 /* _tlbw_hazard_x is handled differently.  */
186 UASM_L_LA(_split)
187 UASM_L_LA(_tlbl_goaround1)
188 UASM_L_LA(_tlbl_goaround2)
189 UASM_L_LA(_nopage_tlbl)
190 UASM_L_LA(_nopage_tlbs)
191 UASM_L_LA(_nopage_tlbm)
192 UASM_L_LA(_smp_pgtable_change)
193 UASM_L_LA(_r3000_write_probe_fail)
194 UASM_L_LA(_large_segbits_fault)
195 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
196 UASM_L_LA(_tlb_huge_update)
197 #endif
198 
199 static int hazard_instance;
200 
201 static void uasm_bgezl_hazard(u32 **p, struct uasm_reloc **r, int instance)
202 {
203 	switch (instance) {
204 	case 0 ... 7:
205 		uasm_il_bgezl(p, r, 0, label_tlbw_hazard_0 + instance);
206 		return;
207 	default:
208 		BUG();
209 	}
210 }
211 
212 static void uasm_bgezl_label(struct uasm_label **l, u32 **p, int instance)
213 {
214 	switch (instance) {
215 	case 0 ... 7:
216 		uasm_build_label(l, *p, label_tlbw_hazard_0 + instance);
217 		break;
218 	default:
219 		BUG();
220 	}
221 }
222 
223 /*
224  * pgtable bits are assigned dynamically depending on processor feature
225  * and statically based on kernel configuration.  This spits out the actual
226  * values the kernel is using.	Required to make sense from disassembled
227  * TLB exception handlers.
228  */
229 static void output_pgtable_bits_defines(void)
230 {
231 #define pr_define(fmt, ...)					\
232 	pr_debug("#define " fmt, ##__VA_ARGS__)
233 
234 	pr_debug("#include <asm/asm.h>\n");
235 	pr_debug("#include <asm/regdef.h>\n");
236 	pr_debug("\n");
237 
238 	pr_define("_PAGE_PRESENT_SHIFT %d\n", _PAGE_PRESENT_SHIFT);
239 	pr_define("_PAGE_NO_READ_SHIFT %d\n", _PAGE_NO_READ_SHIFT);
240 	pr_define("_PAGE_WRITE_SHIFT %d\n", _PAGE_WRITE_SHIFT);
241 	pr_define("_PAGE_ACCESSED_SHIFT %d\n", _PAGE_ACCESSED_SHIFT);
242 	pr_define("_PAGE_MODIFIED_SHIFT %d\n", _PAGE_MODIFIED_SHIFT);
243 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
244 	pr_define("_PAGE_HUGE_SHIFT %d\n", _PAGE_HUGE_SHIFT);
245 #endif
246 #ifdef _PAGE_NO_EXEC_SHIFT
247 	if (cpu_has_rixi)
248 		pr_define("_PAGE_NO_EXEC_SHIFT %d\n", _PAGE_NO_EXEC_SHIFT);
249 #endif
250 	pr_define("_PAGE_GLOBAL_SHIFT %d\n", _PAGE_GLOBAL_SHIFT);
251 	pr_define("_PAGE_VALID_SHIFT %d\n", _PAGE_VALID_SHIFT);
252 	pr_define("_PAGE_DIRTY_SHIFT %d\n", _PAGE_DIRTY_SHIFT);
253 	pr_define("_PFN_SHIFT %d\n", _PFN_SHIFT);
254 	pr_debug("\n");
255 }
256 
257 static inline void dump_handler(const char *symbol, const u32 *handler, int count)
258 {
259 	int i;
260 
261 	pr_debug("LEAF(%s)\n", symbol);
262 
263 	pr_debug("\t.set push\n");
264 	pr_debug("\t.set noreorder\n");
265 
266 	for (i = 0; i < count; i++)
267 		pr_debug("\t.word\t0x%08x\t\t# %p\n", handler[i], &handler[i]);
268 
269 	pr_debug("\t.set\tpop\n");
270 
271 	pr_debug("\tEND(%s)\n", symbol);
272 }
273 
274 /* The only general purpose registers allowed in TLB handlers. */
275 #define K0		26
276 #define K1		27
277 
278 /* Some CP0 registers */
279 #define C0_INDEX	0, 0
280 #define C0_ENTRYLO0	2, 0
281 #define C0_TCBIND	2, 2
282 #define C0_ENTRYLO1	3, 0
283 #define C0_CONTEXT	4, 0
284 #define C0_PAGEMASK	5, 0
285 #define C0_PWBASE	5, 5
286 #define C0_PWFIELD	5, 6
287 #define C0_PWSIZE	5, 7
288 #define C0_PWCTL	6, 6
289 #define C0_BADVADDR	8, 0
290 #define C0_PGD		9, 7
291 #define C0_ENTRYHI	10, 0
292 #define C0_EPC		14, 0
293 #define C0_XCONTEXT	20, 0
294 
295 #ifdef CONFIG_64BIT
296 # define GET_CONTEXT(buf, reg) UASM_i_MFC0(buf, reg, C0_XCONTEXT)
297 #else
298 # define GET_CONTEXT(buf, reg) UASM_i_MFC0(buf, reg, C0_CONTEXT)
299 #endif
300 
301 /* The worst case length of the handler is around 18 instructions for
302  * R3000-style TLBs and up to 63 instructions for R4000-style TLBs.
303  * Maximum space available is 32 instructions for R3000 and 64
304  * instructions for R4000.
305  *
306  * We deliberately chose a buffer size of 128, so we won't scribble
307  * over anything important on overflow before we panic.
308  */
309 static u32 tlb_handler[128];
310 
311 /* simply assume worst case size for labels and relocs */
312 static struct uasm_label labels[128];
313 static struct uasm_reloc relocs[128];
314 
315 static int check_for_high_segbits;
316 static bool fill_includes_sw_bits;
317 
318 static unsigned int kscratch_used_mask;
319 
320 static inline int __maybe_unused c0_kscratch(void)
321 {
322 	switch (current_cpu_type()) {
323 	case CPU_XLP:
324 	case CPU_XLR:
325 		return 22;
326 	default:
327 		return 31;
328 	}
329 }
330 
331 static int allocate_kscratch(void)
332 {
333 	int r;
334 	unsigned int a = cpu_data[0].kscratch_mask & ~kscratch_used_mask;
335 
336 	r = ffs(a);
337 
338 	if (r == 0)
339 		return -1;
340 
341 	r--; /* make it zero based */
342 
343 	kscratch_used_mask |= (1 << r);
344 
345 	return r;
346 }
347 
348 static int scratch_reg;
349 int pgd_reg;
350 EXPORT_SYMBOL_GPL(pgd_reg);
351 enum vmalloc64_mode {not_refill, refill_scratch, refill_noscratch};
352 
353 static struct work_registers build_get_work_registers(u32 **p)
354 {
355 	struct work_registers r;
356 
357 	if (scratch_reg >= 0) {
358 		/* Save in CPU local C0_KScratch? */
359 		UASM_i_MTC0(p, 1, c0_kscratch(), scratch_reg);
360 		r.r1 = K0;
361 		r.r2 = K1;
362 		r.r3 = 1;
363 		return r;
364 	}
365 
366 	if (num_possible_cpus() > 1) {
367 		/* Get smp_processor_id */
368 		UASM_i_CPUID_MFC0(p, K0, SMP_CPUID_REG);
369 		UASM_i_SRL_SAFE(p, K0, K0, SMP_CPUID_REGSHIFT);
370 
371 		/* handler_reg_save index in K0 */
372 		UASM_i_SLL(p, K0, K0, ilog2(sizeof(struct tlb_reg_save)));
373 
374 		UASM_i_LA(p, K1, (long)&handler_reg_save);
375 		UASM_i_ADDU(p, K0, K0, K1);
376 	} else {
377 		UASM_i_LA(p, K0, (long)&handler_reg_save);
378 	}
379 	/* K0 now points to save area, save $1 and $2  */
380 	UASM_i_SW(p, 1, offsetof(struct tlb_reg_save, a), K0);
381 	UASM_i_SW(p, 2, offsetof(struct tlb_reg_save, b), K0);
382 
383 	r.r1 = K1;
384 	r.r2 = 1;
385 	r.r3 = 2;
386 	return r;
387 }
388 
389 static void build_restore_work_registers(u32 **p)
390 {
391 	if (scratch_reg >= 0) {
392 		UASM_i_MFC0(p, 1, c0_kscratch(), scratch_reg);
393 		return;
394 	}
395 	/* K0 already points to save area, restore $1 and $2  */
396 	UASM_i_LW(p, 1, offsetof(struct tlb_reg_save, a), K0);
397 	UASM_i_LW(p, 2, offsetof(struct tlb_reg_save, b), K0);
398 }
399 
400 #ifndef CONFIG_MIPS_PGD_C0_CONTEXT
401 
402 /*
403  * CONFIG_MIPS_PGD_C0_CONTEXT implies 64 bit and lack of pgd_current,
404  * we cannot do r3000 under these circumstances.
405  *
406  * Declare pgd_current here instead of including mmu_context.h to avoid type
407  * conflicts for tlbmiss_handler_setup_pgd
408  */
409 extern unsigned long pgd_current[];
410 
411 /*
412  * The R3000 TLB handler is simple.
413  */
414 static void build_r3000_tlb_refill_handler(void)
415 {
416 	long pgdc = (long)pgd_current;
417 	u32 *p;
418 
419 	memset(tlb_handler, 0, sizeof(tlb_handler));
420 	p = tlb_handler;
421 
422 	uasm_i_mfc0(&p, K0, C0_BADVADDR);
423 	uasm_i_lui(&p, K1, uasm_rel_hi(pgdc)); /* cp0 delay */
424 	uasm_i_lw(&p, K1, uasm_rel_lo(pgdc), K1);
425 	uasm_i_srl(&p, K0, K0, 22); /* load delay */
426 	uasm_i_sll(&p, K0, K0, 2);
427 	uasm_i_addu(&p, K1, K1, K0);
428 	uasm_i_mfc0(&p, K0, C0_CONTEXT);
429 	uasm_i_lw(&p, K1, 0, K1); /* cp0 delay */
430 	uasm_i_andi(&p, K0, K0, 0xffc); /* load delay */
431 	uasm_i_addu(&p, K1, K1, K0);
432 	uasm_i_lw(&p, K0, 0, K1);
433 	uasm_i_nop(&p); /* load delay */
434 	uasm_i_mtc0(&p, K0, C0_ENTRYLO0);
435 	uasm_i_mfc0(&p, K1, C0_EPC); /* cp0 delay */
436 	uasm_i_tlbwr(&p); /* cp0 delay */
437 	uasm_i_jr(&p, K1);
438 	uasm_i_rfe(&p); /* branch delay */
439 
440 	if (p > tlb_handler + 32)
441 		panic("TLB refill handler space exceeded");
442 
443 	pr_debug("Wrote TLB refill handler (%u instructions).\n",
444 		 (unsigned int)(p - tlb_handler));
445 
446 	memcpy((void *)ebase, tlb_handler, 0x80);
447 	local_flush_icache_range(ebase, ebase + 0x80);
448 
449 	dump_handler("r3000_tlb_refill", (u32 *)ebase, 32);
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_R4300:
553 	case CPU_5KC:
554 	case CPU_TX49XX:
555 	case CPU_PR4450:
556 	case CPU_XLR:
557 		uasm_i_nop(p);
558 		tlbw(p);
559 		break;
560 
561 	case CPU_R10000:
562 	case CPU_R12000:
563 	case CPU_R14000:
564 	case CPU_R16000:
565 	case CPU_4KC:
566 	case CPU_4KEC:
567 	case CPU_M14KC:
568 	case CPU_M14KEC:
569 	case CPU_SB1:
570 	case CPU_SB1A:
571 	case CPU_4KSC:
572 	case CPU_20KC:
573 	case CPU_25KF:
574 	case CPU_BMIPS32:
575 	case CPU_BMIPS3300:
576 	case CPU_BMIPS4350:
577 	case CPU_BMIPS4380:
578 	case CPU_BMIPS5000:
579 	case CPU_LOONGSON2:
580 	case CPU_LOONGSON3:
581 	case CPU_R5500:
582 		if (m4kc_tlbp_war())
583 			uasm_i_nop(p);
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 	case CPU_R5432:
611 		uasm_i_nop(p);
612 		uasm_i_nop(p);
613 		tlbw(p);
614 		break;
615 
616 	case CPU_JZRISC:
617 		tlbw(p);
618 		uasm_i_nop(p);
619 		break;
620 
621 	default:
622 		panic("No TLB refill handler yet (CPU type: %d)",
623 		      current_cpu_type());
624 		break;
625 	}
626 }
627 EXPORT_SYMBOL_GPL(build_tlb_write_entry);
628 
629 static __maybe_unused void build_convert_pte_to_entrylo(u32 **p,
630 							unsigned int reg)
631 {
632 	if (_PAGE_GLOBAL_SHIFT == 0) {
633 		/* pte_t is already in EntryLo format */
634 		return;
635 	}
636 
637 	if (cpu_has_rixi && _PAGE_NO_EXEC) {
638 		if (fill_includes_sw_bits) {
639 			UASM_i_ROTR(p, reg, reg, ilog2(_PAGE_GLOBAL));
640 		} else {
641 			UASM_i_SRL(p, reg, reg, ilog2(_PAGE_NO_EXEC));
642 			UASM_i_ROTR(p, reg, reg,
643 				    ilog2(_PAGE_GLOBAL) - ilog2(_PAGE_NO_EXEC));
644 		}
645 	} else {
646 #ifdef CONFIG_PHYS_ADDR_T_64BIT
647 		uasm_i_dsrl_safe(p, reg, reg, ilog2(_PAGE_GLOBAL));
648 #else
649 		UASM_i_SRL(p, reg, reg, ilog2(_PAGE_GLOBAL));
650 #endif
651 	}
652 }
653 
654 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
655 
656 static void build_restore_pagemask(u32 **p, struct uasm_reloc **r,
657 				   unsigned int tmp, enum label_id lid,
658 				   int restore_scratch)
659 {
660 	if (restore_scratch) {
661 		/* Reset default page size */
662 		if (PM_DEFAULT_MASK >> 16) {
663 			uasm_i_lui(p, tmp, PM_DEFAULT_MASK >> 16);
664 			uasm_i_ori(p, tmp, tmp, PM_DEFAULT_MASK & 0xffff);
665 			uasm_i_mtc0(p, tmp, C0_PAGEMASK);
666 			uasm_il_b(p, r, lid);
667 		} else if (PM_DEFAULT_MASK) {
668 			uasm_i_ori(p, tmp, 0, PM_DEFAULT_MASK);
669 			uasm_i_mtc0(p, tmp, C0_PAGEMASK);
670 			uasm_il_b(p, r, lid);
671 		} else {
672 			uasm_i_mtc0(p, 0, C0_PAGEMASK);
673 			uasm_il_b(p, r, lid);
674 		}
675 		if (scratch_reg >= 0)
676 			UASM_i_MFC0(p, 1, c0_kscratch(), scratch_reg);
677 		else
678 			UASM_i_LW(p, 1, scratchpad_offset(0), 0);
679 	} else {
680 		/* Reset default page size */
681 		if (PM_DEFAULT_MASK >> 16) {
682 			uasm_i_lui(p, tmp, PM_DEFAULT_MASK >> 16);
683 			uasm_i_ori(p, tmp, tmp, PM_DEFAULT_MASK & 0xffff);
684 			uasm_il_b(p, r, lid);
685 			uasm_i_mtc0(p, tmp, C0_PAGEMASK);
686 		} else if (PM_DEFAULT_MASK) {
687 			uasm_i_ori(p, tmp, 0, PM_DEFAULT_MASK);
688 			uasm_il_b(p, r, lid);
689 			uasm_i_mtc0(p, tmp, C0_PAGEMASK);
690 		} else {
691 			uasm_il_b(p, r, lid);
692 			uasm_i_mtc0(p, 0, C0_PAGEMASK);
693 		}
694 	}
695 }
696 
697 static void build_huge_tlb_write_entry(u32 **p, struct uasm_label **l,
698 				       struct uasm_reloc **r,
699 				       unsigned int tmp,
700 				       enum tlb_write_entry wmode,
701 				       int restore_scratch)
702 {
703 	/* Set huge page tlb entry size */
704 	uasm_i_lui(p, tmp, PM_HUGE_MASK >> 16);
705 	uasm_i_ori(p, tmp, tmp, PM_HUGE_MASK & 0xffff);
706 	uasm_i_mtc0(p, tmp, C0_PAGEMASK);
707 
708 	build_tlb_write_entry(p, l, r, wmode);
709 
710 	build_restore_pagemask(p, r, tmp, label_leave, restore_scratch);
711 }
712 
713 /*
714  * Check if Huge PTE is present, if so then jump to LABEL.
715  */
716 static void
717 build_is_huge_pte(u32 **p, struct uasm_reloc **r, unsigned int tmp,
718 		  unsigned int pmd, int lid)
719 {
720 	UASM_i_LW(p, tmp, 0, pmd);
721 	if (use_bbit_insns()) {
722 		uasm_il_bbit1(p, r, tmp, ilog2(_PAGE_HUGE), lid);
723 	} else {
724 		uasm_i_andi(p, tmp, tmp, _PAGE_HUGE);
725 		uasm_il_bnez(p, r, tmp, lid);
726 	}
727 }
728 
729 static void build_huge_update_entries(u32 **p, unsigned int pte,
730 				      unsigned int tmp)
731 {
732 	int small_sequence;
733 
734 	/*
735 	 * A huge PTE describes an area the size of the
736 	 * configured huge page size. This is twice the
737 	 * of the large TLB entry size we intend to use.
738 	 * A TLB entry half the size of the configured
739 	 * huge page size is configured into entrylo0
740 	 * and entrylo1 to cover the contiguous huge PTE
741 	 * address space.
742 	 */
743 	small_sequence = (HPAGE_SIZE >> 7) < 0x10000;
744 
745 	/* We can clobber tmp.	It isn't used after this.*/
746 	if (!small_sequence)
747 		uasm_i_lui(p, tmp, HPAGE_SIZE >> (7 + 16));
748 
749 	build_convert_pte_to_entrylo(p, pte);
750 	UASM_i_MTC0(p, pte, C0_ENTRYLO0); /* load it */
751 	/* convert to entrylo1 */
752 	if (small_sequence)
753 		UASM_i_ADDIU(p, pte, pte, HPAGE_SIZE >> 7);
754 	else
755 		UASM_i_ADDU(p, pte, pte, tmp);
756 
757 	UASM_i_MTC0(p, pte, C0_ENTRYLO1); /* load it */
758 }
759 
760 static void build_huge_handler_tail(u32 **p, struct uasm_reloc **r,
761 				    struct uasm_label **l,
762 				    unsigned int pte,
763 				    unsigned int ptr,
764 				    unsigned int flush)
765 {
766 #ifdef CONFIG_SMP
767 	UASM_i_SC(p, pte, 0, ptr);
768 	uasm_il_beqz(p, r, pte, label_tlb_huge_update);
769 	UASM_i_LW(p, pte, 0, ptr); /* Needed because SC killed our PTE */
770 #else
771 	UASM_i_SW(p, pte, 0, ptr);
772 #endif
773 	if (cpu_has_ftlb && flush) {
774 		BUG_ON(!cpu_has_tlbinv);
775 
776 		UASM_i_MFC0(p, ptr, C0_ENTRYHI);
777 		uasm_i_ori(p, ptr, ptr, MIPS_ENTRYHI_EHINV);
778 		UASM_i_MTC0(p, ptr, C0_ENTRYHI);
779 		build_tlb_write_entry(p, l, r, tlb_indexed);
780 
781 		uasm_i_xori(p, ptr, ptr, MIPS_ENTRYHI_EHINV);
782 		UASM_i_MTC0(p, ptr, C0_ENTRYHI);
783 		build_huge_update_entries(p, pte, ptr);
784 		build_huge_tlb_write_entry(p, l, r, pte, tlb_random, 0);
785 
786 		return;
787 	}
788 
789 	build_huge_update_entries(p, pte, ptr);
790 	build_huge_tlb_write_entry(p, l, r, pte, tlb_indexed, 0);
791 }
792 #endif /* CONFIG_MIPS_HUGE_TLB_SUPPORT */
793 
794 #ifdef CONFIG_64BIT
795 /*
796  * TMP and PTR are scratch.
797  * TMP will be clobbered, PTR will hold the pmd entry.
798  */
799 void build_get_pmde64(u32 **p, struct uasm_label **l, struct uasm_reloc **r,
800 		      unsigned int tmp, unsigned int ptr)
801 {
802 #ifndef CONFIG_MIPS_PGD_C0_CONTEXT
803 	long pgdc = (long)pgd_current;
804 #endif
805 	/*
806 	 * The vmalloc handling is not in the hotpath.
807 	 */
808 	uasm_i_dmfc0(p, tmp, C0_BADVADDR);
809 
810 	if (check_for_high_segbits) {
811 		/*
812 		 * The kernel currently implicitely assumes that the
813 		 * MIPS SEGBITS parameter for the processor is
814 		 * (PGDIR_SHIFT+PGDIR_BITS) or less, and will never
815 		 * allocate virtual addresses outside the maximum
816 		 * range for SEGBITS = (PGDIR_SHIFT+PGDIR_BITS). But
817 		 * that doesn't prevent user code from accessing the
818 		 * higher xuseg addresses.  Here, we make sure that
819 		 * everything but the lower xuseg addresses goes down
820 		 * the module_alloc/vmalloc path.
821 		 */
822 		uasm_i_dsrl_safe(p, ptr, tmp, PGDIR_SHIFT + PGD_ORDER + PAGE_SHIFT - 3);
823 		uasm_il_bnez(p, r, ptr, label_vmalloc);
824 	} else {
825 		uasm_il_bltz(p, r, tmp, label_vmalloc);
826 	}
827 	/* No uasm_i_nop needed here, since the next insn doesn't touch TMP. */
828 
829 	if (pgd_reg != -1) {
830 		/* pgd is in pgd_reg */
831 		if (cpu_has_ldpte)
832 			UASM_i_MFC0(p, ptr, C0_PWBASE);
833 		else
834 			UASM_i_MFC0(p, ptr, c0_kscratch(), pgd_reg);
835 	} else {
836 #if defined(CONFIG_MIPS_PGD_C0_CONTEXT)
837 		/*
838 		 * &pgd << 11 stored in CONTEXT [23..63].
839 		 */
840 		UASM_i_MFC0(p, ptr, C0_CONTEXT);
841 
842 		/* Clear lower 23 bits of context. */
843 		uasm_i_dins(p, ptr, 0, 0, 23);
844 
845 		/* 1 0	1 0 1  << 6  xkphys cached */
846 		uasm_i_ori(p, ptr, ptr, 0x540);
847 		uasm_i_drotr(p, ptr, ptr, 11);
848 #elif defined(CONFIG_SMP)
849 		UASM_i_CPUID_MFC0(p, ptr, SMP_CPUID_REG);
850 		uasm_i_dsrl_safe(p, ptr, ptr, SMP_CPUID_PTRSHIFT);
851 		UASM_i_LA_mostly(p, tmp, pgdc);
852 		uasm_i_daddu(p, ptr, ptr, tmp);
853 		uasm_i_dmfc0(p, tmp, C0_BADVADDR);
854 		uasm_i_ld(p, ptr, uasm_rel_lo(pgdc), ptr);
855 #else
856 		UASM_i_LA_mostly(p, ptr, pgdc);
857 		uasm_i_ld(p, ptr, uasm_rel_lo(pgdc), ptr);
858 #endif
859 	}
860 
861 	uasm_l_vmalloc_done(l, *p);
862 
863 	/* get pgd offset in bytes */
864 	uasm_i_dsrl_safe(p, tmp, tmp, PGDIR_SHIFT - 3);
865 
866 	uasm_i_andi(p, tmp, tmp, (PTRS_PER_PGD - 1)<<3);
867 	uasm_i_daddu(p, ptr, ptr, tmp); /* add in pgd offset */
868 #ifndef __PAGETABLE_PUD_FOLDED
869 	uasm_i_dmfc0(p, tmp, C0_BADVADDR); /* get faulting address */
870 	uasm_i_ld(p, ptr, 0, ptr); /* get pud pointer */
871 	uasm_i_dsrl_safe(p, tmp, tmp, PUD_SHIFT - 3); /* get pud offset in bytes */
872 	uasm_i_andi(p, tmp, tmp, (PTRS_PER_PUD - 1) << 3);
873 	uasm_i_daddu(p, ptr, ptr, tmp); /* add in pud offset */
874 #endif
875 #ifndef __PAGETABLE_PMD_FOLDED
876 	uasm_i_dmfc0(p, tmp, C0_BADVADDR); /* get faulting address */
877 	uasm_i_ld(p, ptr, 0, ptr); /* get pmd pointer */
878 	uasm_i_dsrl_safe(p, tmp, tmp, PMD_SHIFT-3); /* get pmd offset in bytes */
879 	uasm_i_andi(p, tmp, tmp, (PTRS_PER_PMD - 1)<<3);
880 	uasm_i_daddu(p, ptr, ptr, tmp); /* add in pmd offset */
881 #endif
882 }
883 EXPORT_SYMBOL_GPL(build_get_pmde64);
884 
885 /*
886  * BVADDR is the faulting address, PTR is scratch.
887  * PTR will hold the pgd for vmalloc.
888  */
889 static void
890 build_get_pgd_vmalloc64(u32 **p, struct uasm_label **l, struct uasm_reloc **r,
891 			unsigned int bvaddr, unsigned int ptr,
892 			enum vmalloc64_mode mode)
893 {
894 	long swpd = (long)swapper_pg_dir;
895 	int single_insn_swpd;
896 	int did_vmalloc_branch = 0;
897 
898 	single_insn_swpd = uasm_in_compat_space_p(swpd) && !uasm_rel_lo(swpd);
899 
900 	uasm_l_vmalloc(l, *p);
901 
902 	if (mode != not_refill && check_for_high_segbits) {
903 		if (single_insn_swpd) {
904 			uasm_il_bltz(p, r, bvaddr, label_vmalloc_done);
905 			uasm_i_lui(p, ptr, uasm_rel_hi(swpd));
906 			did_vmalloc_branch = 1;
907 			/* fall through */
908 		} else {
909 			uasm_il_bgez(p, r, bvaddr, label_large_segbits_fault);
910 		}
911 	}
912 	if (!did_vmalloc_branch) {
913 		if (single_insn_swpd) {
914 			uasm_il_b(p, r, label_vmalloc_done);
915 			uasm_i_lui(p, ptr, uasm_rel_hi(swpd));
916 		} else {
917 			UASM_i_LA_mostly(p, ptr, swpd);
918 			uasm_il_b(p, r, label_vmalloc_done);
919 			if (uasm_in_compat_space_p(swpd))
920 				uasm_i_addiu(p, ptr, ptr, uasm_rel_lo(swpd));
921 			else
922 				uasm_i_daddiu(p, ptr, ptr, uasm_rel_lo(swpd));
923 		}
924 	}
925 	if (mode != not_refill && check_for_high_segbits) {
926 		uasm_l_large_segbits_fault(l, *p);
927 		/*
928 		 * We get here if we are an xsseg address, or if we are
929 		 * an xuseg address above (PGDIR_SHIFT+PGDIR_BITS) boundary.
930 		 *
931 		 * Ignoring xsseg (assume disabled so would generate
932 		 * (address errors?), the only remaining possibility
933 		 * is the upper xuseg addresses.  On processors with
934 		 * TLB_SEGBITS <= PGDIR_SHIFT+PGDIR_BITS, these
935 		 * addresses would have taken an address error. We try
936 		 * to mimic that here by taking a load/istream page
937 		 * fault.
938 		 */
939 		UASM_i_LA(p, ptr, (unsigned long)tlb_do_page_fault_0);
940 		uasm_i_jr(p, ptr);
941 
942 		if (mode == refill_scratch) {
943 			if (scratch_reg >= 0)
944 				UASM_i_MFC0(p, 1, c0_kscratch(), scratch_reg);
945 			else
946 				UASM_i_LW(p, 1, scratchpad_offset(0), 0);
947 		} else {
948 			uasm_i_nop(p);
949 		}
950 	}
951 }
952 
953 #else /* !CONFIG_64BIT */
954 
955 /*
956  * TMP and PTR are scratch.
957  * TMP will be clobbered, PTR will hold the pgd entry.
958  */
959 void build_get_pgde32(u32 **p, unsigned int tmp, unsigned int ptr)
960 {
961 	if (pgd_reg != -1) {
962 		/* pgd is in pgd_reg */
963 		uasm_i_mfc0(p, ptr, c0_kscratch(), pgd_reg);
964 		uasm_i_mfc0(p, tmp, C0_BADVADDR); /* get faulting address */
965 	} else {
966 		long pgdc = (long)pgd_current;
967 
968 		/* 32 bit SMP has smp_processor_id() stored in CONTEXT. */
969 #ifdef CONFIG_SMP
970 		uasm_i_mfc0(p, ptr, SMP_CPUID_REG);
971 		UASM_i_LA_mostly(p, tmp, pgdc);
972 		uasm_i_srl(p, ptr, ptr, SMP_CPUID_PTRSHIFT);
973 		uasm_i_addu(p, ptr, tmp, ptr);
974 #else
975 		UASM_i_LA_mostly(p, ptr, pgdc);
976 #endif
977 		uasm_i_mfc0(p, tmp, C0_BADVADDR); /* get faulting address */
978 		uasm_i_lw(p, ptr, uasm_rel_lo(pgdc), ptr);
979 	}
980 	uasm_i_srl(p, tmp, tmp, PGDIR_SHIFT); /* get pgd only bits */
981 	uasm_i_sll(p, tmp, tmp, PGD_T_LOG2);
982 	uasm_i_addu(p, ptr, ptr, tmp); /* add in pgd offset */
983 }
984 EXPORT_SYMBOL_GPL(build_get_pgde32);
985 
986 #endif /* !CONFIG_64BIT */
987 
988 static void build_adjust_context(u32 **p, unsigned int ctx)
989 {
990 	unsigned int shift = 4 - (PTE_T_LOG2 + 1) + PAGE_SHIFT - 12;
991 	unsigned int mask = (PTRS_PER_PTE / 2 - 1) << (PTE_T_LOG2 + 1);
992 
993 	switch (current_cpu_type()) {
994 	case CPU_VR41XX:
995 	case CPU_VR4111:
996 	case CPU_VR4121:
997 	case CPU_VR4122:
998 	case CPU_VR4131:
999 	case CPU_VR4181:
1000 	case CPU_VR4181A:
1001 	case CPU_VR4133:
1002 		shift += 2;
1003 		break;
1004 
1005 	default:
1006 		break;
1007 	}
1008 
1009 	if (shift)
1010 		UASM_i_SRL(p, ctx, ctx, shift);
1011 	uasm_i_andi(p, ctx, ctx, mask);
1012 }
1013 
1014 void build_get_ptep(u32 **p, unsigned int tmp, unsigned int ptr)
1015 {
1016 	/*
1017 	 * Bug workaround for the Nevada. It seems as if under certain
1018 	 * circumstances the move from cp0_context might produce a
1019 	 * bogus result when the mfc0 instruction and its consumer are
1020 	 * in a different cacheline or a load instruction, probably any
1021 	 * memory reference, is between them.
1022 	 */
1023 	switch (current_cpu_type()) {
1024 	case CPU_NEVADA:
1025 		UASM_i_LW(p, ptr, 0, ptr);
1026 		GET_CONTEXT(p, tmp); /* get context reg */
1027 		break;
1028 
1029 	default:
1030 		GET_CONTEXT(p, tmp); /* get context reg */
1031 		UASM_i_LW(p, ptr, 0, ptr);
1032 		break;
1033 	}
1034 
1035 	build_adjust_context(p, tmp);
1036 	UASM_i_ADDU(p, ptr, ptr, tmp); /* add in offset */
1037 }
1038 EXPORT_SYMBOL_GPL(build_get_ptep);
1039 
1040 void build_update_entries(u32 **p, unsigned int tmp, unsigned int ptep)
1041 {
1042 	int pte_off_even = 0;
1043 	int pte_off_odd = sizeof(pte_t);
1044 
1045 #if defined(CONFIG_CPU_MIPS32) && defined(CONFIG_PHYS_ADDR_T_64BIT)
1046 	/* The low 32 bits of EntryLo is stored in pte_high */
1047 	pte_off_even += offsetof(pte_t, pte_high);
1048 	pte_off_odd += offsetof(pte_t, pte_high);
1049 #endif
1050 
1051 	if (IS_ENABLED(CONFIG_XPA)) {
1052 		uasm_i_lw(p, tmp, pte_off_even, ptep); /* even pte */
1053 		UASM_i_ROTR(p, tmp, tmp, ilog2(_PAGE_GLOBAL));
1054 		UASM_i_MTC0(p, tmp, C0_ENTRYLO0);
1055 
1056 		if (cpu_has_xpa && !mips_xpa_disabled) {
1057 			uasm_i_lw(p, tmp, 0, ptep);
1058 			uasm_i_ext(p, tmp, tmp, 0, 24);
1059 			uasm_i_mthc0(p, tmp, C0_ENTRYLO0);
1060 		}
1061 
1062 		uasm_i_lw(p, tmp, pte_off_odd, ptep); /* odd pte */
1063 		UASM_i_ROTR(p, tmp, tmp, ilog2(_PAGE_GLOBAL));
1064 		UASM_i_MTC0(p, tmp, C0_ENTRYLO1);
1065 
1066 		if (cpu_has_xpa && !mips_xpa_disabled) {
1067 			uasm_i_lw(p, tmp, sizeof(pte_t), ptep);
1068 			uasm_i_ext(p, tmp, tmp, 0, 24);
1069 			uasm_i_mthc0(p, tmp, C0_ENTRYLO1);
1070 		}
1071 		return;
1072 	}
1073 
1074 	UASM_i_LW(p, tmp, pte_off_even, ptep); /* get even pte */
1075 	UASM_i_LW(p, ptep, pte_off_odd, ptep); /* get odd pte */
1076 	if (r45k_bvahwbug())
1077 		build_tlb_probe_entry(p);
1078 	build_convert_pte_to_entrylo(p, tmp);
1079 	if (r4k_250MHZhwbug())
1080 		UASM_i_MTC0(p, 0, C0_ENTRYLO0);
1081 	UASM_i_MTC0(p, tmp, C0_ENTRYLO0); /* load it */
1082 	build_convert_pte_to_entrylo(p, ptep);
1083 	if (r45k_bvahwbug())
1084 		uasm_i_mfc0(p, tmp, C0_INDEX);
1085 	if (r4k_250MHZhwbug())
1086 		UASM_i_MTC0(p, 0, C0_ENTRYLO1);
1087 	UASM_i_MTC0(p, ptep, C0_ENTRYLO1); /* load it */
1088 }
1089 EXPORT_SYMBOL_GPL(build_update_entries);
1090 
1091 struct mips_huge_tlb_info {
1092 	int huge_pte;
1093 	int restore_scratch;
1094 	bool need_reload_pte;
1095 };
1096 
1097 static struct mips_huge_tlb_info
1098 build_fast_tlb_refill_handler (u32 **p, struct uasm_label **l,
1099 			       struct uasm_reloc **r, unsigned int tmp,
1100 			       unsigned int ptr, int c0_scratch_reg)
1101 {
1102 	struct mips_huge_tlb_info rv;
1103 	unsigned int even, odd;
1104 	int vmalloc_branch_delay_filled = 0;
1105 	const int scratch = 1; /* Our extra working register */
1106 
1107 	rv.huge_pte = scratch;
1108 	rv.restore_scratch = 0;
1109 	rv.need_reload_pte = false;
1110 
1111 	if (check_for_high_segbits) {
1112 		UASM_i_MFC0(p, tmp, C0_BADVADDR);
1113 
1114 		if (pgd_reg != -1)
1115 			UASM_i_MFC0(p, ptr, c0_kscratch(), pgd_reg);
1116 		else
1117 			UASM_i_MFC0(p, ptr, C0_CONTEXT);
1118 
1119 		if (c0_scratch_reg >= 0)
1120 			UASM_i_MTC0(p, scratch, c0_kscratch(), c0_scratch_reg);
1121 		else
1122 			UASM_i_SW(p, scratch, scratchpad_offset(0), 0);
1123 
1124 		uasm_i_dsrl_safe(p, scratch, tmp,
1125 				 PGDIR_SHIFT + PGD_ORDER + PAGE_SHIFT - 3);
1126 		uasm_il_bnez(p, r, scratch, label_vmalloc);
1127 
1128 		if (pgd_reg == -1) {
1129 			vmalloc_branch_delay_filled = 1;
1130 			/* Clear lower 23 bits of context. */
1131 			uasm_i_dins(p, ptr, 0, 0, 23);
1132 		}
1133 	} else {
1134 		if (pgd_reg != -1)
1135 			UASM_i_MFC0(p, ptr, c0_kscratch(), pgd_reg);
1136 		else
1137 			UASM_i_MFC0(p, ptr, C0_CONTEXT);
1138 
1139 		UASM_i_MFC0(p, tmp, C0_BADVADDR);
1140 
1141 		if (c0_scratch_reg >= 0)
1142 			UASM_i_MTC0(p, scratch, c0_kscratch(), c0_scratch_reg);
1143 		else
1144 			UASM_i_SW(p, scratch, scratchpad_offset(0), 0);
1145 
1146 		if (pgd_reg == -1)
1147 			/* Clear lower 23 bits of context. */
1148 			uasm_i_dins(p, ptr, 0, 0, 23);
1149 
1150 		uasm_il_bltz(p, r, tmp, label_vmalloc);
1151 	}
1152 
1153 	if (pgd_reg == -1) {
1154 		vmalloc_branch_delay_filled = 1;
1155 		/* 1 0	1 0 1  << 6  xkphys cached */
1156 		uasm_i_ori(p, ptr, ptr, 0x540);
1157 		uasm_i_drotr(p, ptr, ptr, 11);
1158 	}
1159 
1160 #ifdef __PAGETABLE_PMD_FOLDED
1161 #define LOC_PTEP scratch
1162 #else
1163 #define LOC_PTEP ptr
1164 #endif
1165 
1166 	if (!vmalloc_branch_delay_filled)
1167 		/* get pgd offset in bytes */
1168 		uasm_i_dsrl_safe(p, scratch, tmp, PGDIR_SHIFT - 3);
1169 
1170 	uasm_l_vmalloc_done(l, *p);
1171 
1172 	/*
1173 	 *			   tmp		ptr
1174 	 * fall-through case =	 badvaddr  *pgd_current
1175 	 * vmalloc case	     =	 badvaddr  swapper_pg_dir
1176 	 */
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 #ifdef __PAGETABLE_PMD_FOLDED
1183 	GET_CONTEXT(p, tmp); /* get context reg */
1184 #endif
1185 	uasm_i_andi(p, scratch, scratch, (PTRS_PER_PGD - 1) << 3);
1186 
1187 	if (use_lwx_insns()) {
1188 		UASM_i_LWX(p, LOC_PTEP, scratch, ptr);
1189 	} else {
1190 		uasm_i_daddu(p, ptr, ptr, scratch); /* add in pgd offset */
1191 		uasm_i_ld(p, LOC_PTEP, 0, ptr); /* get pmd pointer */
1192 	}
1193 
1194 #ifndef __PAGETABLE_PUD_FOLDED
1195 	/* get pud offset in bytes */
1196 	uasm_i_dsrl_safe(p, scratch, tmp, PUD_SHIFT - 3);
1197 	uasm_i_andi(p, scratch, scratch, (PTRS_PER_PUD - 1) << 3);
1198 
1199 	if (use_lwx_insns()) {
1200 		UASM_i_LWX(p, ptr, scratch, ptr);
1201 	} else {
1202 		uasm_i_daddu(p, ptr, ptr, scratch); /* add in pmd offset */
1203 		UASM_i_LW(p, ptr, 0, ptr);
1204 	}
1205 	/* ptr contains a pointer to PMD entry */
1206 	/* tmp contains the address */
1207 #endif
1208 
1209 #ifndef __PAGETABLE_PMD_FOLDED
1210 	/* get pmd offset in bytes */
1211 	uasm_i_dsrl_safe(p, scratch, tmp, PMD_SHIFT - 3);
1212 	uasm_i_andi(p, scratch, scratch, (PTRS_PER_PMD - 1) << 3);
1213 	GET_CONTEXT(p, tmp); /* get context reg */
1214 
1215 	if (use_lwx_insns()) {
1216 		UASM_i_LWX(p, scratch, scratch, ptr);
1217 	} else {
1218 		uasm_i_daddu(p, ptr, ptr, scratch); /* add in pmd offset */
1219 		UASM_i_LW(p, scratch, 0, ptr);
1220 	}
1221 #endif
1222 	/* Adjust the context during the load latency. */
1223 	build_adjust_context(p, tmp);
1224 
1225 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
1226 	uasm_il_bbit1(p, r, scratch, ilog2(_PAGE_HUGE), label_tlb_huge_update);
1227 	/*
1228 	 * The in the LWX case we don't want to do the load in the
1229 	 * delay slot.	It cannot issue in the same cycle and may be
1230 	 * speculative and unneeded.
1231 	 */
1232 	if (use_lwx_insns())
1233 		uasm_i_nop(p);
1234 #endif /* CONFIG_MIPS_HUGE_TLB_SUPPORT */
1235 
1236 
1237 	/* build_update_entries */
1238 	if (use_lwx_insns()) {
1239 		even = ptr;
1240 		odd = tmp;
1241 		UASM_i_LWX(p, even, scratch, tmp);
1242 		UASM_i_ADDIU(p, tmp, tmp, sizeof(pte_t));
1243 		UASM_i_LWX(p, odd, scratch, tmp);
1244 	} else {
1245 		UASM_i_ADDU(p, ptr, scratch, tmp); /* add in offset */
1246 		even = tmp;
1247 		odd = ptr;
1248 		UASM_i_LW(p, even, 0, ptr); /* get even pte */
1249 		UASM_i_LW(p, odd, sizeof(pte_t), ptr); /* get odd pte */
1250 	}
1251 	if (cpu_has_rixi) {
1252 		uasm_i_drotr(p, even, even, ilog2(_PAGE_GLOBAL));
1253 		UASM_i_MTC0(p, even, C0_ENTRYLO0); /* load it */
1254 		uasm_i_drotr(p, odd, odd, ilog2(_PAGE_GLOBAL));
1255 	} else {
1256 		uasm_i_dsrl_safe(p, even, even, ilog2(_PAGE_GLOBAL));
1257 		UASM_i_MTC0(p, even, C0_ENTRYLO0); /* load it */
1258 		uasm_i_dsrl_safe(p, odd, odd, ilog2(_PAGE_GLOBAL));
1259 	}
1260 	UASM_i_MTC0(p, odd, C0_ENTRYLO1); /* load it */
1261 
1262 	if (c0_scratch_reg >= 0) {
1263 		UASM_i_MFC0(p, scratch, c0_kscratch(), c0_scratch_reg);
1264 		build_tlb_write_entry(p, l, r, tlb_random);
1265 		uasm_l_leave(l, *p);
1266 		rv.restore_scratch = 1;
1267 	} else if (PAGE_SHIFT == 14 || PAGE_SHIFT == 13)  {
1268 		build_tlb_write_entry(p, l, r, tlb_random);
1269 		uasm_l_leave(l, *p);
1270 		UASM_i_LW(p, scratch, scratchpad_offset(0), 0);
1271 	} else {
1272 		UASM_i_LW(p, scratch, scratchpad_offset(0), 0);
1273 		build_tlb_write_entry(p, l, r, tlb_random);
1274 		uasm_l_leave(l, *p);
1275 		rv.restore_scratch = 1;
1276 	}
1277 
1278 	uasm_i_eret(p); /* return from trap */
1279 
1280 	return rv;
1281 }
1282 
1283 /*
1284  * For a 64-bit kernel, we are using the 64-bit XTLB refill exception
1285  * because EXL == 0.  If we wrap, we can also use the 32 instruction
1286  * slots before the XTLB refill exception handler which belong to the
1287  * unused TLB refill exception.
1288  */
1289 #define MIPS64_REFILL_INSNS 32
1290 
1291 static void build_r4000_tlb_refill_handler(void)
1292 {
1293 	u32 *p = tlb_handler;
1294 	struct uasm_label *l = labels;
1295 	struct uasm_reloc *r = relocs;
1296 	u32 *f;
1297 	unsigned int final_len;
1298 	struct mips_huge_tlb_info htlb_info __maybe_unused;
1299 	enum vmalloc64_mode vmalloc_mode __maybe_unused;
1300 
1301 	memset(tlb_handler, 0, sizeof(tlb_handler));
1302 	memset(labels, 0, sizeof(labels));
1303 	memset(relocs, 0, sizeof(relocs));
1304 	memset(final_handler, 0, sizeof(final_handler));
1305 
1306 	if (IS_ENABLED(CONFIG_64BIT) && (scratch_reg >= 0 || scratchpad_available()) && use_bbit_insns()) {
1307 		htlb_info = build_fast_tlb_refill_handler(&p, &l, &r, K0, K1,
1308 							  scratch_reg);
1309 		vmalloc_mode = refill_scratch;
1310 	} else {
1311 		htlb_info.huge_pte = K0;
1312 		htlb_info.restore_scratch = 0;
1313 		htlb_info.need_reload_pte = true;
1314 		vmalloc_mode = refill_noscratch;
1315 		/*
1316 		 * create the plain linear handler
1317 		 */
1318 		if (bcm1250_m3_war()) {
1319 			unsigned int segbits = 44;
1320 
1321 			uasm_i_dmfc0(&p, K0, C0_BADVADDR);
1322 			uasm_i_dmfc0(&p, K1, C0_ENTRYHI);
1323 			uasm_i_xor(&p, K0, K0, K1);
1324 			uasm_i_dsrl_safe(&p, K1, K0, 62);
1325 			uasm_i_dsrl_safe(&p, K0, K0, 12 + 1);
1326 			uasm_i_dsll_safe(&p, K0, K0, 64 + 12 + 1 - segbits);
1327 			uasm_i_or(&p, K0, K0, K1);
1328 			uasm_il_bnez(&p, &r, K0, label_leave);
1329 			/* No need for uasm_i_nop */
1330 		}
1331 
1332 #ifdef CONFIG_64BIT
1333 		build_get_pmde64(&p, &l, &r, K0, K1); /* get pmd in K1 */
1334 #else
1335 		build_get_pgde32(&p, K0, K1); /* get pgd in K1 */
1336 #endif
1337 
1338 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
1339 		build_is_huge_pte(&p, &r, K0, K1, label_tlb_huge_update);
1340 #endif
1341 
1342 		build_get_ptep(&p, K0, K1);
1343 		build_update_entries(&p, K0, K1);
1344 		build_tlb_write_entry(&p, &l, &r, tlb_random);
1345 		uasm_l_leave(&l, p);
1346 		uasm_i_eret(&p); /* return from trap */
1347 	}
1348 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
1349 	uasm_l_tlb_huge_update(&l, p);
1350 	if (htlb_info.need_reload_pte)
1351 		UASM_i_LW(&p, htlb_info.huge_pte, 0, K1);
1352 	build_huge_update_entries(&p, htlb_info.huge_pte, K1);
1353 	build_huge_tlb_write_entry(&p, &l, &r, K0, tlb_random,
1354 				   htlb_info.restore_scratch);
1355 #endif
1356 
1357 #ifdef CONFIG_64BIT
1358 	build_get_pgd_vmalloc64(&p, &l, &r, K0, K1, vmalloc_mode);
1359 #endif
1360 
1361 	/*
1362 	 * Overflow check: For the 64bit handler, we need at least one
1363 	 * free instruction slot for the wrap-around branch. In worst
1364 	 * case, if the intended insertion point is a delay slot, we
1365 	 * need three, with the second nop'ed and the third being
1366 	 * unused.
1367 	 */
1368 	switch (boot_cpu_type()) {
1369 	default:
1370 		if (sizeof(long) == 4) {
1371 	case CPU_LOONGSON2:
1372 		/* Loongson2 ebase is different than r4k, we have more space */
1373 			if ((p - tlb_handler) > 64)
1374 				panic("TLB refill handler space exceeded");
1375 			/*
1376 			 * Now fold the handler in the TLB refill handler space.
1377 			 */
1378 			f = final_handler;
1379 			/* Simplest case, just copy the handler. */
1380 			uasm_copy_handler(relocs, labels, tlb_handler, p, f);
1381 			final_len = p - tlb_handler;
1382 			break;
1383 		} else {
1384 			if (((p - tlb_handler) > (MIPS64_REFILL_INSNS * 2) - 1)
1385 			    || (((p - tlb_handler) > (MIPS64_REFILL_INSNS * 2) - 3)
1386 				&& uasm_insn_has_bdelay(relocs,
1387 							tlb_handler + MIPS64_REFILL_INSNS - 3)))
1388 				panic("TLB refill handler space exceeded");
1389 			/*
1390 			 * Now fold the handler in the TLB refill handler space.
1391 			 */
1392 			f = final_handler + MIPS64_REFILL_INSNS;
1393 			if ((p - tlb_handler) <= MIPS64_REFILL_INSNS) {
1394 				/* Just copy the handler. */
1395 				uasm_copy_handler(relocs, labels, tlb_handler, p, f);
1396 				final_len = p - tlb_handler;
1397 			} else {
1398 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
1399 				const enum label_id ls = label_tlb_huge_update;
1400 #else
1401 				const enum label_id ls = label_vmalloc;
1402 #endif
1403 				u32 *split;
1404 				int ov = 0;
1405 				int i;
1406 
1407 				for (i = 0; i < ARRAY_SIZE(labels) && labels[i].lab != ls; i++)
1408 					;
1409 				BUG_ON(i == ARRAY_SIZE(labels));
1410 				split = labels[i].addr;
1411 
1412 				/*
1413 				 * See if we have overflown one way or the other.
1414 				 */
1415 				if (split > tlb_handler + MIPS64_REFILL_INSNS ||
1416 				    split < p - MIPS64_REFILL_INSNS)
1417 					ov = 1;
1418 
1419 				if (ov) {
1420 					/*
1421 					 * Split two instructions before the end.  One
1422 					 * for the branch and one for the instruction
1423 					 * in the delay slot.
1424 					 */
1425 					split = tlb_handler + MIPS64_REFILL_INSNS - 2;
1426 
1427 					/*
1428 					 * If the branch would fall in a delay slot,
1429 					 * we must back up an additional instruction
1430 					 * so that it is no longer in a delay slot.
1431 					 */
1432 					if (uasm_insn_has_bdelay(relocs, split - 1))
1433 						split--;
1434 				}
1435 				/* Copy first part of the handler. */
1436 				uasm_copy_handler(relocs, labels, tlb_handler, split, f);
1437 				f += split - tlb_handler;
1438 
1439 				if (ov) {
1440 					/* Insert branch. */
1441 					uasm_l_split(&l, final_handler);
1442 					uasm_il_b(&f, &r, label_split);
1443 					if (uasm_insn_has_bdelay(relocs, split))
1444 						uasm_i_nop(&f);
1445 					else {
1446 						uasm_copy_handler(relocs, labels,
1447 								  split, split + 1, f);
1448 						uasm_move_labels(labels, f, f + 1, -1);
1449 						f++;
1450 						split++;
1451 					}
1452 				}
1453 
1454 				/* Copy the rest of the handler. */
1455 				uasm_copy_handler(relocs, labels, split, p, final_handler);
1456 				final_len = (f - (final_handler + MIPS64_REFILL_INSNS)) +
1457 					    (p - split);
1458 			}
1459 		}
1460 		break;
1461 	}
1462 
1463 	uasm_resolve_relocs(relocs, labels);
1464 	pr_debug("Wrote TLB refill handler (%u instructions).\n",
1465 		 final_len);
1466 
1467 	memcpy((void *)ebase, final_handler, 0x100);
1468 	local_flush_icache_range(ebase, ebase + 0x100);
1469 
1470 	dump_handler("r4000_tlb_refill", (u32 *)ebase, 64);
1471 }
1472 
1473 static void setup_pw(void)
1474 {
1475 	unsigned long pgd_i, pgd_w;
1476 #ifndef __PAGETABLE_PMD_FOLDED
1477 	unsigned long pmd_i, pmd_w;
1478 #endif
1479 	unsigned long pt_i, pt_w;
1480 	unsigned long pte_i, pte_w;
1481 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
1482 	unsigned long psn;
1483 
1484 	psn = ilog2(_PAGE_HUGE);     /* bit used to indicate huge page */
1485 #endif
1486 	pgd_i = PGDIR_SHIFT;  /* 1st level PGD */
1487 #ifndef __PAGETABLE_PMD_FOLDED
1488 	pgd_w = PGDIR_SHIFT - PMD_SHIFT + PGD_ORDER;
1489 
1490 	pmd_i = PMD_SHIFT;    /* 2nd level PMD */
1491 	pmd_w = PMD_SHIFT - PAGE_SHIFT;
1492 #else
1493 	pgd_w = PGDIR_SHIFT - PAGE_SHIFT + PGD_ORDER;
1494 #endif
1495 
1496 	pt_i  = PAGE_SHIFT;    /* 3rd level PTE */
1497 	pt_w  = PAGE_SHIFT - 3;
1498 
1499 	pte_i = ilog2(_PAGE_GLOBAL);
1500 	pte_w = 0;
1501 
1502 #ifndef __PAGETABLE_PMD_FOLDED
1503 	write_c0_pwfield(pgd_i << 24 | pmd_i << 12 | pt_i << 6 | pte_i);
1504 	write_c0_pwsize(1 << 30 | pgd_w << 24 | pmd_w << 12 | pt_w << 6 | pte_w);
1505 #else
1506 	write_c0_pwfield(pgd_i << 24 | pt_i << 6 | pte_i);
1507 	write_c0_pwsize(1 << 30 | pgd_w << 24 | pt_w << 6 | pte_w);
1508 #endif
1509 
1510 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
1511 	write_c0_pwctl(1 << 6 | psn);
1512 #endif
1513 	write_c0_kpgd(swapper_pg_dir);
1514 	kscratch_used_mask |= (1 << 7); /* KScratch6 is used for KPGD */
1515 }
1516 
1517 static void build_loongson3_tlb_refill_handler(void)
1518 {
1519 	u32 *p = tlb_handler;
1520 	struct uasm_label *l = labels;
1521 	struct uasm_reloc *r = relocs;
1522 
1523 	memset(labels, 0, sizeof(labels));
1524 	memset(relocs, 0, sizeof(relocs));
1525 	memset(tlb_handler, 0, sizeof(tlb_handler));
1526 
1527 	if (check_for_high_segbits) {
1528 		uasm_i_dmfc0(&p, K0, C0_BADVADDR);
1529 		uasm_i_dsrl_safe(&p, K1, K0, PGDIR_SHIFT + PGD_ORDER + PAGE_SHIFT - 3);
1530 		uasm_il_beqz(&p, &r, K1, label_vmalloc);
1531 		uasm_i_nop(&p);
1532 
1533 		uasm_il_bgez(&p, &r, K0, label_large_segbits_fault);
1534 		uasm_i_nop(&p);
1535 		uasm_l_vmalloc(&l, p);
1536 	}
1537 
1538 	uasm_i_dmfc0(&p, K1, C0_PGD);
1539 
1540 	uasm_i_lddir(&p, K0, K1, 3);  /* global page dir */
1541 #ifndef __PAGETABLE_PMD_FOLDED
1542 	uasm_i_lddir(&p, K1, K0, 1);  /* middle page dir */
1543 #endif
1544 	uasm_i_ldpte(&p, K1, 0);      /* even */
1545 	uasm_i_ldpte(&p, K1, 1);      /* odd */
1546 	uasm_i_tlbwr(&p);
1547 
1548 	/* restore page mask */
1549 	if (PM_DEFAULT_MASK >> 16) {
1550 		uasm_i_lui(&p, K0, PM_DEFAULT_MASK >> 16);
1551 		uasm_i_ori(&p, K0, K0, PM_DEFAULT_MASK & 0xffff);
1552 		uasm_i_mtc0(&p, K0, C0_PAGEMASK);
1553 	} else if (PM_DEFAULT_MASK) {
1554 		uasm_i_ori(&p, K0, 0, PM_DEFAULT_MASK);
1555 		uasm_i_mtc0(&p, K0, C0_PAGEMASK);
1556 	} else {
1557 		uasm_i_mtc0(&p, 0, C0_PAGEMASK);
1558 	}
1559 
1560 	uasm_i_eret(&p);
1561 
1562 	if (check_for_high_segbits) {
1563 		uasm_l_large_segbits_fault(&l, p);
1564 		UASM_i_LA(&p, K1, (unsigned long)tlb_do_page_fault_0);
1565 		uasm_i_jr(&p, K1);
1566 		uasm_i_nop(&p);
1567 	}
1568 
1569 	uasm_resolve_relocs(relocs, labels);
1570 	memcpy((void *)(ebase + 0x80), tlb_handler, 0x80);
1571 	local_flush_icache_range(ebase + 0x80, ebase + 0x100);
1572 	dump_handler("loongson3_tlb_refill", (u32 *)(ebase + 0x80), 32);
1573 }
1574 
1575 extern u32 handle_tlbl[], handle_tlbl_end[];
1576 extern u32 handle_tlbs[], handle_tlbs_end[];
1577 extern u32 handle_tlbm[], handle_tlbm_end[];
1578 extern u32 tlbmiss_handler_setup_pgd_start[];
1579 extern u32 tlbmiss_handler_setup_pgd[];
1580 EXPORT_SYMBOL_GPL(tlbmiss_handler_setup_pgd);
1581 extern u32 tlbmiss_handler_setup_pgd_end[];
1582 
1583 static void build_setup_pgd(void)
1584 {
1585 	const int a0 = 4;
1586 	const int __maybe_unused a1 = 5;
1587 	const int __maybe_unused a2 = 6;
1588 	u32 *p = tlbmiss_handler_setup_pgd_start;
1589 	const int tlbmiss_handler_setup_pgd_size =
1590 		tlbmiss_handler_setup_pgd_end - tlbmiss_handler_setup_pgd_start;
1591 #ifndef CONFIG_MIPS_PGD_C0_CONTEXT
1592 	long pgdc = (long)pgd_current;
1593 #endif
1594 
1595 	memset(tlbmiss_handler_setup_pgd, 0, tlbmiss_handler_setup_pgd_size *
1596 					sizeof(tlbmiss_handler_setup_pgd[0]));
1597 	memset(labels, 0, sizeof(labels));
1598 	memset(relocs, 0, sizeof(relocs));
1599 	pgd_reg = allocate_kscratch();
1600 #ifdef CONFIG_MIPS_PGD_C0_CONTEXT
1601 	if (pgd_reg == -1) {
1602 		struct uasm_label *l = labels;
1603 		struct uasm_reloc *r = relocs;
1604 
1605 		/* PGD << 11 in c0_Context */
1606 		/*
1607 		 * If it is a ckseg0 address, convert to a physical
1608 		 * address.  Shifting right by 29 and adding 4 will
1609 		 * result in zero for these addresses.
1610 		 *
1611 		 */
1612 		UASM_i_SRA(&p, a1, a0, 29);
1613 		UASM_i_ADDIU(&p, a1, a1, 4);
1614 		uasm_il_bnez(&p, &r, a1, label_tlbl_goaround1);
1615 		uasm_i_nop(&p);
1616 		uasm_i_dinsm(&p, a0, 0, 29, 64 - 29);
1617 		uasm_l_tlbl_goaround1(&l, p);
1618 		UASM_i_SLL(&p, a0, a0, 11);
1619 		uasm_i_jr(&p, 31);
1620 		UASM_i_MTC0(&p, a0, C0_CONTEXT);
1621 	} else {
1622 		/* PGD in c0_KScratch */
1623 		uasm_i_jr(&p, 31);
1624 		if (cpu_has_ldpte)
1625 			UASM_i_MTC0(&p, a0, C0_PWBASE);
1626 		else
1627 			UASM_i_MTC0(&p, a0, c0_kscratch(), pgd_reg);
1628 	}
1629 #else
1630 #ifdef CONFIG_SMP
1631 	/* Save PGD to pgd_current[smp_processor_id()] */
1632 	UASM_i_CPUID_MFC0(&p, a1, SMP_CPUID_REG);
1633 	UASM_i_SRL_SAFE(&p, a1, a1, SMP_CPUID_PTRSHIFT);
1634 	UASM_i_LA_mostly(&p, a2, pgdc);
1635 	UASM_i_ADDU(&p, a2, a2, a1);
1636 	UASM_i_SW(&p, a0, uasm_rel_lo(pgdc), a2);
1637 #else
1638 	UASM_i_LA_mostly(&p, a2, pgdc);
1639 	UASM_i_SW(&p, a0, uasm_rel_lo(pgdc), a2);
1640 #endif /* SMP */
1641 	uasm_i_jr(&p, 31);
1642 
1643 	/* if pgd_reg is allocated, save PGD also to scratch register */
1644 	if (pgd_reg != -1)
1645 		UASM_i_MTC0(&p, a0, c0_kscratch(), pgd_reg);
1646 	else
1647 		uasm_i_nop(&p);
1648 #endif
1649 	if (p >= tlbmiss_handler_setup_pgd_end)
1650 		panic("tlbmiss_handler_setup_pgd space exceeded");
1651 
1652 	uasm_resolve_relocs(relocs, labels);
1653 	pr_debug("Wrote tlbmiss_handler_setup_pgd (%u instructions).\n",
1654 		 (unsigned int)(p - tlbmiss_handler_setup_pgd));
1655 
1656 	dump_handler("tlbmiss_handler", tlbmiss_handler_setup_pgd,
1657 					tlbmiss_handler_setup_pgd_size);
1658 }
1659 
1660 static void
1661 iPTE_LW(u32 **p, unsigned int pte, unsigned int ptr)
1662 {
1663 #ifdef CONFIG_SMP
1664 # ifdef CONFIG_PHYS_ADDR_T_64BIT
1665 	if (cpu_has_64bits)
1666 		uasm_i_lld(p, pte, 0, ptr);
1667 	else
1668 # endif
1669 		UASM_i_LL(p, pte, 0, ptr);
1670 #else
1671 # ifdef CONFIG_PHYS_ADDR_T_64BIT
1672 	if (cpu_has_64bits)
1673 		uasm_i_ld(p, pte, 0, ptr);
1674 	else
1675 # endif
1676 		UASM_i_LW(p, pte, 0, ptr);
1677 #endif
1678 }
1679 
1680 static void
1681 iPTE_SW(u32 **p, struct uasm_reloc **r, unsigned int pte, unsigned int ptr,
1682 	unsigned int mode, unsigned int scratch)
1683 {
1684 	unsigned int hwmode = mode & (_PAGE_VALID | _PAGE_DIRTY);
1685 	unsigned int swmode = mode & ~hwmode;
1686 
1687 	if (IS_ENABLED(CONFIG_XPA) && !cpu_has_64bits) {
1688 		uasm_i_lui(p, scratch, swmode >> 16);
1689 		uasm_i_or(p, pte, pte, scratch);
1690 		BUG_ON(swmode & 0xffff);
1691 	} else {
1692 		uasm_i_ori(p, pte, pte, mode);
1693 	}
1694 
1695 #ifdef CONFIG_SMP
1696 # ifdef CONFIG_PHYS_ADDR_T_64BIT
1697 	if (cpu_has_64bits)
1698 		uasm_i_scd(p, pte, 0, ptr);
1699 	else
1700 # endif
1701 		UASM_i_SC(p, pte, 0, ptr);
1702 
1703 	if (r10000_llsc_war())
1704 		uasm_il_beqzl(p, r, pte, label_smp_pgtable_change);
1705 	else
1706 		uasm_il_beqz(p, r, pte, label_smp_pgtable_change);
1707 
1708 # ifdef CONFIG_PHYS_ADDR_T_64BIT
1709 	if (!cpu_has_64bits) {
1710 		/* no uasm_i_nop needed */
1711 		uasm_i_ll(p, pte, sizeof(pte_t) / 2, ptr);
1712 		uasm_i_ori(p, pte, pte, hwmode);
1713 		BUG_ON(hwmode & ~0xffff);
1714 		uasm_i_sc(p, pte, sizeof(pte_t) / 2, ptr);
1715 		uasm_il_beqz(p, r, pte, label_smp_pgtable_change);
1716 		/* no uasm_i_nop needed */
1717 		uasm_i_lw(p, pte, 0, ptr);
1718 	} else
1719 		uasm_i_nop(p);
1720 # else
1721 	uasm_i_nop(p);
1722 # endif
1723 #else
1724 # ifdef CONFIG_PHYS_ADDR_T_64BIT
1725 	if (cpu_has_64bits)
1726 		uasm_i_sd(p, pte, 0, ptr);
1727 	else
1728 # endif
1729 		UASM_i_SW(p, pte, 0, ptr);
1730 
1731 # ifdef CONFIG_PHYS_ADDR_T_64BIT
1732 	if (!cpu_has_64bits) {
1733 		uasm_i_lw(p, pte, sizeof(pte_t) / 2, ptr);
1734 		uasm_i_ori(p, pte, pte, hwmode);
1735 		BUG_ON(hwmode & ~0xffff);
1736 		uasm_i_sw(p, pte, sizeof(pte_t) / 2, ptr);
1737 		uasm_i_lw(p, pte, 0, ptr);
1738 	}
1739 # endif
1740 #endif
1741 }
1742 
1743 /*
1744  * Check if PTE is present, if not then jump to LABEL. PTR points to
1745  * the page table where this PTE is located, PTE will be re-loaded
1746  * with it's original value.
1747  */
1748 static void
1749 build_pte_present(u32 **p, struct uasm_reloc **r,
1750 		  int pte, int ptr, int scratch, enum label_id lid)
1751 {
1752 	int t = scratch >= 0 ? scratch : pte;
1753 	int cur = pte;
1754 
1755 	if (cpu_has_rixi) {
1756 		if (use_bbit_insns()) {
1757 			uasm_il_bbit0(p, r, pte, ilog2(_PAGE_PRESENT), lid);
1758 			uasm_i_nop(p);
1759 		} else {
1760 			if (_PAGE_PRESENT_SHIFT) {
1761 				uasm_i_srl(p, t, cur, _PAGE_PRESENT_SHIFT);
1762 				cur = t;
1763 			}
1764 			uasm_i_andi(p, t, cur, 1);
1765 			uasm_il_beqz(p, r, t, lid);
1766 			if (pte == t)
1767 				/* You lose the SMP race :-(*/
1768 				iPTE_LW(p, pte, ptr);
1769 		}
1770 	} else {
1771 		if (_PAGE_PRESENT_SHIFT) {
1772 			uasm_i_srl(p, t, cur, _PAGE_PRESENT_SHIFT);
1773 			cur = t;
1774 		}
1775 		uasm_i_andi(p, t, cur,
1776 			(_PAGE_PRESENT | _PAGE_NO_READ) >> _PAGE_PRESENT_SHIFT);
1777 		uasm_i_xori(p, t, t, _PAGE_PRESENT >> _PAGE_PRESENT_SHIFT);
1778 		uasm_il_bnez(p, r, t, lid);
1779 		if (pte == t)
1780 			/* You lose the SMP race :-(*/
1781 			iPTE_LW(p, pte, ptr);
1782 	}
1783 }
1784 
1785 /* Make PTE valid, store result in PTR. */
1786 static void
1787 build_make_valid(u32 **p, struct uasm_reloc **r, unsigned int pte,
1788 		 unsigned int ptr, unsigned int scratch)
1789 {
1790 	unsigned int mode = _PAGE_VALID | _PAGE_ACCESSED;
1791 
1792 	iPTE_SW(p, r, pte, ptr, mode, scratch);
1793 }
1794 
1795 /*
1796  * Check if PTE can be written to, if not branch to LABEL. Regardless
1797  * restore PTE with value from PTR when done.
1798  */
1799 static void
1800 build_pte_writable(u32 **p, struct uasm_reloc **r,
1801 		   unsigned int pte, unsigned int ptr, int scratch,
1802 		   enum label_id lid)
1803 {
1804 	int t = scratch >= 0 ? scratch : pte;
1805 	int cur = pte;
1806 
1807 	if (_PAGE_PRESENT_SHIFT) {
1808 		uasm_i_srl(p, t, cur, _PAGE_PRESENT_SHIFT);
1809 		cur = t;
1810 	}
1811 	uasm_i_andi(p, t, cur,
1812 		    (_PAGE_PRESENT | _PAGE_WRITE) >> _PAGE_PRESENT_SHIFT);
1813 	uasm_i_xori(p, t, t,
1814 		    (_PAGE_PRESENT | _PAGE_WRITE) >> _PAGE_PRESENT_SHIFT);
1815 	uasm_il_bnez(p, r, t, lid);
1816 	if (pte == t)
1817 		/* You lose the SMP race :-(*/
1818 		iPTE_LW(p, pte, ptr);
1819 	else
1820 		uasm_i_nop(p);
1821 }
1822 
1823 /* Make PTE writable, update software status bits as well, then store
1824  * at PTR.
1825  */
1826 static void
1827 build_make_write(u32 **p, struct uasm_reloc **r, unsigned int pte,
1828 		 unsigned int ptr, unsigned int scratch)
1829 {
1830 	unsigned int mode = (_PAGE_ACCESSED | _PAGE_MODIFIED | _PAGE_VALID
1831 			     | _PAGE_DIRTY);
1832 
1833 	iPTE_SW(p, r, pte, ptr, mode, scratch);
1834 }
1835 
1836 /*
1837  * Check if PTE can be modified, if not branch to LABEL. Regardless
1838  * restore PTE with value from PTR when done.
1839  */
1840 static void
1841 build_pte_modifiable(u32 **p, struct uasm_reloc **r,
1842 		     unsigned int pte, unsigned int ptr, int scratch,
1843 		     enum label_id lid)
1844 {
1845 	if (use_bbit_insns()) {
1846 		uasm_il_bbit0(p, r, pte, ilog2(_PAGE_WRITE), lid);
1847 		uasm_i_nop(p);
1848 	} else {
1849 		int t = scratch >= 0 ? scratch : pte;
1850 		uasm_i_srl(p, t, pte, _PAGE_WRITE_SHIFT);
1851 		uasm_i_andi(p, t, t, 1);
1852 		uasm_il_beqz(p, r, t, lid);
1853 		if (pte == t)
1854 			/* You lose the SMP race :-(*/
1855 			iPTE_LW(p, pte, ptr);
1856 	}
1857 }
1858 
1859 #ifndef CONFIG_MIPS_PGD_C0_CONTEXT
1860 
1861 
1862 /*
1863  * R3000 style TLB load/store/modify handlers.
1864  */
1865 
1866 /*
1867  * This places the pte into ENTRYLO0 and writes it with tlbwi.
1868  * Then it returns.
1869  */
1870 static void
1871 build_r3000_pte_reload_tlbwi(u32 **p, unsigned int pte, unsigned int tmp)
1872 {
1873 	uasm_i_mtc0(p, pte, C0_ENTRYLO0); /* cp0 delay */
1874 	uasm_i_mfc0(p, tmp, C0_EPC); /* cp0 delay */
1875 	uasm_i_tlbwi(p);
1876 	uasm_i_jr(p, tmp);
1877 	uasm_i_rfe(p); /* branch delay */
1878 }
1879 
1880 /*
1881  * This places the pte into ENTRYLO0 and writes it with tlbwi
1882  * or tlbwr as appropriate.  This is because the index register
1883  * may have the probe fail bit set as a result of a trap on a
1884  * kseg2 access, i.e. without refill.  Then it returns.
1885  */
1886 static void
1887 build_r3000_tlb_reload_write(u32 **p, struct uasm_label **l,
1888 			     struct uasm_reloc **r, unsigned int pte,
1889 			     unsigned int tmp)
1890 {
1891 	uasm_i_mfc0(p, tmp, C0_INDEX);
1892 	uasm_i_mtc0(p, pte, C0_ENTRYLO0); /* cp0 delay */
1893 	uasm_il_bltz(p, r, tmp, label_r3000_write_probe_fail); /* cp0 delay */
1894 	uasm_i_mfc0(p, tmp, C0_EPC); /* branch delay */
1895 	uasm_i_tlbwi(p); /* cp0 delay */
1896 	uasm_i_jr(p, tmp);
1897 	uasm_i_rfe(p); /* branch delay */
1898 	uasm_l_r3000_write_probe_fail(l, *p);
1899 	uasm_i_tlbwr(p); /* cp0 delay */
1900 	uasm_i_jr(p, tmp);
1901 	uasm_i_rfe(p); /* branch delay */
1902 }
1903 
1904 static void
1905 build_r3000_tlbchange_handler_head(u32 **p, unsigned int pte,
1906 				   unsigned int ptr)
1907 {
1908 	long pgdc = (long)pgd_current;
1909 
1910 	uasm_i_mfc0(p, pte, C0_BADVADDR);
1911 	uasm_i_lui(p, ptr, uasm_rel_hi(pgdc)); /* cp0 delay */
1912 	uasm_i_lw(p, ptr, uasm_rel_lo(pgdc), ptr);
1913 	uasm_i_srl(p, pte, pte, 22); /* load delay */
1914 	uasm_i_sll(p, pte, pte, 2);
1915 	uasm_i_addu(p, ptr, ptr, pte);
1916 	uasm_i_mfc0(p, pte, C0_CONTEXT);
1917 	uasm_i_lw(p, ptr, 0, ptr); /* cp0 delay */
1918 	uasm_i_andi(p, pte, pte, 0xffc); /* load delay */
1919 	uasm_i_addu(p, ptr, ptr, pte);
1920 	uasm_i_lw(p, pte, 0, ptr);
1921 	uasm_i_tlbp(p); /* load delay */
1922 }
1923 
1924 static void build_r3000_tlb_load_handler(void)
1925 {
1926 	u32 *p = handle_tlbl;
1927 	const int handle_tlbl_size = handle_tlbl_end - handle_tlbl;
1928 	struct uasm_label *l = labels;
1929 	struct uasm_reloc *r = relocs;
1930 
1931 	memset(handle_tlbl, 0, handle_tlbl_size * sizeof(handle_tlbl[0]));
1932 	memset(labels, 0, sizeof(labels));
1933 	memset(relocs, 0, sizeof(relocs));
1934 
1935 	build_r3000_tlbchange_handler_head(&p, K0, K1);
1936 	build_pte_present(&p, &r, K0, K1, -1, label_nopage_tlbl);
1937 	uasm_i_nop(&p); /* load delay */
1938 	build_make_valid(&p, &r, K0, K1, -1);
1939 	build_r3000_tlb_reload_write(&p, &l, &r, K0, K1);
1940 
1941 	uasm_l_nopage_tlbl(&l, p);
1942 	uasm_i_j(&p, (unsigned long)tlb_do_page_fault_0 & 0x0fffffff);
1943 	uasm_i_nop(&p);
1944 
1945 	if (p >= handle_tlbl_end)
1946 		panic("TLB load handler fastpath space exceeded");
1947 
1948 	uasm_resolve_relocs(relocs, labels);
1949 	pr_debug("Wrote TLB load handler fastpath (%u instructions).\n",
1950 		 (unsigned int)(p - handle_tlbl));
1951 
1952 	dump_handler("r3000_tlb_load", handle_tlbl, handle_tlbl_size);
1953 }
1954 
1955 static void build_r3000_tlb_store_handler(void)
1956 {
1957 	u32 *p = handle_tlbs;
1958 	const int handle_tlbs_size = handle_tlbs_end - handle_tlbs;
1959 	struct uasm_label *l = labels;
1960 	struct uasm_reloc *r = relocs;
1961 
1962 	memset(handle_tlbs, 0, handle_tlbs_size * sizeof(handle_tlbs[0]));
1963 	memset(labels, 0, sizeof(labels));
1964 	memset(relocs, 0, sizeof(relocs));
1965 
1966 	build_r3000_tlbchange_handler_head(&p, K0, K1);
1967 	build_pte_writable(&p, &r, K0, K1, -1, label_nopage_tlbs);
1968 	uasm_i_nop(&p); /* load delay */
1969 	build_make_write(&p, &r, K0, K1, -1);
1970 	build_r3000_tlb_reload_write(&p, &l, &r, K0, K1);
1971 
1972 	uasm_l_nopage_tlbs(&l, p);
1973 	uasm_i_j(&p, (unsigned long)tlb_do_page_fault_1 & 0x0fffffff);
1974 	uasm_i_nop(&p);
1975 
1976 	if (p >= handle_tlbs_end)
1977 		panic("TLB store handler fastpath space exceeded");
1978 
1979 	uasm_resolve_relocs(relocs, labels);
1980 	pr_debug("Wrote TLB store handler fastpath (%u instructions).\n",
1981 		 (unsigned int)(p - handle_tlbs));
1982 
1983 	dump_handler("r3000_tlb_store", handle_tlbs, handle_tlbs_size);
1984 }
1985 
1986 static void build_r3000_tlb_modify_handler(void)
1987 {
1988 	u32 *p = handle_tlbm;
1989 	const int handle_tlbm_size = handle_tlbm_end - handle_tlbm;
1990 	struct uasm_label *l = labels;
1991 	struct uasm_reloc *r = relocs;
1992 
1993 	memset(handle_tlbm, 0, handle_tlbm_size * sizeof(handle_tlbm[0]));
1994 	memset(labels, 0, sizeof(labels));
1995 	memset(relocs, 0, sizeof(relocs));
1996 
1997 	build_r3000_tlbchange_handler_head(&p, K0, K1);
1998 	build_pte_modifiable(&p, &r, K0, K1,  -1, label_nopage_tlbm);
1999 	uasm_i_nop(&p); /* load delay */
2000 	build_make_write(&p, &r, K0, K1, -1);
2001 	build_r3000_pte_reload_tlbwi(&p, K0, K1);
2002 
2003 	uasm_l_nopage_tlbm(&l, p);
2004 	uasm_i_j(&p, (unsigned long)tlb_do_page_fault_1 & 0x0fffffff);
2005 	uasm_i_nop(&p);
2006 
2007 	if (p >= handle_tlbm_end)
2008 		panic("TLB modify handler fastpath space exceeded");
2009 
2010 	uasm_resolve_relocs(relocs, labels);
2011 	pr_debug("Wrote TLB modify handler fastpath (%u instructions).\n",
2012 		 (unsigned int)(p - handle_tlbm));
2013 
2014 	dump_handler("r3000_tlb_modify", handle_tlbm, handle_tlbm_size);
2015 }
2016 #endif /* CONFIG_MIPS_PGD_C0_CONTEXT */
2017 
2018 /*
2019  * R4000 style TLB load/store/modify handlers.
2020  */
2021 static struct work_registers
2022 build_r4000_tlbchange_handler_head(u32 **p, struct uasm_label **l,
2023 				   struct uasm_reloc **r)
2024 {
2025 	struct work_registers wr = build_get_work_registers(p);
2026 
2027 #ifdef CONFIG_64BIT
2028 	build_get_pmde64(p, l, r, wr.r1, wr.r2); /* get pmd in ptr */
2029 #else
2030 	build_get_pgde32(p, wr.r1, wr.r2); /* get pgd in ptr */
2031 #endif
2032 
2033 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
2034 	/*
2035 	 * For huge tlb entries, pmd doesn't contain an address but
2036 	 * instead contains the tlb pte. Check the PAGE_HUGE bit and
2037 	 * see if we need to jump to huge tlb processing.
2038 	 */
2039 	build_is_huge_pte(p, r, wr.r1, wr.r2, label_tlb_huge_update);
2040 #endif
2041 
2042 	UASM_i_MFC0(p, wr.r1, C0_BADVADDR);
2043 	UASM_i_LW(p, wr.r2, 0, wr.r2);
2044 	UASM_i_SRL(p, wr.r1, wr.r1, PAGE_SHIFT + PTE_ORDER - PTE_T_LOG2);
2045 	uasm_i_andi(p, wr.r1, wr.r1, (PTRS_PER_PTE - 1) << PTE_T_LOG2);
2046 	UASM_i_ADDU(p, wr.r2, wr.r2, wr.r1);
2047 
2048 #ifdef CONFIG_SMP
2049 	uasm_l_smp_pgtable_change(l, *p);
2050 #endif
2051 	iPTE_LW(p, wr.r1, wr.r2); /* get even pte */
2052 	if (!m4kc_tlbp_war()) {
2053 		build_tlb_probe_entry(p);
2054 		if (cpu_has_htw) {
2055 			/* race condition happens, leaving */
2056 			uasm_i_ehb(p);
2057 			uasm_i_mfc0(p, wr.r3, C0_INDEX);
2058 			uasm_il_bltz(p, r, wr.r3, label_leave);
2059 			uasm_i_nop(p);
2060 		}
2061 	}
2062 	return wr;
2063 }
2064 
2065 static void
2066 build_r4000_tlbchange_handler_tail(u32 **p, struct uasm_label **l,
2067 				   struct uasm_reloc **r, unsigned int tmp,
2068 				   unsigned int ptr)
2069 {
2070 	uasm_i_ori(p, ptr, ptr, sizeof(pte_t));
2071 	uasm_i_xori(p, ptr, ptr, sizeof(pte_t));
2072 	build_update_entries(p, tmp, ptr);
2073 	build_tlb_write_entry(p, l, r, tlb_indexed);
2074 	uasm_l_leave(l, *p);
2075 	build_restore_work_registers(p);
2076 	uasm_i_eret(p); /* return from trap */
2077 
2078 #ifdef CONFIG_64BIT
2079 	build_get_pgd_vmalloc64(p, l, r, tmp, ptr, not_refill);
2080 #endif
2081 }
2082 
2083 static void build_r4000_tlb_load_handler(void)
2084 {
2085 	u32 *p = (u32 *)msk_isa16_mode((ulong)handle_tlbl);
2086 	const int handle_tlbl_size = handle_tlbl_end - handle_tlbl;
2087 	struct uasm_label *l = labels;
2088 	struct uasm_reloc *r = relocs;
2089 	struct work_registers wr;
2090 
2091 	memset(handle_tlbl, 0, handle_tlbl_size * sizeof(handle_tlbl[0]));
2092 	memset(labels, 0, sizeof(labels));
2093 	memset(relocs, 0, sizeof(relocs));
2094 
2095 	if (bcm1250_m3_war()) {
2096 		unsigned int segbits = 44;
2097 
2098 		uasm_i_dmfc0(&p, K0, C0_BADVADDR);
2099 		uasm_i_dmfc0(&p, K1, C0_ENTRYHI);
2100 		uasm_i_xor(&p, K0, K0, K1);
2101 		uasm_i_dsrl_safe(&p, K1, K0, 62);
2102 		uasm_i_dsrl_safe(&p, K0, K0, 12 + 1);
2103 		uasm_i_dsll_safe(&p, K0, K0, 64 + 12 + 1 - segbits);
2104 		uasm_i_or(&p, K0, K0, K1);
2105 		uasm_il_bnez(&p, &r, K0, label_leave);
2106 		/* No need for uasm_i_nop */
2107 	}
2108 
2109 	wr = build_r4000_tlbchange_handler_head(&p, &l, &r);
2110 	build_pte_present(&p, &r, wr.r1, wr.r2, wr.r3, label_nopage_tlbl);
2111 	if (m4kc_tlbp_war())
2112 		build_tlb_probe_entry(&p);
2113 
2114 	if (cpu_has_rixi && !cpu_has_rixiex) {
2115 		/*
2116 		 * If the page is not _PAGE_VALID, RI or XI could not
2117 		 * have triggered it.  Skip the expensive test..
2118 		 */
2119 		if (use_bbit_insns()) {
2120 			uasm_il_bbit0(&p, &r, wr.r1, ilog2(_PAGE_VALID),
2121 				      label_tlbl_goaround1);
2122 		} else {
2123 			uasm_i_andi(&p, wr.r3, wr.r1, _PAGE_VALID);
2124 			uasm_il_beqz(&p, &r, wr.r3, label_tlbl_goaround1);
2125 		}
2126 		uasm_i_nop(&p);
2127 
2128 		uasm_i_tlbr(&p);
2129 
2130 		switch (current_cpu_type()) {
2131 		default:
2132 			if (cpu_has_mips_r2_exec_hazard) {
2133 				uasm_i_ehb(&p);
2134 
2135 		case CPU_CAVIUM_OCTEON:
2136 		case CPU_CAVIUM_OCTEON_PLUS:
2137 		case CPU_CAVIUM_OCTEON2:
2138 				break;
2139 			}
2140 		}
2141 
2142 		/* Examine  entrylo 0 or 1 based on ptr. */
2143 		if (use_bbit_insns()) {
2144 			uasm_i_bbit0(&p, wr.r2, ilog2(sizeof(pte_t)), 8);
2145 		} else {
2146 			uasm_i_andi(&p, wr.r3, wr.r2, sizeof(pte_t));
2147 			uasm_i_beqz(&p, wr.r3, 8);
2148 		}
2149 		/* load it in the delay slot*/
2150 		UASM_i_MFC0(&p, wr.r3, C0_ENTRYLO0);
2151 		/* load it if ptr is odd */
2152 		UASM_i_MFC0(&p, wr.r3, C0_ENTRYLO1);
2153 		/*
2154 		 * If the entryLo (now in wr.r3) is valid (bit 1), RI or
2155 		 * XI must have triggered it.
2156 		 */
2157 		if (use_bbit_insns()) {
2158 			uasm_il_bbit1(&p, &r, wr.r3, 1, label_nopage_tlbl);
2159 			uasm_i_nop(&p);
2160 			uasm_l_tlbl_goaround1(&l, p);
2161 		} else {
2162 			uasm_i_andi(&p, wr.r3, wr.r3, 2);
2163 			uasm_il_bnez(&p, &r, wr.r3, label_nopage_tlbl);
2164 			uasm_i_nop(&p);
2165 		}
2166 		uasm_l_tlbl_goaround1(&l, p);
2167 	}
2168 	build_make_valid(&p, &r, wr.r1, wr.r2, wr.r3);
2169 	build_r4000_tlbchange_handler_tail(&p, &l, &r, wr.r1, wr.r2);
2170 
2171 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
2172 	/*
2173 	 * This is the entry point when build_r4000_tlbchange_handler_head
2174 	 * spots a huge page.
2175 	 */
2176 	uasm_l_tlb_huge_update(&l, p);
2177 	iPTE_LW(&p, wr.r1, wr.r2);
2178 	build_pte_present(&p, &r, wr.r1, wr.r2, wr.r3, label_nopage_tlbl);
2179 	build_tlb_probe_entry(&p);
2180 
2181 	if (cpu_has_rixi && !cpu_has_rixiex) {
2182 		/*
2183 		 * If the page is not _PAGE_VALID, RI or XI could not
2184 		 * have triggered it.  Skip the expensive test..
2185 		 */
2186 		if (use_bbit_insns()) {
2187 			uasm_il_bbit0(&p, &r, wr.r1, ilog2(_PAGE_VALID),
2188 				      label_tlbl_goaround2);
2189 		} else {
2190 			uasm_i_andi(&p, wr.r3, wr.r1, _PAGE_VALID);
2191 			uasm_il_beqz(&p, &r, wr.r3, label_tlbl_goaround2);
2192 		}
2193 		uasm_i_nop(&p);
2194 
2195 		uasm_i_tlbr(&p);
2196 
2197 		switch (current_cpu_type()) {
2198 		default:
2199 			if (cpu_has_mips_r2_exec_hazard) {
2200 				uasm_i_ehb(&p);
2201 
2202 		case CPU_CAVIUM_OCTEON:
2203 		case CPU_CAVIUM_OCTEON_PLUS:
2204 		case CPU_CAVIUM_OCTEON2:
2205 				break;
2206 			}
2207 		}
2208 
2209 		/* Examine  entrylo 0 or 1 based on ptr. */
2210 		if (use_bbit_insns()) {
2211 			uasm_i_bbit0(&p, wr.r2, ilog2(sizeof(pte_t)), 8);
2212 		} else {
2213 			uasm_i_andi(&p, wr.r3, wr.r2, sizeof(pte_t));
2214 			uasm_i_beqz(&p, wr.r3, 8);
2215 		}
2216 		/* load it in the delay slot*/
2217 		UASM_i_MFC0(&p, wr.r3, C0_ENTRYLO0);
2218 		/* load it if ptr is odd */
2219 		UASM_i_MFC0(&p, wr.r3, C0_ENTRYLO1);
2220 		/*
2221 		 * If the entryLo (now in wr.r3) is valid (bit 1), RI or
2222 		 * XI must have triggered it.
2223 		 */
2224 		if (use_bbit_insns()) {
2225 			uasm_il_bbit0(&p, &r, wr.r3, 1, label_tlbl_goaround2);
2226 		} else {
2227 			uasm_i_andi(&p, wr.r3, wr.r3, 2);
2228 			uasm_il_beqz(&p, &r, wr.r3, label_tlbl_goaround2);
2229 		}
2230 		if (PM_DEFAULT_MASK == 0)
2231 			uasm_i_nop(&p);
2232 		/*
2233 		 * We clobbered C0_PAGEMASK, restore it.  On the other branch
2234 		 * it is restored in build_huge_tlb_write_entry.
2235 		 */
2236 		build_restore_pagemask(&p, &r, wr.r3, label_nopage_tlbl, 0);
2237 
2238 		uasm_l_tlbl_goaround2(&l, p);
2239 	}
2240 	uasm_i_ori(&p, wr.r1, wr.r1, (_PAGE_ACCESSED | _PAGE_VALID));
2241 	build_huge_handler_tail(&p, &r, &l, wr.r1, wr.r2, 1);
2242 #endif
2243 
2244 	uasm_l_nopage_tlbl(&l, p);
2245 	build_restore_work_registers(&p);
2246 #ifdef CONFIG_CPU_MICROMIPS
2247 	if ((unsigned long)tlb_do_page_fault_0 & 1) {
2248 		uasm_i_lui(&p, K0, uasm_rel_hi((long)tlb_do_page_fault_0));
2249 		uasm_i_addiu(&p, K0, K0, uasm_rel_lo((long)tlb_do_page_fault_0));
2250 		uasm_i_jr(&p, K0);
2251 	} else
2252 #endif
2253 	uasm_i_j(&p, (unsigned long)tlb_do_page_fault_0 & 0x0fffffff);
2254 	uasm_i_nop(&p);
2255 
2256 	if (p >= handle_tlbl_end)
2257 		panic("TLB load handler fastpath space exceeded");
2258 
2259 	uasm_resolve_relocs(relocs, labels);
2260 	pr_debug("Wrote TLB load handler fastpath (%u instructions).\n",
2261 		 (unsigned int)(p - handle_tlbl));
2262 
2263 	dump_handler("r4000_tlb_load", handle_tlbl, handle_tlbl_size);
2264 }
2265 
2266 static void build_r4000_tlb_store_handler(void)
2267 {
2268 	u32 *p = (u32 *)msk_isa16_mode((ulong)handle_tlbs);
2269 	const int handle_tlbs_size = handle_tlbs_end - handle_tlbs;
2270 	struct uasm_label *l = labels;
2271 	struct uasm_reloc *r = relocs;
2272 	struct work_registers wr;
2273 
2274 	memset(handle_tlbs, 0, handle_tlbs_size * sizeof(handle_tlbs[0]));
2275 	memset(labels, 0, sizeof(labels));
2276 	memset(relocs, 0, sizeof(relocs));
2277 
2278 	wr = build_r4000_tlbchange_handler_head(&p, &l, &r);
2279 	build_pte_writable(&p, &r, wr.r1, wr.r2, wr.r3, label_nopage_tlbs);
2280 	if (m4kc_tlbp_war())
2281 		build_tlb_probe_entry(&p);
2282 	build_make_write(&p, &r, wr.r1, wr.r2, wr.r3);
2283 	build_r4000_tlbchange_handler_tail(&p, &l, &r, wr.r1, wr.r2);
2284 
2285 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
2286 	/*
2287 	 * This is the entry point when
2288 	 * build_r4000_tlbchange_handler_head spots a huge page.
2289 	 */
2290 	uasm_l_tlb_huge_update(&l, p);
2291 	iPTE_LW(&p, wr.r1, wr.r2);
2292 	build_pte_writable(&p, &r, wr.r1, wr.r2, wr.r3, label_nopage_tlbs);
2293 	build_tlb_probe_entry(&p);
2294 	uasm_i_ori(&p, wr.r1, wr.r1,
2295 		   _PAGE_ACCESSED | _PAGE_MODIFIED | _PAGE_VALID | _PAGE_DIRTY);
2296 	build_huge_handler_tail(&p, &r, &l, wr.r1, wr.r2, 1);
2297 #endif
2298 
2299 	uasm_l_nopage_tlbs(&l, p);
2300 	build_restore_work_registers(&p);
2301 #ifdef CONFIG_CPU_MICROMIPS
2302 	if ((unsigned long)tlb_do_page_fault_1 & 1) {
2303 		uasm_i_lui(&p, K0, uasm_rel_hi((long)tlb_do_page_fault_1));
2304 		uasm_i_addiu(&p, K0, K0, uasm_rel_lo((long)tlb_do_page_fault_1));
2305 		uasm_i_jr(&p, K0);
2306 	} else
2307 #endif
2308 	uasm_i_j(&p, (unsigned long)tlb_do_page_fault_1 & 0x0fffffff);
2309 	uasm_i_nop(&p);
2310 
2311 	if (p >= handle_tlbs_end)
2312 		panic("TLB store handler fastpath space exceeded");
2313 
2314 	uasm_resolve_relocs(relocs, labels);
2315 	pr_debug("Wrote TLB store handler fastpath (%u instructions).\n",
2316 		 (unsigned int)(p - handle_tlbs));
2317 
2318 	dump_handler("r4000_tlb_store", handle_tlbs, handle_tlbs_size);
2319 }
2320 
2321 static void build_r4000_tlb_modify_handler(void)
2322 {
2323 	u32 *p = (u32 *)msk_isa16_mode((ulong)handle_tlbm);
2324 	const int handle_tlbm_size = handle_tlbm_end - handle_tlbm;
2325 	struct uasm_label *l = labels;
2326 	struct uasm_reloc *r = relocs;
2327 	struct work_registers wr;
2328 
2329 	memset(handle_tlbm, 0, handle_tlbm_size * sizeof(handle_tlbm[0]));
2330 	memset(labels, 0, sizeof(labels));
2331 	memset(relocs, 0, sizeof(relocs));
2332 
2333 	wr = build_r4000_tlbchange_handler_head(&p, &l, &r);
2334 	build_pte_modifiable(&p, &r, wr.r1, wr.r2, wr.r3, label_nopage_tlbm);
2335 	if (m4kc_tlbp_war())
2336 		build_tlb_probe_entry(&p);
2337 	/* Present and writable bits set, set accessed and dirty bits. */
2338 	build_make_write(&p, &r, wr.r1, wr.r2, wr.r3);
2339 	build_r4000_tlbchange_handler_tail(&p, &l, &r, wr.r1, wr.r2);
2340 
2341 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
2342 	/*
2343 	 * This is the entry point when
2344 	 * build_r4000_tlbchange_handler_head spots a huge page.
2345 	 */
2346 	uasm_l_tlb_huge_update(&l, p);
2347 	iPTE_LW(&p, wr.r1, wr.r2);
2348 	build_pte_modifiable(&p, &r, wr.r1, wr.r2,  wr.r3, label_nopage_tlbm);
2349 	build_tlb_probe_entry(&p);
2350 	uasm_i_ori(&p, wr.r1, wr.r1,
2351 		   _PAGE_ACCESSED | _PAGE_MODIFIED | _PAGE_VALID | _PAGE_DIRTY);
2352 	build_huge_handler_tail(&p, &r, &l, wr.r1, wr.r2, 0);
2353 #endif
2354 
2355 	uasm_l_nopage_tlbm(&l, p);
2356 	build_restore_work_registers(&p);
2357 #ifdef CONFIG_CPU_MICROMIPS
2358 	if ((unsigned long)tlb_do_page_fault_1 & 1) {
2359 		uasm_i_lui(&p, K0, uasm_rel_hi((long)tlb_do_page_fault_1));
2360 		uasm_i_addiu(&p, K0, K0, uasm_rel_lo((long)tlb_do_page_fault_1));
2361 		uasm_i_jr(&p, K0);
2362 	} else
2363 #endif
2364 	uasm_i_j(&p, (unsigned long)tlb_do_page_fault_1 & 0x0fffffff);
2365 	uasm_i_nop(&p);
2366 
2367 	if (p >= handle_tlbm_end)
2368 		panic("TLB modify handler fastpath space exceeded");
2369 
2370 	uasm_resolve_relocs(relocs, labels);
2371 	pr_debug("Wrote TLB modify handler fastpath (%u instructions).\n",
2372 		 (unsigned int)(p - handle_tlbm));
2373 
2374 	dump_handler("r4000_tlb_modify", handle_tlbm, handle_tlbm_size);
2375 }
2376 
2377 static void flush_tlb_handlers(void)
2378 {
2379 	local_flush_icache_range((unsigned long)handle_tlbl,
2380 			   (unsigned long)handle_tlbl_end);
2381 	local_flush_icache_range((unsigned long)handle_tlbs,
2382 			   (unsigned long)handle_tlbs_end);
2383 	local_flush_icache_range((unsigned long)handle_tlbm,
2384 			   (unsigned long)handle_tlbm_end);
2385 	local_flush_icache_range((unsigned long)tlbmiss_handler_setup_pgd,
2386 			   (unsigned long)tlbmiss_handler_setup_pgd_end);
2387 }
2388 
2389 static void print_htw_config(void)
2390 {
2391 	unsigned long config;
2392 	unsigned int pwctl;
2393 	const int field = 2 * sizeof(unsigned long);
2394 
2395 	config = read_c0_pwfield();
2396 	pr_debug("PWField (0x%0*lx): GDI: 0x%02lx  UDI: 0x%02lx  MDI: 0x%02lx  PTI: 0x%02lx  PTEI: 0x%02lx\n",
2397 		field, config,
2398 		(config & MIPS_PWFIELD_GDI_MASK) >> MIPS_PWFIELD_GDI_SHIFT,
2399 		(config & MIPS_PWFIELD_UDI_MASK) >> MIPS_PWFIELD_UDI_SHIFT,
2400 		(config & MIPS_PWFIELD_MDI_MASK) >> MIPS_PWFIELD_MDI_SHIFT,
2401 		(config & MIPS_PWFIELD_PTI_MASK) >> MIPS_PWFIELD_PTI_SHIFT,
2402 		(config & MIPS_PWFIELD_PTEI_MASK) >> MIPS_PWFIELD_PTEI_SHIFT);
2403 
2404 	config = read_c0_pwsize();
2405 	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",
2406 		field, config,
2407 		(config & MIPS_PWSIZE_PS_MASK) >> MIPS_PWSIZE_PS_SHIFT,
2408 		(config & MIPS_PWSIZE_GDW_MASK) >> MIPS_PWSIZE_GDW_SHIFT,
2409 		(config & MIPS_PWSIZE_UDW_MASK) >> MIPS_PWSIZE_UDW_SHIFT,
2410 		(config & MIPS_PWSIZE_MDW_MASK) >> MIPS_PWSIZE_MDW_SHIFT,
2411 		(config & MIPS_PWSIZE_PTW_MASK) >> MIPS_PWSIZE_PTW_SHIFT,
2412 		(config & MIPS_PWSIZE_PTEW_MASK) >> MIPS_PWSIZE_PTEW_SHIFT);
2413 
2414 	pwctl = read_c0_pwctl();
2415 	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",
2416 		pwctl,
2417 		(pwctl & MIPS_PWCTL_PWEN_MASK) >> MIPS_PWCTL_PWEN_SHIFT,
2418 		(pwctl & MIPS_PWCTL_XK_MASK) >> MIPS_PWCTL_XK_SHIFT,
2419 		(pwctl & MIPS_PWCTL_XS_MASK) >> MIPS_PWCTL_XS_SHIFT,
2420 		(pwctl & MIPS_PWCTL_XU_MASK) >> MIPS_PWCTL_XU_SHIFT,
2421 		(pwctl & MIPS_PWCTL_DPH_MASK) >> MIPS_PWCTL_DPH_SHIFT,
2422 		(pwctl & MIPS_PWCTL_HUGEPG_MASK) >> MIPS_PWCTL_HUGEPG_SHIFT,
2423 		(pwctl & MIPS_PWCTL_PSN_MASK) >> MIPS_PWCTL_PSN_SHIFT);
2424 }
2425 
2426 static void config_htw_params(void)
2427 {
2428 	unsigned long pwfield, pwsize, ptei;
2429 	unsigned int config;
2430 
2431 	/*
2432 	 * We are using 2-level page tables, so we only need to
2433 	 * setup GDW and PTW appropriately. UDW and MDW will remain 0.
2434 	 * The default value of GDI/UDI/MDI/PTI is 0xc. It is illegal to
2435 	 * write values less than 0xc in these fields because the entire
2436 	 * write will be dropped. As a result of which, we must preserve
2437 	 * the original reset values and overwrite only what we really want.
2438 	 */
2439 
2440 	pwfield = read_c0_pwfield();
2441 	/* re-initialize the GDI field */
2442 	pwfield &= ~MIPS_PWFIELD_GDI_MASK;
2443 	pwfield |= PGDIR_SHIFT << MIPS_PWFIELD_GDI_SHIFT;
2444 	/* re-initialize the PTI field including the even/odd bit */
2445 	pwfield &= ~MIPS_PWFIELD_PTI_MASK;
2446 	pwfield |= PAGE_SHIFT << MIPS_PWFIELD_PTI_SHIFT;
2447 	if (CONFIG_PGTABLE_LEVELS >= 3) {
2448 		pwfield &= ~MIPS_PWFIELD_MDI_MASK;
2449 		pwfield |= PMD_SHIFT << MIPS_PWFIELD_MDI_SHIFT;
2450 	}
2451 	/* Set the PTEI right shift */
2452 	ptei = _PAGE_GLOBAL_SHIFT << MIPS_PWFIELD_PTEI_SHIFT;
2453 	pwfield |= ptei;
2454 	write_c0_pwfield(pwfield);
2455 	/* Check whether the PTEI value is supported */
2456 	back_to_back_c0_hazard();
2457 	pwfield = read_c0_pwfield();
2458 	if (((pwfield & MIPS_PWFIELD_PTEI_MASK) << MIPS_PWFIELD_PTEI_SHIFT)
2459 		!= ptei) {
2460 		pr_warn("Unsupported PTEI field value: 0x%lx. HTW will not be enabled",
2461 			ptei);
2462 		/*
2463 		 * Drop option to avoid HTW being enabled via another path
2464 		 * (eg htw_reset())
2465 		 */
2466 		current_cpu_data.options &= ~MIPS_CPU_HTW;
2467 		return;
2468 	}
2469 
2470 	pwsize = ilog2(PTRS_PER_PGD) << MIPS_PWSIZE_GDW_SHIFT;
2471 	pwsize |= ilog2(PTRS_PER_PTE) << MIPS_PWSIZE_PTW_SHIFT;
2472 	if (CONFIG_PGTABLE_LEVELS >= 3)
2473 		pwsize |= ilog2(PTRS_PER_PMD) << MIPS_PWSIZE_MDW_SHIFT;
2474 
2475 	/* Set pointer size to size of directory pointers */
2476 	if (IS_ENABLED(CONFIG_64BIT))
2477 		pwsize |= MIPS_PWSIZE_PS_MASK;
2478 	/* PTEs may be multiple pointers long (e.g. with XPA) */
2479 	pwsize |= ((PTE_T_LOG2 - PGD_T_LOG2) << MIPS_PWSIZE_PTEW_SHIFT)
2480 			& MIPS_PWSIZE_PTEW_MASK;
2481 
2482 	write_c0_pwsize(pwsize);
2483 
2484 	/* Make sure everything is set before we enable the HTW */
2485 	back_to_back_c0_hazard();
2486 
2487 	/*
2488 	 * Enable HTW (and only for XUSeg on 64-bit), and disable the rest of
2489 	 * the pwctl fields.
2490 	 */
2491 	config = 1 << MIPS_PWCTL_PWEN_SHIFT;
2492 	if (IS_ENABLED(CONFIG_64BIT))
2493 		config |= MIPS_PWCTL_XU_MASK;
2494 	write_c0_pwctl(config);
2495 	pr_info("Hardware Page Table Walker enabled\n");
2496 
2497 	print_htw_config();
2498 }
2499 
2500 static void config_xpa_params(void)
2501 {
2502 #ifdef CONFIG_XPA
2503 	unsigned int pagegrain;
2504 
2505 	if (mips_xpa_disabled) {
2506 		pr_info("Extended Physical Addressing (XPA) disabled\n");
2507 		return;
2508 	}
2509 
2510 	pagegrain = read_c0_pagegrain();
2511 	write_c0_pagegrain(pagegrain | PG_ELPA);
2512 	back_to_back_c0_hazard();
2513 	pagegrain = read_c0_pagegrain();
2514 
2515 	if (pagegrain & PG_ELPA)
2516 		pr_info("Extended Physical Addressing (XPA) enabled\n");
2517 	else
2518 		panic("Extended Physical Addressing (XPA) disabled");
2519 #endif
2520 }
2521 
2522 static void check_pabits(void)
2523 {
2524 	unsigned long entry;
2525 	unsigned pabits, fillbits;
2526 
2527 	if (!cpu_has_rixi || !_PAGE_NO_EXEC) {
2528 		/*
2529 		 * We'll only be making use of the fact that we can rotate bits
2530 		 * into the fill if the CPU supports RIXI, so don't bother
2531 		 * probing this for CPUs which don't.
2532 		 */
2533 		return;
2534 	}
2535 
2536 	write_c0_entrylo0(~0ul);
2537 	back_to_back_c0_hazard();
2538 	entry = read_c0_entrylo0();
2539 
2540 	/* clear all non-PFN bits */
2541 	entry &= ~((1 << MIPS_ENTRYLO_PFN_SHIFT) - 1);
2542 	entry &= ~(MIPS_ENTRYLO_RI | MIPS_ENTRYLO_XI);
2543 
2544 	/* find a lower bound on PABITS, and upper bound on fill bits */
2545 	pabits = fls_long(entry) + 6;
2546 	fillbits = max_t(int, (int)BITS_PER_LONG - pabits, 0);
2547 
2548 	/* minus the RI & XI bits */
2549 	fillbits -= min_t(unsigned, fillbits, 2);
2550 
2551 	if (fillbits >= ilog2(_PAGE_NO_EXEC))
2552 		fill_includes_sw_bits = true;
2553 
2554 	pr_debug("Entry* registers contain %u fill bits\n", fillbits);
2555 }
2556 
2557 void build_tlb_refill_handler(void)
2558 {
2559 	/*
2560 	 * The refill handler is generated per-CPU, multi-node systems
2561 	 * may have local storage for it. The other handlers are only
2562 	 * needed once.
2563 	 */
2564 	static int run_once = 0;
2565 
2566 	if (IS_ENABLED(CONFIG_XPA) && !cpu_has_rixi)
2567 		panic("Kernels supporting XPA currently require CPUs with RIXI");
2568 
2569 	output_pgtable_bits_defines();
2570 	check_pabits();
2571 
2572 #ifdef CONFIG_64BIT
2573 	check_for_high_segbits = current_cpu_data.vmbits > (PGDIR_SHIFT + PGD_ORDER + PAGE_SHIFT - 3);
2574 #endif
2575 
2576 	switch (current_cpu_type()) {
2577 	case CPU_R2000:
2578 	case CPU_R3000:
2579 	case CPU_R3000A:
2580 	case CPU_R3081E:
2581 	case CPU_TX3912:
2582 	case CPU_TX3922:
2583 	case CPU_TX3927:
2584 #ifndef CONFIG_MIPS_PGD_C0_CONTEXT
2585 		if (cpu_has_local_ebase)
2586 			build_r3000_tlb_refill_handler();
2587 		if (!run_once) {
2588 			if (!cpu_has_local_ebase)
2589 				build_r3000_tlb_refill_handler();
2590 			build_setup_pgd();
2591 			build_r3000_tlb_load_handler();
2592 			build_r3000_tlb_store_handler();
2593 			build_r3000_tlb_modify_handler();
2594 			flush_tlb_handlers();
2595 			run_once++;
2596 		}
2597 #else
2598 		panic("No R3000 TLB refill handler");
2599 #endif
2600 		break;
2601 
2602 	case CPU_R6000:
2603 	case CPU_R6000A:
2604 		panic("No R6000 TLB refill handler yet");
2605 		break;
2606 
2607 	case CPU_R8000:
2608 		panic("No R8000 TLB refill handler yet");
2609 		break;
2610 
2611 	default:
2612 		if (cpu_has_ldpte)
2613 			setup_pw();
2614 
2615 		if (!run_once) {
2616 			scratch_reg = allocate_kscratch();
2617 			build_setup_pgd();
2618 			build_r4000_tlb_load_handler();
2619 			build_r4000_tlb_store_handler();
2620 			build_r4000_tlb_modify_handler();
2621 			if (cpu_has_ldpte)
2622 				build_loongson3_tlb_refill_handler();
2623 			else if (!cpu_has_local_ebase)
2624 				build_r4000_tlb_refill_handler();
2625 			flush_tlb_handlers();
2626 			run_once++;
2627 		}
2628 		if (cpu_has_local_ebase)
2629 			build_r4000_tlb_refill_handler();
2630 		if (cpu_has_xpa)
2631 			config_xpa_params();
2632 		if (cpu_has_htw)
2633 			config_htw_params();
2634 	}
2635 }
2636