xref: /openbmc/linux/arch/mips/mm/tlbex.c (revision a2818ee4)
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/mmu_context.h>
35 #include <asm/pgtable.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 static inline int __maybe_unused bcm1250_m3_war(void)
87 {
88 	return BCM1250_M3_WAR;
89 }
90 
91 static inline int __maybe_unused r10000_llsc_war(void)
92 {
93 	return R10000_LLSC_WAR;
94 }
95 
96 static int use_bbit_insns(void)
97 {
98 	switch (current_cpu_type()) {
99 	case CPU_CAVIUM_OCTEON:
100 	case CPU_CAVIUM_OCTEON_PLUS:
101 	case CPU_CAVIUM_OCTEON2:
102 	case CPU_CAVIUM_OCTEON3:
103 		return 1;
104 	default:
105 		return 0;
106 	}
107 }
108 
109 static int use_lwx_insns(void)
110 {
111 	switch (current_cpu_type()) {
112 	case CPU_CAVIUM_OCTEON2:
113 	case CPU_CAVIUM_OCTEON3:
114 		return 1;
115 	default:
116 		return 0;
117 	}
118 }
119 #if defined(CONFIG_CAVIUM_OCTEON_CVMSEG_SIZE) && \
120     CONFIG_CAVIUM_OCTEON_CVMSEG_SIZE > 0
121 static bool scratchpad_available(void)
122 {
123 	return true;
124 }
125 static int scratchpad_offset(int i)
126 {
127 	/*
128 	 * CVMSEG starts at address -32768 and extends for
129 	 * CAVIUM_OCTEON_CVMSEG_SIZE 128 byte cache lines.
130 	 */
131 	i += 1; /* Kernel use starts at the top and works down. */
132 	return CONFIG_CAVIUM_OCTEON_CVMSEG_SIZE * 128 - (8 * i) - 32768;
133 }
134 #else
135 static bool scratchpad_available(void)
136 {
137 	return false;
138 }
139 static int scratchpad_offset(int i)
140 {
141 	BUG();
142 	/* Really unreachable, but evidently some GCC want this. */
143 	return 0;
144 }
145 #endif
146 /*
147  * Found by experiment: At least some revisions of the 4kc throw under
148  * some circumstances a machine check exception, triggered by invalid
149  * values in the index register.  Delaying the tlbp instruction until
150  * after the next branch,  plus adding an additional nop in front of
151  * tlbwi/tlbwr avoids the invalid index register values. Nobody knows
152  * why; it's not an issue caused by the core RTL.
153  *
154  */
155 static int m4kc_tlbp_war(void)
156 {
157 	return current_cpu_type() == CPU_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 void *start, const void *end)
258 {
259 	unsigned int count = (end - start) / sizeof(u32);
260 	const u32 *handler = start;
261 	int i;
262 
263 	pr_debug("LEAF(%s)\n", symbol);
264 
265 	pr_debug("\t.set push\n");
266 	pr_debug("\t.set noreorder\n");
267 
268 	for (i = 0; i < count; i++)
269 		pr_debug("\t.word\t0x%08x\t\t# %p\n", handler[i], &handler[i]);
270 
271 	pr_debug("\t.set\tpop\n");
272 
273 	pr_debug("\tEND(%s)\n", symbol);
274 }
275 
276 /* The only general purpose registers allowed in TLB handlers. */
277 #define K0		26
278 #define K1		27
279 
280 /* Some CP0 registers */
281 #define C0_INDEX	0, 0
282 #define C0_ENTRYLO0	2, 0
283 #define C0_TCBIND	2, 2
284 #define C0_ENTRYLO1	3, 0
285 #define C0_CONTEXT	4, 0
286 #define C0_PAGEMASK	5, 0
287 #define C0_PWBASE	5, 5
288 #define C0_PWFIELD	5, 6
289 #define C0_PWSIZE	5, 7
290 #define C0_PWCTL	6, 6
291 #define C0_BADVADDR	8, 0
292 #define C0_PGD		9, 7
293 #define C0_ENTRYHI	10, 0
294 #define C0_EPC		14, 0
295 #define C0_XCONTEXT	20, 0
296 
297 #ifdef CONFIG_64BIT
298 # define GET_CONTEXT(buf, reg) UASM_i_MFC0(buf, reg, C0_XCONTEXT)
299 #else
300 # define GET_CONTEXT(buf, reg) UASM_i_MFC0(buf, reg, C0_CONTEXT)
301 #endif
302 
303 /* The worst case length of the handler is around 18 instructions for
304  * R3000-style TLBs and up to 63 instructions for R4000-style TLBs.
305  * Maximum space available is 32 instructions for R3000 and 64
306  * instructions for R4000.
307  *
308  * We deliberately chose a buffer size of 128, so we won't scribble
309  * over anything important on overflow before we panic.
310  */
311 static u32 tlb_handler[128];
312 
313 /* simply assume worst case size for labels and relocs */
314 static struct uasm_label labels[128];
315 static struct uasm_reloc relocs[128];
316 
317 static int check_for_high_segbits;
318 static bool fill_includes_sw_bits;
319 
320 static unsigned int kscratch_used_mask;
321 
322 static inline int __maybe_unused c0_kscratch(void)
323 {
324 	switch (current_cpu_type()) {
325 	case CPU_XLP:
326 	case CPU_XLR:
327 		return 22;
328 	default:
329 		return 31;
330 	}
331 }
332 
333 static int allocate_kscratch(void)
334 {
335 	int r;
336 	unsigned int a = cpu_data[0].kscratch_mask & ~kscratch_used_mask;
337 
338 	r = ffs(a);
339 
340 	if (r == 0)
341 		return -1;
342 
343 	r--; /* make it zero based */
344 
345 	kscratch_used_mask |= (1 << r);
346 
347 	return r;
348 }
349 
350 static int scratch_reg;
351 int pgd_reg;
352 EXPORT_SYMBOL_GPL(pgd_reg);
353 enum vmalloc64_mode {not_refill, refill_scratch, refill_noscratch};
354 
355 static struct work_registers build_get_work_registers(u32 **p)
356 {
357 	struct work_registers r;
358 
359 	if (scratch_reg >= 0) {
360 		/* Save in CPU local C0_KScratch? */
361 		UASM_i_MTC0(p, 1, c0_kscratch(), scratch_reg);
362 		r.r1 = K0;
363 		r.r2 = K1;
364 		r.r3 = 1;
365 		return r;
366 	}
367 
368 	if (num_possible_cpus() > 1) {
369 		/* Get smp_processor_id */
370 		UASM_i_CPUID_MFC0(p, K0, SMP_CPUID_REG);
371 		UASM_i_SRL_SAFE(p, K0, K0, SMP_CPUID_REGSHIFT);
372 
373 		/* handler_reg_save index in K0 */
374 		UASM_i_SLL(p, K0, K0, ilog2(sizeof(struct tlb_reg_save)));
375 
376 		UASM_i_LA(p, K1, (long)&handler_reg_save);
377 		UASM_i_ADDU(p, K0, K0, K1);
378 	} else {
379 		UASM_i_LA(p, K0, (long)&handler_reg_save);
380 	}
381 	/* K0 now points to save area, save $1 and $2  */
382 	UASM_i_SW(p, 1, offsetof(struct tlb_reg_save, a), K0);
383 	UASM_i_SW(p, 2, offsetof(struct tlb_reg_save, b), K0);
384 
385 	r.r1 = K1;
386 	r.r2 = 1;
387 	r.r3 = 2;
388 	return r;
389 }
390 
391 static void build_restore_work_registers(u32 **p)
392 {
393 	if (scratch_reg >= 0) {
394 		UASM_i_MFC0(p, 1, c0_kscratch(), scratch_reg);
395 		return;
396 	}
397 	/* K0 already points to save area, restore $1 and $2  */
398 	UASM_i_LW(p, 1, offsetof(struct tlb_reg_save, a), K0);
399 	UASM_i_LW(p, 2, offsetof(struct tlb_reg_save, b), K0);
400 }
401 
402 #ifndef CONFIG_MIPS_PGD_C0_CONTEXT
403 
404 /*
405  * CONFIG_MIPS_PGD_C0_CONTEXT implies 64 bit and lack of pgd_current,
406  * we cannot do r3000 under these circumstances.
407  *
408  * The R3000 TLB handler is simple.
409  */
410 static void build_r3000_tlb_refill_handler(void)
411 {
412 	long pgdc = (long)pgd_current;
413 	u32 *p;
414 
415 	memset(tlb_handler, 0, sizeof(tlb_handler));
416 	p = tlb_handler;
417 
418 	uasm_i_mfc0(&p, K0, C0_BADVADDR);
419 	uasm_i_lui(&p, K1, uasm_rel_hi(pgdc)); /* cp0 delay */
420 	uasm_i_lw(&p, K1, uasm_rel_lo(pgdc), K1);
421 	uasm_i_srl(&p, K0, K0, 22); /* load delay */
422 	uasm_i_sll(&p, K0, K0, 2);
423 	uasm_i_addu(&p, K1, K1, K0);
424 	uasm_i_mfc0(&p, K0, C0_CONTEXT);
425 	uasm_i_lw(&p, K1, 0, K1); /* cp0 delay */
426 	uasm_i_andi(&p, K0, K0, 0xffc); /* load delay */
427 	uasm_i_addu(&p, K1, K1, K0);
428 	uasm_i_lw(&p, K0, 0, K1);
429 	uasm_i_nop(&p); /* load delay */
430 	uasm_i_mtc0(&p, K0, C0_ENTRYLO0);
431 	uasm_i_mfc0(&p, K1, C0_EPC); /* cp0 delay */
432 	uasm_i_tlbwr(&p); /* cp0 delay */
433 	uasm_i_jr(&p, K1);
434 	uasm_i_rfe(&p); /* branch delay */
435 
436 	if (p > tlb_handler + 32)
437 		panic("TLB refill handler space exceeded");
438 
439 	pr_debug("Wrote TLB refill handler (%u instructions).\n",
440 		 (unsigned int)(p - tlb_handler));
441 
442 	memcpy((void *)ebase, tlb_handler, 0x80);
443 	local_flush_icache_range(ebase, ebase + 0x80);
444 	dump_handler("r3000_tlb_refill", (u32 *)ebase, (u32 *)(ebase + 0x80));
445 }
446 #endif /* CONFIG_MIPS_PGD_C0_CONTEXT */
447 
448 /*
449  * The R4000 TLB handler is much more complicated. We have two
450  * consecutive handler areas with 32 instructions space each.
451  * Since they aren't used at the same time, we can overflow in the
452  * other one.To keep things simple, we first assume linear space,
453  * then we relocate it to the final handler layout as needed.
454  */
455 static u32 final_handler[64];
456 
457 /*
458  * Hazards
459  *
460  * From the IDT errata for the QED RM5230 (Nevada), processor revision 1.0:
461  * 2. A timing hazard exists for the TLBP instruction.
462  *
463  *	stalling_instruction
464  *	TLBP
465  *
466  * The JTLB is being read for the TLBP throughout the stall generated by the
467  * previous instruction. This is not really correct as the stalling instruction
468  * can modify the address used to access the JTLB.  The failure symptom is that
469  * the TLBP instruction will use an address created for the stalling instruction
470  * and not the address held in C0_ENHI and thus report the wrong results.
471  *
472  * The software work-around is to not allow the instruction preceding the TLBP
473  * to stall - make it an NOP or some other instruction guaranteed not to stall.
474  *
475  * Errata 2 will not be fixed.	This errata is also on the R5000.
476  *
477  * As if we MIPS hackers wouldn't know how to nop pipelines happy ...
478  */
479 static void __maybe_unused build_tlb_probe_entry(u32 **p)
480 {
481 	switch (current_cpu_type()) {
482 	/* Found by experiment: R4600 v2.0/R4700 needs this, too.  */
483 	case CPU_R4600:
484 	case CPU_R4700:
485 	case CPU_R5000:
486 	case CPU_NEVADA:
487 		uasm_i_nop(p);
488 		uasm_i_tlbp(p);
489 		break;
490 
491 	default:
492 		uasm_i_tlbp(p);
493 		break;
494 	}
495 }
496 
497 void build_tlb_write_entry(u32 **p, struct uasm_label **l,
498 			   struct uasm_reloc **r,
499 			   enum tlb_write_entry wmode)
500 {
501 	void(*tlbw)(u32 **) = NULL;
502 
503 	switch (wmode) {
504 	case tlb_random: tlbw = uasm_i_tlbwr; break;
505 	case tlb_indexed: tlbw = uasm_i_tlbwi; break;
506 	}
507 
508 	if (cpu_has_mips_r2_r6) {
509 		if (cpu_has_mips_r2_exec_hazard)
510 			uasm_i_ehb(p);
511 		tlbw(p);
512 		return;
513 	}
514 
515 	switch (current_cpu_type()) {
516 	case CPU_R4000PC:
517 	case CPU_R4000SC:
518 	case CPU_R4000MC:
519 	case CPU_R4400PC:
520 	case CPU_R4400SC:
521 	case CPU_R4400MC:
522 		/*
523 		 * This branch uses up a mtc0 hazard nop slot and saves
524 		 * two nops after the tlbw instruction.
525 		 */
526 		uasm_bgezl_hazard(p, r, hazard_instance);
527 		tlbw(p);
528 		uasm_bgezl_label(l, p, hazard_instance);
529 		hazard_instance++;
530 		uasm_i_nop(p);
531 		break;
532 
533 	case CPU_R4600:
534 	case CPU_R4700:
535 		uasm_i_nop(p);
536 		tlbw(p);
537 		uasm_i_nop(p);
538 		break;
539 
540 	case CPU_R5000:
541 	case CPU_NEVADA:
542 		uasm_i_nop(p); /* QED specifies 2 nops hazard */
543 		uasm_i_nop(p); /* QED specifies 2 nops hazard */
544 		tlbw(p);
545 		break;
546 
547 	case CPU_R4300:
548 	case CPU_5KC:
549 	case CPU_TX49XX:
550 	case CPU_PR4450:
551 	case CPU_XLR:
552 		uasm_i_nop(p);
553 		tlbw(p);
554 		break;
555 
556 	case CPU_R10000:
557 	case CPU_R12000:
558 	case CPU_R14000:
559 	case CPU_R16000:
560 	case CPU_4KC:
561 	case CPU_4KEC:
562 	case CPU_M14KC:
563 	case CPU_M14KEC:
564 	case CPU_SB1:
565 	case CPU_SB1A:
566 	case CPU_4KSC:
567 	case CPU_20KC:
568 	case CPU_25KF:
569 	case CPU_BMIPS32:
570 	case CPU_BMIPS3300:
571 	case CPU_BMIPS4350:
572 	case CPU_BMIPS4380:
573 	case CPU_BMIPS5000:
574 	case CPU_LOONGSON2:
575 	case CPU_LOONGSON3:
576 	case CPU_R5500:
577 		if (m4kc_tlbp_war())
578 			uasm_i_nop(p);
579 		/* fall through */
580 	case CPU_ALCHEMY:
581 		tlbw(p);
582 		break;
583 
584 	case CPU_RM7000:
585 		uasm_i_nop(p);
586 		uasm_i_nop(p);
587 		uasm_i_nop(p);
588 		uasm_i_nop(p);
589 		tlbw(p);
590 		break;
591 
592 	case CPU_VR4111:
593 	case CPU_VR4121:
594 	case CPU_VR4122:
595 	case CPU_VR4181:
596 	case CPU_VR4181A:
597 		uasm_i_nop(p);
598 		uasm_i_nop(p);
599 		tlbw(p);
600 		uasm_i_nop(p);
601 		uasm_i_nop(p);
602 		break;
603 
604 	case CPU_VR4131:
605 	case CPU_VR4133:
606 	case CPU_R5432:
607 		uasm_i_nop(p);
608 		uasm_i_nop(p);
609 		tlbw(p);
610 		break;
611 
612 	case CPU_JZRISC:
613 		tlbw(p);
614 		uasm_i_nop(p);
615 		break;
616 
617 	default:
618 		panic("No TLB refill handler yet (CPU type: %d)",
619 		      current_cpu_type());
620 		break;
621 	}
622 }
623 EXPORT_SYMBOL_GPL(build_tlb_write_entry);
624 
625 static __maybe_unused void build_convert_pte_to_entrylo(u32 **p,
626 							unsigned int reg)
627 {
628 	if (_PAGE_GLOBAL_SHIFT == 0) {
629 		/* pte_t is already in EntryLo format */
630 		return;
631 	}
632 
633 	if (cpu_has_rixi && _PAGE_NO_EXEC) {
634 		if (fill_includes_sw_bits) {
635 			UASM_i_ROTR(p, reg, reg, ilog2(_PAGE_GLOBAL));
636 		} else {
637 			UASM_i_SRL(p, reg, reg, ilog2(_PAGE_NO_EXEC));
638 			UASM_i_ROTR(p, reg, reg,
639 				    ilog2(_PAGE_GLOBAL) - ilog2(_PAGE_NO_EXEC));
640 		}
641 	} else {
642 #ifdef CONFIG_PHYS_ADDR_T_64BIT
643 		uasm_i_dsrl_safe(p, reg, reg, ilog2(_PAGE_GLOBAL));
644 #else
645 		UASM_i_SRL(p, reg, reg, ilog2(_PAGE_GLOBAL));
646 #endif
647 	}
648 }
649 
650 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
651 
652 static void build_restore_pagemask(u32 **p, struct uasm_reloc **r,
653 				   unsigned int tmp, enum label_id lid,
654 				   int restore_scratch)
655 {
656 	if (restore_scratch) {
657 		/* Reset default page size */
658 		if (PM_DEFAULT_MASK >> 16) {
659 			uasm_i_lui(p, tmp, PM_DEFAULT_MASK >> 16);
660 			uasm_i_ori(p, tmp, tmp, PM_DEFAULT_MASK & 0xffff);
661 			uasm_i_mtc0(p, tmp, C0_PAGEMASK);
662 			uasm_il_b(p, r, lid);
663 		} else if (PM_DEFAULT_MASK) {
664 			uasm_i_ori(p, tmp, 0, PM_DEFAULT_MASK);
665 			uasm_i_mtc0(p, tmp, C0_PAGEMASK);
666 			uasm_il_b(p, r, lid);
667 		} else {
668 			uasm_i_mtc0(p, 0, C0_PAGEMASK);
669 			uasm_il_b(p, r, lid);
670 		}
671 		if (scratch_reg >= 0)
672 			UASM_i_MFC0(p, 1, c0_kscratch(), scratch_reg);
673 		else
674 			UASM_i_LW(p, 1, scratchpad_offset(0), 0);
675 	} else {
676 		/* Reset default page size */
677 		if (PM_DEFAULT_MASK >> 16) {
678 			uasm_i_lui(p, tmp, PM_DEFAULT_MASK >> 16);
679 			uasm_i_ori(p, tmp, tmp, PM_DEFAULT_MASK & 0xffff);
680 			uasm_il_b(p, r, lid);
681 			uasm_i_mtc0(p, tmp, C0_PAGEMASK);
682 		} else if (PM_DEFAULT_MASK) {
683 			uasm_i_ori(p, tmp, 0, PM_DEFAULT_MASK);
684 			uasm_il_b(p, r, lid);
685 			uasm_i_mtc0(p, tmp, C0_PAGEMASK);
686 		} else {
687 			uasm_il_b(p, r, lid);
688 			uasm_i_mtc0(p, 0, C0_PAGEMASK);
689 		}
690 	}
691 }
692 
693 static void build_huge_tlb_write_entry(u32 **p, struct uasm_label **l,
694 				       struct uasm_reloc **r,
695 				       unsigned int tmp,
696 				       enum tlb_write_entry wmode,
697 				       int restore_scratch)
698 {
699 	/* Set huge page tlb entry size */
700 	uasm_i_lui(p, tmp, PM_HUGE_MASK >> 16);
701 	uasm_i_ori(p, tmp, tmp, PM_HUGE_MASK & 0xffff);
702 	uasm_i_mtc0(p, tmp, C0_PAGEMASK);
703 
704 	build_tlb_write_entry(p, l, r, wmode);
705 
706 	build_restore_pagemask(p, r, tmp, label_leave, restore_scratch);
707 }
708 
709 /*
710  * Check if Huge PTE is present, if so then jump to LABEL.
711  */
712 static void
713 build_is_huge_pte(u32 **p, struct uasm_reloc **r, unsigned int tmp,
714 		  unsigned int pmd, int lid)
715 {
716 	UASM_i_LW(p, tmp, 0, pmd);
717 	if (use_bbit_insns()) {
718 		uasm_il_bbit1(p, r, tmp, ilog2(_PAGE_HUGE), lid);
719 	} else {
720 		uasm_i_andi(p, tmp, tmp, _PAGE_HUGE);
721 		uasm_il_bnez(p, r, tmp, lid);
722 	}
723 }
724 
725 static void build_huge_update_entries(u32 **p, unsigned int pte,
726 				      unsigned int tmp)
727 {
728 	int small_sequence;
729 
730 	/*
731 	 * A huge PTE describes an area the size of the
732 	 * configured huge page size. This is twice the
733 	 * of the large TLB entry size we intend to use.
734 	 * A TLB entry half the size of the configured
735 	 * huge page size is configured into entrylo0
736 	 * and entrylo1 to cover the contiguous huge PTE
737 	 * address space.
738 	 */
739 	small_sequence = (HPAGE_SIZE >> 7) < 0x10000;
740 
741 	/* We can clobber tmp.	It isn't used after this.*/
742 	if (!small_sequence)
743 		uasm_i_lui(p, tmp, HPAGE_SIZE >> (7 + 16));
744 
745 	build_convert_pte_to_entrylo(p, pte);
746 	UASM_i_MTC0(p, pte, C0_ENTRYLO0); /* load it */
747 	/* convert to entrylo1 */
748 	if (small_sequence)
749 		UASM_i_ADDIU(p, pte, pte, HPAGE_SIZE >> 7);
750 	else
751 		UASM_i_ADDU(p, pte, pte, tmp);
752 
753 	UASM_i_MTC0(p, pte, C0_ENTRYLO1); /* load it */
754 }
755 
756 static void build_huge_handler_tail(u32 **p, struct uasm_reloc **r,
757 				    struct uasm_label **l,
758 				    unsigned int pte,
759 				    unsigned int ptr,
760 				    unsigned int flush)
761 {
762 #ifdef CONFIG_SMP
763 	UASM_i_SC(p, pte, 0, ptr);
764 	uasm_il_beqz(p, r, pte, label_tlb_huge_update);
765 	UASM_i_LW(p, pte, 0, ptr); /* Needed because SC killed our PTE */
766 #else
767 	UASM_i_SW(p, pte, 0, ptr);
768 #endif
769 	if (cpu_has_ftlb && flush) {
770 		BUG_ON(!cpu_has_tlbinv);
771 
772 		UASM_i_MFC0(p, ptr, C0_ENTRYHI);
773 		uasm_i_ori(p, ptr, ptr, MIPS_ENTRYHI_EHINV);
774 		UASM_i_MTC0(p, ptr, C0_ENTRYHI);
775 		build_tlb_write_entry(p, l, r, tlb_indexed);
776 
777 		uasm_i_xori(p, ptr, ptr, MIPS_ENTRYHI_EHINV);
778 		UASM_i_MTC0(p, ptr, C0_ENTRYHI);
779 		build_huge_update_entries(p, pte, ptr);
780 		build_huge_tlb_write_entry(p, l, r, pte, tlb_random, 0);
781 
782 		return;
783 	}
784 
785 	build_huge_update_entries(p, pte, ptr);
786 	build_huge_tlb_write_entry(p, l, r, pte, tlb_indexed, 0);
787 }
788 #endif /* CONFIG_MIPS_HUGE_TLB_SUPPORT */
789 
790 #ifdef CONFIG_64BIT
791 /*
792  * TMP and PTR are scratch.
793  * TMP will be clobbered, PTR will hold the pmd entry.
794  */
795 void build_get_pmde64(u32 **p, struct uasm_label **l, struct uasm_reloc **r,
796 		      unsigned int tmp, unsigned int ptr)
797 {
798 #ifndef CONFIG_MIPS_PGD_C0_CONTEXT
799 	long pgdc = (long)pgd_current;
800 #endif
801 	/*
802 	 * The vmalloc handling is not in the hotpath.
803 	 */
804 	uasm_i_dmfc0(p, tmp, C0_BADVADDR);
805 
806 	if (check_for_high_segbits) {
807 		/*
808 		 * The kernel currently implicitely assumes that the
809 		 * MIPS SEGBITS parameter for the processor is
810 		 * (PGDIR_SHIFT+PGDIR_BITS) or less, and will never
811 		 * allocate virtual addresses outside the maximum
812 		 * range for SEGBITS = (PGDIR_SHIFT+PGDIR_BITS). But
813 		 * that doesn't prevent user code from accessing the
814 		 * higher xuseg addresses.  Here, we make sure that
815 		 * everything but the lower xuseg addresses goes down
816 		 * the module_alloc/vmalloc path.
817 		 */
818 		uasm_i_dsrl_safe(p, ptr, tmp, PGDIR_SHIFT + PGD_ORDER + PAGE_SHIFT - 3);
819 		uasm_il_bnez(p, r, ptr, label_vmalloc);
820 	} else {
821 		uasm_il_bltz(p, r, tmp, label_vmalloc);
822 	}
823 	/* No uasm_i_nop needed here, since the next insn doesn't touch TMP. */
824 
825 	if (pgd_reg != -1) {
826 		/* pgd is in pgd_reg */
827 		if (cpu_has_ldpte)
828 			UASM_i_MFC0(p, ptr, C0_PWBASE);
829 		else
830 			UASM_i_MFC0(p, ptr, c0_kscratch(), pgd_reg);
831 	} else {
832 #if defined(CONFIG_MIPS_PGD_C0_CONTEXT)
833 		/*
834 		 * &pgd << 11 stored in CONTEXT [23..63].
835 		 */
836 		UASM_i_MFC0(p, ptr, C0_CONTEXT);
837 
838 		/* Clear lower 23 bits of context. */
839 		uasm_i_dins(p, ptr, 0, 0, 23);
840 
841 		/* 1 0	1 0 1  << 6  xkphys cached */
842 		uasm_i_ori(p, ptr, ptr, 0x540);
843 		uasm_i_drotr(p, ptr, ptr, 11);
844 #elif defined(CONFIG_SMP)
845 		UASM_i_CPUID_MFC0(p, ptr, SMP_CPUID_REG);
846 		uasm_i_dsrl_safe(p, ptr, ptr, SMP_CPUID_PTRSHIFT);
847 		UASM_i_LA_mostly(p, tmp, pgdc);
848 		uasm_i_daddu(p, ptr, ptr, tmp);
849 		uasm_i_dmfc0(p, tmp, C0_BADVADDR);
850 		uasm_i_ld(p, ptr, uasm_rel_lo(pgdc), ptr);
851 #else
852 		UASM_i_LA_mostly(p, ptr, pgdc);
853 		uasm_i_ld(p, ptr, uasm_rel_lo(pgdc), ptr);
854 #endif
855 	}
856 
857 	uasm_l_vmalloc_done(l, *p);
858 
859 	/* get pgd offset in bytes */
860 	uasm_i_dsrl_safe(p, tmp, tmp, PGDIR_SHIFT - 3);
861 
862 	uasm_i_andi(p, tmp, tmp, (PTRS_PER_PGD - 1)<<3);
863 	uasm_i_daddu(p, ptr, ptr, tmp); /* add in pgd offset */
864 #ifndef __PAGETABLE_PUD_FOLDED
865 	uasm_i_dmfc0(p, tmp, C0_BADVADDR); /* get faulting address */
866 	uasm_i_ld(p, ptr, 0, ptr); /* get pud pointer */
867 	uasm_i_dsrl_safe(p, tmp, tmp, PUD_SHIFT - 3); /* get pud offset in bytes */
868 	uasm_i_andi(p, tmp, tmp, (PTRS_PER_PUD - 1) << 3);
869 	uasm_i_daddu(p, ptr, ptr, tmp); /* add in pud offset */
870 #endif
871 #ifndef __PAGETABLE_PMD_FOLDED
872 	uasm_i_dmfc0(p, tmp, C0_BADVADDR); /* get faulting address */
873 	uasm_i_ld(p, ptr, 0, ptr); /* get pmd pointer */
874 	uasm_i_dsrl_safe(p, tmp, tmp, PMD_SHIFT-3); /* get pmd offset in bytes */
875 	uasm_i_andi(p, tmp, tmp, (PTRS_PER_PMD - 1)<<3);
876 	uasm_i_daddu(p, ptr, ptr, tmp); /* add in pmd offset */
877 #endif
878 }
879 EXPORT_SYMBOL_GPL(build_get_pmde64);
880 
881 /*
882  * BVADDR is the faulting address, PTR is scratch.
883  * PTR will hold the pgd for vmalloc.
884  */
885 static void
886 build_get_pgd_vmalloc64(u32 **p, struct uasm_label **l, struct uasm_reloc **r,
887 			unsigned int bvaddr, unsigned int ptr,
888 			enum vmalloc64_mode mode)
889 {
890 	long swpd = (long)swapper_pg_dir;
891 	int single_insn_swpd;
892 	int did_vmalloc_branch = 0;
893 
894 	single_insn_swpd = uasm_in_compat_space_p(swpd) && !uasm_rel_lo(swpd);
895 
896 	uasm_l_vmalloc(l, *p);
897 
898 	if (mode != not_refill && check_for_high_segbits) {
899 		if (single_insn_swpd) {
900 			uasm_il_bltz(p, r, bvaddr, label_vmalloc_done);
901 			uasm_i_lui(p, ptr, uasm_rel_hi(swpd));
902 			did_vmalloc_branch = 1;
903 			/* fall through */
904 		} else {
905 			uasm_il_bgez(p, r, bvaddr, label_large_segbits_fault);
906 		}
907 	}
908 	if (!did_vmalloc_branch) {
909 		if (single_insn_swpd) {
910 			uasm_il_b(p, r, label_vmalloc_done);
911 			uasm_i_lui(p, ptr, uasm_rel_hi(swpd));
912 		} else {
913 			UASM_i_LA_mostly(p, ptr, swpd);
914 			uasm_il_b(p, r, label_vmalloc_done);
915 			if (uasm_in_compat_space_p(swpd))
916 				uasm_i_addiu(p, ptr, ptr, uasm_rel_lo(swpd));
917 			else
918 				uasm_i_daddiu(p, ptr, ptr, uasm_rel_lo(swpd));
919 		}
920 	}
921 	if (mode != not_refill && check_for_high_segbits) {
922 		uasm_l_large_segbits_fault(l, *p);
923 		/*
924 		 * We get here if we are an xsseg address, or if we are
925 		 * an xuseg address above (PGDIR_SHIFT+PGDIR_BITS) boundary.
926 		 *
927 		 * Ignoring xsseg (assume disabled so would generate
928 		 * (address errors?), the only remaining possibility
929 		 * is the upper xuseg addresses.  On processors with
930 		 * TLB_SEGBITS <= PGDIR_SHIFT+PGDIR_BITS, these
931 		 * addresses would have taken an address error. We try
932 		 * to mimic that here by taking a load/istream page
933 		 * fault.
934 		 */
935 		UASM_i_LA(p, ptr, (unsigned long)tlb_do_page_fault_0);
936 		uasm_i_jr(p, ptr);
937 
938 		if (mode == refill_scratch) {
939 			if (scratch_reg >= 0)
940 				UASM_i_MFC0(p, 1, c0_kscratch(), scratch_reg);
941 			else
942 				UASM_i_LW(p, 1, scratchpad_offset(0), 0);
943 		} else {
944 			uasm_i_nop(p);
945 		}
946 	}
947 }
948 
949 #else /* !CONFIG_64BIT */
950 
951 /*
952  * TMP and PTR are scratch.
953  * TMP will be clobbered, PTR will hold the pgd entry.
954  */
955 void build_get_pgde32(u32 **p, unsigned int tmp, unsigned int ptr)
956 {
957 	if (pgd_reg != -1) {
958 		/* pgd is in pgd_reg */
959 		uasm_i_mfc0(p, ptr, c0_kscratch(), pgd_reg);
960 		uasm_i_mfc0(p, tmp, C0_BADVADDR); /* get faulting address */
961 	} else {
962 		long pgdc = (long)pgd_current;
963 
964 		/* 32 bit SMP has smp_processor_id() stored in CONTEXT. */
965 #ifdef CONFIG_SMP
966 		uasm_i_mfc0(p, ptr, SMP_CPUID_REG);
967 		UASM_i_LA_mostly(p, tmp, pgdc);
968 		uasm_i_srl(p, ptr, ptr, SMP_CPUID_PTRSHIFT);
969 		uasm_i_addu(p, ptr, tmp, ptr);
970 #else
971 		UASM_i_LA_mostly(p, ptr, pgdc);
972 #endif
973 		uasm_i_mfc0(p, tmp, C0_BADVADDR); /* get faulting address */
974 		uasm_i_lw(p, ptr, uasm_rel_lo(pgdc), ptr);
975 	}
976 	uasm_i_srl(p, tmp, tmp, PGDIR_SHIFT); /* get pgd only bits */
977 	uasm_i_sll(p, tmp, tmp, PGD_T_LOG2);
978 	uasm_i_addu(p, ptr, ptr, tmp); /* add in pgd offset */
979 }
980 EXPORT_SYMBOL_GPL(build_get_pgde32);
981 
982 #endif /* !CONFIG_64BIT */
983 
984 static void build_adjust_context(u32 **p, unsigned int ctx)
985 {
986 	unsigned int shift = 4 - (PTE_T_LOG2 + 1) + PAGE_SHIFT - 12;
987 	unsigned int mask = (PTRS_PER_PTE / 2 - 1) << (PTE_T_LOG2 + 1);
988 
989 	switch (current_cpu_type()) {
990 	case CPU_VR41XX:
991 	case CPU_VR4111:
992 	case CPU_VR4121:
993 	case CPU_VR4122:
994 	case CPU_VR4131:
995 	case CPU_VR4181:
996 	case CPU_VR4181A:
997 	case CPU_VR4133:
998 		shift += 2;
999 		break;
1000 
1001 	default:
1002 		break;
1003 	}
1004 
1005 	if (shift)
1006 		UASM_i_SRL(p, ctx, ctx, shift);
1007 	uasm_i_andi(p, ctx, ctx, mask);
1008 }
1009 
1010 void build_get_ptep(u32 **p, unsigned int tmp, unsigned int ptr)
1011 {
1012 	/*
1013 	 * Bug workaround for the Nevada. It seems as if under certain
1014 	 * circumstances the move from cp0_context might produce a
1015 	 * bogus result when the mfc0 instruction and its consumer are
1016 	 * in a different cacheline or a load instruction, probably any
1017 	 * memory reference, is between them.
1018 	 */
1019 	switch (current_cpu_type()) {
1020 	case CPU_NEVADA:
1021 		UASM_i_LW(p, ptr, 0, ptr);
1022 		GET_CONTEXT(p, tmp); /* get context reg */
1023 		break;
1024 
1025 	default:
1026 		GET_CONTEXT(p, tmp); /* get context reg */
1027 		UASM_i_LW(p, ptr, 0, ptr);
1028 		break;
1029 	}
1030 
1031 	build_adjust_context(p, tmp);
1032 	UASM_i_ADDU(p, ptr, ptr, tmp); /* add in offset */
1033 }
1034 EXPORT_SYMBOL_GPL(build_get_ptep);
1035 
1036 void build_update_entries(u32 **p, unsigned int tmp, unsigned int ptep)
1037 {
1038 	int pte_off_even = 0;
1039 	int pte_off_odd = sizeof(pte_t);
1040 
1041 #if defined(CONFIG_CPU_MIPS32) && defined(CONFIG_PHYS_ADDR_T_64BIT)
1042 	/* The low 32 bits of EntryLo is stored in pte_high */
1043 	pte_off_even += offsetof(pte_t, pte_high);
1044 	pte_off_odd += offsetof(pte_t, pte_high);
1045 #endif
1046 
1047 	if (IS_ENABLED(CONFIG_XPA)) {
1048 		uasm_i_lw(p, tmp, pte_off_even, ptep); /* even pte */
1049 		UASM_i_ROTR(p, tmp, tmp, ilog2(_PAGE_GLOBAL));
1050 		UASM_i_MTC0(p, tmp, C0_ENTRYLO0);
1051 
1052 		if (cpu_has_xpa && !mips_xpa_disabled) {
1053 			uasm_i_lw(p, tmp, 0, ptep);
1054 			uasm_i_ext(p, tmp, tmp, 0, 24);
1055 			uasm_i_mthc0(p, tmp, C0_ENTRYLO0);
1056 		}
1057 
1058 		uasm_i_lw(p, tmp, pte_off_odd, ptep); /* odd pte */
1059 		UASM_i_ROTR(p, tmp, tmp, ilog2(_PAGE_GLOBAL));
1060 		UASM_i_MTC0(p, tmp, C0_ENTRYLO1);
1061 
1062 		if (cpu_has_xpa && !mips_xpa_disabled) {
1063 			uasm_i_lw(p, tmp, sizeof(pte_t), ptep);
1064 			uasm_i_ext(p, tmp, tmp, 0, 24);
1065 			uasm_i_mthc0(p, tmp, C0_ENTRYLO1);
1066 		}
1067 		return;
1068 	}
1069 
1070 	UASM_i_LW(p, tmp, pte_off_even, ptep); /* get even pte */
1071 	UASM_i_LW(p, ptep, pte_off_odd, ptep); /* get odd pte */
1072 	if (r45k_bvahwbug())
1073 		build_tlb_probe_entry(p);
1074 	build_convert_pte_to_entrylo(p, tmp);
1075 	if (r4k_250MHZhwbug())
1076 		UASM_i_MTC0(p, 0, C0_ENTRYLO0);
1077 	UASM_i_MTC0(p, tmp, C0_ENTRYLO0); /* load it */
1078 	build_convert_pte_to_entrylo(p, ptep);
1079 	if (r45k_bvahwbug())
1080 		uasm_i_mfc0(p, tmp, C0_INDEX);
1081 	if (r4k_250MHZhwbug())
1082 		UASM_i_MTC0(p, 0, C0_ENTRYLO1);
1083 	UASM_i_MTC0(p, ptep, C0_ENTRYLO1); /* load it */
1084 }
1085 EXPORT_SYMBOL_GPL(build_update_entries);
1086 
1087 struct mips_huge_tlb_info {
1088 	int huge_pte;
1089 	int restore_scratch;
1090 	bool need_reload_pte;
1091 };
1092 
1093 static struct mips_huge_tlb_info
1094 build_fast_tlb_refill_handler (u32 **p, struct uasm_label **l,
1095 			       struct uasm_reloc **r, unsigned int tmp,
1096 			       unsigned int ptr, int c0_scratch_reg)
1097 {
1098 	struct mips_huge_tlb_info rv;
1099 	unsigned int even, odd;
1100 	int vmalloc_branch_delay_filled = 0;
1101 	const int scratch = 1; /* Our extra working register */
1102 
1103 	rv.huge_pte = scratch;
1104 	rv.restore_scratch = 0;
1105 	rv.need_reload_pte = false;
1106 
1107 	if (check_for_high_segbits) {
1108 		UASM_i_MFC0(p, tmp, C0_BADVADDR);
1109 
1110 		if (pgd_reg != -1)
1111 			UASM_i_MFC0(p, ptr, c0_kscratch(), pgd_reg);
1112 		else
1113 			UASM_i_MFC0(p, ptr, C0_CONTEXT);
1114 
1115 		if (c0_scratch_reg >= 0)
1116 			UASM_i_MTC0(p, scratch, c0_kscratch(), c0_scratch_reg);
1117 		else
1118 			UASM_i_SW(p, scratch, scratchpad_offset(0), 0);
1119 
1120 		uasm_i_dsrl_safe(p, scratch, tmp,
1121 				 PGDIR_SHIFT + PGD_ORDER + PAGE_SHIFT - 3);
1122 		uasm_il_bnez(p, r, scratch, label_vmalloc);
1123 
1124 		if (pgd_reg == -1) {
1125 			vmalloc_branch_delay_filled = 1;
1126 			/* Clear lower 23 bits of context. */
1127 			uasm_i_dins(p, ptr, 0, 0, 23);
1128 		}
1129 	} else {
1130 		if (pgd_reg != -1)
1131 			UASM_i_MFC0(p, ptr, c0_kscratch(), pgd_reg);
1132 		else
1133 			UASM_i_MFC0(p, ptr, C0_CONTEXT);
1134 
1135 		UASM_i_MFC0(p, tmp, C0_BADVADDR);
1136 
1137 		if (c0_scratch_reg >= 0)
1138 			UASM_i_MTC0(p, scratch, c0_kscratch(), c0_scratch_reg);
1139 		else
1140 			UASM_i_SW(p, scratch, scratchpad_offset(0), 0);
1141 
1142 		if (pgd_reg == -1)
1143 			/* Clear lower 23 bits of context. */
1144 			uasm_i_dins(p, ptr, 0, 0, 23);
1145 
1146 		uasm_il_bltz(p, r, tmp, label_vmalloc);
1147 	}
1148 
1149 	if (pgd_reg == -1) {
1150 		vmalloc_branch_delay_filled = 1;
1151 		/* 1 0	1 0 1  << 6  xkphys cached */
1152 		uasm_i_ori(p, ptr, ptr, 0x540);
1153 		uasm_i_drotr(p, ptr, ptr, 11);
1154 	}
1155 
1156 #ifdef __PAGETABLE_PMD_FOLDED
1157 #define LOC_PTEP scratch
1158 #else
1159 #define LOC_PTEP ptr
1160 #endif
1161 
1162 	if (!vmalloc_branch_delay_filled)
1163 		/* get pgd offset in bytes */
1164 		uasm_i_dsrl_safe(p, scratch, tmp, PGDIR_SHIFT - 3);
1165 
1166 	uasm_l_vmalloc_done(l, *p);
1167 
1168 	/*
1169 	 *			   tmp		ptr
1170 	 * fall-through case =	 badvaddr  *pgd_current
1171 	 * vmalloc case	     =	 badvaddr  swapper_pg_dir
1172 	 */
1173 
1174 	if (vmalloc_branch_delay_filled)
1175 		/* get pgd offset in bytes */
1176 		uasm_i_dsrl_safe(p, scratch, tmp, PGDIR_SHIFT - 3);
1177 
1178 #ifdef __PAGETABLE_PMD_FOLDED
1179 	GET_CONTEXT(p, tmp); /* get context reg */
1180 #endif
1181 	uasm_i_andi(p, scratch, scratch, (PTRS_PER_PGD - 1) << 3);
1182 
1183 	if (use_lwx_insns()) {
1184 		UASM_i_LWX(p, LOC_PTEP, scratch, ptr);
1185 	} else {
1186 		uasm_i_daddu(p, ptr, ptr, scratch); /* add in pgd offset */
1187 		uasm_i_ld(p, LOC_PTEP, 0, ptr); /* get pmd pointer */
1188 	}
1189 
1190 #ifndef __PAGETABLE_PUD_FOLDED
1191 	/* get pud offset in bytes */
1192 	uasm_i_dsrl_safe(p, scratch, tmp, PUD_SHIFT - 3);
1193 	uasm_i_andi(p, scratch, scratch, (PTRS_PER_PUD - 1) << 3);
1194 
1195 	if (use_lwx_insns()) {
1196 		UASM_i_LWX(p, ptr, scratch, ptr);
1197 	} else {
1198 		uasm_i_daddu(p, ptr, ptr, scratch); /* add in pmd offset */
1199 		UASM_i_LW(p, ptr, 0, ptr);
1200 	}
1201 	/* ptr contains a pointer to PMD entry */
1202 	/* tmp contains the address */
1203 #endif
1204 
1205 #ifndef __PAGETABLE_PMD_FOLDED
1206 	/* get pmd offset in bytes */
1207 	uasm_i_dsrl_safe(p, scratch, tmp, PMD_SHIFT - 3);
1208 	uasm_i_andi(p, scratch, scratch, (PTRS_PER_PMD - 1) << 3);
1209 	GET_CONTEXT(p, tmp); /* get context reg */
1210 
1211 	if (use_lwx_insns()) {
1212 		UASM_i_LWX(p, scratch, scratch, ptr);
1213 	} else {
1214 		uasm_i_daddu(p, ptr, ptr, scratch); /* add in pmd offset */
1215 		UASM_i_LW(p, scratch, 0, ptr);
1216 	}
1217 #endif
1218 	/* Adjust the context during the load latency. */
1219 	build_adjust_context(p, tmp);
1220 
1221 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
1222 	uasm_il_bbit1(p, r, scratch, ilog2(_PAGE_HUGE), label_tlb_huge_update);
1223 	/*
1224 	 * The in the LWX case we don't want to do the load in the
1225 	 * delay slot.	It cannot issue in the same cycle and may be
1226 	 * speculative and unneeded.
1227 	 */
1228 	if (use_lwx_insns())
1229 		uasm_i_nop(p);
1230 #endif /* CONFIG_MIPS_HUGE_TLB_SUPPORT */
1231 
1232 
1233 	/* build_update_entries */
1234 	if (use_lwx_insns()) {
1235 		even = ptr;
1236 		odd = tmp;
1237 		UASM_i_LWX(p, even, scratch, tmp);
1238 		UASM_i_ADDIU(p, tmp, tmp, sizeof(pte_t));
1239 		UASM_i_LWX(p, odd, scratch, tmp);
1240 	} else {
1241 		UASM_i_ADDU(p, ptr, scratch, tmp); /* add in offset */
1242 		even = tmp;
1243 		odd = ptr;
1244 		UASM_i_LW(p, even, 0, ptr); /* get even pte */
1245 		UASM_i_LW(p, odd, sizeof(pte_t), ptr); /* get odd pte */
1246 	}
1247 	if (cpu_has_rixi) {
1248 		uasm_i_drotr(p, even, even, ilog2(_PAGE_GLOBAL));
1249 		UASM_i_MTC0(p, even, C0_ENTRYLO0); /* load it */
1250 		uasm_i_drotr(p, odd, odd, ilog2(_PAGE_GLOBAL));
1251 	} else {
1252 		uasm_i_dsrl_safe(p, even, even, ilog2(_PAGE_GLOBAL));
1253 		UASM_i_MTC0(p, even, C0_ENTRYLO0); /* load it */
1254 		uasm_i_dsrl_safe(p, odd, odd, ilog2(_PAGE_GLOBAL));
1255 	}
1256 	UASM_i_MTC0(p, odd, C0_ENTRYLO1); /* load it */
1257 
1258 	if (c0_scratch_reg >= 0) {
1259 		UASM_i_MFC0(p, scratch, c0_kscratch(), c0_scratch_reg);
1260 		build_tlb_write_entry(p, l, r, tlb_random);
1261 		uasm_l_leave(l, *p);
1262 		rv.restore_scratch = 1;
1263 	} else if (PAGE_SHIFT == 14 || PAGE_SHIFT == 13)  {
1264 		build_tlb_write_entry(p, l, r, tlb_random);
1265 		uasm_l_leave(l, *p);
1266 		UASM_i_LW(p, scratch, scratchpad_offset(0), 0);
1267 	} else {
1268 		UASM_i_LW(p, scratch, scratchpad_offset(0), 0);
1269 		build_tlb_write_entry(p, l, r, tlb_random);
1270 		uasm_l_leave(l, *p);
1271 		rv.restore_scratch = 1;
1272 	}
1273 
1274 	uasm_i_eret(p); /* return from trap */
1275 
1276 	return rv;
1277 }
1278 
1279 /*
1280  * For a 64-bit kernel, we are using the 64-bit XTLB refill exception
1281  * because EXL == 0.  If we wrap, we can also use the 32 instruction
1282  * slots before the XTLB refill exception handler which belong to the
1283  * unused TLB refill exception.
1284  */
1285 #define MIPS64_REFILL_INSNS 32
1286 
1287 static void build_r4000_tlb_refill_handler(void)
1288 {
1289 	u32 *p = tlb_handler;
1290 	struct uasm_label *l = labels;
1291 	struct uasm_reloc *r = relocs;
1292 	u32 *f;
1293 	unsigned int final_len;
1294 	struct mips_huge_tlb_info htlb_info __maybe_unused;
1295 	enum vmalloc64_mode vmalloc_mode __maybe_unused;
1296 
1297 	memset(tlb_handler, 0, sizeof(tlb_handler));
1298 	memset(labels, 0, sizeof(labels));
1299 	memset(relocs, 0, sizeof(relocs));
1300 	memset(final_handler, 0, sizeof(final_handler));
1301 
1302 	if (IS_ENABLED(CONFIG_64BIT) && (scratch_reg >= 0 || scratchpad_available()) && use_bbit_insns()) {
1303 		htlb_info = build_fast_tlb_refill_handler(&p, &l, &r, K0, K1,
1304 							  scratch_reg);
1305 		vmalloc_mode = refill_scratch;
1306 	} else {
1307 		htlb_info.huge_pte = K0;
1308 		htlb_info.restore_scratch = 0;
1309 		htlb_info.need_reload_pte = true;
1310 		vmalloc_mode = refill_noscratch;
1311 		/*
1312 		 * create the plain linear handler
1313 		 */
1314 		if (bcm1250_m3_war()) {
1315 			unsigned int segbits = 44;
1316 
1317 			uasm_i_dmfc0(&p, K0, C0_BADVADDR);
1318 			uasm_i_dmfc0(&p, K1, C0_ENTRYHI);
1319 			uasm_i_xor(&p, K0, K0, K1);
1320 			uasm_i_dsrl_safe(&p, K1, K0, 62);
1321 			uasm_i_dsrl_safe(&p, K0, K0, 12 + 1);
1322 			uasm_i_dsll_safe(&p, K0, K0, 64 + 12 + 1 - segbits);
1323 			uasm_i_or(&p, K0, K0, K1);
1324 			uasm_il_bnez(&p, &r, K0, label_leave);
1325 			/* No need for uasm_i_nop */
1326 		}
1327 
1328 #ifdef CONFIG_64BIT
1329 		build_get_pmde64(&p, &l, &r, K0, K1); /* get pmd in K1 */
1330 #else
1331 		build_get_pgde32(&p, K0, K1); /* get pgd in K1 */
1332 #endif
1333 
1334 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
1335 		build_is_huge_pte(&p, &r, K0, K1, label_tlb_huge_update);
1336 #endif
1337 
1338 		build_get_ptep(&p, K0, K1);
1339 		build_update_entries(&p, K0, K1);
1340 		build_tlb_write_entry(&p, &l, &r, tlb_random);
1341 		uasm_l_leave(&l, p);
1342 		uasm_i_eret(&p); /* return from trap */
1343 	}
1344 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
1345 	uasm_l_tlb_huge_update(&l, p);
1346 	if (htlb_info.need_reload_pte)
1347 		UASM_i_LW(&p, htlb_info.huge_pte, 0, K1);
1348 	build_huge_update_entries(&p, htlb_info.huge_pte, K1);
1349 	build_huge_tlb_write_entry(&p, &l, &r, K0, tlb_random,
1350 				   htlb_info.restore_scratch);
1351 #endif
1352 
1353 #ifdef CONFIG_64BIT
1354 	build_get_pgd_vmalloc64(&p, &l, &r, K0, K1, vmalloc_mode);
1355 #endif
1356 
1357 	/*
1358 	 * Overflow check: For the 64bit handler, we need at least one
1359 	 * free instruction slot for the wrap-around branch. In worst
1360 	 * case, if the intended insertion point is a delay slot, we
1361 	 * need three, with the second nop'ed and the third being
1362 	 * unused.
1363 	 */
1364 	switch (boot_cpu_type()) {
1365 	default:
1366 		if (sizeof(long) == 4) {
1367 	case CPU_LOONGSON2:
1368 		/* Loongson2 ebase is different than r4k, we have more space */
1369 			if ((p - tlb_handler) > 64)
1370 				panic("TLB refill handler space exceeded");
1371 			/*
1372 			 * Now fold the handler in the TLB refill handler space.
1373 			 */
1374 			f = final_handler;
1375 			/* Simplest case, just copy the handler. */
1376 			uasm_copy_handler(relocs, labels, tlb_handler, p, f);
1377 			final_len = p - tlb_handler;
1378 			break;
1379 		} else {
1380 			if (((p - tlb_handler) > (MIPS64_REFILL_INSNS * 2) - 1)
1381 			    || (((p - tlb_handler) > (MIPS64_REFILL_INSNS * 2) - 3)
1382 				&& uasm_insn_has_bdelay(relocs,
1383 							tlb_handler + MIPS64_REFILL_INSNS - 3)))
1384 				panic("TLB refill handler space exceeded");
1385 			/*
1386 			 * Now fold the handler in the TLB refill handler space.
1387 			 */
1388 			f = final_handler + MIPS64_REFILL_INSNS;
1389 			if ((p - tlb_handler) <= MIPS64_REFILL_INSNS) {
1390 				/* Just copy the handler. */
1391 				uasm_copy_handler(relocs, labels, tlb_handler, p, f);
1392 				final_len = p - tlb_handler;
1393 			} else {
1394 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
1395 				const enum label_id ls = label_tlb_huge_update;
1396 #else
1397 				const enum label_id ls = label_vmalloc;
1398 #endif
1399 				u32 *split;
1400 				int ov = 0;
1401 				int i;
1402 
1403 				for (i = 0; i < ARRAY_SIZE(labels) && labels[i].lab != ls; i++)
1404 					;
1405 				BUG_ON(i == ARRAY_SIZE(labels));
1406 				split = labels[i].addr;
1407 
1408 				/*
1409 				 * See if we have overflown one way or the other.
1410 				 */
1411 				if (split > tlb_handler + MIPS64_REFILL_INSNS ||
1412 				    split < p - MIPS64_REFILL_INSNS)
1413 					ov = 1;
1414 
1415 				if (ov) {
1416 					/*
1417 					 * Split two instructions before the end.  One
1418 					 * for the branch and one for the instruction
1419 					 * in the delay slot.
1420 					 */
1421 					split = tlb_handler + MIPS64_REFILL_INSNS - 2;
1422 
1423 					/*
1424 					 * If the branch would fall in a delay slot,
1425 					 * we must back up an additional instruction
1426 					 * so that it is no longer in a delay slot.
1427 					 */
1428 					if (uasm_insn_has_bdelay(relocs, split - 1))
1429 						split--;
1430 				}
1431 				/* Copy first part of the handler. */
1432 				uasm_copy_handler(relocs, labels, tlb_handler, split, f);
1433 				f += split - tlb_handler;
1434 
1435 				if (ov) {
1436 					/* Insert branch. */
1437 					uasm_l_split(&l, final_handler);
1438 					uasm_il_b(&f, &r, label_split);
1439 					if (uasm_insn_has_bdelay(relocs, split))
1440 						uasm_i_nop(&f);
1441 					else {
1442 						uasm_copy_handler(relocs, labels,
1443 								  split, split + 1, f);
1444 						uasm_move_labels(labels, f, f + 1, -1);
1445 						f++;
1446 						split++;
1447 					}
1448 				}
1449 
1450 				/* Copy the rest of the handler. */
1451 				uasm_copy_handler(relocs, labels, split, p, final_handler);
1452 				final_len = (f - (final_handler + MIPS64_REFILL_INSNS)) +
1453 					    (p - split);
1454 			}
1455 		}
1456 		break;
1457 	}
1458 
1459 	uasm_resolve_relocs(relocs, labels);
1460 	pr_debug("Wrote TLB refill handler (%u instructions).\n",
1461 		 final_len);
1462 
1463 	memcpy((void *)ebase, final_handler, 0x100);
1464 	local_flush_icache_range(ebase, ebase + 0x100);
1465 	dump_handler("r4000_tlb_refill", (u32 *)ebase, (u32 *)(ebase + 0x100));
1466 }
1467 
1468 static void setup_pw(void)
1469 {
1470 	unsigned long pgd_i, pgd_w;
1471 #ifndef __PAGETABLE_PMD_FOLDED
1472 	unsigned long pmd_i, pmd_w;
1473 #endif
1474 	unsigned long pt_i, pt_w;
1475 	unsigned long pte_i, pte_w;
1476 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
1477 	unsigned long psn;
1478 
1479 	psn = ilog2(_PAGE_HUGE);     /* bit used to indicate huge page */
1480 #endif
1481 	pgd_i = PGDIR_SHIFT;  /* 1st level PGD */
1482 #ifndef __PAGETABLE_PMD_FOLDED
1483 	pgd_w = PGDIR_SHIFT - PMD_SHIFT + PGD_ORDER;
1484 
1485 	pmd_i = PMD_SHIFT;    /* 2nd level PMD */
1486 	pmd_w = PMD_SHIFT - PAGE_SHIFT;
1487 #else
1488 	pgd_w = PGDIR_SHIFT - PAGE_SHIFT + PGD_ORDER;
1489 #endif
1490 
1491 	pt_i  = PAGE_SHIFT;    /* 3rd level PTE */
1492 	pt_w  = PAGE_SHIFT - 3;
1493 
1494 	pte_i = ilog2(_PAGE_GLOBAL);
1495 	pte_w = 0;
1496 
1497 #ifndef __PAGETABLE_PMD_FOLDED
1498 	write_c0_pwfield(pgd_i << 24 | pmd_i << 12 | pt_i << 6 | pte_i);
1499 	write_c0_pwsize(1 << 30 | pgd_w << 24 | pmd_w << 12 | pt_w << 6 | pte_w);
1500 #else
1501 	write_c0_pwfield(pgd_i << 24 | pt_i << 6 | pte_i);
1502 	write_c0_pwsize(1 << 30 | pgd_w << 24 | pt_w << 6 | pte_w);
1503 #endif
1504 
1505 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
1506 	write_c0_pwctl(1 << 6 | psn);
1507 #endif
1508 	write_c0_kpgd((long)swapper_pg_dir);
1509 	kscratch_used_mask |= (1 << 7); /* KScratch6 is used for KPGD */
1510 }
1511 
1512 static void build_loongson3_tlb_refill_handler(void)
1513 {
1514 	u32 *p = tlb_handler;
1515 	struct uasm_label *l = labels;
1516 	struct uasm_reloc *r = relocs;
1517 
1518 	memset(labels, 0, sizeof(labels));
1519 	memset(relocs, 0, sizeof(relocs));
1520 	memset(tlb_handler, 0, sizeof(tlb_handler));
1521 
1522 	if (check_for_high_segbits) {
1523 		uasm_i_dmfc0(&p, K0, C0_BADVADDR);
1524 		uasm_i_dsrl_safe(&p, K1, K0, PGDIR_SHIFT + PGD_ORDER + PAGE_SHIFT - 3);
1525 		uasm_il_beqz(&p, &r, K1, label_vmalloc);
1526 		uasm_i_nop(&p);
1527 
1528 		uasm_il_bgez(&p, &r, K0, label_large_segbits_fault);
1529 		uasm_i_nop(&p);
1530 		uasm_l_vmalloc(&l, p);
1531 	}
1532 
1533 	uasm_i_dmfc0(&p, K1, C0_PGD);
1534 
1535 	uasm_i_lddir(&p, K0, K1, 3);  /* global page dir */
1536 #ifndef __PAGETABLE_PMD_FOLDED
1537 	uasm_i_lddir(&p, K1, K0, 1);  /* middle page dir */
1538 #endif
1539 	uasm_i_ldpte(&p, K1, 0);      /* even */
1540 	uasm_i_ldpte(&p, K1, 1);      /* odd */
1541 	uasm_i_tlbwr(&p);
1542 
1543 	/* restore page mask */
1544 	if (PM_DEFAULT_MASK >> 16) {
1545 		uasm_i_lui(&p, K0, PM_DEFAULT_MASK >> 16);
1546 		uasm_i_ori(&p, K0, K0, PM_DEFAULT_MASK & 0xffff);
1547 		uasm_i_mtc0(&p, K0, C0_PAGEMASK);
1548 	} else if (PM_DEFAULT_MASK) {
1549 		uasm_i_ori(&p, K0, 0, PM_DEFAULT_MASK);
1550 		uasm_i_mtc0(&p, K0, C0_PAGEMASK);
1551 	} else {
1552 		uasm_i_mtc0(&p, 0, C0_PAGEMASK);
1553 	}
1554 
1555 	uasm_i_eret(&p);
1556 
1557 	if (check_for_high_segbits) {
1558 		uasm_l_large_segbits_fault(&l, p);
1559 		UASM_i_LA(&p, K1, (unsigned long)tlb_do_page_fault_0);
1560 		uasm_i_jr(&p, K1);
1561 		uasm_i_nop(&p);
1562 	}
1563 
1564 	uasm_resolve_relocs(relocs, labels);
1565 	memcpy((void *)(ebase + 0x80), tlb_handler, 0x80);
1566 	local_flush_icache_range(ebase + 0x80, ebase + 0x100);
1567 	dump_handler("loongson3_tlb_refill",
1568 		     (u32 *)(ebase + 0x80), (u32 *)(ebase + 0x100));
1569 }
1570 
1571 static void build_setup_pgd(void)
1572 {
1573 	const int a0 = 4;
1574 	const int __maybe_unused a1 = 5;
1575 	const int __maybe_unused a2 = 6;
1576 	u32 *p = (u32 *)msk_isa16_mode((ulong)tlbmiss_handler_setup_pgd);
1577 #ifndef CONFIG_MIPS_PGD_C0_CONTEXT
1578 	long pgdc = (long)pgd_current;
1579 #endif
1580 
1581 	memset(p, 0, tlbmiss_handler_setup_pgd_end - (char *)p);
1582 	memset(labels, 0, sizeof(labels));
1583 	memset(relocs, 0, sizeof(relocs));
1584 	pgd_reg = allocate_kscratch();
1585 #ifdef CONFIG_MIPS_PGD_C0_CONTEXT
1586 	if (pgd_reg == -1) {
1587 		struct uasm_label *l = labels;
1588 		struct uasm_reloc *r = relocs;
1589 
1590 		/* PGD << 11 in c0_Context */
1591 		/*
1592 		 * If it is a ckseg0 address, convert to a physical
1593 		 * address.  Shifting right by 29 and adding 4 will
1594 		 * result in zero for these addresses.
1595 		 *
1596 		 */
1597 		UASM_i_SRA(&p, a1, a0, 29);
1598 		UASM_i_ADDIU(&p, a1, a1, 4);
1599 		uasm_il_bnez(&p, &r, a1, label_tlbl_goaround1);
1600 		uasm_i_nop(&p);
1601 		uasm_i_dinsm(&p, a0, 0, 29, 64 - 29);
1602 		uasm_l_tlbl_goaround1(&l, p);
1603 		UASM_i_SLL(&p, a0, a0, 11);
1604 		uasm_i_jr(&p, 31);
1605 		UASM_i_MTC0(&p, a0, C0_CONTEXT);
1606 	} else {
1607 		/* PGD in c0_KScratch */
1608 		uasm_i_jr(&p, 31);
1609 		if (cpu_has_ldpte)
1610 			UASM_i_MTC0(&p, a0, C0_PWBASE);
1611 		else
1612 			UASM_i_MTC0(&p, a0, c0_kscratch(), pgd_reg);
1613 	}
1614 #else
1615 #ifdef CONFIG_SMP
1616 	/* Save PGD to pgd_current[smp_processor_id()] */
1617 	UASM_i_CPUID_MFC0(&p, a1, SMP_CPUID_REG);
1618 	UASM_i_SRL_SAFE(&p, a1, a1, SMP_CPUID_PTRSHIFT);
1619 	UASM_i_LA_mostly(&p, a2, pgdc);
1620 	UASM_i_ADDU(&p, a2, a2, a1);
1621 	UASM_i_SW(&p, a0, uasm_rel_lo(pgdc), a2);
1622 #else
1623 	UASM_i_LA_mostly(&p, a2, pgdc);
1624 	UASM_i_SW(&p, a0, uasm_rel_lo(pgdc), a2);
1625 #endif /* SMP */
1626 	uasm_i_jr(&p, 31);
1627 
1628 	/* if pgd_reg is allocated, save PGD also to scratch register */
1629 	if (pgd_reg != -1)
1630 		UASM_i_MTC0(&p, a0, c0_kscratch(), pgd_reg);
1631 	else
1632 		uasm_i_nop(&p);
1633 #endif
1634 	if (p >= (u32 *)tlbmiss_handler_setup_pgd_end)
1635 		panic("tlbmiss_handler_setup_pgd space exceeded");
1636 
1637 	uasm_resolve_relocs(relocs, labels);
1638 	pr_debug("Wrote tlbmiss_handler_setup_pgd (%u instructions).\n",
1639 		 (unsigned int)(p - (u32 *)tlbmiss_handler_setup_pgd));
1640 
1641 	dump_handler("tlbmiss_handler", tlbmiss_handler_setup_pgd,
1642 					tlbmiss_handler_setup_pgd_end);
1643 }
1644 
1645 static void
1646 iPTE_LW(u32 **p, unsigned int pte, unsigned int ptr)
1647 {
1648 #ifdef CONFIG_SMP
1649 # ifdef CONFIG_PHYS_ADDR_T_64BIT
1650 	if (cpu_has_64bits)
1651 		uasm_i_lld(p, pte, 0, ptr);
1652 	else
1653 # endif
1654 		UASM_i_LL(p, pte, 0, ptr);
1655 #else
1656 # ifdef CONFIG_PHYS_ADDR_T_64BIT
1657 	if (cpu_has_64bits)
1658 		uasm_i_ld(p, pte, 0, ptr);
1659 	else
1660 # endif
1661 		UASM_i_LW(p, pte, 0, ptr);
1662 #endif
1663 }
1664 
1665 static void
1666 iPTE_SW(u32 **p, struct uasm_reloc **r, unsigned int pte, unsigned int ptr,
1667 	unsigned int mode, unsigned int scratch)
1668 {
1669 	unsigned int hwmode = mode & (_PAGE_VALID | _PAGE_DIRTY);
1670 	unsigned int swmode = mode & ~hwmode;
1671 
1672 	if (IS_ENABLED(CONFIG_XPA) && !cpu_has_64bits) {
1673 		uasm_i_lui(p, scratch, swmode >> 16);
1674 		uasm_i_or(p, pte, pte, scratch);
1675 		BUG_ON(swmode & 0xffff);
1676 	} else {
1677 		uasm_i_ori(p, pte, pte, mode);
1678 	}
1679 
1680 #ifdef CONFIG_SMP
1681 # ifdef CONFIG_PHYS_ADDR_T_64BIT
1682 	if (cpu_has_64bits)
1683 		uasm_i_scd(p, pte, 0, ptr);
1684 	else
1685 # endif
1686 		UASM_i_SC(p, pte, 0, ptr);
1687 
1688 	if (r10000_llsc_war())
1689 		uasm_il_beqzl(p, r, pte, label_smp_pgtable_change);
1690 	else
1691 		uasm_il_beqz(p, r, pte, label_smp_pgtable_change);
1692 
1693 # ifdef CONFIG_PHYS_ADDR_T_64BIT
1694 	if (!cpu_has_64bits) {
1695 		/* no uasm_i_nop needed */
1696 		uasm_i_ll(p, pte, sizeof(pte_t) / 2, ptr);
1697 		uasm_i_ori(p, pte, pte, hwmode);
1698 		BUG_ON(hwmode & ~0xffff);
1699 		uasm_i_sc(p, pte, sizeof(pte_t) / 2, ptr);
1700 		uasm_il_beqz(p, r, pte, label_smp_pgtable_change);
1701 		/* no uasm_i_nop needed */
1702 		uasm_i_lw(p, pte, 0, ptr);
1703 	} else
1704 		uasm_i_nop(p);
1705 # else
1706 	uasm_i_nop(p);
1707 # endif
1708 #else
1709 # ifdef CONFIG_PHYS_ADDR_T_64BIT
1710 	if (cpu_has_64bits)
1711 		uasm_i_sd(p, pte, 0, ptr);
1712 	else
1713 # endif
1714 		UASM_i_SW(p, pte, 0, ptr);
1715 
1716 # ifdef CONFIG_PHYS_ADDR_T_64BIT
1717 	if (!cpu_has_64bits) {
1718 		uasm_i_lw(p, pte, sizeof(pte_t) / 2, ptr);
1719 		uasm_i_ori(p, pte, pte, hwmode);
1720 		BUG_ON(hwmode & ~0xffff);
1721 		uasm_i_sw(p, pte, sizeof(pte_t) / 2, ptr);
1722 		uasm_i_lw(p, pte, 0, ptr);
1723 	}
1724 # endif
1725 #endif
1726 }
1727 
1728 /*
1729  * Check if PTE is present, if not then jump to LABEL. PTR points to
1730  * the page table where this PTE is located, PTE will be re-loaded
1731  * with it's original value.
1732  */
1733 static void
1734 build_pte_present(u32 **p, struct uasm_reloc **r,
1735 		  int pte, int ptr, int scratch, enum label_id lid)
1736 {
1737 	int t = scratch >= 0 ? scratch : pte;
1738 	int cur = pte;
1739 
1740 	if (cpu_has_rixi) {
1741 		if (use_bbit_insns()) {
1742 			uasm_il_bbit0(p, r, pte, ilog2(_PAGE_PRESENT), lid);
1743 			uasm_i_nop(p);
1744 		} else {
1745 			if (_PAGE_PRESENT_SHIFT) {
1746 				uasm_i_srl(p, t, cur, _PAGE_PRESENT_SHIFT);
1747 				cur = t;
1748 			}
1749 			uasm_i_andi(p, t, cur, 1);
1750 			uasm_il_beqz(p, r, t, lid);
1751 			if (pte == t)
1752 				/* You lose the SMP race :-(*/
1753 				iPTE_LW(p, pte, ptr);
1754 		}
1755 	} else {
1756 		if (_PAGE_PRESENT_SHIFT) {
1757 			uasm_i_srl(p, t, cur, _PAGE_PRESENT_SHIFT);
1758 			cur = t;
1759 		}
1760 		uasm_i_andi(p, t, cur,
1761 			(_PAGE_PRESENT | _PAGE_NO_READ) >> _PAGE_PRESENT_SHIFT);
1762 		uasm_i_xori(p, t, t, _PAGE_PRESENT >> _PAGE_PRESENT_SHIFT);
1763 		uasm_il_bnez(p, r, t, lid);
1764 		if (pte == t)
1765 			/* You lose the SMP race :-(*/
1766 			iPTE_LW(p, pte, ptr);
1767 	}
1768 }
1769 
1770 /* Make PTE valid, store result in PTR. */
1771 static void
1772 build_make_valid(u32 **p, struct uasm_reloc **r, unsigned int pte,
1773 		 unsigned int ptr, unsigned int scratch)
1774 {
1775 	unsigned int mode = _PAGE_VALID | _PAGE_ACCESSED;
1776 
1777 	iPTE_SW(p, r, pte, ptr, mode, scratch);
1778 }
1779 
1780 /*
1781  * Check if PTE can be written to, if not branch to LABEL. Regardless
1782  * restore PTE with value from PTR when done.
1783  */
1784 static void
1785 build_pte_writable(u32 **p, struct uasm_reloc **r,
1786 		   unsigned int pte, unsigned int ptr, int scratch,
1787 		   enum label_id lid)
1788 {
1789 	int t = scratch >= 0 ? scratch : pte;
1790 	int cur = pte;
1791 
1792 	if (_PAGE_PRESENT_SHIFT) {
1793 		uasm_i_srl(p, t, cur, _PAGE_PRESENT_SHIFT);
1794 		cur = t;
1795 	}
1796 	uasm_i_andi(p, t, cur,
1797 		    (_PAGE_PRESENT | _PAGE_WRITE) >> _PAGE_PRESENT_SHIFT);
1798 	uasm_i_xori(p, t, t,
1799 		    (_PAGE_PRESENT | _PAGE_WRITE) >> _PAGE_PRESENT_SHIFT);
1800 	uasm_il_bnez(p, r, t, lid);
1801 	if (pte == t)
1802 		/* You lose the SMP race :-(*/
1803 		iPTE_LW(p, pte, ptr);
1804 	else
1805 		uasm_i_nop(p);
1806 }
1807 
1808 /* Make PTE writable, update software status bits as well, then store
1809  * at PTR.
1810  */
1811 static void
1812 build_make_write(u32 **p, struct uasm_reloc **r, unsigned int pte,
1813 		 unsigned int ptr, unsigned int scratch)
1814 {
1815 	unsigned int mode = (_PAGE_ACCESSED | _PAGE_MODIFIED | _PAGE_VALID
1816 			     | _PAGE_DIRTY);
1817 
1818 	iPTE_SW(p, r, pte, ptr, mode, scratch);
1819 }
1820 
1821 /*
1822  * Check if PTE can be modified, if not branch to LABEL. Regardless
1823  * restore PTE with value from PTR when done.
1824  */
1825 static void
1826 build_pte_modifiable(u32 **p, struct uasm_reloc **r,
1827 		     unsigned int pte, unsigned int ptr, int scratch,
1828 		     enum label_id lid)
1829 {
1830 	if (use_bbit_insns()) {
1831 		uasm_il_bbit0(p, r, pte, ilog2(_PAGE_WRITE), lid);
1832 		uasm_i_nop(p);
1833 	} else {
1834 		int t = scratch >= 0 ? scratch : pte;
1835 		uasm_i_srl(p, t, pte, _PAGE_WRITE_SHIFT);
1836 		uasm_i_andi(p, t, t, 1);
1837 		uasm_il_beqz(p, r, t, lid);
1838 		if (pte == t)
1839 			/* You lose the SMP race :-(*/
1840 			iPTE_LW(p, pte, ptr);
1841 	}
1842 }
1843 
1844 #ifndef CONFIG_MIPS_PGD_C0_CONTEXT
1845 
1846 
1847 /*
1848  * R3000 style TLB load/store/modify handlers.
1849  */
1850 
1851 /*
1852  * This places the pte into ENTRYLO0 and writes it with tlbwi.
1853  * Then it returns.
1854  */
1855 static void
1856 build_r3000_pte_reload_tlbwi(u32 **p, unsigned int pte, unsigned int tmp)
1857 {
1858 	uasm_i_mtc0(p, pte, C0_ENTRYLO0); /* cp0 delay */
1859 	uasm_i_mfc0(p, tmp, C0_EPC); /* cp0 delay */
1860 	uasm_i_tlbwi(p);
1861 	uasm_i_jr(p, tmp);
1862 	uasm_i_rfe(p); /* branch delay */
1863 }
1864 
1865 /*
1866  * This places the pte into ENTRYLO0 and writes it with tlbwi
1867  * or tlbwr as appropriate.  This is because the index register
1868  * may have the probe fail bit set as a result of a trap on a
1869  * kseg2 access, i.e. without refill.  Then it returns.
1870  */
1871 static void
1872 build_r3000_tlb_reload_write(u32 **p, struct uasm_label **l,
1873 			     struct uasm_reloc **r, unsigned int pte,
1874 			     unsigned int tmp)
1875 {
1876 	uasm_i_mfc0(p, tmp, C0_INDEX);
1877 	uasm_i_mtc0(p, pte, C0_ENTRYLO0); /* cp0 delay */
1878 	uasm_il_bltz(p, r, tmp, label_r3000_write_probe_fail); /* cp0 delay */
1879 	uasm_i_mfc0(p, tmp, C0_EPC); /* branch delay */
1880 	uasm_i_tlbwi(p); /* cp0 delay */
1881 	uasm_i_jr(p, tmp);
1882 	uasm_i_rfe(p); /* branch delay */
1883 	uasm_l_r3000_write_probe_fail(l, *p);
1884 	uasm_i_tlbwr(p); /* cp0 delay */
1885 	uasm_i_jr(p, tmp);
1886 	uasm_i_rfe(p); /* branch delay */
1887 }
1888 
1889 static void
1890 build_r3000_tlbchange_handler_head(u32 **p, unsigned int pte,
1891 				   unsigned int ptr)
1892 {
1893 	long pgdc = (long)pgd_current;
1894 
1895 	uasm_i_mfc0(p, pte, C0_BADVADDR);
1896 	uasm_i_lui(p, ptr, uasm_rel_hi(pgdc)); /* cp0 delay */
1897 	uasm_i_lw(p, ptr, uasm_rel_lo(pgdc), ptr);
1898 	uasm_i_srl(p, pte, pte, 22); /* load delay */
1899 	uasm_i_sll(p, pte, pte, 2);
1900 	uasm_i_addu(p, ptr, ptr, pte);
1901 	uasm_i_mfc0(p, pte, C0_CONTEXT);
1902 	uasm_i_lw(p, ptr, 0, ptr); /* cp0 delay */
1903 	uasm_i_andi(p, pte, pte, 0xffc); /* load delay */
1904 	uasm_i_addu(p, ptr, ptr, pte);
1905 	uasm_i_lw(p, pte, 0, ptr);
1906 	uasm_i_tlbp(p); /* load delay */
1907 }
1908 
1909 static void build_r3000_tlb_load_handler(void)
1910 {
1911 	u32 *p = (u32 *)handle_tlbl;
1912 	struct uasm_label *l = labels;
1913 	struct uasm_reloc *r = relocs;
1914 
1915 	memset(p, 0, handle_tlbl_end - (char *)p);
1916 	memset(labels, 0, sizeof(labels));
1917 	memset(relocs, 0, sizeof(relocs));
1918 
1919 	build_r3000_tlbchange_handler_head(&p, K0, K1);
1920 	build_pte_present(&p, &r, K0, K1, -1, label_nopage_tlbl);
1921 	uasm_i_nop(&p); /* load delay */
1922 	build_make_valid(&p, &r, K0, K1, -1);
1923 	build_r3000_tlb_reload_write(&p, &l, &r, K0, K1);
1924 
1925 	uasm_l_nopage_tlbl(&l, p);
1926 	uasm_i_j(&p, (unsigned long)tlb_do_page_fault_0 & 0x0fffffff);
1927 	uasm_i_nop(&p);
1928 
1929 	if (p >= (u32 *)handle_tlbl_end)
1930 		panic("TLB load handler fastpath space exceeded");
1931 
1932 	uasm_resolve_relocs(relocs, labels);
1933 	pr_debug("Wrote TLB load handler fastpath (%u instructions).\n",
1934 		 (unsigned int)(p - (u32 *)handle_tlbl));
1935 
1936 	dump_handler("r3000_tlb_load", handle_tlbl, handle_tlbl_end);
1937 }
1938 
1939 static void build_r3000_tlb_store_handler(void)
1940 {
1941 	u32 *p = (u32 *)handle_tlbs;
1942 	struct uasm_label *l = labels;
1943 	struct uasm_reloc *r = relocs;
1944 
1945 	memset(p, 0, handle_tlbs_end - (char *)p);
1946 	memset(labels, 0, sizeof(labels));
1947 	memset(relocs, 0, sizeof(relocs));
1948 
1949 	build_r3000_tlbchange_handler_head(&p, K0, K1);
1950 	build_pte_writable(&p, &r, K0, K1, -1, label_nopage_tlbs);
1951 	uasm_i_nop(&p); /* load delay */
1952 	build_make_write(&p, &r, K0, K1, -1);
1953 	build_r3000_tlb_reload_write(&p, &l, &r, K0, K1);
1954 
1955 	uasm_l_nopage_tlbs(&l, p);
1956 	uasm_i_j(&p, (unsigned long)tlb_do_page_fault_1 & 0x0fffffff);
1957 	uasm_i_nop(&p);
1958 
1959 	if (p >= (u32 *)handle_tlbs_end)
1960 		panic("TLB store handler fastpath space exceeded");
1961 
1962 	uasm_resolve_relocs(relocs, labels);
1963 	pr_debug("Wrote TLB store handler fastpath (%u instructions).\n",
1964 		 (unsigned int)(p - (u32 *)handle_tlbs));
1965 
1966 	dump_handler("r3000_tlb_store", handle_tlbs, handle_tlbs_end);
1967 }
1968 
1969 static void build_r3000_tlb_modify_handler(void)
1970 {
1971 	u32 *p = (u32 *)handle_tlbm;
1972 	struct uasm_label *l = labels;
1973 	struct uasm_reloc *r = relocs;
1974 
1975 	memset(p, 0, handle_tlbm_end - (char *)p);
1976 	memset(labels, 0, sizeof(labels));
1977 	memset(relocs, 0, sizeof(relocs));
1978 
1979 	build_r3000_tlbchange_handler_head(&p, K0, K1);
1980 	build_pte_modifiable(&p, &r, K0, K1,  -1, label_nopage_tlbm);
1981 	uasm_i_nop(&p); /* load delay */
1982 	build_make_write(&p, &r, K0, K1, -1);
1983 	build_r3000_pte_reload_tlbwi(&p, K0, K1);
1984 
1985 	uasm_l_nopage_tlbm(&l, p);
1986 	uasm_i_j(&p, (unsigned long)tlb_do_page_fault_1 & 0x0fffffff);
1987 	uasm_i_nop(&p);
1988 
1989 	if (p >= (u32 *)handle_tlbm_end)
1990 		panic("TLB modify handler fastpath space exceeded");
1991 
1992 	uasm_resolve_relocs(relocs, labels);
1993 	pr_debug("Wrote TLB modify handler fastpath (%u instructions).\n",
1994 		 (unsigned int)(p - (u32 *)handle_tlbm));
1995 
1996 	dump_handler("r3000_tlb_modify", handle_tlbm, handle_tlbm_end);
1997 }
1998 #endif /* CONFIG_MIPS_PGD_C0_CONTEXT */
1999 
2000 static bool cpu_has_tlbex_tlbp_race(void)
2001 {
2002 	/*
2003 	 * When a Hardware Table Walker is running it can replace TLB entries
2004 	 * at any time, leading to a race between it & the CPU.
2005 	 */
2006 	if (cpu_has_htw)
2007 		return true;
2008 
2009 	/*
2010 	 * If the CPU shares FTLB RAM with its siblings then our entry may be
2011 	 * replaced at any time by a sibling performing a write to the FTLB.
2012 	 */
2013 	if (cpu_has_shared_ftlb_ram)
2014 		return true;
2015 
2016 	/* In all other cases there ought to be no race condition to handle */
2017 	return false;
2018 }
2019 
2020 /*
2021  * R4000 style TLB load/store/modify handlers.
2022  */
2023 static struct work_registers
2024 build_r4000_tlbchange_handler_head(u32 **p, struct uasm_label **l,
2025 				   struct uasm_reloc **r)
2026 {
2027 	struct work_registers wr = build_get_work_registers(p);
2028 
2029 #ifdef CONFIG_64BIT
2030 	build_get_pmde64(p, l, r, wr.r1, wr.r2); /* get pmd in ptr */
2031 #else
2032 	build_get_pgde32(p, wr.r1, wr.r2); /* get pgd in ptr */
2033 #endif
2034 
2035 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
2036 	/*
2037 	 * For huge tlb entries, pmd doesn't contain an address but
2038 	 * instead contains the tlb pte. Check the PAGE_HUGE bit and
2039 	 * see if we need to jump to huge tlb processing.
2040 	 */
2041 	build_is_huge_pte(p, r, wr.r1, wr.r2, label_tlb_huge_update);
2042 #endif
2043 
2044 	UASM_i_MFC0(p, wr.r1, C0_BADVADDR);
2045 	UASM_i_LW(p, wr.r2, 0, wr.r2);
2046 	UASM_i_SRL(p, wr.r1, wr.r1, PAGE_SHIFT + PTE_ORDER - PTE_T_LOG2);
2047 	uasm_i_andi(p, wr.r1, wr.r1, (PTRS_PER_PTE - 1) << PTE_T_LOG2);
2048 	UASM_i_ADDU(p, wr.r2, wr.r2, wr.r1);
2049 
2050 #ifdef CONFIG_SMP
2051 	uasm_l_smp_pgtable_change(l, *p);
2052 #endif
2053 	iPTE_LW(p, wr.r1, wr.r2); /* get even pte */
2054 	if (!m4kc_tlbp_war()) {
2055 		build_tlb_probe_entry(p);
2056 		if (cpu_has_tlbex_tlbp_race()) {
2057 			/* race condition happens, leaving */
2058 			uasm_i_ehb(p);
2059 			uasm_i_mfc0(p, wr.r3, C0_INDEX);
2060 			uasm_il_bltz(p, r, wr.r3, label_leave);
2061 			uasm_i_nop(p);
2062 		}
2063 	}
2064 	return wr;
2065 }
2066 
2067 static void
2068 build_r4000_tlbchange_handler_tail(u32 **p, struct uasm_label **l,
2069 				   struct uasm_reloc **r, unsigned int tmp,
2070 				   unsigned int ptr)
2071 {
2072 	uasm_i_ori(p, ptr, ptr, sizeof(pte_t));
2073 	uasm_i_xori(p, ptr, ptr, sizeof(pte_t));
2074 	build_update_entries(p, tmp, ptr);
2075 	build_tlb_write_entry(p, l, r, tlb_indexed);
2076 	uasm_l_leave(l, *p);
2077 	build_restore_work_registers(p);
2078 	uasm_i_eret(p); /* return from trap */
2079 
2080 #ifdef CONFIG_64BIT
2081 	build_get_pgd_vmalloc64(p, l, r, tmp, ptr, not_refill);
2082 #endif
2083 }
2084 
2085 static void build_r4000_tlb_load_handler(void)
2086 {
2087 	u32 *p = (u32 *)msk_isa16_mode((ulong)handle_tlbl);
2088 	struct uasm_label *l = labels;
2089 	struct uasm_reloc *r = relocs;
2090 	struct work_registers wr;
2091 
2092 	memset(p, 0, handle_tlbl_end - (char *)p);
2093 	memset(labels, 0, sizeof(labels));
2094 	memset(relocs, 0, sizeof(relocs));
2095 
2096 	if (bcm1250_m3_war()) {
2097 		unsigned int segbits = 44;
2098 
2099 		uasm_i_dmfc0(&p, K0, C0_BADVADDR);
2100 		uasm_i_dmfc0(&p, K1, C0_ENTRYHI);
2101 		uasm_i_xor(&p, K0, K0, K1);
2102 		uasm_i_dsrl_safe(&p, K1, K0, 62);
2103 		uasm_i_dsrl_safe(&p, K0, K0, 12 + 1);
2104 		uasm_i_dsll_safe(&p, K0, K0, 64 + 12 + 1 - segbits);
2105 		uasm_i_or(&p, K0, K0, K1);
2106 		uasm_il_bnez(&p, &r, K0, label_leave);
2107 		/* No need for uasm_i_nop */
2108 	}
2109 
2110 	wr = build_r4000_tlbchange_handler_head(&p, &l, &r);
2111 	build_pte_present(&p, &r, wr.r1, wr.r2, wr.r3, label_nopage_tlbl);
2112 	if (m4kc_tlbp_war())
2113 		build_tlb_probe_entry(&p);
2114 
2115 	if (cpu_has_rixi && !cpu_has_rixiex) {
2116 		/*
2117 		 * If the page is not _PAGE_VALID, RI or XI could not
2118 		 * have triggered it.  Skip the expensive test..
2119 		 */
2120 		if (use_bbit_insns()) {
2121 			uasm_il_bbit0(&p, &r, wr.r1, ilog2(_PAGE_VALID),
2122 				      label_tlbl_goaround1);
2123 		} else {
2124 			uasm_i_andi(&p, wr.r3, wr.r1, _PAGE_VALID);
2125 			uasm_il_beqz(&p, &r, wr.r3, label_tlbl_goaround1);
2126 		}
2127 		uasm_i_nop(&p);
2128 
2129 		/*
2130 		 * Warn if something may race with us & replace the TLB entry
2131 		 * before we read it here. Everything with such races should
2132 		 * also have dedicated RiXi exception handlers, so this
2133 		 * shouldn't be hit.
2134 		 */
2135 		WARN(cpu_has_tlbex_tlbp_race(), "Unhandled race in RiXi path");
2136 
2137 		uasm_i_tlbr(&p);
2138 
2139 		switch (current_cpu_type()) {
2140 		default:
2141 			if (cpu_has_mips_r2_exec_hazard) {
2142 				uasm_i_ehb(&p);
2143 
2144 		case CPU_CAVIUM_OCTEON:
2145 		case CPU_CAVIUM_OCTEON_PLUS:
2146 		case CPU_CAVIUM_OCTEON2:
2147 				break;
2148 			}
2149 		}
2150 
2151 		/* Examine  entrylo 0 or 1 based on ptr. */
2152 		if (use_bbit_insns()) {
2153 			uasm_i_bbit0(&p, wr.r2, ilog2(sizeof(pte_t)), 8);
2154 		} else {
2155 			uasm_i_andi(&p, wr.r3, wr.r2, sizeof(pte_t));
2156 			uasm_i_beqz(&p, wr.r3, 8);
2157 		}
2158 		/* load it in the delay slot*/
2159 		UASM_i_MFC0(&p, wr.r3, C0_ENTRYLO0);
2160 		/* load it if ptr is odd */
2161 		UASM_i_MFC0(&p, wr.r3, C0_ENTRYLO1);
2162 		/*
2163 		 * If the entryLo (now in wr.r3) is valid (bit 1), RI or
2164 		 * XI must have triggered it.
2165 		 */
2166 		if (use_bbit_insns()) {
2167 			uasm_il_bbit1(&p, &r, wr.r3, 1, label_nopage_tlbl);
2168 			uasm_i_nop(&p);
2169 			uasm_l_tlbl_goaround1(&l, p);
2170 		} else {
2171 			uasm_i_andi(&p, wr.r3, wr.r3, 2);
2172 			uasm_il_bnez(&p, &r, wr.r3, label_nopage_tlbl);
2173 			uasm_i_nop(&p);
2174 		}
2175 		uasm_l_tlbl_goaround1(&l, p);
2176 	}
2177 	build_make_valid(&p, &r, wr.r1, wr.r2, wr.r3);
2178 	build_r4000_tlbchange_handler_tail(&p, &l, &r, wr.r1, wr.r2);
2179 
2180 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
2181 	/*
2182 	 * This is the entry point when build_r4000_tlbchange_handler_head
2183 	 * spots a huge page.
2184 	 */
2185 	uasm_l_tlb_huge_update(&l, p);
2186 	iPTE_LW(&p, wr.r1, wr.r2);
2187 	build_pte_present(&p, &r, wr.r1, wr.r2, wr.r3, label_nopage_tlbl);
2188 	build_tlb_probe_entry(&p);
2189 
2190 	if (cpu_has_rixi && !cpu_has_rixiex) {
2191 		/*
2192 		 * If the page is not _PAGE_VALID, RI or XI could not
2193 		 * have triggered it.  Skip the expensive test..
2194 		 */
2195 		if (use_bbit_insns()) {
2196 			uasm_il_bbit0(&p, &r, wr.r1, ilog2(_PAGE_VALID),
2197 				      label_tlbl_goaround2);
2198 		} else {
2199 			uasm_i_andi(&p, wr.r3, wr.r1, _PAGE_VALID);
2200 			uasm_il_beqz(&p, &r, wr.r3, label_tlbl_goaround2);
2201 		}
2202 		uasm_i_nop(&p);
2203 
2204 		/*
2205 		 * Warn if something may race with us & replace the TLB entry
2206 		 * before we read it here. Everything with such races should
2207 		 * also have dedicated RiXi exception handlers, so this
2208 		 * shouldn't be hit.
2209 		 */
2210 		WARN(cpu_has_tlbex_tlbp_race(), "Unhandled race in RiXi path");
2211 
2212 		uasm_i_tlbr(&p);
2213 
2214 		switch (current_cpu_type()) {
2215 		default:
2216 			if (cpu_has_mips_r2_exec_hazard) {
2217 				uasm_i_ehb(&p);
2218 
2219 		case CPU_CAVIUM_OCTEON:
2220 		case CPU_CAVIUM_OCTEON_PLUS:
2221 		case CPU_CAVIUM_OCTEON2:
2222 				break;
2223 			}
2224 		}
2225 
2226 		/* Examine  entrylo 0 or 1 based on ptr. */
2227 		if (use_bbit_insns()) {
2228 			uasm_i_bbit0(&p, wr.r2, ilog2(sizeof(pte_t)), 8);
2229 		} else {
2230 			uasm_i_andi(&p, wr.r3, wr.r2, sizeof(pte_t));
2231 			uasm_i_beqz(&p, wr.r3, 8);
2232 		}
2233 		/* load it in the delay slot*/
2234 		UASM_i_MFC0(&p, wr.r3, C0_ENTRYLO0);
2235 		/* load it if ptr is odd */
2236 		UASM_i_MFC0(&p, wr.r3, C0_ENTRYLO1);
2237 		/*
2238 		 * If the entryLo (now in wr.r3) is valid (bit 1), RI or
2239 		 * XI must have triggered it.
2240 		 */
2241 		if (use_bbit_insns()) {
2242 			uasm_il_bbit0(&p, &r, wr.r3, 1, label_tlbl_goaround2);
2243 		} else {
2244 			uasm_i_andi(&p, wr.r3, wr.r3, 2);
2245 			uasm_il_beqz(&p, &r, wr.r3, label_tlbl_goaround2);
2246 		}
2247 		if (PM_DEFAULT_MASK == 0)
2248 			uasm_i_nop(&p);
2249 		/*
2250 		 * We clobbered C0_PAGEMASK, restore it.  On the other branch
2251 		 * it is restored in build_huge_tlb_write_entry.
2252 		 */
2253 		build_restore_pagemask(&p, &r, wr.r3, label_nopage_tlbl, 0);
2254 
2255 		uasm_l_tlbl_goaround2(&l, p);
2256 	}
2257 	uasm_i_ori(&p, wr.r1, wr.r1, (_PAGE_ACCESSED | _PAGE_VALID));
2258 	build_huge_handler_tail(&p, &r, &l, wr.r1, wr.r2, 1);
2259 #endif
2260 
2261 	uasm_l_nopage_tlbl(&l, p);
2262 	build_restore_work_registers(&p);
2263 #ifdef CONFIG_CPU_MICROMIPS
2264 	if ((unsigned long)tlb_do_page_fault_0 & 1) {
2265 		uasm_i_lui(&p, K0, uasm_rel_hi((long)tlb_do_page_fault_0));
2266 		uasm_i_addiu(&p, K0, K0, uasm_rel_lo((long)tlb_do_page_fault_0));
2267 		uasm_i_jr(&p, K0);
2268 	} else
2269 #endif
2270 	uasm_i_j(&p, (unsigned long)tlb_do_page_fault_0 & 0x0fffffff);
2271 	uasm_i_nop(&p);
2272 
2273 	if (p >= (u32 *)handle_tlbl_end)
2274 		panic("TLB load handler fastpath space exceeded");
2275 
2276 	uasm_resolve_relocs(relocs, labels);
2277 	pr_debug("Wrote TLB load handler fastpath (%u instructions).\n",
2278 		 (unsigned int)(p - (u32 *)handle_tlbl));
2279 
2280 	dump_handler("r4000_tlb_load", handle_tlbl, handle_tlbl_end);
2281 }
2282 
2283 static void build_r4000_tlb_store_handler(void)
2284 {
2285 	u32 *p = (u32 *)msk_isa16_mode((ulong)handle_tlbs);
2286 	struct uasm_label *l = labels;
2287 	struct uasm_reloc *r = relocs;
2288 	struct work_registers wr;
2289 
2290 	memset(p, 0, handle_tlbs_end - (char *)p);
2291 	memset(labels, 0, sizeof(labels));
2292 	memset(relocs, 0, sizeof(relocs));
2293 
2294 	wr = build_r4000_tlbchange_handler_head(&p, &l, &r);
2295 	build_pte_writable(&p, &r, wr.r1, wr.r2, wr.r3, label_nopage_tlbs);
2296 	if (m4kc_tlbp_war())
2297 		build_tlb_probe_entry(&p);
2298 	build_make_write(&p, &r, wr.r1, wr.r2, wr.r3);
2299 	build_r4000_tlbchange_handler_tail(&p, &l, &r, wr.r1, wr.r2);
2300 
2301 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
2302 	/*
2303 	 * This is the entry point when
2304 	 * build_r4000_tlbchange_handler_head spots a huge page.
2305 	 */
2306 	uasm_l_tlb_huge_update(&l, p);
2307 	iPTE_LW(&p, wr.r1, wr.r2);
2308 	build_pte_writable(&p, &r, wr.r1, wr.r2, wr.r3, label_nopage_tlbs);
2309 	build_tlb_probe_entry(&p);
2310 	uasm_i_ori(&p, wr.r1, wr.r1,
2311 		   _PAGE_ACCESSED | _PAGE_MODIFIED | _PAGE_VALID | _PAGE_DIRTY);
2312 	build_huge_handler_tail(&p, &r, &l, wr.r1, wr.r2, 1);
2313 #endif
2314 
2315 	uasm_l_nopage_tlbs(&l, p);
2316 	build_restore_work_registers(&p);
2317 #ifdef CONFIG_CPU_MICROMIPS
2318 	if ((unsigned long)tlb_do_page_fault_1 & 1) {
2319 		uasm_i_lui(&p, K0, uasm_rel_hi((long)tlb_do_page_fault_1));
2320 		uasm_i_addiu(&p, K0, K0, uasm_rel_lo((long)tlb_do_page_fault_1));
2321 		uasm_i_jr(&p, K0);
2322 	} else
2323 #endif
2324 	uasm_i_j(&p, (unsigned long)tlb_do_page_fault_1 & 0x0fffffff);
2325 	uasm_i_nop(&p);
2326 
2327 	if (p >= (u32 *)handle_tlbs_end)
2328 		panic("TLB store handler fastpath space exceeded");
2329 
2330 	uasm_resolve_relocs(relocs, labels);
2331 	pr_debug("Wrote TLB store handler fastpath (%u instructions).\n",
2332 		 (unsigned int)(p - (u32 *)handle_tlbs));
2333 
2334 	dump_handler("r4000_tlb_store", handle_tlbs, handle_tlbs_end);
2335 }
2336 
2337 static void build_r4000_tlb_modify_handler(void)
2338 {
2339 	u32 *p = (u32 *)msk_isa16_mode((ulong)handle_tlbm);
2340 	struct uasm_label *l = labels;
2341 	struct uasm_reloc *r = relocs;
2342 	struct work_registers wr;
2343 
2344 	memset(p, 0, handle_tlbm_end - (char *)p);
2345 	memset(labels, 0, sizeof(labels));
2346 	memset(relocs, 0, sizeof(relocs));
2347 
2348 	wr = build_r4000_tlbchange_handler_head(&p, &l, &r);
2349 	build_pte_modifiable(&p, &r, wr.r1, wr.r2, wr.r3, label_nopage_tlbm);
2350 	if (m4kc_tlbp_war())
2351 		build_tlb_probe_entry(&p);
2352 	/* Present and writable bits set, set accessed and dirty bits. */
2353 	build_make_write(&p, &r, wr.r1, wr.r2, wr.r3);
2354 	build_r4000_tlbchange_handler_tail(&p, &l, &r, wr.r1, wr.r2);
2355 
2356 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
2357 	/*
2358 	 * This is the entry point when
2359 	 * build_r4000_tlbchange_handler_head spots a huge page.
2360 	 */
2361 	uasm_l_tlb_huge_update(&l, p);
2362 	iPTE_LW(&p, wr.r1, wr.r2);
2363 	build_pte_modifiable(&p, &r, wr.r1, wr.r2,  wr.r3, label_nopage_tlbm);
2364 	build_tlb_probe_entry(&p);
2365 	uasm_i_ori(&p, wr.r1, wr.r1,
2366 		   _PAGE_ACCESSED | _PAGE_MODIFIED | _PAGE_VALID | _PAGE_DIRTY);
2367 	build_huge_handler_tail(&p, &r, &l, wr.r1, wr.r2, 0);
2368 #endif
2369 
2370 	uasm_l_nopage_tlbm(&l, p);
2371 	build_restore_work_registers(&p);
2372 #ifdef CONFIG_CPU_MICROMIPS
2373 	if ((unsigned long)tlb_do_page_fault_1 & 1) {
2374 		uasm_i_lui(&p, K0, uasm_rel_hi((long)tlb_do_page_fault_1));
2375 		uasm_i_addiu(&p, K0, K0, uasm_rel_lo((long)tlb_do_page_fault_1));
2376 		uasm_i_jr(&p, K0);
2377 	} else
2378 #endif
2379 	uasm_i_j(&p, (unsigned long)tlb_do_page_fault_1 & 0x0fffffff);
2380 	uasm_i_nop(&p);
2381 
2382 	if (p >= (u32 *)handle_tlbm_end)
2383 		panic("TLB modify handler fastpath space exceeded");
2384 
2385 	uasm_resolve_relocs(relocs, labels);
2386 	pr_debug("Wrote TLB modify handler fastpath (%u instructions).\n",
2387 		 (unsigned int)(p - (u32 *)handle_tlbm));
2388 
2389 	dump_handler("r4000_tlb_modify", handle_tlbm, handle_tlbm_end);
2390 }
2391 
2392 static void flush_tlb_handlers(void)
2393 {
2394 	local_flush_icache_range((unsigned long)handle_tlbl,
2395 			   (unsigned long)handle_tlbl_end);
2396 	local_flush_icache_range((unsigned long)handle_tlbs,
2397 			   (unsigned long)handle_tlbs_end);
2398 	local_flush_icache_range((unsigned long)handle_tlbm,
2399 			   (unsigned long)handle_tlbm_end);
2400 	local_flush_icache_range((unsigned long)tlbmiss_handler_setup_pgd,
2401 			   (unsigned long)tlbmiss_handler_setup_pgd_end);
2402 }
2403 
2404 static void print_htw_config(void)
2405 {
2406 	unsigned long config;
2407 	unsigned int pwctl;
2408 	const int field = 2 * sizeof(unsigned long);
2409 
2410 	config = read_c0_pwfield();
2411 	pr_debug("PWField (0x%0*lx): GDI: 0x%02lx  UDI: 0x%02lx  MDI: 0x%02lx  PTI: 0x%02lx  PTEI: 0x%02lx\n",
2412 		field, config,
2413 		(config & MIPS_PWFIELD_GDI_MASK) >> MIPS_PWFIELD_GDI_SHIFT,
2414 		(config & MIPS_PWFIELD_UDI_MASK) >> MIPS_PWFIELD_UDI_SHIFT,
2415 		(config & MIPS_PWFIELD_MDI_MASK) >> MIPS_PWFIELD_MDI_SHIFT,
2416 		(config & MIPS_PWFIELD_PTI_MASK) >> MIPS_PWFIELD_PTI_SHIFT,
2417 		(config & MIPS_PWFIELD_PTEI_MASK) >> MIPS_PWFIELD_PTEI_SHIFT);
2418 
2419 	config = read_c0_pwsize();
2420 	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",
2421 		field, config,
2422 		(config & MIPS_PWSIZE_PS_MASK) >> MIPS_PWSIZE_PS_SHIFT,
2423 		(config & MIPS_PWSIZE_GDW_MASK) >> MIPS_PWSIZE_GDW_SHIFT,
2424 		(config & MIPS_PWSIZE_UDW_MASK) >> MIPS_PWSIZE_UDW_SHIFT,
2425 		(config & MIPS_PWSIZE_MDW_MASK) >> MIPS_PWSIZE_MDW_SHIFT,
2426 		(config & MIPS_PWSIZE_PTW_MASK) >> MIPS_PWSIZE_PTW_SHIFT,
2427 		(config & MIPS_PWSIZE_PTEW_MASK) >> MIPS_PWSIZE_PTEW_SHIFT);
2428 
2429 	pwctl = read_c0_pwctl();
2430 	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",
2431 		pwctl,
2432 		(pwctl & MIPS_PWCTL_PWEN_MASK) >> MIPS_PWCTL_PWEN_SHIFT,
2433 		(pwctl & MIPS_PWCTL_XK_MASK) >> MIPS_PWCTL_XK_SHIFT,
2434 		(pwctl & MIPS_PWCTL_XS_MASK) >> MIPS_PWCTL_XS_SHIFT,
2435 		(pwctl & MIPS_PWCTL_XU_MASK) >> MIPS_PWCTL_XU_SHIFT,
2436 		(pwctl & MIPS_PWCTL_DPH_MASK) >> MIPS_PWCTL_DPH_SHIFT,
2437 		(pwctl & MIPS_PWCTL_HUGEPG_MASK) >> MIPS_PWCTL_HUGEPG_SHIFT,
2438 		(pwctl & MIPS_PWCTL_PSN_MASK) >> MIPS_PWCTL_PSN_SHIFT);
2439 }
2440 
2441 static void config_htw_params(void)
2442 {
2443 	unsigned long pwfield, pwsize, ptei;
2444 	unsigned int config;
2445 
2446 	/*
2447 	 * We are using 2-level page tables, so we only need to
2448 	 * setup GDW and PTW appropriately. UDW and MDW will remain 0.
2449 	 * The default value of GDI/UDI/MDI/PTI is 0xc. It is illegal to
2450 	 * write values less than 0xc in these fields because the entire
2451 	 * write will be dropped. As a result of which, we must preserve
2452 	 * the original reset values and overwrite only what we really want.
2453 	 */
2454 
2455 	pwfield = read_c0_pwfield();
2456 	/* re-initialize the GDI field */
2457 	pwfield &= ~MIPS_PWFIELD_GDI_MASK;
2458 	pwfield |= PGDIR_SHIFT << MIPS_PWFIELD_GDI_SHIFT;
2459 	/* re-initialize the PTI field including the even/odd bit */
2460 	pwfield &= ~MIPS_PWFIELD_PTI_MASK;
2461 	pwfield |= PAGE_SHIFT << MIPS_PWFIELD_PTI_SHIFT;
2462 	if (CONFIG_PGTABLE_LEVELS >= 3) {
2463 		pwfield &= ~MIPS_PWFIELD_MDI_MASK;
2464 		pwfield |= PMD_SHIFT << MIPS_PWFIELD_MDI_SHIFT;
2465 	}
2466 	/* Set the PTEI right shift */
2467 	ptei = _PAGE_GLOBAL_SHIFT << MIPS_PWFIELD_PTEI_SHIFT;
2468 	pwfield |= ptei;
2469 	write_c0_pwfield(pwfield);
2470 	/* Check whether the PTEI value is supported */
2471 	back_to_back_c0_hazard();
2472 	pwfield = read_c0_pwfield();
2473 	if (((pwfield & MIPS_PWFIELD_PTEI_MASK) << MIPS_PWFIELD_PTEI_SHIFT)
2474 		!= ptei) {
2475 		pr_warn("Unsupported PTEI field value: 0x%lx. HTW will not be enabled",
2476 			ptei);
2477 		/*
2478 		 * Drop option to avoid HTW being enabled via another path
2479 		 * (eg htw_reset())
2480 		 */
2481 		current_cpu_data.options &= ~MIPS_CPU_HTW;
2482 		return;
2483 	}
2484 
2485 	pwsize = ilog2(PTRS_PER_PGD) << MIPS_PWSIZE_GDW_SHIFT;
2486 	pwsize |= ilog2(PTRS_PER_PTE) << MIPS_PWSIZE_PTW_SHIFT;
2487 	if (CONFIG_PGTABLE_LEVELS >= 3)
2488 		pwsize |= ilog2(PTRS_PER_PMD) << MIPS_PWSIZE_MDW_SHIFT;
2489 
2490 	/* Set pointer size to size of directory pointers */
2491 	if (IS_ENABLED(CONFIG_64BIT))
2492 		pwsize |= MIPS_PWSIZE_PS_MASK;
2493 	/* PTEs may be multiple pointers long (e.g. with XPA) */
2494 	pwsize |= ((PTE_T_LOG2 - PGD_T_LOG2) << MIPS_PWSIZE_PTEW_SHIFT)
2495 			& MIPS_PWSIZE_PTEW_MASK;
2496 
2497 	write_c0_pwsize(pwsize);
2498 
2499 	/* Make sure everything is set before we enable the HTW */
2500 	back_to_back_c0_hazard();
2501 
2502 	/*
2503 	 * Enable HTW (and only for XUSeg on 64-bit), and disable the rest of
2504 	 * the pwctl fields.
2505 	 */
2506 	config = 1 << MIPS_PWCTL_PWEN_SHIFT;
2507 	if (IS_ENABLED(CONFIG_64BIT))
2508 		config |= MIPS_PWCTL_XU_MASK;
2509 	write_c0_pwctl(config);
2510 	pr_info("Hardware Page Table Walker enabled\n");
2511 
2512 	print_htw_config();
2513 }
2514 
2515 static void config_xpa_params(void)
2516 {
2517 #ifdef CONFIG_XPA
2518 	unsigned int pagegrain;
2519 
2520 	if (mips_xpa_disabled) {
2521 		pr_info("Extended Physical Addressing (XPA) disabled\n");
2522 		return;
2523 	}
2524 
2525 	pagegrain = read_c0_pagegrain();
2526 	write_c0_pagegrain(pagegrain | PG_ELPA);
2527 	back_to_back_c0_hazard();
2528 	pagegrain = read_c0_pagegrain();
2529 
2530 	if (pagegrain & PG_ELPA)
2531 		pr_info("Extended Physical Addressing (XPA) enabled\n");
2532 	else
2533 		panic("Extended Physical Addressing (XPA) disabled");
2534 #endif
2535 }
2536 
2537 static void check_pabits(void)
2538 {
2539 	unsigned long entry;
2540 	unsigned pabits, fillbits;
2541 
2542 	if (!cpu_has_rixi || !_PAGE_NO_EXEC) {
2543 		/*
2544 		 * We'll only be making use of the fact that we can rotate bits
2545 		 * into the fill if the CPU supports RIXI, so don't bother
2546 		 * probing this for CPUs which don't.
2547 		 */
2548 		return;
2549 	}
2550 
2551 	write_c0_entrylo0(~0ul);
2552 	back_to_back_c0_hazard();
2553 	entry = read_c0_entrylo0();
2554 
2555 	/* clear all non-PFN bits */
2556 	entry &= ~((1 << MIPS_ENTRYLO_PFN_SHIFT) - 1);
2557 	entry &= ~(MIPS_ENTRYLO_RI | MIPS_ENTRYLO_XI);
2558 
2559 	/* find a lower bound on PABITS, and upper bound on fill bits */
2560 	pabits = fls_long(entry) + 6;
2561 	fillbits = max_t(int, (int)BITS_PER_LONG - pabits, 0);
2562 
2563 	/* minus the RI & XI bits */
2564 	fillbits -= min_t(unsigned, fillbits, 2);
2565 
2566 	if (fillbits >= ilog2(_PAGE_NO_EXEC))
2567 		fill_includes_sw_bits = true;
2568 
2569 	pr_debug("Entry* registers contain %u fill bits\n", fillbits);
2570 }
2571 
2572 void build_tlb_refill_handler(void)
2573 {
2574 	/*
2575 	 * The refill handler is generated per-CPU, multi-node systems
2576 	 * may have local storage for it. The other handlers are only
2577 	 * needed once.
2578 	 */
2579 	static int run_once = 0;
2580 
2581 	if (IS_ENABLED(CONFIG_XPA) && !cpu_has_rixi)
2582 		panic("Kernels supporting XPA currently require CPUs with RIXI");
2583 
2584 	output_pgtable_bits_defines();
2585 	check_pabits();
2586 
2587 #ifdef CONFIG_64BIT
2588 	check_for_high_segbits = current_cpu_data.vmbits > (PGDIR_SHIFT + PGD_ORDER + PAGE_SHIFT - 3);
2589 #endif
2590 
2591 	switch (current_cpu_type()) {
2592 	case CPU_R2000:
2593 	case CPU_R3000:
2594 	case CPU_R3000A:
2595 	case CPU_R3081E:
2596 	case CPU_TX3912:
2597 	case CPU_TX3922:
2598 	case CPU_TX3927:
2599 #ifndef CONFIG_MIPS_PGD_C0_CONTEXT
2600 		if (cpu_has_local_ebase)
2601 			build_r3000_tlb_refill_handler();
2602 		if (!run_once) {
2603 			if (!cpu_has_local_ebase)
2604 				build_r3000_tlb_refill_handler();
2605 			build_setup_pgd();
2606 			build_r3000_tlb_load_handler();
2607 			build_r3000_tlb_store_handler();
2608 			build_r3000_tlb_modify_handler();
2609 			flush_tlb_handlers();
2610 			run_once++;
2611 		}
2612 #else
2613 		panic("No R3000 TLB refill handler");
2614 #endif
2615 		break;
2616 
2617 	case CPU_R8000:
2618 		panic("No R8000 TLB refill handler yet");
2619 		break;
2620 
2621 	default:
2622 		if (cpu_has_ldpte)
2623 			setup_pw();
2624 
2625 		if (!run_once) {
2626 			scratch_reg = allocate_kscratch();
2627 			build_setup_pgd();
2628 			build_r4000_tlb_load_handler();
2629 			build_r4000_tlb_store_handler();
2630 			build_r4000_tlb_modify_handler();
2631 			if (cpu_has_ldpte)
2632 				build_loongson3_tlb_refill_handler();
2633 			else if (!cpu_has_local_ebase)
2634 				build_r4000_tlb_refill_handler();
2635 			flush_tlb_handlers();
2636 			run_once++;
2637 		}
2638 		if (cpu_has_local_ebase)
2639 			build_r4000_tlb_refill_handler();
2640 		if (cpu_has_xpa)
2641 			config_xpa_params();
2642 		if (cpu_has_htw)
2643 			config_htw_params();
2644 	}
2645 }
2646