diff options
Diffstat (limited to 'arch/sparc64/kernel/tsb.S')
-rw-r--r-- | arch/sparc64/kernel/tsb.S | 442 |
1 files changed, 442 insertions, 0 deletions
diff --git a/arch/sparc64/kernel/tsb.S b/arch/sparc64/kernel/tsb.S new file mode 100644 index 000000000000..118baea44f69 --- /dev/null +++ b/arch/sparc64/kernel/tsb.S | |||
@@ -0,0 +1,442 @@ | |||
1 | /* tsb.S: Sparc64 TSB table handling. | ||
2 | * | ||
3 | * Copyright (C) 2006 David S. Miller <davem@davemloft.net> | ||
4 | */ | ||
5 | |||
6 | #include <asm/tsb.h> | ||
7 | #include <asm/hypervisor.h> | ||
8 | |||
9 | .text | ||
10 | .align 32 | ||
11 | |||
12 | /* Invoked from TLB miss handler, we are in the | ||
13 | * MMU global registers and they are setup like | ||
14 | * this: | ||
15 | * | ||
16 | * %g1: TSB entry pointer | ||
17 | * %g2: available temporary | ||
18 | * %g3: FAULT_CODE_{D,I}TLB | ||
19 | * %g4: available temporary | ||
20 | * %g5: available temporary | ||
21 | * %g6: TAG TARGET | ||
22 | * %g7: available temporary, will be loaded by us with | ||
23 | * the physical address base of the linux page | ||
24 | * tables for the current address space | ||
25 | */ | ||
26 | tsb_miss_dtlb: | ||
27 | mov TLB_TAG_ACCESS, %g4 | ||
28 | ba,pt %xcc, tsb_miss_page_table_walk | ||
29 | ldxa [%g4] ASI_DMMU, %g4 | ||
30 | |||
31 | tsb_miss_itlb: | ||
32 | mov TLB_TAG_ACCESS, %g4 | ||
33 | ba,pt %xcc, tsb_miss_page_table_walk | ||
34 | ldxa [%g4] ASI_IMMU, %g4 | ||
35 | |||
36 | /* At this point we have: | ||
37 | * %g1 -- TSB entry address | ||
38 | * %g3 -- FAULT_CODE_{D,I}TLB | ||
39 | * %g4 -- missing virtual address | ||
40 | * %g6 -- TAG TARGET (vaddr >> 22) | ||
41 | */ | ||
42 | tsb_miss_page_table_walk: | ||
43 | TRAP_LOAD_PGD_PHYS(%g7, %g5) | ||
44 | |||
45 | /* And now we have the PGD base physical address in %g7. */ | ||
46 | tsb_miss_page_table_walk_sun4v_fastpath: | ||
47 | USER_PGTABLE_WALK_TL1(%g4, %g7, %g5, %g2, tsb_do_fault) | ||
48 | |||
49 | /* At this point we have: | ||
50 | * %g1 -- TSB entry address | ||
51 | * %g3 -- FAULT_CODE_{D,I}TLB | ||
52 | * %g5 -- physical address of PTE in Linux page tables | ||
53 | * %g6 -- TAG TARGET (vaddr >> 22) | ||
54 | */ | ||
55 | tsb_reload: | ||
56 | TSB_LOCK_TAG(%g1, %g2, %g7) | ||
57 | |||
58 | /* Load and check PTE. */ | ||
59 | ldxa [%g5] ASI_PHYS_USE_EC, %g5 | ||
60 | mov 1, %g7 | ||
61 | sllx %g7, TSB_TAG_INVALID_BIT, %g7 | ||
62 | brgez,a,pn %g5, tsb_do_fault | ||
63 | TSB_STORE(%g1, %g7) | ||
64 | |||
65 | TSB_WRITE(%g1, %g5, %g6) | ||
66 | |||
67 | /* Finally, load TLB and return from trap. */ | ||
68 | tsb_tlb_reload: | ||
69 | cmp %g3, FAULT_CODE_DTLB | ||
70 | bne,pn %xcc, tsb_itlb_load | ||
71 | nop | ||
72 | |||
73 | tsb_dtlb_load: | ||
74 | |||
75 | 661: stxa %g5, [%g0] ASI_DTLB_DATA_IN | ||
76 | retry | ||
77 | .section .sun4v_2insn_patch, "ax" | ||
78 | .word 661b | ||
79 | nop | ||
80 | nop | ||
81 | .previous | ||
82 | |||
83 | /* For sun4v the ASI_DTLB_DATA_IN store and the retry | ||
84 | * instruction get nop'd out and we get here to branch | ||
85 | * to the sun4v tlb load code. The registers are setup | ||
86 | * as follows: | ||
87 | * | ||
88 | * %g4: vaddr | ||
89 | * %g5: PTE | ||
90 | * %g6: TAG | ||
91 | * | ||
92 | * The sun4v TLB load wants the PTE in %g3 so we fix that | ||
93 | * up here. | ||
94 | */ | ||
95 | ba,pt %xcc, sun4v_dtlb_load | ||
96 | mov %g5, %g3 | ||
97 | |||
98 | tsb_itlb_load: | ||
99 | /* Executable bit must be set. */ | ||
100 | 661: andcc %g5, _PAGE_EXEC_4U, %g0 | ||
101 | .section .sun4v_1insn_patch, "ax" | ||
102 | .word 661b | ||
103 | andcc %g5, _PAGE_EXEC_4V, %g0 | ||
104 | .previous | ||
105 | |||
106 | be,pn %xcc, tsb_do_fault | ||
107 | nop | ||
108 | |||
109 | 661: stxa %g5, [%g0] ASI_ITLB_DATA_IN | ||
110 | retry | ||
111 | .section .sun4v_2insn_patch, "ax" | ||
112 | .word 661b | ||
113 | nop | ||
114 | nop | ||
115 | .previous | ||
116 | |||
117 | /* For sun4v the ASI_ITLB_DATA_IN store and the retry | ||
118 | * instruction get nop'd out and we get here to branch | ||
119 | * to the sun4v tlb load code. The registers are setup | ||
120 | * as follows: | ||
121 | * | ||
122 | * %g4: vaddr | ||
123 | * %g5: PTE | ||
124 | * %g6: TAG | ||
125 | * | ||
126 | * The sun4v TLB load wants the PTE in %g3 so we fix that | ||
127 | * up here. | ||
128 | */ | ||
129 | ba,pt %xcc, sun4v_itlb_load | ||
130 | mov %g5, %g3 | ||
131 | |||
132 | /* No valid entry in the page tables, do full fault | ||
133 | * processing. | ||
134 | */ | ||
135 | |||
136 | .globl tsb_do_fault | ||
137 | tsb_do_fault: | ||
138 | cmp %g3, FAULT_CODE_DTLB | ||
139 | |||
140 | 661: rdpr %pstate, %g5 | ||
141 | wrpr %g5, PSTATE_AG | PSTATE_MG, %pstate | ||
142 | .section .sun4v_2insn_patch, "ax" | ||
143 | .word 661b | ||
144 | SET_GL(1) | ||
145 | ldxa [%g0] ASI_SCRATCHPAD, %g4 | ||
146 | .previous | ||
147 | |||
148 | bne,pn %xcc, tsb_do_itlb_fault | ||
149 | nop | ||
150 | |||
151 | tsb_do_dtlb_fault: | ||
152 | rdpr %tl, %g3 | ||
153 | cmp %g3, 1 | ||
154 | |||
155 | 661: mov TLB_TAG_ACCESS, %g4 | ||
156 | ldxa [%g4] ASI_DMMU, %g5 | ||
157 | .section .sun4v_2insn_patch, "ax" | ||
158 | .word 661b | ||
159 | ldx [%g4 + HV_FAULT_D_ADDR_OFFSET], %g5 | ||
160 | nop | ||
161 | .previous | ||
162 | |||
163 | be,pt %xcc, sparc64_realfault_common | ||
164 | mov FAULT_CODE_DTLB, %g4 | ||
165 | ba,pt %xcc, winfix_trampoline | ||
166 | nop | ||
167 | |||
168 | tsb_do_itlb_fault: | ||
169 | rdpr %tpc, %g5 | ||
170 | ba,pt %xcc, sparc64_realfault_common | ||
171 | mov FAULT_CODE_ITLB, %g4 | ||
172 | |||
173 | .globl sparc64_realfault_common | ||
174 | sparc64_realfault_common: | ||
175 | /* fault code in %g4, fault address in %g5, etrap will | ||
176 | * preserve these two values in %l4 and %l5 respectively | ||
177 | */ | ||
178 | ba,pt %xcc, etrap ! Save trap state | ||
179 | 1: rd %pc, %g7 ! ... | ||
180 | stb %l4, [%g6 + TI_FAULT_CODE] ! Save fault code | ||
181 | stx %l5, [%g6 + TI_FAULT_ADDR] ! Save fault address | ||
182 | call do_sparc64_fault ! Call fault handler | ||
183 | add %sp, PTREGS_OFF, %o0 ! Compute pt_regs arg | ||
184 | ba,pt %xcc, rtrap_clr_l6 ! Restore cpu state | ||
185 | nop ! Delay slot (fill me) | ||
186 | |||
187 | winfix_trampoline: | ||
188 | rdpr %tpc, %g3 ! Prepare winfixup TNPC | ||
189 | or %g3, 0x7c, %g3 ! Compute branch offset | ||
190 | wrpr %g3, %tnpc ! Write it into TNPC | ||
191 | done ! Trap return | ||
192 | |||
193 | /* Insert an entry into the TSB. | ||
194 | * | ||
195 | * %o0: TSB entry pointer (virt or phys address) | ||
196 | * %o1: tag | ||
197 | * %o2: pte | ||
198 | */ | ||
199 | .align 32 | ||
200 | .globl __tsb_insert | ||
201 | __tsb_insert: | ||
202 | rdpr %pstate, %o5 | ||
203 | wrpr %o5, PSTATE_IE, %pstate | ||
204 | TSB_LOCK_TAG(%o0, %g2, %g3) | ||
205 | TSB_WRITE(%o0, %o2, %o1) | ||
206 | wrpr %o5, %pstate | ||
207 | retl | ||
208 | nop | ||
209 | .size __tsb_insert, .-__tsb_insert | ||
210 | |||
211 | /* Flush the given TSB entry if it has the matching | ||
212 | * tag. | ||
213 | * | ||
214 | * %o0: TSB entry pointer (virt or phys address) | ||
215 | * %o1: tag | ||
216 | */ | ||
217 | .align 32 | ||
218 | .globl tsb_flush | ||
219 | .type tsb_flush,#function | ||
220 | tsb_flush: | ||
221 | sethi %hi(TSB_TAG_LOCK_HIGH), %g2 | ||
222 | 1: TSB_LOAD_TAG(%o0, %g1) | ||
223 | srlx %g1, 32, %o3 | ||
224 | andcc %o3, %g2, %g0 | ||
225 | bne,pn %icc, 1b | ||
226 | membar #LoadLoad | ||
227 | cmp %g1, %o1 | ||
228 | mov 1, %o3 | ||
229 | bne,pt %xcc, 2f | ||
230 | sllx %o3, TSB_TAG_INVALID_BIT, %o3 | ||
231 | TSB_CAS_TAG(%o0, %g1, %o3) | ||
232 | cmp %g1, %o3 | ||
233 | bne,pn %xcc, 1b | ||
234 | nop | ||
235 | 2: retl | ||
236 | TSB_MEMBAR | ||
237 | .size tsb_flush, .-tsb_flush | ||
238 | |||
239 | /* Reload MMU related context switch state at | ||
240 | * schedule() time. | ||
241 | * | ||
242 | * %o0: page table physical address | ||
243 | * %o1: TSB register value | ||
244 | * %o2: TSB virtual address | ||
245 | * %o3: TSB mapping locked PTE | ||
246 | * %o4: Hypervisor TSB descriptor physical address | ||
247 | * | ||
248 | * We have to run this whole thing with interrupts | ||
249 | * disabled so that the current cpu doesn't change | ||
250 | * due to preemption. | ||
251 | */ | ||
252 | .align 32 | ||
253 | .globl __tsb_context_switch | ||
254 | .type __tsb_context_switch,#function | ||
255 | __tsb_context_switch: | ||
256 | rdpr %pstate, %o5 | ||
257 | wrpr %o5, PSTATE_IE, %pstate | ||
258 | |||
259 | ldub [%g6 + TI_CPU], %g1 | ||
260 | sethi %hi(trap_block), %g2 | ||
261 | sllx %g1, TRAP_BLOCK_SZ_SHIFT, %g1 | ||
262 | or %g2, %lo(trap_block), %g2 | ||
263 | add %g2, %g1, %g2 | ||
264 | stx %o0, [%g2 + TRAP_PER_CPU_PGD_PADDR] | ||
265 | |||
266 | sethi %hi(tlb_type), %g1 | ||
267 | lduw [%g1 + %lo(tlb_type)], %g1 | ||
268 | cmp %g1, 3 | ||
269 | bne,pt %icc, 1f | ||
270 | nop | ||
271 | |||
272 | /* Hypervisor TSB switch. */ | ||
273 | mov SCRATCHPAD_UTSBREG1, %g1 | ||
274 | stxa %o1, [%g1] ASI_SCRATCHPAD | ||
275 | mov -1, %g2 | ||
276 | mov SCRATCHPAD_UTSBREG2, %g1 | ||
277 | stxa %g2, [%g1] ASI_SCRATCHPAD | ||
278 | |||
279 | /* Save away %o5's %pstate, we have to use %o5 for | ||
280 | * the hypervisor call. | ||
281 | */ | ||
282 | mov %o5, %g1 | ||
283 | |||
284 | mov HV_FAST_MMU_TSB_CTXNON0, %o5 | ||
285 | mov 1, %o0 | ||
286 | mov %o4, %o1 | ||
287 | ta HV_FAST_TRAP | ||
288 | |||
289 | /* Finish up and restore %o5. */ | ||
290 | ba,pt %xcc, 9f | ||
291 | mov %g1, %o5 | ||
292 | |||
293 | /* SUN4U TSB switch. */ | ||
294 | 1: mov TSB_REG, %g1 | ||
295 | stxa %o1, [%g1] ASI_DMMU | ||
296 | membar #Sync | ||
297 | stxa %o1, [%g1] ASI_IMMU | ||
298 | membar #Sync | ||
299 | |||
300 | 2: brz %o2, 9f | ||
301 | nop | ||
302 | |||
303 | sethi %hi(sparc64_highest_unlocked_tlb_ent), %g2 | ||
304 | mov TLB_TAG_ACCESS, %g1 | ||
305 | lduw [%g2 + %lo(sparc64_highest_unlocked_tlb_ent)], %g2 | ||
306 | stxa %o2, [%g1] ASI_DMMU | ||
307 | membar #Sync | ||
308 | sllx %g2, 3, %g2 | ||
309 | stxa %o3, [%g2] ASI_DTLB_DATA_ACCESS | ||
310 | membar #Sync | ||
311 | 9: | ||
312 | wrpr %o5, %pstate | ||
313 | |||
314 | retl | ||
315 | nop | ||
316 | .size __tsb_context_switch, .-__tsb_context_switch | ||
317 | |||
318 | #define TSB_PASS_BITS ((1 << TSB_TAG_LOCK_BIT) | \ | ||
319 | (1 << TSB_TAG_INVALID_BIT)) | ||
320 | |||
321 | .align 32 | ||
322 | .globl copy_tsb | ||
323 | .type copy_tsb,#function | ||
324 | copy_tsb: /* %o0=old_tsb_base, %o1=old_tsb_size | ||
325 | * %o2=new_tsb_base, %o3=new_tsb_size | ||
326 | */ | ||
327 | sethi %uhi(TSB_PASS_BITS), %g7 | ||
328 | srlx %o3, 4, %o3 | ||
329 | add %o0, %o1, %g1 /* end of old tsb */ | ||
330 | sllx %g7, 32, %g7 | ||
331 | sub %o3, 1, %o3 /* %o3 == new tsb hash mask */ | ||
332 | |||
333 | 661: prefetcha [%o0] ASI_N, #one_read | ||
334 | .section .tsb_phys_patch, "ax" | ||
335 | .word 661b | ||
336 | prefetcha [%o0] ASI_PHYS_USE_EC, #one_read | ||
337 | .previous | ||
338 | |||
339 | 90: andcc %o0, (64 - 1), %g0 | ||
340 | bne 1f | ||
341 | add %o0, 64, %o5 | ||
342 | |||
343 | 661: prefetcha [%o5] ASI_N, #one_read | ||
344 | .section .tsb_phys_patch, "ax" | ||
345 | .word 661b | ||
346 | prefetcha [%o5] ASI_PHYS_USE_EC, #one_read | ||
347 | .previous | ||
348 | |||
349 | 1: TSB_LOAD_QUAD(%o0, %g2) /* %g2/%g3 == TSB entry */ | ||
350 | andcc %g2, %g7, %g0 /* LOCK or INVALID set? */ | ||
351 | bne,pn %xcc, 80f /* Skip it */ | ||
352 | sllx %g2, 22, %o4 /* TAG --> VADDR */ | ||
353 | |||
354 | /* This can definitely be computed faster... */ | ||
355 | srlx %o0, 4, %o5 /* Build index */ | ||
356 | and %o5, 511, %o5 /* Mask index */ | ||
357 | sllx %o5, PAGE_SHIFT, %o5 /* Put into vaddr position */ | ||
358 | or %o4, %o5, %o4 /* Full VADDR. */ | ||
359 | srlx %o4, PAGE_SHIFT, %o4 /* Shift down to create index */ | ||
360 | and %o4, %o3, %o4 /* Mask with new_tsb_nents-1 */ | ||
361 | sllx %o4, 4, %o4 /* Shift back up into tsb ent offset */ | ||
362 | TSB_STORE(%o2 + %o4, %g2) /* Store TAG */ | ||
363 | add %o4, 0x8, %o4 /* Advance to TTE */ | ||
364 | TSB_STORE(%o2 + %o4, %g3) /* Store TTE */ | ||
365 | |||
366 | 80: add %o0, 16, %o0 | ||
367 | cmp %o0, %g1 | ||
368 | bne,pt %xcc, 90b | ||
369 | nop | ||
370 | |||
371 | retl | ||
372 | TSB_MEMBAR | ||
373 | .size copy_tsb, .-copy_tsb | ||
374 | |||
375 | /* Set the invalid bit in all TSB entries. */ | ||
376 | .align 32 | ||
377 | .globl tsb_init | ||
378 | .type tsb_init,#function | ||
379 | tsb_init: /* %o0 = TSB vaddr, %o1 = size in bytes */ | ||
380 | prefetch [%o0 + 0x000], #n_writes | ||
381 | mov 1, %g1 | ||
382 | prefetch [%o0 + 0x040], #n_writes | ||
383 | sllx %g1, TSB_TAG_INVALID_BIT, %g1 | ||
384 | prefetch [%o0 + 0x080], #n_writes | ||
385 | 1: prefetch [%o0 + 0x0c0], #n_writes | ||
386 | stx %g1, [%o0 + 0x00] | ||
387 | stx %g1, [%o0 + 0x10] | ||
388 | stx %g1, [%o0 + 0x20] | ||
389 | stx %g1, [%o0 + 0x30] | ||
390 | prefetch [%o0 + 0x100], #n_writes | ||
391 | stx %g1, [%o0 + 0x40] | ||
392 | stx %g1, [%o0 + 0x50] | ||
393 | stx %g1, [%o0 + 0x60] | ||
394 | stx %g1, [%o0 + 0x70] | ||
395 | prefetch [%o0 + 0x140], #n_writes | ||
396 | stx %g1, [%o0 + 0x80] | ||
397 | stx %g1, [%o0 + 0x90] | ||
398 | stx %g1, [%o0 + 0xa0] | ||
399 | stx %g1, [%o0 + 0xb0] | ||
400 | prefetch [%o0 + 0x180], #n_writes | ||
401 | stx %g1, [%o0 + 0xc0] | ||
402 | stx %g1, [%o0 + 0xd0] | ||
403 | stx %g1, [%o0 + 0xe0] | ||
404 | stx %g1, [%o0 + 0xf0] | ||
405 | subcc %o1, 0x100, %o1 | ||
406 | bne,pt %xcc, 1b | ||
407 | add %o0, 0x100, %o0 | ||
408 | retl | ||
409 | nop | ||
410 | nop | ||
411 | nop | ||
412 | .size tsb_init, .-tsb_init | ||
413 | |||
414 | .globl NGtsb_init | ||
415 | .type NGtsb_init,#function | ||
416 | NGtsb_init: | ||
417 | rd %asi, %g2 | ||
418 | mov 1, %g1 | ||
419 | wr %g0, ASI_BLK_INIT_QUAD_LDD_P, %asi | ||
420 | sllx %g1, TSB_TAG_INVALID_BIT, %g1 | ||
421 | 1: stxa %g1, [%o0 + 0x00] %asi | ||
422 | stxa %g1, [%o0 + 0x10] %asi | ||
423 | stxa %g1, [%o0 + 0x20] %asi | ||
424 | stxa %g1, [%o0 + 0x30] %asi | ||
425 | stxa %g1, [%o0 + 0x40] %asi | ||
426 | stxa %g1, [%o0 + 0x50] %asi | ||
427 | stxa %g1, [%o0 + 0x60] %asi | ||
428 | stxa %g1, [%o0 + 0x70] %asi | ||
429 | stxa %g1, [%o0 + 0x80] %asi | ||
430 | stxa %g1, [%o0 + 0x90] %asi | ||
431 | stxa %g1, [%o0 + 0xa0] %asi | ||
432 | stxa %g1, [%o0 + 0xb0] %asi | ||
433 | stxa %g1, [%o0 + 0xc0] %asi | ||
434 | stxa %g1, [%o0 + 0xd0] %asi | ||
435 | stxa %g1, [%o0 + 0xe0] %asi | ||
436 | stxa %g1, [%o0 + 0xf0] %asi | ||
437 | subcc %o1, 0x100, %o1 | ||
438 | bne,pt %xcc, 1b | ||
439 | add %o0, 0x100, %o0 | ||
440 | retl | ||
441 | wr %g2, 0x0, %asi | ||
442 | .size NGtsb_init, .-NGtsb_init | ||