1/*
2 * hypersparc.S: High speed Hypersparc mmu/cache operations.
3 *
4 * Copyright (C) 1997 David S. Miller (davem@caip.rutgers.edu)
5 */
6
7#include <asm/ptrace.h>
8#include <asm/psr.h>
9#include <asm/asm-offsets.h>
10#include <asm/asi.h>
11#include <asm/page.h>
12#include <asm/pgtsrmmu.h>
13#include <linux/init.h>
14
15	.text
16	.align	4
17
18	.globl	hypersparc_flush_cache_all, hypersparc_flush_cache_mm
19	.globl	hypersparc_flush_cache_range, hypersparc_flush_cache_page
20	.globl	hypersparc_flush_page_to_ram
21	.globl	hypersparc_flush_page_for_dma, hypersparc_flush_sig_insns
22	.globl	hypersparc_flush_tlb_all, hypersparc_flush_tlb_mm
23	.globl	hypersparc_flush_tlb_range, hypersparc_flush_tlb_page
24
25hypersparc_flush_cache_all:
26	WINDOW_FLUSH(%g4, %g5)
27	sethi	%hi(vac_cache_size), %g4
28	ld	[%g4 + %lo(vac_cache_size)], %g5
29	sethi	%hi(vac_line_size), %g1
30	ld	[%g1 + %lo(vac_line_size)], %g2
311:
32	subcc	%g5, %g2, %g5			! hyper_flush_unconditional_combined
33	bne	1b
34	 sta	%g0, [%g5] ASI_M_FLUSH_CTX
35	retl
36	 sta	%g0, [%g0] ASI_M_FLUSH_IWHOLE	! hyper_flush_whole_icache
37
38	/* We expand the window flush to get maximum performance. */
39hypersparc_flush_cache_mm:
40#ifndef CONFIG_SMP
41	ld	[%o0 + AOFF_mm_context], %g1
42	cmp	%g1, -1
43	be	hypersparc_flush_cache_mm_out
44#endif
45	WINDOW_FLUSH(%g4, %g5)
46
47	sethi	%hi(vac_line_size), %g1
48	ld	[%g1 + %lo(vac_line_size)], %o1
49	sethi	%hi(vac_cache_size), %g2
50	ld	[%g2 + %lo(vac_cache_size)], %o0
51	add	%o1, %o1, %g1
52	add	%o1, %g1, %g2
53	add	%o1, %g2, %g3
54	add	%o1, %g3, %g4
55	add	%o1, %g4, %g5
56	add	%o1, %g5, %o4
57	add	%o1, %o4, %o5
58
59	/* BLAMMO! */
601:
61	subcc	%o0, %o5, %o0				! hyper_flush_cache_user
62	sta	%g0, [%o0 + %g0] ASI_M_FLUSH_USER
63	sta	%g0, [%o0 + %o1] ASI_M_FLUSH_USER
64	sta	%g0, [%o0 + %g1] ASI_M_FLUSH_USER
65	sta	%g0, [%o0 + %g2] ASI_M_FLUSH_USER
66	sta	%g0, [%o0 + %g3] ASI_M_FLUSH_USER
67	sta	%g0, [%o0 + %g4] ASI_M_FLUSH_USER
68	sta	%g0, [%o0 + %g5] ASI_M_FLUSH_USER
69	bne	1b
70	 sta	%g0, [%o0 + %o4] ASI_M_FLUSH_USER
71hypersparc_flush_cache_mm_out:
72	retl
73	 nop
74
75	/* The things we do for performance... */
76hypersparc_flush_cache_range:
77	ld	[%o0 + 0x0], %o0		/* XXX vma->vm_mm, GROSS XXX */
78#ifndef CONFIG_SMP
79	ld	[%o0 + AOFF_mm_context], %g1
80	cmp	%g1, -1
81	be	hypersparc_flush_cache_range_out
82#endif
83	WINDOW_FLUSH(%g4, %g5)
84
85	sethi	%hi(vac_line_size), %g1
86	ld	[%g1 + %lo(vac_line_size)], %o4
87	sethi	%hi(vac_cache_size), %g2
88	ld	[%g2 + %lo(vac_cache_size)], %o3
89
90	/* Here comes the fun part... */
91	add	%o2, (PAGE_SIZE - 1), %o2
92	andn	%o1, (PAGE_SIZE - 1), %o1
93	add	%o4, %o4, %o5
94	andn	%o2, (PAGE_SIZE - 1), %o2
95	add	%o4, %o5, %g1
96	sub	%o2, %o1, %g4
97	add	%o4, %g1, %g2
98	sll	%o3, 2, %g5
99	add	%o4, %g2, %g3
100	cmp	%g4, %g5
101	add	%o4, %g3, %g4
102	blu	0f
103	 add	%o4, %g4, %g5
104	add	%o4, %g5, %g7
105
106	/* Flush entire user space, believe it or not this is quicker
107	 * than page at a time flushings for range > (cache_size<<2).
108	 */
1091:
110	subcc	%o3, %g7, %o3
111	sta	%g0, [%o3 + %g0] ASI_M_FLUSH_USER
112	sta	%g0, [%o3 + %o4] ASI_M_FLUSH_USER
113	sta	%g0, [%o3 + %o5] ASI_M_FLUSH_USER
114	sta	%g0, [%o3 + %g1] ASI_M_FLUSH_USER
115	sta	%g0, [%o3 + %g2] ASI_M_FLUSH_USER
116	sta	%g0, [%o3 + %g3] ASI_M_FLUSH_USER
117	sta	%g0, [%o3 + %g4] ASI_M_FLUSH_USER
118	bne	1b
119	 sta	%g0, [%o3 + %g5] ASI_M_FLUSH_USER
120	retl
121	 nop
122
123	/* Below our threshold, flush one page at a time. */
1240:
125	ld	[%o0 + AOFF_mm_context], %o0
126	mov	SRMMU_CTX_REG, %g7
127	lda	[%g7] ASI_M_MMUREGS, %o3
128	sta	%o0, [%g7] ASI_M_MMUREGS
129	add	%o2, -PAGE_SIZE, %o0
1301:
131	or	%o0, 0x400, %g7
132	lda	[%g7] ASI_M_FLUSH_PROBE, %g7
133	orcc	%g7, 0, %g0
134	be,a	3f
135	 mov	%o0, %o2
136	add	%o4, %g5, %g7
1372:
138	sub	%o2, %g7, %o2
139	sta	%g0, [%o2 + %g0] ASI_M_FLUSH_PAGE
140	sta	%g0, [%o2 + %o4] ASI_M_FLUSH_PAGE
141	sta	%g0, [%o2 + %o5] ASI_M_FLUSH_PAGE
142	sta	%g0, [%o2 + %g1] ASI_M_FLUSH_PAGE
143	sta	%g0, [%o2 + %g2] ASI_M_FLUSH_PAGE
144	sta	%g0, [%o2 + %g3] ASI_M_FLUSH_PAGE
145	andcc	%o2, 0xffc, %g0
146	sta	%g0, [%o2 + %g4] ASI_M_FLUSH_PAGE
147	bne	2b
148	 sta	%g0, [%o2 + %g5] ASI_M_FLUSH_PAGE
1493:
150	cmp	%o2, %o1
151	bne	1b
152	 add	%o2, -PAGE_SIZE, %o0
153	mov	SRMMU_FAULT_STATUS, %g5
154	lda	[%g5] ASI_M_MMUREGS, %g0
155	mov	SRMMU_CTX_REG, %g7
156	sta	%o3, [%g7] ASI_M_MMUREGS
157hypersparc_flush_cache_range_out:
158	retl
159	 nop
160
161	/* HyperSparc requires a valid mapping where we are about to flush
162	 * in order to check for a physical tag match during the flush.
163	 */
164	/* Verified, my ass... */
165hypersparc_flush_cache_page:
166	ld	[%o0 + 0x0], %o0		/* XXX vma->vm_mm, GROSS XXX */
167	ld	[%o0 + AOFF_mm_context], %g2
168#ifndef CONFIG_SMP
169	cmp	%g2, -1
170	be	hypersparc_flush_cache_page_out
171#endif
172	WINDOW_FLUSH(%g4, %g5)
173
174	sethi	%hi(vac_line_size), %g1
175	ld	[%g1 + %lo(vac_line_size)], %o4
176	mov	SRMMU_CTX_REG, %o3
177	andn	%o1, (PAGE_SIZE - 1), %o1
178	lda	[%o3] ASI_M_MMUREGS, %o2
179	sta	%g2, [%o3] ASI_M_MMUREGS
180	or	%o1, 0x400, %o5
181	lda	[%o5] ASI_M_FLUSH_PROBE, %g1
182	orcc	%g0, %g1, %g0
183	be	2f
184	 add	%o4, %o4, %o5
185	sub	%o1, -PAGE_SIZE, %o1
186	add	%o4, %o5, %g1
187	add	%o4, %g1, %g2
188	add	%o4, %g2, %g3
189	add	%o4, %g3, %g4
190	add	%o4, %g4, %g5
191	add	%o4, %g5, %g7
192
193	/* BLAMMO! */
1941:
195	sub	%o1, %g7, %o1
196	sta	%g0, [%o1 + %g0] ASI_M_FLUSH_PAGE
197	sta	%g0, [%o1 + %o4] ASI_M_FLUSH_PAGE
198	sta	%g0, [%o1 + %o5] ASI_M_FLUSH_PAGE
199	sta	%g0, [%o1 + %g1] ASI_M_FLUSH_PAGE
200	sta	%g0, [%o1 + %g2] ASI_M_FLUSH_PAGE
201	sta	%g0, [%o1 + %g3] ASI_M_FLUSH_PAGE
202	andcc	%o1, 0xffc, %g0
203	sta	%g0, [%o1 + %g4] ASI_M_FLUSH_PAGE
204	bne	1b
205	 sta	%g0, [%o1 + %g5] ASI_M_FLUSH_PAGE
2062:
207	mov	SRMMU_FAULT_STATUS, %g7
208	mov	SRMMU_CTX_REG, %g4
209	lda	[%g7] ASI_M_MMUREGS, %g0
210	sta	%o2, [%g4] ASI_M_MMUREGS
211hypersparc_flush_cache_page_out:
212	retl
213	 nop
214
215hypersparc_flush_sig_insns:
216	flush	%o1
217	retl
218	 flush	%o1 + 4
219
220	/* HyperSparc is copy-back. */
221hypersparc_flush_page_to_ram:
222	sethi	%hi(vac_line_size), %g1
223	ld	[%g1 + %lo(vac_line_size)], %o4
224	andn	%o0, (PAGE_SIZE - 1), %o0
225	add	%o4, %o4, %o5
226	or	%o0, 0x400, %g7
227	lda	[%g7] ASI_M_FLUSH_PROBE, %g5
228	add	%o4, %o5, %g1
229	orcc	%g5, 0, %g0
230	be	2f
231	 add	%o4, %g1, %g2
232	add	%o4, %g2, %g3
233	sub	%o0, -PAGE_SIZE, %o0
234	add	%o4, %g3, %g4
235	add	%o4, %g4, %g5
236	add	%o4, %g5, %g7
237
238	/* BLAMMO! */
2391:
240	sub	%o0, %g7, %o0
241	sta	%g0, [%o0 + %g0] ASI_M_FLUSH_PAGE
242	sta	%g0, [%o0 + %o4] ASI_M_FLUSH_PAGE
243	sta	%g0, [%o0 + %o5] ASI_M_FLUSH_PAGE
244	sta	%g0, [%o0 + %g1] ASI_M_FLUSH_PAGE
245	sta	%g0, [%o0 + %g2] ASI_M_FLUSH_PAGE
246	sta	%g0, [%o0 + %g3] ASI_M_FLUSH_PAGE
247	andcc	%o0, 0xffc, %g0
248	sta	%g0, [%o0 + %g4] ASI_M_FLUSH_PAGE
249	bne	1b
250	 sta	%g0, [%o0 + %g5] ASI_M_FLUSH_PAGE
2512:
252	mov	SRMMU_FAULT_STATUS, %g1
253	retl
254	 lda	[%g1] ASI_M_MMUREGS, %g0
255
256	/* HyperSparc is IO cache coherent. */
257hypersparc_flush_page_for_dma:
258	retl
259	 nop
260
261	/* It was noted that at boot time a TLB flush all in a delay slot
262	 * can deliver an illegal instruction to the processor if the timing
263	 * is just right...
264	 */
265hypersparc_flush_tlb_all:
266	mov	0x400, %g1
267	sta	%g0, [%g1] ASI_M_FLUSH_PROBE
268	retl
269	 nop
270
271hypersparc_flush_tlb_mm:
272	mov	SRMMU_CTX_REG, %g1
273	ld	[%o0 + AOFF_mm_context], %o1
274	lda	[%g1] ASI_M_MMUREGS, %g5
275#ifndef CONFIG_SMP
276	cmp	%o1, -1
277	be	hypersparc_flush_tlb_mm_out
278#endif
279	 mov	0x300, %g2
280	sta	%o1, [%g1] ASI_M_MMUREGS
281	sta	%g0, [%g2] ASI_M_FLUSH_PROBE
282hypersparc_flush_tlb_mm_out:
283	retl
284	 sta	%g5, [%g1] ASI_M_MMUREGS
285
286hypersparc_flush_tlb_range:
287	ld	[%o0 + 0x00], %o0	/* XXX vma->vm_mm GROSS XXX */
288	mov	SRMMU_CTX_REG, %g1
289	ld	[%o0 + AOFF_mm_context], %o3
290	lda	[%g1] ASI_M_MMUREGS, %g5
291#ifndef CONFIG_SMP
292	cmp	%o3, -1
293	be	hypersparc_flush_tlb_range_out
294#endif
295	 sethi	%hi(~((1 << SRMMU_PGDIR_SHIFT) - 1)), %o4
296	sta	%o3, [%g1] ASI_M_MMUREGS
297	and	%o1, %o4, %o1
298	add	%o1, 0x200, %o1
299	sta	%g0, [%o1] ASI_M_FLUSH_PROBE
3001:
301	sub	%o1, %o4, %o1
302	cmp	%o1, %o2
303	blu,a	1b
304	 sta	%g0, [%o1] ASI_M_FLUSH_PROBE
305hypersparc_flush_tlb_range_out:
306	retl
307	 sta	%g5, [%g1] ASI_M_MMUREGS
308
309hypersparc_flush_tlb_page:
310	ld	[%o0 + 0x00], %o0	/* XXX vma->vm_mm GROSS XXX */
311	mov	SRMMU_CTX_REG, %g1
312	ld	[%o0 + AOFF_mm_context], %o3
313	andn	%o1, (PAGE_SIZE - 1), %o1
314#ifndef CONFIG_SMP
315	cmp	%o3, -1
316	be	hypersparc_flush_tlb_page_out
317#endif
318	 lda	[%g1] ASI_M_MMUREGS, %g5
319	sta	%o3, [%g1] ASI_M_MMUREGS
320	sta	%g0, [%o1] ASI_M_FLUSH_PROBE
321hypersparc_flush_tlb_page_out:
322	retl
323	 sta	%g5, [%g1] ASI_M_MMUREGS
324
325	__INIT
326
327	/* High speed page clear/copy. */
328hypersparc_bzero_1page:
329/* NOTE: This routine has to be shorter than 40insns --jj */
330	clr	%g1
331	mov	32, %g2
332	mov	64, %g3
333	mov	96, %g4
334	mov	128, %g5
335	mov	160, %g7
336	mov	192, %o2
337	mov	224, %o3
338	mov	16, %o1
3391:
340	stda	%g0, [%o0 + %g0] ASI_M_BFILL
341	stda	%g0, [%o0 + %g2] ASI_M_BFILL
342	stda	%g0, [%o0 + %g3] ASI_M_BFILL
343	stda	%g0, [%o0 + %g4] ASI_M_BFILL
344	stda	%g0, [%o0 + %g5] ASI_M_BFILL
345	stda	%g0, [%o0 + %g7] ASI_M_BFILL
346	stda	%g0, [%o0 + %o2] ASI_M_BFILL
347	stda	%g0, [%o0 + %o3] ASI_M_BFILL
348	subcc	%o1, 1, %o1
349	bne	1b
350	 add	%o0, 256, %o0
351
352	retl
353	 nop
354
355hypersparc_copy_1page:
356/* NOTE: This routine has to be shorter than 70insns --jj */
357	sub	%o1, %o0, %o2		! difference
358	mov	16, %g1
3591:
360	sta	%o0, [%o0 + %o2] ASI_M_BCOPY
361	add	%o0, 32, %o0
362	sta	%o0, [%o0 + %o2] ASI_M_BCOPY
363	add	%o0, 32, %o0
364	sta	%o0, [%o0 + %o2] ASI_M_BCOPY
365	add	%o0, 32, %o0
366	sta	%o0, [%o0 + %o2] ASI_M_BCOPY
367	add	%o0, 32, %o0
368	sta	%o0, [%o0 + %o2] ASI_M_BCOPY
369	add	%o0, 32, %o0
370	sta	%o0, [%o0 + %o2] ASI_M_BCOPY
371	add	%o0, 32, %o0
372	sta	%o0, [%o0 + %o2] ASI_M_BCOPY
373	add	%o0, 32, %o0
374	sta	%o0, [%o0 + %o2] ASI_M_BCOPY
375	subcc	%g1, 1, %g1
376	bne	1b
377	 add	%o0, 32, %o0
378
379	retl
380	 nop
381
382	.globl	hypersparc_setup_blockops
383hypersparc_setup_blockops:
384	sethi	%hi(bzero_1page), %o0
385	or	%o0, %lo(bzero_1page), %o0
386	sethi	%hi(hypersparc_bzero_1page), %o1
387	or	%o1, %lo(hypersparc_bzero_1page), %o1
388	sethi	%hi(hypersparc_copy_1page), %o2
389	or	%o2, %lo(hypersparc_copy_1page), %o2
390	ld	[%o1], %o4
3911:
392	add	%o1, 4, %o1
393	st	%o4, [%o0]
394	add	%o0, 4, %o0
395	cmp	%o1, %o2
396	bne	1b
397	 ld	[%o1], %o4
398	sethi	%hi(__copy_1page), %o0
399	or	%o0, %lo(__copy_1page), %o0
400	sethi	%hi(hypersparc_setup_blockops), %o2
401	or	%o2, %lo(hypersparc_setup_blockops), %o2
402	ld	[%o1], %o4
4031:
404	add	%o1, 4, %o1
405	st	%o4, [%o0]
406	add	%o0, 4, %o0
407	cmp	%o1, %o2
408	bne	1b
409	 ld	[%o1], %o4
410	sta	%g0, [%g0] ASI_M_FLUSH_IWHOLE
411	retl
412	 nop
413