sh: TLB fast path optimizations for load/store exceptions.
[linux-block.git] / arch / sh / kernel / head_64.S
CommitLineData
1da177e4 1/*
a23ba435 2 * arch/sh/kernel/head_64.S
1da177e4
LT
3 *
4 * Copyright (C) 2000, 2001 Paolo Alberelli
5 * Copyright (C) 2003, 2004 Paul Mundt
6 *
a23ba435
PM
7 * This file is subject to the terms and conditions of the GNU General Public
8 * License. See the file "COPYING" in the main directory of this archive
9 * for more details.
1da177e4 10 */
bbe215c2
TA
11
12#include <linux/init.h>
13
1da177e4 14#include <asm/page.h>
1da177e4
LT
15#include <asm/cache.h>
16#include <asm/tlb.h>
f15cbe6f
PM
17#include <cpu/registers.h>
18#include <cpu/mmu_context.h>
1da177e4
LT
19#include <asm/thread_info.h>
20
21/*
22 * MMU defines: TLB boundaries.
23 */
24
25#define MMUIR_FIRST ITLB_FIXED
26#define MMUIR_END ITLB_LAST_VAR_UNRESTRICTED+TLB_STEP
27#define MMUIR_STEP TLB_STEP
28
29#define MMUDR_FIRST DTLB_FIXED
30#define MMUDR_END DTLB_LAST_VAR_UNRESTRICTED+TLB_STEP
31#define MMUDR_STEP TLB_STEP
32
36763b22
PM
33/* Safety check : CONFIG_PAGE_OFFSET has to be a multiple of 512Mb */
34#if (CONFIG_PAGE_OFFSET & ((1UL<<29)-1))
35#error "CONFIG_PAGE_OFFSET must be a multiple of 512Mb"
1da177e4
LT
36#endif
37
38/*
39 * MMU defines: Fixed TLBs.
40 */
41/* Deal safely with the case where the base of RAM is not 512Mb aligned */
42
43#define ALIGN_512M_MASK (0xffffffffe0000000)
36763b22 44#define ALIGNED_EFFECTIVE ((CONFIG_PAGE_OFFSET + CONFIG_MEMORY_START) & ALIGN_512M_MASK)
1da177e4
LT
45#define ALIGNED_PHYSICAL (CONFIG_MEMORY_START & ALIGN_512M_MASK)
46
47#define MMUIR_TEXT_H (0x0000000000000003 | ALIGNED_EFFECTIVE)
48 /* Enabled, Shared, ASID 0, Eff. Add. 0xA0000000 */
49
50#define MMUIR_TEXT_L (0x000000000000009a | ALIGNED_PHYSICAL)
51 /* 512 Mb, Cacheable, Write-back, execute, Not User, Ph. Add. */
52
53#define MMUDR_CACHED_H 0x0000000000000003 | ALIGNED_EFFECTIVE
54 /* Enabled, Shared, ASID 0, Eff. Add. 0xA0000000 */
55#define MMUDR_CACHED_L 0x000000000000015a | ALIGNED_PHYSICAL
56 /* 512 Mb, Cacheable, Write-back, read/write, Not User, Ph. Add. */
57
c96bcf95 58#ifdef CONFIG_CACHE_OFF
1da177e4
LT
59#define ICCR0_INIT_VAL ICCR0_OFF /* ICACHE off */
60#else
61#define ICCR0_INIT_VAL ICCR0_ON | ICCR0_ICI /* ICE + ICI */
62#endif
63#define ICCR1_INIT_VAL ICCR1_NOLOCK /* No locking */
64
c96bcf95 65#if defined (CONFIG_CACHE_OFF)
1da177e4 66#define OCCR0_INIT_VAL OCCR0_OFF /* D-cache: off */
c96bcf95 67#elif defined (CONFIG_CACHE_WRITETHROUGH)
1da177e4
LT
68#define OCCR0_INIT_VAL OCCR0_ON | OCCR0_OCI | OCCR0_WT /* D-cache: on, */
69 /* WT, invalidate */
c96bcf95 70#elif defined (CONFIG_CACHE_WRITEBACK)
1da177e4
LT
71#define OCCR0_INIT_VAL OCCR0_ON | OCCR0_OCI | OCCR0_WB /* D-cache: on, */
72 /* WB, invalidate */
73#else
c96bcf95 74#error preprocessor flag CONFIG_CACHE_... not recognized!
1da177e4
LT
75#endif
76
77#define OCCR1_INIT_VAL OCCR1_NOLOCK /* No locking */
78
79 .section .empty_zero_page, "aw"
80 .global empty_zero_page
81
82empty_zero_page:
83 .long 1 /* MOUNT_ROOT_RDONLY */
84 .long 0 /* RAMDISK_FLAGS */
85 .long 0x0200 /* ORIG_ROOT_DEV */
86 .long 1 /* LOADER_TYPE */
87 .long 0x00800000 /* INITRD_START */
88 .long 0x00800000 /* INITRD_SIZE */
89 .long 0
90
91 .text
92 .balign 4096,0,4096
93
94 .section .data, "aw"
95 .balign PAGE_SIZE
96
97 .section .data, "aw"
98 .balign PAGE_SIZE
99
061854fd
PM
100 .global mmu_pdtp_cache
101mmu_pdtp_cache:
1da177e4
LT
102 .space PAGE_SIZE, 0
103
104 .global empty_bad_page
105empty_bad_page:
106 .space PAGE_SIZE, 0
107
108 .global empty_bad_pte_table
109empty_bad_pte_table:
110 .space PAGE_SIZE, 0
111
112 .global fpu_in_use
113fpu_in_use: .quad 0
114
115
bbe215c2 116 __HEAD
1da177e4
LT
117 .balign L1_CACHE_BYTES
118/*
119 * Condition at the entry of __stext:
120 * . Reset state:
121 * . SR.FD = 1 (FPU disabled)
122 * . SR.BL = 1 (Exceptions disabled)
123 * . SR.MD = 1 (Privileged Mode)
124 * . SR.MMU = 0 (MMU Disabled)
125 * . SR.CD = 0 (CTC User Visible)
126 * . SR.IMASK = Undefined (Interrupt Mask)
127 *
128 * Operations supposed to be performed by __stext:
129 * . prevent speculative fetch onto device memory while MMU is off
130 * . reflect as much as possible SH5 ABI (r15, r26, r27, r18)
131 * . first, save CPU state and set it to something harmless
132 * . any CPU detection and/or endianness settings (?)
133 * . initialize EMI/LMI (but not TMU/RTC/INTC/SCIF): TBD
134 * . set initial TLB entries for cached and uncached regions
135 * (no fine granularity paging)
136 * . set initial cache state
137 * . enable MMU and caches
138 * . set CPU to a consistent state
139 * . registers (including stack pointer and current/KCR0)
140 * . NOT expecting to set Exception handling nor VBR/RESVEC/DCR
141 * at this stage. This is all to later Linux initialization steps.
142 * . initialize FPU
143 * . clear BSS
144 * . jump into start_kernel()
145 * . be prepared to hopeless start_kernel() returns.
146 *
147 */
148 .global _stext
149_stext:
150 /*
151 * Prevent speculative fetch on device memory due to
152 * uninitialized target registers.
153 */
154 ptabs/u ZERO, tr0
155 ptabs/u ZERO, tr1
156 ptabs/u ZERO, tr2
157 ptabs/u ZERO, tr3
158 ptabs/u ZERO, tr4
159 ptabs/u ZERO, tr5
160 ptabs/u ZERO, tr6
161 ptabs/u ZERO, tr7
162 synci
163
164 /*
165 * Read/Set CPU state. After this block:
166 * r29 = Initial SR
167 */
168 getcon SR, r29
169 movi SR_HARMLESS, r20
170 putcon r20, SR
171
172 /*
173 * Initialize EMI/LMI. To Be Done.
174 */
175
176 /*
177 * CPU detection and/or endianness settings (?). To Be Done.
178 * Pure PIC code here, please ! Just save state into r30.
179 * After this block:
180 * r30 = CPU type/Platform Endianness
181 */
182
183 /*
184 * Set initial TLB entries for cached and uncached regions.
185 * Note: PTA/BLINK is PIC code, PTABS/BLINK isn't !
186 */
187 /* Clear ITLBs */
188 pta clear_ITLB, tr1
189 movi MMUIR_FIRST, r21
190 movi MMUIR_END, r22
191clear_ITLB:
192 putcfg r21, 0, ZERO /* Clear MMUIR[n].PTEH.V */
193 addi r21, MMUIR_STEP, r21
194 bne r21, r22, tr1
195
196 /* Clear DTLBs */
197 pta clear_DTLB, tr1
198 movi MMUDR_FIRST, r21
199 movi MMUDR_END, r22
200clear_DTLB:
201 putcfg r21, 0, ZERO /* Clear MMUDR[n].PTEH.V */
202 addi r21, MMUDR_STEP, r21
203 bne r21, r22, tr1
204
205 /* Map one big (512Mb) page for ITLB */
206 movi MMUIR_FIRST, r21
207 movi MMUIR_TEXT_L, r22 /* PTEL first */
208 add.l r22, r63, r22 /* Sign extend */
209 putcfg r21, 1, r22 /* Set MMUIR[0].PTEL */
210 movi MMUIR_TEXT_H, r22 /* PTEH last */
211 add.l r22, r63, r22 /* Sign extend */
212 putcfg r21, 0, r22 /* Set MMUIR[0].PTEH */
213
214 /* Map one big CACHED (512Mb) page for DTLB */
215 movi MMUDR_FIRST, r21
216 movi MMUDR_CACHED_L, r22 /* PTEL first */
217 add.l r22, r63, r22 /* Sign extend */
218 putcfg r21, 1, r22 /* Set MMUDR[0].PTEL */
219 movi MMUDR_CACHED_H, r22 /* PTEH last */
220 add.l r22, r63, r22 /* Sign extend */
221 putcfg r21, 0, r22 /* Set MMUDR[0].PTEH */
222
223#ifdef CONFIG_EARLY_PRINTK
224 /*
225 * Setup a DTLB translation for SCIF phys.
226 */
227 addi r21, MMUDR_STEP, r21
228 movi 0x0a03, r22 /* SCIF phys */
229 shori 0x0148, r22
230 putcfg r21, 1, r22 /* PTEL first */
231 movi 0xfa03, r22 /* 0xfa030000, fixed SCIF virt */
232 shori 0x0003, r22
233 putcfg r21, 0, r22 /* PTEH last */
234#endif
235
236 /*
237 * Set cache behaviours.
238 */
239 /* ICache */
240 movi ICCR_BASE, r21
241 movi ICCR0_INIT_VAL, r22
242 movi ICCR1_INIT_VAL, r23
243 putcfg r21, ICCR_REG0, r22
244 putcfg r21, ICCR_REG1, r23
245
246 /* OCache */
247 movi OCCR_BASE, r21
248 movi OCCR0_INIT_VAL, r22
249 movi OCCR1_INIT_VAL, r23
250 putcfg r21, OCCR_REG0, r22
251 putcfg r21, OCCR_REG1, r23
252
253
254 /*
255 * Enable Caches and MMU. Do the first non-PIC jump.
256 * Now head.S global variables, constants and externs
257 * can be used.
258 */
259 getcon SR, r21
260 movi SR_ENABLE_MMU, r22
261 or r21, r22, r21
262 putcon r21, SSR
263 movi hyperspace, r22
264 ori r22, 1, r22 /* Make it SHmedia, not required but..*/
265 putcon r22, SPC
266 synco
267 rte /* And now go into the hyperspace ... */
268hyperspace: /* ... that's the next instruction ! */
269
270 /*
271 * Set CPU to a consistent state.
272 * r31 = FPU support flag
273 * tr0/tr7 in use. Others give a chance to loop somewhere safe
274 */
275 movi start_kernel, r32
276 ori r32, 1, r32
277
278 ptabs r32, tr0 /* r32 = _start_kernel address */
279 pta/u hopeless, tr1
280 pta/u hopeless, tr2
281 pta/u hopeless, tr3
282 pta/u hopeless, tr4
283 pta/u hopeless, tr5
284 pta/u hopeless, tr6
285 pta/u hopeless, tr7
286 gettr tr1, r28 /* r28 = hopeless address */
287
288 /* Set initial stack pointer */
289 movi init_thread_union, SP
290 putcon SP, KCR0 /* Set current to init_task */
291 movi THREAD_SIZE, r22 /* Point to the end */
292 add SP, r22, SP
293
294 /*
295 * Initialize FPU.
296 * Keep FPU flag in r31. After this block:
297 * r31 = FPU flag
298 */
299 movi fpu_in_use, r31 /* Temporary */
300
301#ifdef CONFIG_SH_FPU
302 getcon SR, r21
303 movi SR_ENABLE_FPU, r22
304 and r21, r22, r22
305 putcon r22, SR /* Try to enable */
306 getcon SR, r22
307 xor r21, r22, r21
308 shlri r21, 15, r21 /* Supposedly 0/1 */
309 st.q r31, 0 , r21 /* Set fpu_in_use */
310#else
311 movi 0, r21
312 st.q r31, 0 , r21 /* Set fpu_in_use */
313#endif
314 or r21, ZERO, r31 /* Set FPU flag at last */
315
316#ifndef CONFIG_SH_NO_BSS_INIT
317/* Don't clear BSS if running on slow platforms such as an RTL simulation,
318 remote memory via SHdebug link, etc. For these the memory can be guaranteed
319 to be all zero on boot anyway. */
320 /*
321 * Clear bss
322 */
323 pta clear_quad, tr1
324 movi __bss_start, r22
325 movi _end, r23
326clear_quad:
327 st.q r22, 0, ZERO
328 addi r22, 8, r22
329 bne r22, r23, tr1 /* Both quad aligned, see vmlinux.lds.S */
330#endif
331 pta/u hopeless, tr1
332
333 /* Say bye to head.S but be prepared to wrongly get back ... */
334 blink tr0, LINK
335
336 /* If we ever get back here through LINK/tr1-tr7 */
337 pta/u hopeless, tr7
338
339hopeless:
340 /*
341 * Something's badly wrong here. Loop endlessly,
342 * there's nothing more we can do about it.
343 *
344 * Note on hopeless: it can be jumped into invariably
345 * before or after jumping into hyperspace. The only
346 * requirement is to be PIC called (PTA) before and
347 * any way (PTA/PTABS) after. According to Virtual
348 * to Physical mapping a simulator/emulator can easily
349 * tell where we came here from just looking at hopeless
350 * (PC) address.
351 *
352 * For debugging purposes:
353 * (r28) hopeless/loop address
354 * (r29) Original SR
355 * (r30) CPU type/Platform endianness
356 * (r31) FPU Support
357 * (r32) _start_kernel address
358 */
359 blink tr7, ZERO