MIPS: Use per-mm page to execute branch delay slot instructions
[cascardo/linux.git] / arch / mips / include / asm / processor.h
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  * Copyright (C) 1994 Waldorf GMBH
7  * Copyright (C) 1995, 1996, 1997, 1998, 1999, 2001, 2002, 2003 Ralf Baechle
8  * Copyright (C) 1996 Paul M. Antoine
9  * Copyright (C) 1999, 2000 Silicon Graphics, Inc.
10  */
11 #ifndef _ASM_PROCESSOR_H
12 #define _ASM_PROCESSOR_H
13
14 #include <linux/atomic.h>
15 #include <linux/cpumask.h>
16 #include <linux/threads.h>
17
18 #include <asm/cachectl.h>
19 #include <asm/cpu.h>
20 #include <asm/cpu-info.h>
21 #include <asm/dsemul.h>
22 #include <asm/mipsregs.h>
23 #include <asm/prefetch.h>
24
25 /*
26  * Return current * instruction pointer ("program counter").
27  */
28 #define current_text_addr() ({ __label__ _l; _l: &&_l;})
29
30 /*
31  * System setup and hardware flags..
32  */
33
34 extern unsigned int vced_count, vcei_count;
35
36 /*
37  * MIPS does have an arch_pick_mmap_layout()
38  */
39 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
40
41 #ifdef CONFIG_32BIT
42 #ifdef CONFIG_KVM_GUEST
43 /* User space process size is limited to 1GB in KVM Guest Mode */
44 #define TASK_SIZE       0x3fff8000UL
45 #else
46 /*
47  * User space process size: 2GB. This is hardcoded into a few places,
48  * so don't change it unless you know what you are doing.
49  */
50 #define TASK_SIZE       0x80000000UL
51 #endif
52
53 #define STACK_TOP_MAX   TASK_SIZE
54
55 #define TASK_IS_32BIT_ADDR 1
56
57 #endif
58
59 #ifdef CONFIG_64BIT
60 /*
61  * User space process size: 1TB. This is hardcoded into a few places,
62  * so don't change it unless you know what you are doing.  TASK_SIZE
63  * is limited to 1TB by the R4000 architecture; R10000 and better can
64  * support 16TB; the architectural reserve for future expansion is
65  * 8192EB ...
66  */
67 #define TASK_SIZE32     0x7fff8000UL
68 #ifdef CONFIG_MIPS_VA_BITS_48
69 #define TASK_SIZE64     (0x1UL << ((cpu_data[0].vmbits>48)?48:cpu_data[0].vmbits))
70 #else
71 #define TASK_SIZE64     0x10000000000UL
72 #endif
73 #define TASK_SIZE (test_thread_flag(TIF_32BIT_ADDR) ? TASK_SIZE32 : TASK_SIZE64)
74 #define STACK_TOP_MAX   TASK_SIZE64
75
76 #define TASK_SIZE_OF(tsk)                                               \
77         (test_tsk_thread_flag(tsk, TIF_32BIT_ADDR) ? TASK_SIZE32 : TASK_SIZE64)
78
79 #define TASK_IS_32BIT_ADDR test_thread_flag(TIF_32BIT_ADDR)
80
81 #endif
82
83 /*
84  * One page above the stack is used for branch delay slot "emulation".
85  * See dsemul.c for details.
86  */
87 #define STACK_TOP       ((TASK_SIZE & PAGE_MASK) - PAGE_SIZE)
88
89 /*
90  * This decides where the kernel will search for a free chunk of vm
91  * space during mmap's.
92  */
93 #define TASK_UNMAPPED_BASE PAGE_ALIGN(TASK_SIZE / 3)
94
95
96 #define NUM_FPU_REGS    32
97
98 #ifdef CONFIG_CPU_HAS_MSA
99 # define FPU_REG_WIDTH  128
100 #else
101 # define FPU_REG_WIDTH  64
102 #endif
103
104 union fpureg {
105         __u32   val32[FPU_REG_WIDTH / 32];
106         __u64   val64[FPU_REG_WIDTH / 64];
107 };
108
109 #ifdef CONFIG_CPU_LITTLE_ENDIAN
110 # define FPR_IDX(width, idx)    (idx)
111 #else
112 # define FPR_IDX(width, idx)    ((idx) ^ ((64 / (width)) - 1))
113 #endif
114
115 #define BUILD_FPR_ACCESS(width) \
116 static inline u##width get_fpr##width(union fpureg *fpr, unsigned idx)  \
117 {                                                                       \
118         return fpr->val##width[FPR_IDX(width, idx)];                    \
119 }                                                                       \
120                                                                         \
121 static inline void set_fpr##width(union fpureg *fpr, unsigned idx,      \
122                                   u##width val)                         \
123 {                                                                       \
124         fpr->val##width[FPR_IDX(width, idx)] = val;                     \
125 }
126
127 BUILD_FPR_ACCESS(32)
128 BUILD_FPR_ACCESS(64)
129
130 /*
131  * It would be nice to add some more fields for emulator statistics,
132  * the additional information is private to the FPU emulator for now.
133  * See arch/mips/include/asm/fpu_emulator.h.
134  */
135
136 struct mips_fpu_struct {
137         union fpureg    fpr[NUM_FPU_REGS];
138         unsigned int    fcr31;
139         unsigned int    msacsr;
140 };
141
142 #define NUM_DSP_REGS   6
143
144 typedef __u32 dspreg_t;
145
146 struct mips_dsp_state {
147         dspreg_t        dspr[NUM_DSP_REGS];
148         unsigned int    dspcontrol;
149 };
150
151 #define INIT_CPUMASK { \
152         {0,} \
153 }
154
155 struct mips3264_watch_reg_state {
156         /* The width of watchlo is 32 in a 32 bit kernel and 64 in a
157            64 bit kernel.  We use unsigned long as it has the same
158            property. */
159         unsigned long watchlo[NUM_WATCH_REGS];
160         /* Only the mask and IRW bits from watchhi. */
161         u16 watchhi[NUM_WATCH_REGS];
162 };
163
164 union mips_watch_reg_state {
165         struct mips3264_watch_reg_state mips3264;
166 };
167
168 #if defined(CONFIG_CPU_CAVIUM_OCTEON)
169
170 struct octeon_cop2_state {
171         /* DMFC2 rt, 0x0201 */
172         unsigned long   cop2_crc_iv;
173         /* DMFC2 rt, 0x0202 (Set with DMTC2 rt, 0x1202) */
174         unsigned long   cop2_crc_length;
175         /* DMFC2 rt, 0x0200 (set with DMTC2 rt, 0x4200) */
176         unsigned long   cop2_crc_poly;
177         /* DMFC2 rt, 0x0402; DMFC2 rt, 0x040A */
178         unsigned long   cop2_llm_dat[2];
179        /* DMFC2 rt, 0x0084 */
180         unsigned long   cop2_3des_iv;
181         /* DMFC2 rt, 0x0080; DMFC2 rt, 0x0081; DMFC2 rt, 0x0082 */
182         unsigned long   cop2_3des_key[3];
183         /* DMFC2 rt, 0x0088 (Set with DMTC2 rt, 0x0098) */
184         unsigned long   cop2_3des_result;
185         /* DMFC2 rt, 0x0111 (FIXME: Read Pass1 Errata) */
186         unsigned long   cop2_aes_inp0;
187         /* DMFC2 rt, 0x0102; DMFC2 rt, 0x0103 */
188         unsigned long   cop2_aes_iv[2];
189         /* DMFC2 rt, 0x0104; DMFC2 rt, 0x0105; DMFC2 rt, 0x0106; DMFC2
190          * rt, 0x0107 */
191         unsigned long   cop2_aes_key[4];
192         /* DMFC2 rt, 0x0110 */
193         unsigned long   cop2_aes_keylen;
194         /* DMFC2 rt, 0x0100; DMFC2 rt, 0x0101 */
195         unsigned long   cop2_aes_result[2];
196         /* DMFC2 rt, 0x0240; DMFC2 rt, 0x0241; DMFC2 rt, 0x0242; DMFC2
197          * rt, 0x0243; DMFC2 rt, 0x0244; DMFC2 rt, 0x0245; DMFC2 rt,
198          * 0x0246; DMFC2 rt, 0x0247; DMFC2 rt, 0x0248; DMFC2 rt,
199          * 0x0249; DMFC2 rt, 0x024A; DMFC2 rt, 0x024B; DMFC2 rt,
200          * 0x024C; DMFC2 rt, 0x024D; DMFC2 rt, 0x024E - Pass2 */
201         unsigned long   cop2_hsh_datw[15];
202         /* DMFC2 rt, 0x0250; DMFC2 rt, 0x0251; DMFC2 rt, 0x0252; DMFC2
203          * rt, 0x0253; DMFC2 rt, 0x0254; DMFC2 rt, 0x0255; DMFC2 rt,
204          * 0x0256; DMFC2 rt, 0x0257 - Pass2 */
205         unsigned long   cop2_hsh_ivw[8];
206         /* DMFC2 rt, 0x0258; DMFC2 rt, 0x0259 - Pass2 */
207         unsigned long   cop2_gfm_mult[2];
208         /* DMFC2 rt, 0x025E - Pass2 */
209         unsigned long   cop2_gfm_poly;
210         /* DMFC2 rt, 0x025A; DMFC2 rt, 0x025B - Pass2 */
211         unsigned long   cop2_gfm_result[2];
212         /* DMFC2 rt, 0x24F, DMFC2 rt, 0x50, OCTEON III */
213         unsigned long   cop2_sha3[2];
214 };
215 #define COP2_INIT                                               \
216         .cp2                    = {0,},
217
218 struct octeon_cvmseg_state {
219         unsigned long cvmseg[CONFIG_CAVIUM_OCTEON_CVMSEG_SIZE]
220                             [cpu_dcache_line_size() / sizeof(unsigned long)];
221 };
222
223 #elif defined(CONFIG_CPU_XLP)
224 struct nlm_cop2_state {
225         u64     rx[4];
226         u64     tx[4];
227         u32     tx_msg_status;
228         u32     rx_msg_status;
229 };
230
231 #define COP2_INIT                                               \
232         .cp2                    = {{0}, {0}, 0, 0},
233 #else
234 #define COP2_INIT
235 #endif
236
237 typedef struct {
238         unsigned long seg;
239 } mm_segment_t;
240
241 #ifdef CONFIG_CPU_HAS_MSA
242 # define ARCH_MIN_TASKALIGN     16
243 # define FPU_ALIGN              __aligned(16)
244 #else
245 # define ARCH_MIN_TASKALIGN     8
246 # define FPU_ALIGN
247 #endif
248
249 struct mips_abi;
250
251 /*
252  * If you change thread_struct remember to change the #defines below too!
253  */
254 struct thread_struct {
255         /* Saved main processor registers. */
256         unsigned long reg16;
257         unsigned long reg17, reg18, reg19, reg20, reg21, reg22, reg23;
258         unsigned long reg29, reg30, reg31;
259
260         /* Saved cp0 stuff. */
261         unsigned long cp0_status;
262
263         /* Saved fpu/fpu emulator stuff. */
264         struct mips_fpu_struct fpu FPU_ALIGN;
265         /* Assigned branch delay slot 'emulation' frame */
266         atomic_t bd_emu_frame;
267         /* PC of the branch from a branch delay slot 'emulation' */
268         unsigned long bd_emu_branch_pc;
269         /* PC to continue from following a branch delay slot 'emulation' */
270         unsigned long bd_emu_cont_pc;
271 #ifdef CONFIG_MIPS_MT_FPAFF
272         /* Emulated instruction count */
273         unsigned long emulated_fp;
274         /* Saved per-thread scheduler affinity mask */
275         cpumask_t user_cpus_allowed;
276 #endif /* CONFIG_MIPS_MT_FPAFF */
277
278         /* Saved state of the DSP ASE, if available. */
279         struct mips_dsp_state dsp;
280
281         /* Saved watch register state, if available. */
282         union mips_watch_reg_state watch;
283
284         /* Other stuff associated with the thread. */
285         unsigned long cp0_badvaddr;     /* Last user fault */
286         unsigned long cp0_baduaddr;     /* Last kernel fault accessing USEG */
287         unsigned long error_code;
288         unsigned long trap_nr;
289 #ifdef CONFIG_CPU_CAVIUM_OCTEON
290         struct octeon_cop2_state cp2 __attribute__ ((__aligned__(128)));
291         struct octeon_cvmseg_state cvmseg __attribute__ ((__aligned__(128)));
292 #endif
293 #ifdef CONFIG_CPU_XLP
294         struct nlm_cop2_state cp2;
295 #endif
296         struct mips_abi *abi;
297 };
298
299 #ifdef CONFIG_MIPS_MT_FPAFF
300 #define FPAFF_INIT                                              \
301         .emulated_fp                    = 0,                    \
302         .user_cpus_allowed              = INIT_CPUMASK,
303 #else
304 #define FPAFF_INIT
305 #endif /* CONFIG_MIPS_MT_FPAFF */
306
307 #define INIT_THREAD  {                                          \
308         /*                                                      \
309          * Saved main processor registers                       \
310          */                                                     \
311         .reg16                  = 0,                            \
312         .reg17                  = 0,                            \
313         .reg18                  = 0,                            \
314         .reg19                  = 0,                            \
315         .reg20                  = 0,                            \
316         .reg21                  = 0,                            \
317         .reg22                  = 0,                            \
318         .reg23                  = 0,                            \
319         .reg29                  = 0,                            \
320         .reg30                  = 0,                            \
321         .reg31                  = 0,                            \
322         /*                                                      \
323          * Saved cp0 stuff                                      \
324          */                                                     \
325         .cp0_status             = 0,                            \
326         /*                                                      \
327          * Saved FPU/FPU emulator stuff                         \
328          */                                                     \
329         .fpu                    = {                             \
330                 .fpr            = {{{0,},},},                   \
331                 .fcr31          = 0,                            \
332                 .msacsr         = 0,                            \
333         },                                                      \
334         /*                                                      \
335          * FPU affinity state (null if not FPAFF)               \
336          */                                                     \
337         FPAFF_INIT                                              \
338         /* Delay slot emulation */                              \
339         .bd_emu_frame = ATOMIC_INIT(BD_EMUFRAME_NONE),          \
340         .bd_emu_branch_pc = 0,                                  \
341         .bd_emu_cont_pc = 0,                                    \
342         /*                                                      \
343          * Saved DSP stuff                                      \
344          */                                                     \
345         .dsp                    = {                             \
346                 .dspr           = {0, },                        \
347                 .dspcontrol     = 0,                            \
348         },                                                      \
349         /*                                                      \
350          * saved watch register stuff                           \
351          */                                                     \
352         .watch = {{{0,},},},                                    \
353         /*                                                      \
354          * Other stuff associated with the process              \
355          */                                                     \
356         .cp0_badvaddr           = 0,                            \
357         .cp0_baduaddr           = 0,                            \
358         .error_code             = 0,                            \
359         .trap_nr                = 0,                            \
360         /*                                                      \
361          * Platform specific cop2 registers(null if no COP2)    \
362          */                                                     \
363         COP2_INIT                                               \
364 }
365
366 struct task_struct;
367
368 /* Free all resources held by a thread. */
369 #define release_thread(thread) do { } while(0)
370
371 extern unsigned long thread_saved_pc(struct task_struct *tsk);
372
373 /*
374  * Do necessary setup to start up a newly executed thread.
375  */
376 extern void start_thread(struct pt_regs * regs, unsigned long pc, unsigned long sp);
377
378 static inline void flush_thread(void)
379 {
380 }
381
382 unsigned long get_wchan(struct task_struct *p);
383
384 #define __KSTK_TOS(tsk) ((unsigned long)task_stack_page(tsk) + \
385                          THREAD_SIZE - 32 - sizeof(struct pt_regs))
386 #define task_pt_regs(tsk) ((struct pt_regs *)__KSTK_TOS(tsk))
387 #define KSTK_EIP(tsk) (task_pt_regs(tsk)->cp0_epc)
388 #define KSTK_ESP(tsk) (task_pt_regs(tsk)->regs[29])
389 #define KSTK_STATUS(tsk) (task_pt_regs(tsk)->cp0_status)
390
391 #define cpu_relax()     barrier()
392 #define cpu_relax_lowlatency() cpu_relax()
393
394 /*
395  * Return_address is a replacement for __builtin_return_address(count)
396  * which on certain architectures cannot reasonably be implemented in GCC
397  * (MIPS, Alpha) or is unusable with -fomit-frame-pointer (i386).
398  * Note that __builtin_return_address(x>=1) is forbidden because GCC
399  * aborts compilation on some CPUs.  It's simply not possible to unwind
400  * some CPU's stackframes.
401  *
402  * __builtin_return_address works only for non-leaf functions.  We avoid the
403  * overhead of a function call by forcing the compiler to save the return
404  * address register on the stack.
405  */
406 #define return_address() ({__asm__ __volatile__("":::"$31");__builtin_return_address(0);})
407
408 #ifdef CONFIG_CPU_HAS_PREFETCH
409
410 #define ARCH_HAS_PREFETCH
411 #define prefetch(x) __builtin_prefetch((x), 0, 1)
412
413 #define ARCH_HAS_PREFETCHW
414 #define prefetchw(x) __builtin_prefetch((x), 1, 1)
415
416 #endif
417
418 /*
419  * Functions & macros implementing the PR_GET_FP_MODE & PR_SET_FP_MODE options
420  * to the prctl syscall.
421  */
422 extern int mips_get_process_fp_mode(struct task_struct *task);
423 extern int mips_set_process_fp_mode(struct task_struct *task,
424                                     unsigned int value);
425
426 #define GET_FP_MODE(task)               mips_get_process_fp_mode(task)
427 #define SET_FP_MODE(task,value)         mips_set_process_fp_mode(task, value)
428
429 #endif /* _ASM_PROCESSOR_H */