xref: /linux/arch/sparc/include/asm/processor_64.h (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
1 /*
2  * include/asm/processor.h
3  *
4  * Copyright (C) 1996 David S. Miller (davem@caip.rutgers.edu)
5  */
6 
7 #ifndef __ASM_SPARC64_PROCESSOR_H
8 #define __ASM_SPARC64_PROCESSOR_H
9 
10 /*
11  * Sparc64 implementation of macro that returns current
12  * instruction pointer ("program counter").
13  */
14 #define current_text_addr() ({ void *pc; __asm__("rd %%pc, %0" : "=r" (pc)); pc; })
15 
16 #include <asm/asi.h>
17 #include <asm/pstate.h>
18 #include <asm/ptrace.h>
19 #include <asm/page.h>
20 
21 /* The sparc has no problems with write protection */
22 #define wp_works_ok 1
23 #define wp_works_ok__is_a_macro /* for versions in ksyms.c */
24 
25 /*
26  * User lives in his very own context, and cannot reference us. Note
27  * that TASK_SIZE is a misnomer, it really gives maximum user virtual
28  * address that the kernel will allocate out.
29  *
30  * XXX No longer using virtual page tables, kill this upper limit...
31  */
32 #define VA_BITS		44
33 #ifndef __ASSEMBLY__
34 #define VPTE_SIZE	(1UL << (VA_BITS - PAGE_SHIFT + 3))
35 #else
36 #define VPTE_SIZE	(1 << (VA_BITS - PAGE_SHIFT + 3))
37 #endif
38 
39 #define TASK_SIZE_OF(tsk) \
40 	(test_tsk_thread_flag(tsk,TIF_32BIT) ? \
41 	 (1UL << 32UL) : ((unsigned long)-VPTE_SIZE))
42 #define TASK_SIZE \
43 	(test_thread_flag(TIF_32BIT) ? \
44 	 (1UL << 32UL) : ((unsigned long)-VPTE_SIZE))
45 #ifdef __KERNEL__
46 
47 #define STACK_TOP32	((1UL << 32UL) - PAGE_SIZE)
48 #define STACK_TOP64	(0x0000080000000000UL - (1UL << 32UL))
49 
50 #define STACK_TOP	(test_thread_flag(TIF_32BIT) ? \
51 			 STACK_TOP32 : STACK_TOP64)
52 
53 #define STACK_TOP_MAX	STACK_TOP64
54 
55 #endif
56 
57 #ifndef __ASSEMBLY__
58 
59 typedef struct {
60 	unsigned char seg;
61 } mm_segment_t;
62 
63 /* The Sparc processor specific thread struct. */
64 /* XXX This should die, everything can go into thread_info now. */
65 struct thread_struct {
66 #ifdef CONFIG_DEBUG_SPINLOCK
67 	/* How many spinlocks held by this thread.
68 	 * Used with spin lock debugging to catch tasks
69 	 * sleeping illegally with locks held.
70 	 */
71 	int smp_lock_count;
72 	unsigned int smp_lock_pc;
73 #else
74 	int dummy; /* f'in gcc bug... */
75 #endif
76 };
77 
78 #endif /* !(__ASSEMBLY__) */
79 
80 #ifndef CONFIG_DEBUG_SPINLOCK
81 #define INIT_THREAD  {			\
82 	0,				\
83 }
84 #else /* CONFIG_DEBUG_SPINLOCK */
85 #define INIT_THREAD  {					\
86 /* smp_lock_count, smp_lock_pc, */			\
87    0,		   0,					\
88 }
89 #endif /* !(CONFIG_DEBUG_SPINLOCK) */
90 
91 #ifndef __ASSEMBLY__
92 
93 #include <linux/types.h>
94 #include <asm/fpumacro.h>
95 
96 /* Return saved PC of a blocked thread. */
97 struct task_struct;
98 unsigned long thread_saved_pc(struct task_struct *);
99 
100 /* On Uniprocessor, even in RMO processes see TSO semantics */
101 #ifdef CONFIG_SMP
102 #define TSTATE_INITIAL_MM	TSTATE_TSO
103 #else
104 #define TSTATE_INITIAL_MM	TSTATE_RMO
105 #endif
106 
107 /* Do necessary setup to start up a newly executed thread. */
108 #define start_thread(regs, pc, sp) \
109 do { \
110 	unsigned long __asi = ASI_PNF; \
111 	regs->tstate = (regs->tstate & (TSTATE_CWP)) | (TSTATE_INITIAL_MM|TSTATE_IE) | (__asi << 24UL); \
112 	regs->tpc = ((pc & (~3)) - 4); \
113 	regs->tnpc = regs->tpc + 4; \
114 	regs->y = 0; \
115 	set_thread_wstate(1 << 3); \
116 	if (current_thread_info()->utraps) { \
117 		if (*(current_thread_info()->utraps) < 2) \
118 			kfree(current_thread_info()->utraps); \
119 		else \
120 			(*(current_thread_info()->utraps))--; \
121 		current_thread_info()->utraps = NULL; \
122 	} \
123 	__asm__ __volatile__( \
124 	"stx		%%g0, [%0 + %2 + 0x00]\n\t" \
125 	"stx		%%g0, [%0 + %2 + 0x08]\n\t" \
126 	"stx		%%g0, [%0 + %2 + 0x10]\n\t" \
127 	"stx		%%g0, [%0 + %2 + 0x18]\n\t" \
128 	"stx		%%g0, [%0 + %2 + 0x20]\n\t" \
129 	"stx		%%g0, [%0 + %2 + 0x28]\n\t" \
130 	"stx		%%g0, [%0 + %2 + 0x30]\n\t" \
131 	"stx		%%g0, [%0 + %2 + 0x38]\n\t" \
132 	"stx		%%g0, [%0 + %2 + 0x40]\n\t" \
133 	"stx		%%g0, [%0 + %2 + 0x48]\n\t" \
134 	"stx		%%g0, [%0 + %2 + 0x50]\n\t" \
135 	"stx		%%g0, [%0 + %2 + 0x58]\n\t" \
136 	"stx		%%g0, [%0 + %2 + 0x60]\n\t" \
137 	"stx		%%g0, [%0 + %2 + 0x68]\n\t" \
138 	"stx		%1,   [%0 + %2 + 0x70]\n\t" \
139 	"stx		%%g0, [%0 + %2 + 0x78]\n\t" \
140 	"wrpr		%%g0, (1 << 3), %%wstate\n\t" \
141 	: \
142 	: "r" (regs), "r" (sp - sizeof(struct reg_window) - STACK_BIAS), \
143 	  "i" ((const unsigned long)(&((struct pt_regs *)0)->u_regs[0]))); \
144 	fprs_write(0);	\
145 	current_thread_info()->xfsr[0] = 0;	\
146 	current_thread_info()->fpsaved[0] = 0;	\
147 	regs->tstate &= ~TSTATE_PEF;	\
148 } while (0)
149 
150 #define start_thread32(regs, pc, sp) \
151 do { \
152 	unsigned long __asi = ASI_PNF; \
153 	pc &= 0x00000000ffffffffUL; \
154 	sp &= 0x00000000ffffffffUL; \
155 	regs->tstate = (regs->tstate & (TSTATE_CWP))|(TSTATE_INITIAL_MM|TSTATE_IE|TSTATE_AM) | (__asi << 24UL); \
156 	regs->tpc = ((pc & (~3)) - 4); \
157 	regs->tnpc = regs->tpc + 4; \
158 	regs->y = 0; \
159 	set_thread_wstate(2 << 3); \
160 	if (current_thread_info()->utraps) { \
161 		if (*(current_thread_info()->utraps) < 2) \
162 			kfree(current_thread_info()->utraps); \
163 		else \
164 			(*(current_thread_info()->utraps))--; \
165 		current_thread_info()->utraps = NULL; \
166 	} \
167 	__asm__ __volatile__( \
168 	"stx		%%g0, [%0 + %2 + 0x00]\n\t" \
169 	"stx		%%g0, [%0 + %2 + 0x08]\n\t" \
170 	"stx		%%g0, [%0 + %2 + 0x10]\n\t" \
171 	"stx		%%g0, [%0 + %2 + 0x18]\n\t" \
172 	"stx		%%g0, [%0 + %2 + 0x20]\n\t" \
173 	"stx		%%g0, [%0 + %2 + 0x28]\n\t" \
174 	"stx		%%g0, [%0 + %2 + 0x30]\n\t" \
175 	"stx		%%g0, [%0 + %2 + 0x38]\n\t" \
176 	"stx		%%g0, [%0 + %2 + 0x40]\n\t" \
177 	"stx		%%g0, [%0 + %2 + 0x48]\n\t" \
178 	"stx		%%g0, [%0 + %2 + 0x50]\n\t" \
179 	"stx		%%g0, [%0 + %2 + 0x58]\n\t" \
180 	"stx		%%g0, [%0 + %2 + 0x60]\n\t" \
181 	"stx		%%g0, [%0 + %2 + 0x68]\n\t" \
182 	"stx		%1,   [%0 + %2 + 0x70]\n\t" \
183 	"stx		%%g0, [%0 + %2 + 0x78]\n\t" \
184 	"wrpr		%%g0, (2 << 3), %%wstate\n\t" \
185 	: \
186 	: "r" (regs), "r" (sp - sizeof(struct reg_window32)), \
187 	  "i" ((const unsigned long)(&((struct pt_regs *)0)->u_regs[0]))); \
188 	fprs_write(0);	\
189 	current_thread_info()->xfsr[0] = 0;	\
190 	current_thread_info()->fpsaved[0] = 0;	\
191 	regs->tstate &= ~TSTATE_PEF;	\
192 } while (0)
193 
194 /* Free all resources held by a thread. */
195 #define release_thread(tsk)		do { } while (0)
196 
197 unsigned long get_wchan(struct task_struct *task);
198 
199 #define task_pt_regs(tsk) (task_thread_info(tsk)->kregs)
200 #define KSTK_EIP(tsk)  (task_pt_regs(tsk)->tpc)
201 #define KSTK_ESP(tsk)  (task_pt_regs(tsk)->u_regs[UREG_FP])
202 
203 /* Please see the commentary in asm/backoff.h for a description of
204  * what these instructions are doing and how they have been choosen.
205  * To make a long story short, we are trying to yield the current cpu
206  * strand during busy loops.
207  */
208 #define cpu_relax()	asm volatile("\n99:\n\t"			\
209 				     "rd	%%ccr, %%g0\n\t"	\
210 				     "rd	%%ccr, %%g0\n\t"	\
211 				     "rd	%%ccr, %%g0\n\t"	\
212 				     ".section	.pause_3insn_patch,\"ax\"\n\t"\
213 				     ".word	99b\n\t"		\
214 				     "wr	%%g0, 128, %%asr27\n\t"	\
215 				     "nop\n\t"				\
216 				     "nop\n\t"				\
217 				     ".previous"			\
218 				     ::: "memory")
219 #define cpu_relax_lowlatency() cpu_relax()
220 
221 /* Prefetch support.  This is tuned for UltraSPARC-III and later.
222  * UltraSPARC-I will treat these as nops, and UltraSPARC-II has
223  * a shallower prefetch queue than later chips.
224  */
225 #define ARCH_HAS_PREFETCH
226 #define ARCH_HAS_PREFETCHW
227 #define ARCH_HAS_SPINLOCK_PREFETCH
228 
229 static inline void prefetch(const void *x)
230 {
231 	/* We do not use the read prefetch mnemonic because that
232 	 * prefetches into the prefetch-cache which only is accessible
233 	 * by floating point operations in UltraSPARC-III and later.
234 	 * By contrast, "#one_write" prefetches into the L2 cache
235 	 * in shared state.
236 	 */
237 	__asm__ __volatile__("prefetch [%0], #one_write"
238 			     : /* no outputs */
239 			     : "r" (x));
240 }
241 
242 static inline void prefetchw(const void *x)
243 {
244 	/* The most optimal prefetch to use for writes is
245 	 * "#n_writes".  This brings the cacheline into the
246 	 * L2 cache in "owned" state.
247 	 */
248 	__asm__ __volatile__("prefetch [%0], #n_writes"
249 			     : /* no outputs */
250 			     : "r" (x));
251 }
252 
253 #define spin_lock_prefetch(x)	prefetchw(x)
254 
255 #define HAVE_ARCH_PICK_MMAP_LAYOUT
256 
257 int do_mathemu(struct pt_regs *regs, struct fpustate *f, bool illegal_insn_trap);
258 
259 #endif /* !(__ASSEMBLY__) */
260 
261 #endif /* !(__ASM_SPARC64_PROCESSOR_H) */
262