2 * PowerPC64 SLB support.
4 * Copyright (C) 2004 David Gibson <dwg@au.ibm.com>, IBM
5 * Based on earlier code written by:
6 * Dave Engebretsen and Mike Corrigan {engebret|mikejc}@us.ibm.com
7 * Copyright (c) 2001 Dave Engebretsen
8 * Copyright (C) 2002 Anton Blanchard <anton@au.ibm.com>, IBM
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License
13 * as published by the Free Software Foundation; either version
14 * 2 of the License, or (at your option) any later version.
17 #include <asm/pgtable.h>
19 #include <asm/mmu_context.h>
21 #include <asm/cputable.h>
22 #include <asm/cacheflush.h>
24 #include <linux/compiler.h>
25 #include <linux/context_tracking.h>
26 #include <linux/mm_types.h>
29 #include <asm/code-patching.h>
32 LINEAR_INDEX
= 0, /* Kernel linear map (0xc000000000000000) */
33 VMALLOC_INDEX
= 1, /* Kernel virtual map (0xd000000000000000) */
34 KSTACK_INDEX
= 2, /* Kernel stack map */
37 extern void slb_allocate(unsigned long ea
);
39 #define slb_esid_mask(ssize) \
40 (((ssize) == MMU_SEGSIZE_256M)? ESID_MASK: ESID_MASK_1T)
42 static inline unsigned long mk_esid_data(unsigned long ea
, int ssize
,
45 return (ea
& slb_esid_mask(ssize
)) | SLB_ESID_V
| index
;
48 static inline unsigned long mk_vsid_data(unsigned long ea
, int ssize
,
51 return (get_kernel_vsid(ea
, ssize
) << slb_vsid_shift(ssize
)) | flags
|
52 ((unsigned long) ssize
<< SLB_VSID_SSIZE_SHIFT
);
55 static inline void slb_shadow_update(unsigned long ea
, int ssize
,
59 struct slb_shadow
*p
= get_slb_shadow();
62 * Clear the ESID first so the entry is not valid while we are
63 * updating it. No write barriers are needed here, provided
64 * we only update the current CPU's SLB shadow buffer.
66 WRITE_ONCE(p
->save_area
[index
].esid
, 0);
67 WRITE_ONCE(p
->save_area
[index
].vsid
, cpu_to_be64(mk_vsid_data(ea
, ssize
, flags
)));
68 WRITE_ONCE(p
->save_area
[index
].esid
, cpu_to_be64(mk_esid_data(ea
, ssize
, index
)));
71 static inline void slb_shadow_clear(enum slb_index index
)
73 WRITE_ONCE(get_slb_shadow()->save_area
[index
].esid
, 0);
76 static inline void create_shadowed_slbe(unsigned long ea
, int ssize
,
81 * Updating the shadow buffer before writing the SLB ensures
82 * we don't get a stale entry here if we get preempted by PHYP
83 * between these two statements.
85 slb_shadow_update(ea
, ssize
, flags
, index
);
87 asm volatile("slbmte %0,%1" :
88 : "r" (mk_vsid_data(ea
, ssize
, flags
)),
89 "r" (mk_esid_data(ea
, ssize
, index
))
94 * Insert bolted entries into SLB (which may not be empty, so don't clear
97 void __slb_restore_bolted_realmode(void)
99 struct slb_shadow
*p
= get_slb_shadow();
100 enum slb_index index
;
102 /* No isync needed because realmode. */
103 for (index
= 0; index
< SLB_NUM_BOLTED
; index
++) {
104 asm volatile("slbmte %0,%1" :
105 : "r" (be64_to_cpu(p
->save_area
[index
].vsid
)),
106 "r" (be64_to_cpu(p
->save_area
[index
].esid
)));
111 * Insert the bolted entries into an empty SLB.
112 * This is not the same as rebolt because the bolted segments are not
113 * changed, just loaded from the shadow area.
115 void slb_restore_bolted_realmode(void)
117 __slb_restore_bolted_realmode();
118 get_paca()->slb_cache_ptr
= 0;
122 * This flushes all SLB entries including 0, so it must be realmode.
124 void slb_flush_all_realmode(void)
127 * This flushes all SLB entries including 0, so it must be realmode.
129 asm volatile("slbmte %0,%0; slbia" : : "r" (0));
132 static void __slb_flush_and_rebolt(void)
134 /* If you change this make sure you change SLB_NUM_BOLTED
135 * and PR KVM appropriately too. */
136 unsigned long linear_llp
, vmalloc_llp
, lflags
, vflags
;
137 unsigned long ksp_esid_data
, ksp_vsid_data
;
139 linear_llp
= mmu_psize_defs
[mmu_linear_psize
].sllp
;
140 vmalloc_llp
= mmu_psize_defs
[mmu_vmalloc_psize
].sllp
;
141 lflags
= SLB_VSID_KERNEL
| linear_llp
;
142 vflags
= SLB_VSID_KERNEL
| vmalloc_llp
;
144 ksp_esid_data
= mk_esid_data(get_paca()->kstack
, mmu_kernel_ssize
, KSTACK_INDEX
);
145 if ((ksp_esid_data
& ~0xfffffffUL
) <= PAGE_OFFSET
) {
146 ksp_esid_data
&= ~SLB_ESID_V
;
148 slb_shadow_clear(KSTACK_INDEX
);
150 /* Update stack entry; others don't change */
151 slb_shadow_update(get_paca()->kstack
, mmu_kernel_ssize
, lflags
, KSTACK_INDEX
);
153 be64_to_cpu(get_slb_shadow()->save_area
[KSTACK_INDEX
].vsid
);
156 /* We need to do this all in asm, so we're sure we don't touch
157 * the stack between the slbia and rebolting it. */
158 asm volatile("isync\n"
160 /* Slot 1 - first VMALLOC segment */
162 /* Slot 2 - kernel stack */
165 :: "r"(mk_vsid_data(VMALLOC_START
, mmu_kernel_ssize
, vflags
)),
166 "r"(mk_esid_data(VMALLOC_START
, mmu_kernel_ssize
, VMALLOC_INDEX
)),
172 void slb_flush_and_rebolt(void)
175 WARN_ON(!irqs_disabled());
178 * We can't take a PMU exception in the following code, so hard
179 * disable interrupts.
183 __slb_flush_and_rebolt();
184 get_paca()->slb_cache_ptr
= 0;
187 void slb_vmalloc_update(void)
189 unsigned long vflags
;
191 vflags
= SLB_VSID_KERNEL
| mmu_psize_defs
[mmu_vmalloc_psize
].sllp
;
192 slb_shadow_update(VMALLOC_START
, mmu_kernel_ssize
, vflags
, VMALLOC_INDEX
);
193 slb_flush_and_rebolt();
196 /* Helper function to compare esids. There are four cases to handle.
197 * 1. The system is not 1T segment size capable. Use the GET_ESID compare.
198 * 2. The system is 1T capable, both addresses are < 1T, use the GET_ESID compare.
199 * 3. The system is 1T capable, only one of the two addresses is > 1T. This is not a match.
200 * 4. The system is 1T capable, both addresses are > 1T, use the GET_ESID_1T macro to compare.
202 static inline int esids_match(unsigned long addr1
, unsigned long addr2
)
206 /* System is not 1T segment size capable. */
207 if (!mmu_has_feature(MMU_FTR_1T_SEGMENT
))
208 return (GET_ESID(addr1
) == GET_ESID(addr2
));
210 esid_1t_count
= (((addr1
>> SID_SHIFT_1T
) != 0) +
211 ((addr2
>> SID_SHIFT_1T
) != 0));
213 /* both addresses are < 1T */
214 if (esid_1t_count
== 0)
215 return (GET_ESID(addr1
) == GET_ESID(addr2
));
217 /* One address < 1T, the other > 1T. Not a match */
218 if (esid_1t_count
== 1)
221 /* Both addresses are > 1T. */
222 return (GET_ESID_1T(addr1
) == GET_ESID_1T(addr2
));
225 /* Flush all user entries from the segment table of the current processor. */
226 void switch_slb(struct task_struct
*tsk
, struct mm_struct
*mm
)
228 unsigned long offset
;
229 unsigned long slbie_data
= 0;
230 unsigned long pc
= KSTK_EIP(tsk
);
231 unsigned long stack
= KSTK_ESP(tsk
);
232 unsigned long exec_base
;
235 * We need interrupts hard-disabled here, not just soft-disabled,
236 * so that a PMU interrupt can't occur, which might try to access
237 * user memory (to get a stack trace) and possible cause an SLB miss
238 * which would update the slb_cache/slb_cache_ptr fields in the PACA.
241 offset
= get_paca()->slb_cache_ptr
;
242 if (!mmu_has_feature(MMU_FTR_NO_SLBIE_B
) &&
243 offset
<= SLB_CACHE_ENTRIES
) {
245 asm volatile("isync" : : : "memory");
246 for (i
= 0; i
< offset
; i
++) {
247 slbie_data
= (unsigned long)get_paca()->slb_cache
[i
]
248 << SID_SHIFT
; /* EA */
249 slbie_data
|= user_segment_size(slbie_data
)
250 << SLBIE_SSIZE_SHIFT
;
251 slbie_data
|= SLBIE_C
; /* C set for user addresses */
252 asm volatile("slbie %0" : : "r" (slbie_data
));
254 asm volatile("isync" : : : "memory");
256 __slb_flush_and_rebolt();
259 /* Workaround POWER5 < DD2.1 issue */
260 if (offset
== 1 || offset
> SLB_CACHE_ENTRIES
)
261 asm volatile("slbie %0" : : "r" (slbie_data
));
263 get_paca()->slb_cache_ptr
= 0;
267 * preload some userspace segments into the SLB.
268 * Almost all 32 and 64bit PowerPC executables are linked at
269 * 0x10000000 so it makes sense to preload this segment.
271 exec_base
= 0x10000000;
273 if (is_kernel_addr(pc
) || is_kernel_addr(stack
) ||
274 is_kernel_addr(exec_base
))
279 if (!esids_match(pc
, stack
))
282 if (!esids_match(pc
, exec_base
) &&
283 !esids_match(stack
, exec_base
))
284 slb_allocate(exec_base
);
287 static inline void patch_slb_encoding(unsigned int *insn_addr
,
292 * This function patches either an li or a cmpldi instruction with
293 * a new immediate value. This relies on the fact that both li
294 * (which is actually addi) and cmpldi both take a 16-bit immediate
295 * value, and it is situated in the same location in the instruction,
296 * ie. bits 16-31 (Big endian bit order) or the lower 16 bits.
297 * The signedness of the immediate operand differs between the two
298 * instructions however this code is only ever patching a small value,
299 * much less than 1 << 15, so we can get away with it.
300 * To patch the value we read the existing instruction, clear the
301 * immediate value, and or in our new value, then write the instruction
304 unsigned int insn
= (*insn_addr
& 0xffff0000) | immed
;
305 patch_instruction(insn_addr
, insn
);
308 extern u32 slb_miss_kernel_load_linear
[];
309 extern u32 slb_miss_kernel_load_io
[];
310 extern u32 slb_compare_rr_to_size
[];
311 extern u32 slb_miss_kernel_load_vmemmap
[];
313 void slb_set_size(u16 size
)
315 if (mmu_slb_size
== size
)
319 patch_slb_encoding(slb_compare_rr_to_size
, mmu_slb_size
);
322 void slb_initialize(void)
324 unsigned long linear_llp
, vmalloc_llp
, io_llp
;
325 unsigned long lflags
, vflags
;
326 static int slb_encoding_inited
;
327 #ifdef CONFIG_SPARSEMEM_VMEMMAP
328 unsigned long vmemmap_llp
;
331 /* Prepare our SLB miss handler based on our page size */
332 linear_llp
= mmu_psize_defs
[mmu_linear_psize
].sllp
;
333 io_llp
= mmu_psize_defs
[mmu_io_psize
].sllp
;
334 vmalloc_llp
= mmu_psize_defs
[mmu_vmalloc_psize
].sllp
;
335 get_paca()->vmalloc_sllp
= SLB_VSID_KERNEL
| vmalloc_llp
;
336 #ifdef CONFIG_SPARSEMEM_VMEMMAP
337 vmemmap_llp
= mmu_psize_defs
[mmu_vmemmap_psize
].sllp
;
339 if (!slb_encoding_inited
) {
340 slb_encoding_inited
= 1;
341 patch_slb_encoding(slb_miss_kernel_load_linear
,
342 SLB_VSID_KERNEL
| linear_llp
);
343 patch_slb_encoding(slb_miss_kernel_load_io
,
344 SLB_VSID_KERNEL
| io_llp
);
345 patch_slb_encoding(slb_compare_rr_to_size
,
348 pr_devel("SLB: linear LLP = %04lx\n", linear_llp
);
349 pr_devel("SLB: io LLP = %04lx\n", io_llp
);
351 #ifdef CONFIG_SPARSEMEM_VMEMMAP
352 patch_slb_encoding(slb_miss_kernel_load_vmemmap
,
353 SLB_VSID_KERNEL
| vmemmap_llp
);
354 pr_devel("SLB: vmemmap LLP = %04lx\n", vmemmap_llp
);
358 get_paca()->stab_rr
= SLB_NUM_BOLTED
;
360 lflags
= SLB_VSID_KERNEL
| linear_llp
;
361 vflags
= SLB_VSID_KERNEL
| vmalloc_llp
;
363 /* Invalidate the entire SLB (even entry 0) & all the ERATS */
364 asm volatile("isync":::"memory");
365 asm volatile("slbmte %0,%0"::"r" (0) : "memory");
366 asm volatile("isync; slbia; isync":::"memory");
367 create_shadowed_slbe(PAGE_OFFSET
, mmu_kernel_ssize
, lflags
, LINEAR_INDEX
);
368 create_shadowed_slbe(VMALLOC_START
, mmu_kernel_ssize
, vflags
, VMALLOC_INDEX
);
370 /* For the boot cpu, we're running on the stack in init_thread_union,
371 * which is in the first segment of the linear mapping, and also
372 * get_paca()->kstack hasn't been initialized yet.
373 * For secondary cpus, we need to bolt the kernel stack entry now.
375 slb_shadow_clear(KSTACK_INDEX
);
376 if (raw_smp_processor_id() != boot_cpuid
&&
377 (get_paca()->kstack
& slb_esid_mask(mmu_kernel_ssize
)) > PAGE_OFFSET
)
378 create_shadowed_slbe(get_paca()->kstack
,
379 mmu_kernel_ssize
, lflags
, KSTACK_INDEX
);
381 asm volatile("isync":::"memory");
384 static void insert_slb_entry(unsigned long vsid
, unsigned long ea
,
385 int bpsize
, int ssize
)
387 unsigned long flags
, vsid_data
, esid_data
;
388 enum slb_index index
;
392 * We are irq disabled, hence should be safe to access PACA.
394 VM_WARN_ON(!irqs_disabled());
397 * We can't take a PMU exception in the following code, so hard
398 * disable interrupts.
402 index
= get_paca()->stab_rr
;
405 * simple round-robin replacement of slb starting at SLB_NUM_BOLTED.
407 if (index
< (mmu_slb_size
- 1))
410 index
= SLB_NUM_BOLTED
;
412 get_paca()->stab_rr
= index
;
414 flags
= SLB_VSID_USER
| mmu_psize_defs
[bpsize
].sllp
;
415 vsid_data
= (vsid
<< slb_vsid_shift(ssize
)) | flags
|
416 ((unsigned long) ssize
<< SLB_VSID_SSIZE_SHIFT
);
417 esid_data
= mk_esid_data(ea
, ssize
, index
);
420 * No need for an isync before or after this slbmte. The exception
421 * we enter with and the rfid we exit with are context synchronizing.
422 * Also we only handle user segments here.
424 asm volatile("slbmte %0, %1" : : "r" (vsid_data
), "r" (esid_data
)
428 * Now update slb cache entries
430 slb_cache_index
= get_paca()->slb_cache_ptr
;
431 if (slb_cache_index
< SLB_CACHE_ENTRIES
) {
433 * We have space in slb cache for optimized switch_slb().
434 * Top 36 bits from esid_data as per ISA
436 get_paca()->slb_cache
[slb_cache_index
++] = esid_data
>> 28;
437 get_paca()->slb_cache_ptr
++;
440 * Our cache is full and the current cache content strictly
441 * doesn't indicate the active SLB conents. Bump the ptr
442 * so that switch_slb() will ignore the cache.
444 get_paca()->slb_cache_ptr
= SLB_CACHE_ENTRIES
+ 1;
448 static void handle_multi_context_slb_miss(int context_id
, unsigned long ea
)
450 struct mm_struct
*mm
= current
->mm
;
455 * We are always above 1TB, hence use high user segment size.
457 vsid
= get_vsid(context_id
, ea
, mmu_highuser_ssize
);
458 bpsize
= get_slice_psize(mm
, ea
);
459 insert_slb_entry(vsid
, ea
, bpsize
, mmu_highuser_ssize
);
462 void slb_miss_large_addr(struct pt_regs
*regs
)
464 enum ctx_state prev_state
= exception_enter();
465 unsigned long ea
= regs
->dar
;
468 if (REGION_ID(ea
) != USER_REGION_ID
)
472 * Are we beyound what the page table layout supports ?
474 if ((ea
& ~REGION_MASK
) >= H_PGTABLE_RANGE
)
477 /* Lower address should have been handled by asm code */
478 if (ea
< (1UL << MAX_EA_BITS_PER_CONTEXT
))
482 * consider this as bad access if we take a SLB miss
483 * on an address above addr limit.
485 if (ea
>= current
->mm
->context
.slb_addr_limit
)
488 context
= get_ea_context(¤t
->mm
->context
, ea
);
492 handle_multi_context_slb_miss(context
, ea
);
493 exception_exit(prev_state
);
498 _exception(SIGSEGV
, regs
, SEGV_BNDERR
, ea
);
500 bad_page_fault(regs
, ea
, SIGSEGV
);
501 exception_exit(prev_state
);