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
8 * Copyright (C) 2000, 2001 Paolo Alberelli
9 * Copyright (C) 2003 Richard Curnow (/proc/tlb, bug fixes)
10 * Copyright (C) 2003 Paul Mundt
14 #include <linux/signal.h>
15 #include <linux/rwsem.h>
16 #include <linux/sched.h>
17 #include <linux/kernel.h>
18 #include <linux/errno.h>
19 #include <linux/string.h>
20 #include <linux/types.h>
21 #include <linux/ptrace.h>
22 #include <linux/mman.h>
24 #include <linux/smp.h>
25 #include <linux/interrupt.h>
27 #include <asm/system.h>
30 #include <asm/uaccess.h>
31 #include <asm/pgalloc.h>
32 #include <asm/mmu_context.h>
33 #include <asm/registers.h> /* required by inline asm statements */
35 #if defined(CONFIG_SH64_PROC_TLB)
36 #include <linux/init.h>
37 #include <linux/proc_fs.h>
38 /* Count numbers of tlb refills in each region */
39 static unsigned long long calls_to_update_mmu_cache
= 0ULL;
40 static unsigned long long calls_to_flush_tlb_page
= 0ULL;
41 static unsigned long long calls_to_flush_tlb_range
= 0ULL;
42 static unsigned long long calls_to_flush_tlb_mm
= 0ULL;
43 static unsigned long long calls_to_flush_tlb_all
= 0ULL;
44 unsigned long long calls_to_do_slow_page_fault
= 0ULL;
45 unsigned long long calls_to_do_fast_page_fault
= 0ULL;
47 /* Count size of ranges for flush_tlb_range */
48 static unsigned long long flush_tlb_range_1
= 0ULL;
49 static unsigned long long flush_tlb_range_2
= 0ULL;
50 static unsigned long long flush_tlb_range_3_4
= 0ULL;
51 static unsigned long long flush_tlb_range_5_7
= 0ULL;
52 static unsigned long long flush_tlb_range_8_11
= 0ULL;
53 static unsigned long long flush_tlb_range_12_15
= 0ULL;
54 static unsigned long long flush_tlb_range_16_up
= 0ULL;
56 static unsigned long long page_not_present
= 0ULL;
60 extern void die(const char *,struct pt_regs
*,long);
62 #define PFLAG(val,flag) (( (val) & (flag) ) ? #flag : "" )
63 #define PPROT(flag) PFLAG(pgprot_val(prot),flag)
65 static inline void print_prots(pgprot_t prot
)
67 printk("prot is 0x%08lx\n",pgprot_val(prot
));
69 printk("%s %s %s %s %s\n",PPROT(_PAGE_SHARED
),PPROT(_PAGE_READ
),
70 PPROT(_PAGE_EXECUTE
),PPROT(_PAGE_WRITE
),PPROT(_PAGE_USER
));
73 static inline void print_vma(struct vm_area_struct
*vma
)
75 printk("vma start 0x%08lx\n", vma
->vm_start
);
76 printk("vma end 0x%08lx\n", vma
->vm_end
);
78 print_prots(vma
->vm_page_prot
);
79 printk("vm_flags 0x%08lx\n", vma
->vm_flags
);
82 static inline void print_task(struct task_struct
*tsk
)
84 printk("Task pid %d\n", task_pid_nr(tsk
));
87 static pte_t
*lookup_pte(struct mm_struct
*mm
, unsigned long address
)
94 dir
= pgd_offset(mm
, address
);
99 pmd
= pmd_offset(dir
, address
);
100 if (pmd_none(*pmd
)) {
104 pte
= pte_offset_kernel(pmd
, address
);
107 if (pte_none(entry
)) {
110 if (!pte_present(entry
)) {
118 * This routine handles page faults. It determines the address,
119 * and the problem, and then passes it off to one of the appropriate
122 asmlinkage
void do_page_fault(struct pt_regs
*regs
, unsigned long writeaccess
,
123 unsigned long textaccess
, unsigned long address
)
125 struct task_struct
*tsk
;
126 struct mm_struct
*mm
;
127 struct vm_area_struct
* vma
;
128 const struct exception_table_entry
*fixup
;
132 #if defined(CONFIG_SH64_PROC_TLB)
133 ++calls_to_do_slow_page_fault
;
137 * Note this is now called with interrupts still disabled
138 * This is to cope with being called for a missing IO port
139 * address with interrupts disabled. This should be fixed as
140 * soon as we have a better 'fast path' miss handler.
142 * Plus take care how you try and debug this stuff.
143 * For example, writing debug data to a port which you
144 * have just faulted on is not going to work.
150 /* Not an IO address, so reenable interrupts */
154 * If we're in an interrupt or have no user
155 * context, we must not take the fault..
157 if (in_atomic() || !mm
)
160 /* TLB misses upon some cache flushes get done under cli() */
161 down_read(&mm
->mmap_sem
);
163 vma
= find_vma(mm
, address
);
168 printk("%s:%d fault, address is 0x%08x PC %016Lx textaccess %d writeaccess %d\n",
169 __FUNCTION__
,__LINE__
,
170 address
,regs
->pc
,textaccess
,writeaccess
);
175 if (vma
->vm_start
<= address
) {
179 if (!(vma
->vm_flags
& VM_GROWSDOWN
)) {
182 printk("%s:%d fault, address is 0x%08x PC %016Lx textaccess %d writeaccess %d\n",
183 __FUNCTION__
,__LINE__
,
184 address
,regs
->pc
,textaccess
,writeaccess
);
191 if (expand_stack(vma
, address
)) {
194 printk("%s:%d fault, address is 0x%08x PC %016Lx textaccess %d writeaccess %d\n",
195 __FUNCTION__
,__LINE__
,
196 address
,regs
->pc
,textaccess
,writeaccess
);
202 * Ok, we have a good vm_area for this memory access, so
207 if (!(vma
->vm_flags
& VM_EXEC
))
211 if (!(vma
->vm_flags
& VM_WRITE
))
214 if (!(vma
->vm_flags
& VM_READ
))
220 * If for any reason at all we couldn't handle the fault,
221 * make sure we exit gracefully rather than endlessly redo
225 fault
= handle_mm_fault(mm
, vma
, address
, writeaccess
);
226 if (unlikely(fault
& VM_FAULT_ERROR
)) {
227 if (fault
& VM_FAULT_OOM
)
229 else if (fault
& VM_FAULT_SIGBUS
)
233 if (fault
& VM_FAULT_MAJOR
)
238 /* If we get here, the page fault has been handled. Do the TLB refill
239 now from the newly-setup PTE, to avoid having to fault again right
240 away on the same instruction. */
241 pte
= lookup_pte (mm
, address
);
243 /* From empirical evidence, we can get here, due to
244 !pte_present(pte). (e.g. if a swap-in occurs, and the page
245 is swapped back out again before the process that wanted it
246 gets rescheduled?) */
250 __do_tlb_refill(address
, textaccess
, pte
);
254 up_read(&mm
->mmap_sem
);
258 * Something tried to access memory that isn't in our memory map..
259 * Fix it, but check if it's kernel or user first..
263 printk("fault:bad area\n");
265 up_read(&mm
->mmap_sem
);
267 if (user_mode(regs
)) {
271 /* This is really to help debug faults when starting
272 * usermode, so only need a few */
274 printk("user mode bad_area address=%08lx pid=%d (%s) pc=%08lx\n",
275 address
, task_pid_nr(current
), current
->comm
,
276 (unsigned long) regs
->pc
);
281 if (is_global_init(tsk
)) {
282 panic("INIT had user mode bad_area\n");
284 tsk
->thread
.address
= address
;
285 tsk
->thread
.error_code
= writeaccess
;
286 info
.si_signo
= SIGSEGV
;
288 info
.si_addr
= (void *) address
;
289 force_sig_info(SIGSEGV
, &info
, tsk
);
295 printk("fault:No context\n");
297 /* Are we prepared to handle this kernel fault? */
298 fixup
= search_exception_tables(regs
->pc
);
300 regs
->pc
= fixup
->fixup
;
305 * Oops. The kernel tried to access some bad page. We'll have to
306 * terminate things with extreme prejudice.
309 if (address
< PAGE_SIZE
)
310 printk(KERN_ALERT
"Unable to handle kernel NULL pointer dereference");
312 printk(KERN_ALERT
"Unable to handle kernel paging request");
313 printk(" at virtual address %08lx\n", address
);
314 printk(KERN_ALERT
"pc = %08Lx%08Lx\n", regs
->pc
>> 32, regs
->pc
& 0xffffffff);
315 die("Oops", regs
, writeaccess
);
319 * We ran out of memory, or some other thing happened to us that made
320 * us unable to handle the page fault gracefully.
323 if (is_global_init(current
)) {
324 panic("INIT out of memory\n");
328 printk("fault:Out of memory\n");
329 up_read(&mm
->mmap_sem
);
330 if (is_global_init(current
)) {
332 down_read(&mm
->mmap_sem
);
335 printk("VM: killing process %s\n", tsk
->comm
);
337 do_group_exit(SIGKILL
);
341 printk("fault:Do sigbus\n");
342 up_read(&mm
->mmap_sem
);
345 * Send a sigbus, regardless of whether we were in kernel
348 tsk
->thread
.address
= address
;
349 tsk
->thread
.error_code
= writeaccess
;
350 tsk
->thread
.trap_no
= 14;
351 force_sig(SIGBUS
, tsk
);
353 /* Kernel mode? Handle exceptions or die */
354 if (!user_mode(regs
))
359 void flush_tlb_all(void);
361 void update_mmu_cache(struct vm_area_struct
* vma
,
362 unsigned long address
, pte_t pte
)
364 #if defined(CONFIG_SH64_PROC_TLB)
365 ++calls_to_update_mmu_cache
;
369 * This appears to get called once for every pte entry that gets
370 * established => I don't think it's efficient to try refilling the
371 * TLBs with the pages - some may not get accessed even. Also, for
372 * executable pages, it is impossible to determine reliably here which
373 * TLB they should be mapped into (or both even).
375 * So, just do nothing here and handle faults on demand. In the
376 * TLBMISS handling case, the refill is now done anyway after the pte
377 * has been fixed up, so that deals with most useful cases.
381 static void __flush_tlb_page(struct vm_area_struct
*vma
, unsigned long page
)
383 unsigned long long match
, pteh
=0, lpage
;
385 struct mm_struct
*mm
;
389 if (mm
->context
== NO_CONTEXT
)
393 * Sign-extend based on neff.
395 lpage
= (page
& NEFF_SIGN
) ? (page
| NEFF_MASK
) : page
;
396 match
= ((mm
->context
& MMU_CONTEXT_ASID_MASK
) << PTEH_ASID_SHIFT
) | PTEH_VALID
;
399 /* Do ITLB : don't bother for pages in non-exectutable VMAs */
400 if (vma
->vm_flags
& VM_EXEC
) {
401 for_each_itlb_entry(tlb
) {
402 asm volatile ("getcfg %1, 0, %0"
407 __flush_tlb_slot(tlb
);
414 /* Do DTLB : any page could potentially be in here. */
415 for_each_dtlb_entry(tlb
) {
416 asm volatile ("getcfg %1, 0, %0"
421 __flush_tlb_slot(tlb
);
428 void flush_tlb_page(struct vm_area_struct
*vma
, unsigned long page
)
432 #if defined(CONFIG_SH64_PROC_TLB)
433 ++calls_to_flush_tlb_page
;
438 local_irq_save(flags
);
439 __flush_tlb_page(vma
, page
);
440 local_irq_restore(flags
);
444 void flush_tlb_range(struct vm_area_struct
*vma
, unsigned long start
,
448 unsigned long long match
, pteh
=0, pteh_epn
, pteh_low
;
450 struct mm_struct
*mm
;
454 #if defined(CONFIG_SH64_PROC_TLB)
455 ++calls_to_flush_tlb_range
;
458 unsigned long size
= (end
- 1) - start
;
459 size
>>= 12; /* divide by PAGE_SIZE */
460 size
++; /* end=start+4096 => 1 page */
462 case 1 : flush_tlb_range_1
++; break;
463 case 2 : flush_tlb_range_2
++; break;
464 case 3 ... 4 : flush_tlb_range_3_4
++; break;
465 case 5 ... 7 : flush_tlb_range_5_7
++; break;
466 case 8 ... 11 : flush_tlb_range_8_11
++; break;
467 case 12 ... 15 : flush_tlb_range_12_15
++; break;
468 default : flush_tlb_range_16_up
++; break;
473 if (mm
->context
== NO_CONTEXT
)
476 local_irq_save(flags
);
481 match
= ((mm
->context
& MMU_CONTEXT_ASID_MASK
) << PTEH_ASID_SHIFT
) | PTEH_VALID
;
484 for_each_itlb_entry(tlb
) {
485 asm volatile ("getcfg %1, 0, %0"
489 pteh_epn
= pteh
& PAGE_MASK
;
490 pteh_low
= pteh
& ~PAGE_MASK
;
492 if (pteh_low
== match
&& pteh_epn
>= start
&& pteh_epn
<= end
)
493 __flush_tlb_slot(tlb
);
497 for_each_dtlb_entry(tlb
) {
498 asm volatile ("getcfg %1, 0, %0"
502 pteh_epn
= pteh
& PAGE_MASK
;
503 pteh_low
= pteh
& ~PAGE_MASK
;
505 if (pteh_low
== match
&& pteh_epn
>= start
&& pteh_epn
<= end
)
506 __flush_tlb_slot(tlb
);
509 local_irq_restore(flags
);
512 void flush_tlb_mm(struct mm_struct
*mm
)
516 #if defined(CONFIG_SH64_PROC_TLB)
517 ++calls_to_flush_tlb_mm
;
520 if (mm
->context
== NO_CONTEXT
)
523 local_irq_save(flags
);
525 mm
->context
=NO_CONTEXT
;
527 activate_context(mm
);
529 local_irq_restore(flags
);
533 void flush_tlb_all(void)
535 /* Invalidate all, including shared pages, excluding fixed TLBs */
537 unsigned long flags
, tlb
;
539 #if defined(CONFIG_SH64_PROC_TLB)
540 ++calls_to_flush_tlb_all
;
543 local_irq_save(flags
);
545 /* Flush each ITLB entry */
546 for_each_itlb_entry(tlb
) {
547 __flush_tlb_slot(tlb
);
550 /* Flush each DTLB entry */
551 for_each_dtlb_entry(tlb
) {
552 __flush_tlb_slot(tlb
);
555 local_irq_restore(flags
);
558 void flush_tlb_kernel_range(unsigned long start
, unsigned long end
)
560 /* FIXME: Optimize this later.. */
564 #if defined(CONFIG_SH64_PROC_TLB)
565 /* Procfs interface to read the performance information */
568 tlb_proc_info(char *buf
, char **start
, off_t fpos
, int length
, int *eof
, void *data
)
571 len
+= sprintf(buf
+len
, "do_fast_page_fault called %12lld times\n", calls_to_do_fast_page_fault
);
572 len
+= sprintf(buf
+len
, "do_slow_page_fault called %12lld times\n", calls_to_do_slow_page_fault
);
573 len
+= sprintf(buf
+len
, "update_mmu_cache called %12lld times\n", calls_to_update_mmu_cache
);
574 len
+= sprintf(buf
+len
, "flush_tlb_page called %12lld times\n", calls_to_flush_tlb_page
);
575 len
+= sprintf(buf
+len
, "flush_tlb_range called %12lld times\n", calls_to_flush_tlb_range
);
576 len
+= sprintf(buf
+len
, "flush_tlb_mm called %12lld times\n", calls_to_flush_tlb_mm
);
577 len
+= sprintf(buf
+len
, "flush_tlb_all called %12lld times\n", calls_to_flush_tlb_all
);
578 len
+= sprintf(buf
+len
, "flush_tlb_range_sizes\n"
586 flush_tlb_range_1
, flush_tlb_range_2
, flush_tlb_range_3_4
,
587 flush_tlb_range_5_7
, flush_tlb_range_8_11
, flush_tlb_range_12_15
,
588 flush_tlb_range_16_up
);
589 len
+= sprintf(buf
+len
, "page not present %12lld times\n", page_not_present
);
594 static int __init
register_proc_tlb(void)
596 create_proc_read_entry("tlb", 0, NULL
, tlb_proc_info
, NULL
);
600 __initcall(register_proc_tlb
);