1 /*
2  * arch/sh/mm/tlb-flush_64.c
3  *
4  * Copyright (C) 2000, 2001  Paolo Alberelli
5  * Copyright (C) 2003  Richard Curnow (/proc/tlb, bug fixes)
6  * Copyright (C) 2003 - 2009 Paul Mundt
7  *
8  * This file is subject to the terms and conditions of the GNU General Public
9  * License.  See the file "COPYING" in the main directory of this archive
10  * for more details.
11  */
12 #include <linux/signal.h>
13 #include <linux/rwsem.h>
14 #include <linux/sched.h>
15 #include <linux/kernel.h>
16 #include <linux/errno.h>
17 #include <linux/string.h>
18 #include <linux/types.h>
19 #include <linux/ptrace.h>
20 #include <linux/mman.h>
21 #include <linux/mm.h>
22 #include <linux/smp.h>
23 #include <linux/perf_event.h>
24 #include <linux/interrupt.h>
25 #include <asm/system.h>
26 #include <asm/io.h>
27 #include <asm/tlb.h>
28 #include <asm/uaccess.h>
29 #include <asm/pgalloc.h>
30 #include <asm/mmu_context.h>
31 
32 extern void die(const char *,struct pt_regs *,long);
33 
34 #define PFLAG(val,flag)   (( (val) & (flag) ) ? #flag : "" )
35 #define PPROT(flag) PFLAG(pgprot_val(prot),flag)
36 
print_prots(pgprot_t prot)37 static inline void print_prots(pgprot_t prot)
38 {
39 	printk("prot is 0x%016llx\n",pgprot_val(prot));
40 
41 	printk("%s %s %s %s %s\n",PPROT(_PAGE_SHARED),PPROT(_PAGE_READ),
42 	       PPROT(_PAGE_EXECUTE),PPROT(_PAGE_WRITE),PPROT(_PAGE_USER));
43 }
44 
print_vma(struct vm_area_struct * vma)45 static inline void print_vma(struct vm_area_struct *vma)
46 {
47 	printk("vma start 0x%08lx\n", vma->vm_start);
48 	printk("vma end   0x%08lx\n", vma->vm_end);
49 
50 	print_prots(vma->vm_page_prot);
51 	printk("vm_flags 0x%08lx\n", vma->vm_flags);
52 }
53 
print_task(struct task_struct * tsk)54 static inline void print_task(struct task_struct *tsk)
55 {
56 	printk("Task pid %d\n", task_pid_nr(tsk));
57 }
58 
lookup_pte(struct mm_struct * mm,unsigned long address)59 static pte_t *lookup_pte(struct mm_struct *mm, unsigned long address)
60 {
61 	pgd_t *dir;
62 	pud_t *pud;
63 	pmd_t *pmd;
64 	pte_t *pte;
65 	pte_t entry;
66 
67 	dir = pgd_offset(mm, address);
68 	if (pgd_none(*dir))
69 		return NULL;
70 
71 	pud = pud_offset(dir, address);
72 	if (pud_none(*pud))
73 		return NULL;
74 
75 	pmd = pmd_offset(pud, address);
76 	if (pmd_none(*pmd))
77 		return NULL;
78 
79 	pte = pte_offset_kernel(pmd, address);
80 	entry = *pte;
81 	if (pte_none(entry) || !pte_present(entry))
82 		return NULL;
83 
84 	return pte;
85 }
86 
87 /*
88  * This routine handles page faults.  It determines the address,
89  * and the problem, and then passes it off to one of the appropriate
90  * routines.
91  */
do_page_fault(struct pt_regs * regs,unsigned long writeaccess,unsigned long textaccess,unsigned long address)92 asmlinkage void do_page_fault(struct pt_regs *regs, unsigned long writeaccess,
93 			      unsigned long textaccess, unsigned long address)
94 {
95 	struct task_struct *tsk;
96 	struct mm_struct *mm;
97 	struct vm_area_struct * vma;
98 	const struct exception_table_entry *fixup;
99 	pte_t *pte;
100 	int fault;
101 
102 	/* SIM
103 	 * Note this is now called with interrupts still disabled
104 	 * This is to cope with being called for a missing IO port
105 	 * address with interrupts disabled. This should be fixed as
106 	 * soon as we have a better 'fast path' miss handler.
107 	 *
108 	 * Plus take care how you try and debug this stuff.
109 	 * For example, writing debug data to a port which you
110 	 * have just faulted on is not going to work.
111 	 */
112 
113 	tsk = current;
114 	mm = tsk->mm;
115 
116 	/* Not an IO address, so reenable interrupts */
117 	local_irq_enable();
118 
119 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
120 
121 	/*
122 	 * If we're in an interrupt or have no user
123 	 * context, we must not take the fault..
124 	 */
125 	if (in_atomic() || !mm)
126 		goto no_context;
127 
128 	/* TLB misses upon some cache flushes get done under cli() */
129 	down_read(&mm->mmap_sem);
130 
131 	vma = find_vma(mm, address);
132 
133 	if (!vma) {
134 #ifdef DEBUG_FAULT
135 		print_task(tsk);
136 		printk("%s:%d fault, address is 0x%08x PC %016Lx textaccess %d writeaccess %d\n",
137 		       __func__, __LINE__,
138 		       address,regs->pc,textaccess,writeaccess);
139 		show_regs(regs);
140 #endif
141 		goto bad_area;
142 	}
143 	if (vma->vm_start <= address) {
144 		goto good_area;
145 	}
146 
147 	if (!(vma->vm_flags & VM_GROWSDOWN)) {
148 #ifdef DEBUG_FAULT
149 		print_task(tsk);
150 		printk("%s:%d fault, address is 0x%08x PC %016Lx textaccess %d writeaccess %d\n",
151 		       __func__, __LINE__,
152 		       address,regs->pc,textaccess,writeaccess);
153 		show_regs(regs);
154 
155 		print_vma(vma);
156 #endif
157 		goto bad_area;
158 	}
159 	if (expand_stack(vma, address)) {
160 #ifdef DEBUG_FAULT
161 		print_task(tsk);
162 		printk("%s:%d fault, address is 0x%08x PC %016Lx textaccess %d writeaccess %d\n",
163 		       __func__, __LINE__,
164 		       address,regs->pc,textaccess,writeaccess);
165 		show_regs(regs);
166 #endif
167 		goto bad_area;
168 	}
169 /*
170  * Ok, we have a good vm_area for this memory access, so
171  * we can handle it..
172  */
173 good_area:
174 	if (textaccess) {
175 		if (!(vma->vm_flags & VM_EXEC))
176 			goto bad_area;
177 	} else {
178 		if (writeaccess) {
179 			if (!(vma->vm_flags & VM_WRITE))
180 				goto bad_area;
181 		} else {
182 			if (!(vma->vm_flags & VM_READ))
183 				goto bad_area;
184 		}
185 	}
186 
187 	/*
188 	 * If for any reason at all we couldn't handle the fault,
189 	 * make sure we exit gracefully rather than endlessly redo
190 	 * the fault.
191 	 */
192 	fault = handle_mm_fault(mm, vma, address, writeaccess ? FAULT_FLAG_WRITE : 0);
193 	if (unlikely(fault & VM_FAULT_ERROR)) {
194 		if (fault & VM_FAULT_OOM)
195 			goto out_of_memory;
196 		else if (fault & VM_FAULT_SIGBUS)
197 			goto do_sigbus;
198 		BUG();
199 	}
200 
201 	if (fault & VM_FAULT_MAJOR) {
202 		tsk->maj_flt++;
203 		perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
204 				     regs, address);
205 	} else {
206 		tsk->min_flt++;
207 		perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
208 				     regs, address);
209 	}
210 
211 	/* If we get here, the page fault has been handled.  Do the TLB refill
212 	   now from the newly-setup PTE, to avoid having to fault again right
213 	   away on the same instruction. */
214 	pte = lookup_pte (mm, address);
215 	if (!pte) {
216 		/* From empirical evidence, we can get here, due to
217 		   !pte_present(pte).  (e.g. if a swap-in occurs, and the page
218 		   is swapped back out again before the process that wanted it
219 		   gets rescheduled?) */
220 		goto no_pte;
221 	}
222 
223 	__do_tlb_refill(address, textaccess, pte);
224 
225 no_pte:
226 
227 	up_read(&mm->mmap_sem);
228 	return;
229 
230 /*
231  * Something tried to access memory that isn't in our memory map..
232  * Fix it, but check if it's kernel or user first..
233  */
234 bad_area:
235 #ifdef DEBUG_FAULT
236 	printk("fault:bad area\n");
237 #endif
238 	up_read(&mm->mmap_sem);
239 
240 	if (user_mode(regs)) {
241 		static int count=0;
242 		siginfo_t info;
243 		if (count < 4) {
244 			/* This is really to help debug faults when starting
245 			 * usermode, so only need a few */
246 			count++;
247 			printk("user mode bad_area address=%08lx pid=%d (%s) pc=%08lx\n",
248 				address, task_pid_nr(current), current->comm,
249 				(unsigned long) regs->pc);
250 #if 0
251 			show_regs(regs);
252 #endif
253 		}
254 		if (is_global_init(tsk)) {
255 			panic("INIT had user mode bad_area\n");
256 		}
257 		tsk->thread.address = address;
258 		tsk->thread.error_code = writeaccess;
259 		info.si_signo = SIGSEGV;
260 		info.si_errno = 0;
261 		info.si_addr = (void *) address;
262 		force_sig_info(SIGSEGV, &info, tsk);
263 		return;
264 	}
265 
266 no_context:
267 #ifdef DEBUG_FAULT
268 	printk("fault:No context\n");
269 #endif
270 	/* Are we prepared to handle this kernel fault?  */
271 	fixup = search_exception_tables(regs->pc);
272 	if (fixup) {
273 		regs->pc = fixup->fixup;
274 		return;
275 	}
276 
277 /*
278  * Oops. The kernel tried to access some bad page. We'll have to
279  * terminate things with extreme prejudice.
280  *
281  */
282 	if (address < PAGE_SIZE)
283 		printk(KERN_ALERT "Unable to handle kernel NULL pointer dereference");
284 	else
285 		printk(KERN_ALERT "Unable to handle kernel paging request");
286 	printk(" at virtual address %08lx\n", address);
287 	printk(KERN_ALERT "pc = %08Lx%08Lx\n", regs->pc >> 32, regs->pc & 0xffffffff);
288 	die("Oops", regs, writeaccess);
289 	do_exit(SIGKILL);
290 
291 /*
292  * We ran out of memory, or some other thing happened to us that made
293  * us unable to handle the page fault gracefully.
294  */
295 out_of_memory:
296 	up_read(&mm->mmap_sem);
297 	if (!user_mode(regs))
298 		goto no_context;
299 	pagefault_out_of_memory();
300 	return;
301 
302 do_sigbus:
303 	printk("fault:Do sigbus\n");
304 	up_read(&mm->mmap_sem);
305 
306 	/*
307 	 * Send a sigbus, regardless of whether we were in kernel
308 	 * or user mode.
309 	 */
310 	tsk->thread.address = address;
311 	tsk->thread.error_code = writeaccess;
312 	tsk->thread.trap_no = 14;
313 	force_sig(SIGBUS, tsk);
314 
315 	/* Kernel mode? Handle exceptions or die */
316 	if (!user_mode(regs))
317 		goto no_context;
318 }
319 
local_flush_tlb_one(unsigned long asid,unsigned long page)320 void local_flush_tlb_one(unsigned long asid, unsigned long page)
321 {
322 	unsigned long long match, pteh=0, lpage;
323 	unsigned long tlb;
324 
325 	/*
326 	 * Sign-extend based on neff.
327 	 */
328 	lpage = neff_sign_extend(page);
329 	match = (asid << PTEH_ASID_SHIFT) | PTEH_VALID;
330 	match |= lpage;
331 
332 	for_each_itlb_entry(tlb) {
333 		asm volatile ("getcfg	%1, 0, %0"
334 			      : "=r" (pteh)
335 			      : "r" (tlb) );
336 
337 		if (pteh == match) {
338 			__flush_tlb_slot(tlb);
339 			break;
340 		}
341 	}
342 
343 	for_each_dtlb_entry(tlb) {
344 		asm volatile ("getcfg	%1, 0, %0"
345 			      : "=r" (pteh)
346 			      : "r" (tlb) );
347 
348 		if (pteh == match) {
349 			__flush_tlb_slot(tlb);
350 			break;
351 		}
352 
353 	}
354 }
355 
local_flush_tlb_page(struct vm_area_struct * vma,unsigned long page)356 void local_flush_tlb_page(struct vm_area_struct *vma, unsigned long page)
357 {
358 	unsigned long flags;
359 
360 	if (vma->vm_mm) {
361 		page &= PAGE_MASK;
362 		local_irq_save(flags);
363 		local_flush_tlb_one(get_asid(), page);
364 		local_irq_restore(flags);
365 	}
366 }
367 
local_flush_tlb_range(struct vm_area_struct * vma,unsigned long start,unsigned long end)368 void local_flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
369 			   unsigned long end)
370 {
371 	unsigned long flags;
372 	unsigned long long match, pteh=0, pteh_epn, pteh_low;
373 	unsigned long tlb;
374 	unsigned int cpu = smp_processor_id();
375 	struct mm_struct *mm;
376 
377 	mm = vma->vm_mm;
378 	if (cpu_context(cpu, mm) == NO_CONTEXT)
379 		return;
380 
381 	local_irq_save(flags);
382 
383 	start &= PAGE_MASK;
384 	end &= PAGE_MASK;
385 
386 	match = (cpu_asid(cpu, mm) << PTEH_ASID_SHIFT) | PTEH_VALID;
387 
388 	/* Flush ITLB */
389 	for_each_itlb_entry(tlb) {
390 		asm volatile ("getcfg	%1, 0, %0"
391 			      : "=r" (pteh)
392 			      : "r" (tlb) );
393 
394 		pteh_epn = pteh & PAGE_MASK;
395 		pteh_low = pteh & ~PAGE_MASK;
396 
397 		if (pteh_low == match && pteh_epn >= start && pteh_epn <= end)
398 			__flush_tlb_slot(tlb);
399 	}
400 
401 	/* Flush DTLB */
402 	for_each_dtlb_entry(tlb) {
403 		asm volatile ("getcfg	%1, 0, %0"
404 			      : "=r" (pteh)
405 			      : "r" (tlb) );
406 
407 		pteh_epn = pteh & PAGE_MASK;
408 		pteh_low = pteh & ~PAGE_MASK;
409 
410 		if (pteh_low == match && pteh_epn >= start && pteh_epn <= end)
411 			__flush_tlb_slot(tlb);
412 	}
413 
414 	local_irq_restore(flags);
415 }
416 
local_flush_tlb_mm(struct mm_struct * mm)417 void local_flush_tlb_mm(struct mm_struct *mm)
418 {
419 	unsigned long flags;
420 	unsigned int cpu = smp_processor_id();
421 
422 	if (cpu_context(cpu, mm) == NO_CONTEXT)
423 		return;
424 
425 	local_irq_save(flags);
426 
427 	cpu_context(cpu, mm) = NO_CONTEXT;
428 	if (mm == current->mm)
429 		activate_context(mm, cpu);
430 
431 	local_irq_restore(flags);
432 }
433 
local_flush_tlb_all(void)434 void local_flush_tlb_all(void)
435 {
436 	/* Invalidate all, including shared pages, excluding fixed TLBs */
437 	unsigned long flags, tlb;
438 
439 	local_irq_save(flags);
440 
441 	/* Flush each ITLB entry */
442 	for_each_itlb_entry(tlb)
443 		__flush_tlb_slot(tlb);
444 
445 	/* Flush each DTLB entry */
446 	for_each_dtlb_entry(tlb)
447 		__flush_tlb_slot(tlb);
448 
449 	local_irq_restore(flags);
450 }
451 
local_flush_tlb_kernel_range(unsigned long start,unsigned long end)452 void local_flush_tlb_kernel_range(unsigned long start, unsigned long end)
453 {
454         /* FIXME: Optimize this later.. */
455         flush_tlb_all();
456 }
457 
__flush_tlb_global(void)458 void __flush_tlb_global(void)
459 {
460 	flush_tlb_all();
461 }
462 
__update_tlb(struct vm_area_struct * vma,unsigned long address,pte_t pte)463 void __update_tlb(struct vm_area_struct *vma, unsigned long address, pte_t pte)
464 {
465 }
466