mm: fix exec activate_mm vs TLB shootdown and lazy tlb switching race
[linux/fpc-iii.git] / arch / sparc / mm / fault_64.c
blob815c03d7a765524424b92866b1567ea2a43695d4
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * arch/sparc64/mm/fault.c: Page fault handlers for the 64-bit Sparc.
5 * Copyright (C) 1996, 2008 David S. Miller (davem@davemloft.net)
6 * Copyright (C) 1997, 1999 Jakub Jelinek (jj@ultra.linux.cz)
7 */
9 #include <asm/head.h>
11 #include <linux/string.h>
12 #include <linux/types.h>
13 #include <linux/sched.h>
14 #include <linux/sched/debug.h>
15 #include <linux/ptrace.h>
16 #include <linux/mman.h>
17 #include <linux/signal.h>
18 #include <linux/mm.h>
19 #include <linux/extable.h>
20 #include <linux/init.h>
21 #include <linux/perf_event.h>
22 #include <linux/interrupt.h>
23 #include <linux/kprobes.h>
24 #include <linux/kdebug.h>
25 #include <linux/percpu.h>
26 #include <linux/context_tracking.h>
27 #include <linux/uaccess.h>
29 #include <asm/page.h>
30 #include <asm/pgtable.h>
31 #include <asm/openprom.h>
32 #include <asm/oplib.h>
33 #include <asm/asi.h>
34 #include <asm/lsu.h>
35 #include <asm/sections.h>
36 #include <asm/mmu_context.h>
37 #include <asm/setup.h>
39 int show_unhandled_signals = 1;
41 static inline __kprobes int notify_page_fault(struct pt_regs *regs)
43 int ret = 0;
45 /* kprobe_running() needs smp_processor_id() */
46 if (kprobes_built_in() && !user_mode(regs)) {
47 preempt_disable();
48 if (kprobe_running() && kprobe_fault_handler(regs, 0))
49 ret = 1;
50 preempt_enable();
52 return ret;
55 static void __kprobes unhandled_fault(unsigned long address,
56 struct task_struct *tsk,
57 struct pt_regs *regs)
59 if ((unsigned long) address < PAGE_SIZE) {
60 printk(KERN_ALERT "Unable to handle kernel NULL "
61 "pointer dereference\n");
62 } else {
63 printk(KERN_ALERT "Unable to handle kernel paging request "
64 "at virtual address %016lx\n", (unsigned long)address);
66 printk(KERN_ALERT "tsk->{mm,active_mm}->context = %016lx\n",
67 (tsk->mm ?
68 CTX_HWBITS(tsk->mm->context) :
69 CTX_HWBITS(tsk->active_mm->context)));
70 printk(KERN_ALERT "tsk->{mm,active_mm}->pgd = %016lx\n",
71 (tsk->mm ? (unsigned long) tsk->mm->pgd :
72 (unsigned long) tsk->active_mm->pgd));
73 die_if_kernel("Oops", regs);
76 static void __kprobes bad_kernel_pc(struct pt_regs *regs, unsigned long vaddr)
78 printk(KERN_CRIT "OOPS: Bogus kernel PC [%016lx] in fault handler\n",
79 regs->tpc);
80 printk(KERN_CRIT "OOPS: RPC [%016lx]\n", regs->u_regs[15]);
81 printk("OOPS: RPC <%pS>\n", (void *) regs->u_regs[15]);
82 printk(KERN_CRIT "OOPS: Fault was to vaddr[%lx]\n", vaddr);
83 dump_stack();
84 unhandled_fault(regs->tpc, current, regs);
88 * We now make sure that mmap_sem is held in all paths that call
89 * this. Additionally, to prevent kswapd from ripping ptes from
90 * under us, raise interrupts around the time that we look at the
91 * pte, kswapd will have to wait to get his smp ipi response from
92 * us. vmtruncate likewise. This saves us having to get pte lock.
94 static unsigned int get_user_insn(unsigned long tpc)
96 pgd_t *pgdp = pgd_offset(current->mm, tpc);
97 pud_t *pudp;
98 pmd_t *pmdp;
99 pte_t *ptep, pte;
100 unsigned long pa;
101 u32 insn = 0;
103 if (pgd_none(*pgdp) || unlikely(pgd_bad(*pgdp)))
104 goto out;
105 pudp = pud_offset(pgdp, tpc);
106 if (pud_none(*pudp) || unlikely(pud_bad(*pudp)))
107 goto out;
109 /* This disables preemption for us as well. */
110 local_irq_disable();
112 pmdp = pmd_offset(pudp, tpc);
113 if (pmd_none(*pmdp) || unlikely(pmd_bad(*pmdp)))
114 goto out_irq_enable;
116 #if defined(CONFIG_HUGETLB_PAGE) || defined(CONFIG_TRANSPARENT_HUGEPAGE)
117 if (is_hugetlb_pmd(*pmdp)) {
118 pa = pmd_pfn(*pmdp) << PAGE_SHIFT;
119 pa += tpc & ~HPAGE_MASK;
121 /* Use phys bypass so we don't pollute dtlb/dcache. */
122 __asm__ __volatile__("lduwa [%1] %2, %0"
123 : "=r" (insn)
124 : "r" (pa), "i" (ASI_PHYS_USE_EC));
125 } else
126 #endif
128 ptep = pte_offset_map(pmdp, tpc);
129 pte = *ptep;
130 if (pte_present(pte)) {
131 pa = (pte_pfn(pte) << PAGE_SHIFT);
132 pa += (tpc & ~PAGE_MASK);
134 /* Use phys bypass so we don't pollute dtlb/dcache. */
135 __asm__ __volatile__("lduwa [%1] %2, %0"
136 : "=r" (insn)
137 : "r" (pa), "i" (ASI_PHYS_USE_EC));
139 pte_unmap(ptep);
141 out_irq_enable:
142 local_irq_enable();
143 out:
144 return insn;
147 static inline void
148 show_signal_msg(struct pt_regs *regs, int sig, int code,
149 unsigned long address, struct task_struct *tsk)
151 if (!unhandled_signal(tsk, sig))
152 return;
154 if (!printk_ratelimit())
155 return;
157 printk("%s%s[%d]: segfault at %lx ip %p (rpc %p) sp %p error %x",
158 task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
159 tsk->comm, task_pid_nr(tsk), address,
160 (void *)regs->tpc, (void *)regs->u_regs[UREG_I7],
161 (void *)regs->u_regs[UREG_FP], code);
163 print_vma_addr(KERN_CONT " in ", regs->tpc);
165 printk(KERN_CONT "\n");
168 static void do_fault_siginfo(int code, int sig, struct pt_regs *regs,
169 unsigned long fault_addr, unsigned int insn,
170 int fault_code)
172 unsigned long addr;
173 siginfo_t info;
175 info.si_code = code;
176 info.si_signo = sig;
177 info.si_errno = 0;
178 if (fault_code & FAULT_CODE_ITLB) {
179 addr = regs->tpc;
180 } else {
181 /* If we were able to probe the faulting instruction, use it
182 * to compute a precise fault address. Otherwise use the fault
183 * time provided address which may only have page granularity.
185 if (insn)
186 addr = compute_effective_address(regs, insn, 0);
187 else
188 addr = fault_addr;
190 info.si_addr = (void __user *) addr;
191 info.si_trapno = 0;
193 if (unlikely(show_unhandled_signals))
194 show_signal_msg(regs, sig, code, addr, current);
196 force_sig_info(sig, &info, current);
199 static unsigned int get_fault_insn(struct pt_regs *regs, unsigned int insn)
201 if (!insn) {
202 if (!regs->tpc || (regs->tpc & 0x3))
203 return 0;
204 if (regs->tstate & TSTATE_PRIV) {
205 insn = *(unsigned int *) regs->tpc;
206 } else {
207 insn = get_user_insn(regs->tpc);
210 return insn;
213 static void __kprobes do_kernel_fault(struct pt_regs *regs, int si_code,
214 int fault_code, unsigned int insn,
215 unsigned long address)
217 unsigned char asi = ASI_P;
219 if ((!insn) && (regs->tstate & TSTATE_PRIV))
220 goto cannot_handle;
222 /* If user insn could be read (thus insn is zero), that
223 * is fine. We will just gun down the process with a signal
224 * in that case.
227 if (!(fault_code & (FAULT_CODE_WRITE|FAULT_CODE_ITLB)) &&
228 (insn & 0xc0800000) == 0xc0800000) {
229 if (insn & 0x2000)
230 asi = (regs->tstate >> 24);
231 else
232 asi = (insn >> 5);
233 if ((asi & 0xf2) == 0x82) {
234 if (insn & 0x1000000) {
235 handle_ldf_stq(insn, regs);
236 } else {
237 /* This was a non-faulting load. Just clear the
238 * destination register(s) and continue with the next
239 * instruction. -jj
241 handle_ld_nf(insn, regs);
243 return;
247 /* Is this in ex_table? */
248 if (regs->tstate & TSTATE_PRIV) {
249 const struct exception_table_entry *entry;
251 entry = search_exception_tables(regs->tpc);
252 if (entry) {
253 regs->tpc = entry->fixup;
254 regs->tnpc = regs->tpc + 4;
255 return;
257 } else {
258 /* The si_code was set to make clear whether
259 * this was a SEGV_MAPERR or SEGV_ACCERR fault.
261 do_fault_siginfo(si_code, SIGSEGV, regs, address, insn, fault_code);
262 return;
265 cannot_handle:
266 unhandled_fault (address, current, regs);
269 static void noinline __kprobes bogus_32bit_fault_tpc(struct pt_regs *regs)
271 static int times;
273 if (times++ < 10)
274 printk(KERN_ERR "FAULT[%s:%d]: 32-bit process reports "
275 "64-bit TPC [%lx]\n",
276 current->comm, current->pid,
277 regs->tpc);
278 show_regs(regs);
281 asmlinkage void __kprobes do_sparc64_fault(struct pt_regs *regs)
283 enum ctx_state prev_state = exception_enter();
284 struct mm_struct *mm = current->mm;
285 struct vm_area_struct *vma;
286 unsigned int insn = 0;
287 int si_code, fault_code, fault;
288 unsigned long address, mm_rss;
289 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
291 fault_code = get_thread_fault_code();
293 if (notify_page_fault(regs))
294 goto exit_exception;
296 si_code = SEGV_MAPERR;
297 address = current_thread_info()->fault_address;
299 if ((fault_code & FAULT_CODE_ITLB) &&
300 (fault_code & FAULT_CODE_DTLB))
301 BUG();
303 if (test_thread_flag(TIF_32BIT)) {
304 if (!(regs->tstate & TSTATE_PRIV)) {
305 if (unlikely((regs->tpc >> 32) != 0)) {
306 bogus_32bit_fault_tpc(regs);
307 goto intr_or_no_mm;
310 if (unlikely((address >> 32) != 0))
311 goto intr_or_no_mm;
314 if (regs->tstate & TSTATE_PRIV) {
315 unsigned long tpc = regs->tpc;
317 /* Sanity check the PC. */
318 if ((tpc >= KERNBASE && tpc < (unsigned long) __init_end) ||
319 (tpc >= MODULES_VADDR && tpc < MODULES_END)) {
320 /* Valid, no problems... */
321 } else {
322 bad_kernel_pc(regs, address);
323 goto exit_exception;
325 } else
326 flags |= FAULT_FLAG_USER;
329 * If we're in an interrupt or have no user
330 * context, we must not take the fault..
332 if (faulthandler_disabled() || !mm)
333 goto intr_or_no_mm;
335 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
337 if (!down_read_trylock(&mm->mmap_sem)) {
338 if ((regs->tstate & TSTATE_PRIV) &&
339 !search_exception_tables(regs->tpc)) {
340 insn = get_fault_insn(regs, insn);
341 goto handle_kernel_fault;
344 retry:
345 down_read(&mm->mmap_sem);
348 if (fault_code & FAULT_CODE_BAD_RA)
349 goto do_sigbus;
351 vma = find_vma(mm, address);
352 if (!vma)
353 goto bad_area;
355 /* Pure DTLB misses do not tell us whether the fault causing
356 * load/store/atomic was a write or not, it only says that there
357 * was no match. So in such a case we (carefully) read the
358 * instruction to try and figure this out. It's an optimization
359 * so it's ok if we can't do this.
361 * Special hack, window spill/fill knows the exact fault type.
363 if (((fault_code &
364 (FAULT_CODE_DTLB | FAULT_CODE_WRITE | FAULT_CODE_WINFIXUP)) == FAULT_CODE_DTLB) &&
365 (vma->vm_flags & VM_WRITE) != 0) {
366 insn = get_fault_insn(regs, 0);
367 if (!insn)
368 goto continue_fault;
369 /* All loads, stores and atomics have bits 30 and 31 both set
370 * in the instruction. Bit 21 is set in all stores, but we
371 * have to avoid prefetches which also have bit 21 set.
373 if ((insn & 0xc0200000) == 0xc0200000 &&
374 (insn & 0x01780000) != 0x01680000) {
375 /* Don't bother updating thread struct value,
376 * because update_mmu_cache only cares which tlb
377 * the access came from.
379 fault_code |= FAULT_CODE_WRITE;
382 continue_fault:
384 if (vma->vm_start <= address)
385 goto good_area;
386 if (!(vma->vm_flags & VM_GROWSDOWN))
387 goto bad_area;
388 if (!(fault_code & FAULT_CODE_WRITE)) {
389 /* Non-faulting loads shouldn't expand stack. */
390 insn = get_fault_insn(regs, insn);
391 if ((insn & 0xc0800000) == 0xc0800000) {
392 unsigned char asi;
394 if (insn & 0x2000)
395 asi = (regs->tstate >> 24);
396 else
397 asi = (insn >> 5);
398 if ((asi & 0xf2) == 0x82)
399 goto bad_area;
402 if (expand_stack(vma, address))
403 goto bad_area;
405 * Ok, we have a good vm_area for this memory access, so
406 * we can handle it..
408 good_area:
409 si_code = SEGV_ACCERR;
411 /* If we took a ITLB miss on a non-executable page, catch
412 * that here.
414 if ((fault_code & FAULT_CODE_ITLB) && !(vma->vm_flags & VM_EXEC)) {
415 WARN(address != regs->tpc,
416 "address (%lx) != regs->tpc (%lx)\n", address, regs->tpc);
417 WARN_ON(regs->tstate & TSTATE_PRIV);
418 goto bad_area;
421 if (fault_code & FAULT_CODE_WRITE) {
422 if (!(vma->vm_flags & VM_WRITE))
423 goto bad_area;
425 /* Spitfire has an icache which does not snoop
426 * processor stores. Later processors do...
428 if (tlb_type == spitfire &&
429 (vma->vm_flags & VM_EXEC) != 0 &&
430 vma->vm_file != NULL)
431 set_thread_fault_code(fault_code |
432 FAULT_CODE_BLKCOMMIT);
434 flags |= FAULT_FLAG_WRITE;
435 } else {
436 /* Allow reads even for write-only mappings */
437 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
438 goto bad_area;
441 fault = handle_mm_fault(vma, address, flags);
443 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
444 goto exit_exception;
446 if (unlikely(fault & VM_FAULT_ERROR)) {
447 if (fault & VM_FAULT_OOM)
448 goto out_of_memory;
449 else if (fault & VM_FAULT_SIGSEGV)
450 goto bad_area;
451 else if (fault & VM_FAULT_SIGBUS)
452 goto do_sigbus;
453 BUG();
456 if (flags & FAULT_FLAG_ALLOW_RETRY) {
457 if (fault & VM_FAULT_MAJOR) {
458 current->maj_flt++;
459 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ,
460 1, regs, address);
461 } else {
462 current->min_flt++;
463 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN,
464 1, regs, address);
466 if (fault & VM_FAULT_RETRY) {
467 flags &= ~FAULT_FLAG_ALLOW_RETRY;
468 flags |= FAULT_FLAG_TRIED;
470 /* No need to up_read(&mm->mmap_sem) as we would
471 * have already released it in __lock_page_or_retry
472 * in mm/filemap.c.
475 goto retry;
478 up_read(&mm->mmap_sem);
480 mm_rss = get_mm_rss(mm);
481 #if defined(CONFIG_TRANSPARENT_HUGEPAGE)
482 mm_rss -= (mm->context.thp_pte_count * (HPAGE_SIZE / PAGE_SIZE));
483 #endif
484 if (unlikely(mm_rss >
485 mm->context.tsb_block[MM_TSB_BASE].tsb_rss_limit))
486 tsb_grow(mm, MM_TSB_BASE, mm_rss);
487 #if defined(CONFIG_HUGETLB_PAGE) || defined(CONFIG_TRANSPARENT_HUGEPAGE)
488 mm_rss = mm->context.hugetlb_pte_count + mm->context.thp_pte_count;
489 mm_rss *= REAL_HPAGE_PER_HPAGE;
490 if (unlikely(mm_rss >
491 mm->context.tsb_block[MM_TSB_HUGE].tsb_rss_limit)) {
492 if (mm->context.tsb_block[MM_TSB_HUGE].tsb)
493 tsb_grow(mm, MM_TSB_HUGE, mm_rss);
494 else
495 hugetlb_setup(regs);
498 #endif
499 exit_exception:
500 exception_exit(prev_state);
501 return;
504 * Something tried to access memory that isn't in our memory map..
505 * Fix it, but check if it's kernel or user first..
507 bad_area:
508 insn = get_fault_insn(regs, insn);
509 up_read(&mm->mmap_sem);
511 handle_kernel_fault:
512 do_kernel_fault(regs, si_code, fault_code, insn, address);
513 goto exit_exception;
516 * We ran out of memory, or some other thing happened to us that made
517 * us unable to handle the page fault gracefully.
519 out_of_memory:
520 insn = get_fault_insn(regs, insn);
521 up_read(&mm->mmap_sem);
522 if (!(regs->tstate & TSTATE_PRIV)) {
523 pagefault_out_of_memory();
524 goto exit_exception;
526 goto handle_kernel_fault;
528 intr_or_no_mm:
529 insn = get_fault_insn(regs, 0);
530 goto handle_kernel_fault;
532 do_sigbus:
533 insn = get_fault_insn(regs, insn);
534 up_read(&mm->mmap_sem);
537 * Send a sigbus, regardless of whether we were in kernel
538 * or user mode.
540 do_fault_siginfo(BUS_ADRERR, SIGBUS, regs, address, insn, fault_code);
542 /* Kernel mode? Handle exceptions or die */
543 if (regs->tstate & TSTATE_PRIV)
544 goto handle_kernel_fault;