csky/mm/fault.c

0001 // SPDX-License-Identifier: GPL-2.0
0002 // Copyright (C) 2018 Hangzhou C-SKY Microsystems co.,ltd.
0003
0004 #include <linux/extable.h>
0005 #include <linux/kprobes.h>
0006 #include <linux/mmu_context.h>
0007 #include <linux/perf_event.h>
0008
0009 int fixup_exception(struct pt_regs *regs)
0010 {
0011     const struct exception_table_entry *fixup;
0012
0013     fixup = search_exception_tables(instruction_pointer(regs));
0014     if (fixup) {
0015         regs->pc = fixup->fixup;
0016
0017         return 1;
0018     }
0019
0020     return 0;
0021 }
0022
0023 static inline bool is_write(struct pt_regs *regs)
0024 {
0025     switch (trap_no(regs)) {
0026     case VEC_TLBINVALIDS:
0027         return true;
0028     case VEC_TLBMODIFIED:
0029         return true;
0030     }
0031
0032     return false;
0033 }
0034
0035 #ifdef CONFIG_CPU_HAS_LDSTEX
0036 static inline void csky_cmpxchg_fixup(struct pt_regs *regs)
0037 {
0038     return;
0039 }
0040 #else
0041 extern unsigned long csky_cmpxchg_ldw;
0042 extern unsigned long csky_cmpxchg_stw;
0043 static inline void csky_cmpxchg_fixup(struct pt_regs *regs)
0044 {
0045     if (trap_no(regs) != VEC_TLBMODIFIED)
0046         return;
0047
0048     if (instruction_pointer(regs) == csky_cmpxchg_stw)
0049         instruction_pointer_set(regs, csky_cmpxchg_ldw);
0050     return;
0051 }
0052 #endif
0053
0054 static inline void no_context(struct pt_regs *regs, unsigned long addr)
0055 {
0056     current->thread.trap_no = trap_no(regs);
0057
0058     /* Are we prepared to handle this kernel fault? */
0059     if (fixup_exception(regs))
0060         return;
0061
0062     /*
0063      * Oops. The kernel tried to access some bad page. We'll have to
0064      * terminate things with extreme prejudice.
0065      */
0066     bust_spinlocks(1);
0067     pr_alert("Unable to handle kernel paging request at virtual "
0068          "addr 0x%08lx, pc: 0x%08lx\n", addr, regs->pc);
0069     die(regs, "Oops");
0070     make_task_dead(SIGKILL);
0071 }
0072
0073 static inline void mm_fault_error(struct pt_regs *regs, unsigned long addr, vm_fault_t fault)
0074 {
0075     current->thread.trap_no = trap_no(regs);
0076
0077     if (fault & VM_FAULT_OOM) {
0078         /*
0079          * We ran out of memory, call the OOM killer, and return the userspace
0080          * (which will retry the fault, or kill us if we got oom-killed).
0081          */
0082         if (!user_mode(regs)) {
0083             no_context(regs, addr);
0084             return;
0085         }
0086         pagefault_out_of_memory();
0087         return;
0088     } else if (fault & VM_FAULT_SIGBUS) {
0089         /* Kernel mode? Handle exceptions or die */
0090         if (!user_mode(regs)) {
0091             no_context(regs, addr);
0092             return;
0093         }
0094         do_trap(regs, SIGBUS, BUS_ADRERR, addr);
0095         return;
0096     }
0097     BUG();
0098 }
0099
0100 static inline void bad_area(struct pt_regs *regs, struct mm_struct *mm, int code, unsigned long addr)
0101 {
0102     /*
0103      * Something tried to access memory that isn't in our memory map.
0104      * Fix it, but check if it's kernel or user first.
0105      */
0106     mmap_read_unlock(mm);
0107     /* User mode accesses just cause a SIGSEGV */
0108     if (user_mode(regs)) {
0109         do_trap(regs, SIGSEGV, code, addr);
0110         return;
0111     }
0112
0113     no_context(regs, addr);
0114 }
0115
0116 static inline void vmalloc_fault(struct pt_regs *regs, int code, unsigned long addr)
0117 {
0118     pgd_t *pgd, *pgd_k;
0119     pud_t *pud, *pud_k;
0120     pmd_t *pmd, *pmd_k;
0121     pte_t *pte_k;
0122     int offset;
0123
0124     /* User mode accesses just cause a SIGSEGV */
0125     if (user_mode(regs)) {
0126         do_trap(regs, SIGSEGV, code, addr);
0127         return;
0128     }
0129
0130     /*
0131      * Synchronize this task's top level page-table
0132      * with the 'reference' page table.
0133      *
0134      * Do _not_ use "tsk" here. We might be inside
0135      * an interrupt in the middle of a task switch..
0136      */
0137     offset = pgd_index(addr);
0138
0139     pgd = get_pgd() + offset;
0140     pgd_k = init_mm.pgd + offset;
0141
0142     if (!pgd_present(*pgd_k)) {
0143         no_context(regs, addr);
0144         return;
0145     }
0146     set_pgd(pgd, *pgd_k);
0147
0148     pud = (pud_t *)pgd;
0149     pud_k = (pud_t *)pgd_k;
0150     if (!pud_present(*pud_k)) {
0151         no_context(regs, addr);
0152         return;
0153     }
0154
0155     pmd = pmd_offset(pud, addr);
0156     pmd_k = pmd_offset(pud_k, addr);
0157     if (!pmd_present(*pmd_k)) {
0158         no_context(regs, addr);
0159         return;
0160     }
0161     set_pmd(pmd, *pmd_k);
0162
0163     pte_k = pte_offset_kernel(pmd_k, addr);
0164     if (!pte_present(*pte_k)) {
0165         no_context(regs, addr);
0166         return;
0167     }
0168
0169     flush_tlb_one(addr);
0170 }
0171
0172 static inline bool access_error(struct pt_regs *regs, struct vm_area_struct *vma)
0173 {
0174     if (is_write(regs)) {
0175         if (!(vma->vm_flags & VM_WRITE))
0176             return true;
0177     } else {
0178         if (unlikely(!vma_is_accessible(vma)))
0179             return true;
0180     }
0181     return false;
0182 }
0183
0184 /*
0185  * This routine handles page faults.  It determines the address and the
0186  * problem, and then passes it off to one of the appropriate routines.
0187  */
0188 asmlinkage void do_page_fault(struct pt_regs *regs)
0189 {
0190     struct task_struct *tsk;
0191     struct vm_area_struct *vma;
0192     struct mm_struct *mm;
0193     unsigned long addr = read_mmu_entryhi() & PAGE_MASK;
0194     unsigned int flags = FAULT_FLAG_DEFAULT;
0195     int code = SEGV_MAPERR;
0196     vm_fault_t fault;
0197
0198     tsk = current;
0199     mm = tsk->mm;
0200
0201     csky_cmpxchg_fixup(regs);
0202
0203     if (kprobe_page_fault(regs, tsk->thread.trap_no))
0204         return;
0205
0206     /*
0207      * Fault-in kernel-space virtual memory on-demand.
0208      * The 'reference' page table is init_mm.pgd.
0209      *
0210      * NOTE! We MUST NOT take any locks for this case. We may
0211      * be in an interrupt or a critical region, and should
0212      * only copy the information from the master page table,
0213      * nothing more.
0214      */
0215     if (unlikely((addr >= VMALLOC_START) && (addr <= VMALLOC_END))) {
0216         vmalloc_fault(regs, code, addr);
0217         return;
0218     }
0219
0220     /* Enable interrupts if they were enabled in the parent context. */
0221     if (likely(regs->sr & BIT(6)))
0222         local_irq_enable();
0223
0224     /*
0225      * If we're in an interrupt, have no user context, or are running
0226      * in an atomic region, then we must not take the fault.
0227      */
0228     if (unlikely(faulthandler_disabled() || !mm)) {
0229         no_context(regs, addr);
0230         return;
0231     }
0232
0233     if (user_mode(regs))
0234         flags |= FAULT_FLAG_USER;
0235
0236     perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
0237
0238     if (is_write(regs))
0239         flags |= FAULT_FLAG_WRITE;
0240 retry:
0241     mmap_read_lock(mm);
0242     vma = find_vma(mm, addr);
0243     if (unlikely(!vma)) {
0244         bad_area(regs, mm, code, addr);
0245         return;
0246     }
0247     if (likely(vma->vm_start <= addr))
0248         goto good_area;
0249     if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) {
0250         bad_area(regs, mm, code, addr);
0251         return;
0252     }
0253     if (unlikely(expand_stack(vma, addr))) {
0254         bad_area(regs, mm, code, addr);
0255         return;
0256     }
0257
0258     /*
0259      * Ok, we have a good vm_area for this memory access, so
0260      * we can handle it.
0261      */
0262 good_area:
0263     code = SEGV_ACCERR;
0264
0265     if (unlikely(access_error(regs, vma))) {
0266         bad_area(regs, mm, code, addr);
0267         return;
0268     }
0269
0270     /*
0271      * If for any reason at all we could not handle the fault,
0272      * make sure we exit gracefully rather than endlessly redo
0273      * the fault.
0274      */
0275     fault = handle_mm_fault(vma, addr, flags, regs);
0276
0277     /*
0278      * If we need to retry but a fatal signal is pending, handle the
0279      * signal first. We do not need to release the mmap_lock because it
0280      * would already be released in __lock_page_or_retry in mm/filemap.c.
0281      */
0282     if (fault_signal_pending(fault, regs)) {
0283         if (!user_mode(regs))
0284             no_context(regs, addr);
0285         return;
0286     }
0287
0288     /* The fault is fully completed (including releasing mmap lock) */
0289     if (fault & VM_FAULT_COMPLETED)
0290         return;
0291
0292     if (unlikely((fault & VM_FAULT_RETRY) && (flags & FAULT_FLAG_ALLOW_RETRY))) {
0293         flags |= FAULT_FLAG_TRIED;
0294
0295         /*
0296          * No need to mmap_read_unlock(mm) as we would
0297          * have already released it in __lock_page_or_retry
0298          * in mm/filemap.c.
0299          */
0300         goto retry;
0301     }
0302
0303     mmap_read_unlock(mm);
0304
0305     if (unlikely(fault & VM_FAULT_ERROR)) {
0306         mm_fault_error(regs, addr, fault);
0307         return;
0308     }
0309     return;
0310 }