arch/riscv/mm/fault.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * Copyright (C) 2009 Sunplus Core Technology Co., Ltd.
   4  *  Lennox Wu <lennox.wu@sunplusct.com>
   5  *  Chen Liqin <liqin.chen@sunplusct.com>
   6  * Copyright (C) 2012 Regents of the University of California
   7  */
   8
   9
  10 #include <linux/mm.h>
  11 #include <linux/kernel.h>
  12 #include <linux/interrupt.h>
  13 #include <linux/perf_event.h>
  14 #include <linux/signal.h>
  15 #include <linux/uaccess.h>
  16 #include <linux/kprobes.h>
  17 #include <linux/kfence.h>
  18 #include <linux/entry-common.h>
  19
  20 #include <asm/ptrace.h>
  21 #include <asm/tlbflush.h>
  22
  23 #include "../kernel/head.h"
  24
  25 static void die_kernel_fault(const char *msg, unsigned long addr,
  26                 struct pt_regs *regs)
  27 {
  28         bust_spinlocks(1);
  29
  30         pr_alert("Unable to handle kernel %s at virtual address " REG_FMT "\n", msg,
  31                 addr);
  32
  33         bust_spinlocks(0);
  34         die(regs, "Oops");
  35         make_task_dead(SIGKILL);
  36 }
  37
  38 static inline void no_context(struct pt_regs *regs, unsigned long addr)
  39 {
  40         const char *msg;
  41
  42         /* Are we prepared to handle this kernel fault? */
  43         if (fixup_exception(regs))
  44                 return;
  45
  46         /*
  47          * Oops. The kernel tried to access some bad page. We'll have to
  48          * terminate things with extreme prejudice.
  49          */
  50         if (addr < PAGE_SIZE)
  51                 msg = "NULL pointer dereference";
  52         else {
  53                 if (kfence_handle_page_fault(addr, regs->cause == EXC_STORE_PAGE_FAULT, regs))
  54                         return;
  55
  56                 msg = "paging request";
  57         }
  58
  59         die_kernel_fault(msg, addr, regs);
  60 }
  61
  62 static inline void mm_fault_error(struct pt_regs *regs, unsigned long addr, vm_fault_t fault)
  63 {
  64         if (!user_mode(regs)) {
  65                 no_context(regs, addr);
  66                 return;
  67         }
  68
  69         if (fault & VM_FAULT_OOM) {
  70                 /*
  71                  * We ran out of memory, call the OOM killer, and return the userspace
  72                  * (which will retry the fault, or kill us if we got oom-killed).
  73                  */
  74                 pagefault_out_of_memory();
  75                 return;
  76         } else if (fault & (VM_FAULT_SIGBUS | VM_FAULT_HWPOISON | VM_FAULT_HWPOISON_LARGE)) {
  77                 /* Kernel mode? Handle exceptions or die */
  78                 do_trap(regs, SIGBUS, BUS_ADRERR, addr);
  79                 return;
  80         } else if (fault & VM_FAULT_SIGSEGV) {
  81                 do_trap(regs, SIGSEGV, SEGV_MAPERR, addr);
  82                 return;
  83         }
  84
  85         BUG();
  86 }
  87
  88 static inline void
  89 bad_area_nosemaphore(struct pt_regs *regs, int code, unsigned long addr)
  90 {
  91         /*
  92          * Something tried to access memory that isn't in our memory map.
  93          * Fix it, but check if it's kernel or user first.
  94          */
  95         /* User mode accesses just cause a SIGSEGV */
  96         if (user_mode(regs)) {
  97                 do_trap(regs, SIGSEGV, code, addr);
  98                 return;
  99         }
 100
 101         no_context(regs, addr);
 102 }
 103
 104 static inline void
 105 bad_area(struct pt_regs *regs, struct mm_struct *mm, int code,
 106          unsigned long addr)
 107 {
 108         mmap_read_unlock(mm);
 109
 110         bad_area_nosemaphore(regs, code, addr);
 111 }
 112
 113 static inline void vmalloc_fault(struct pt_regs *regs, int code, unsigned long addr)
 114 {
 115         pgd_t *pgd, *pgd_k;
 116         pud_t *pud_k;
 117         p4d_t *p4d_k;
 118         pmd_t *pmd_k;
 119         pte_t *pte_k;
 120         int index;
 121         unsigned long pfn;
 122
 123         /* User mode accesses just cause a SIGSEGV */
 124         if (user_mode(regs))
 125                 return do_trap(regs, SIGSEGV, code, addr);
 126
 127         /*
 128          * Synchronize this task's top level page-table
 129          * with the 'reference' page table.
 130          *
 131          * Do _not_ use "tsk->active_mm->pgd" here.
 132          * We might be inside an interrupt in the middle
 133          * of a task switch.
 134          */
 135         index = pgd_index(addr);
 136         pfn = csr_read(CSR_SATP) & SATP_PPN;
 137         pgd = (pgd_t *)pfn_to_virt(pfn) + index;
 138         pgd_k = init_mm.pgd + index;
 139
 140         if (!pgd_present(pgdp_get(pgd_k))) {
 141                 no_context(regs, addr);
 142                 return;
 143         }
 144         set_pgd(pgd, pgdp_get(pgd_k));
 145
 146         p4d_k = p4d_offset(pgd_k, addr);
 147         if (!p4d_present(p4dp_get(p4d_k))) {
 148                 no_context(regs, addr);
 149                 return;
 150         }
 151
 152         pud_k = pud_offset(p4d_k, addr);
 153         if (!pud_present(pudp_get(pud_k))) {
 154                 no_context(regs, addr);
 155                 return;
 156         }
 157         if (pud_leaf(pudp_get(pud_k)))
 158                 goto flush_tlb;
 159
 160         /*
 161          * Since the vmalloc area is global, it is unnecessary
 162          * to copy individual PTEs
 163          */
 164         pmd_k = pmd_offset(pud_k, addr);
 165         if (!pmd_present(pmdp_get(pmd_k))) {
 166                 no_context(regs, addr);
 167                 return;
 168         }
 169         if (pmd_leaf(pmdp_get(pmd_k)))
 170                 goto flush_tlb;
 171
 172         /*
 173          * Make sure the actual PTE exists as well to
 174          * catch kernel vmalloc-area accesses to non-mapped
 175          * addresses. If we don't do this, this will just
 176          * silently loop forever.
 177          */
 178         pte_k = pte_offset_kernel(pmd_k, addr);
 179         if (!pte_present(ptep_get(pte_k))) {
 180                 no_context(regs, addr);
 181                 return;
 182         }
 183
 184         /*
 185          * The kernel assumes that TLBs don't cache invalid
 186          * entries, but in RISC-V, SFENCE.VMA specifies an
 187          * ordering constraint, not a cache flush; it is
 188          * necessary even after writing invalid entries.
 189          */
 190 flush_tlb:
 191         local_flush_tlb_page(addr);
 192 }
 193
 194 static inline bool access_error(unsigned long cause, struct vm_area_struct *vma)
 195 {
 196         switch (cause) {
 197         case EXC_INST_PAGE_FAULT:
 198                 if (!(vma->vm_flags & VM_EXEC)) {
 199                         return true;
 200                 }
 201                 break;
 202         case EXC_LOAD_PAGE_FAULT:
 203                 /* Write implies read */
 204                 if (!(vma->vm_flags & (VM_READ | VM_WRITE))) {
 205                         return true;
 206                 }
 207                 break;
 208         case EXC_STORE_PAGE_FAULT:
 209                 if (!(vma->vm_flags & VM_WRITE)) {
 210                         return true;
 211                 }
 212                 break;
 213         default:
 214                 panic("%s: unhandled cause %lu", __func__, cause);
 215         }
 216         return false;
 217 }
 218
 219 /*
 220  * This routine handles page faults.  It determines the address and the
 221  * problem, and then passes it off to one of the appropriate routines.
 222  */
 223 void handle_page_fault(struct pt_regs *regs)
 224 {
 225         struct task_struct *tsk;
 226         struct vm_area_struct *vma;
 227         struct mm_struct *mm;
 228         unsigned long addr, cause;
 229         unsigned int flags = FAULT_FLAG_DEFAULT;
 230         int code = SEGV_MAPERR;
 231         vm_fault_t fault;
 232
 233         cause = regs->cause;
 234         addr = regs->badaddr;
 235
 236         tsk = current;
 237         mm = tsk->mm;
 238
 239         if (kprobe_page_fault(regs, cause))
 240                 return;
 241
 242         /*
 243          * Fault-in kernel-space virtual memory on-demand.
 244          * The 'reference' page table is init_mm.pgd.
 245          *
 246          * NOTE! We MUST NOT take any locks for this case. We may
 247          * be in an interrupt or a critical region, and should
 248          * only copy the information from the master page table,
 249          * nothing more.
 250          */
 251         if ((!IS_ENABLED(CONFIG_MMU) || !IS_ENABLED(CONFIG_64BIT)) &&
 252             unlikely(addr >= VMALLOC_START && addr < VMALLOC_END)) {
 253                 vmalloc_fault(regs, code, addr);
 254                 return;
 255         }
 256
 257         /* Enable interrupts if they were enabled in the parent context. */
 258         if (!regs_irqs_disabled(regs))
 259                 local_irq_enable();
 260
 261         /*
 262          * If we're in an interrupt, have no user context, or are running
 263          * in an atomic region, then we must not take the fault.
 264          */
 265         if (unlikely(faulthandler_disabled() || !mm)) {
 266                 tsk->thread.bad_cause = cause;
 267                 no_context(regs, addr);
 268                 return;
 269         }
 270
 271         if (user_mode(regs))
 272                 flags |= FAULT_FLAG_USER;
 273
 274         if (!user_mode(regs) && addr < TASK_SIZE && unlikely(!(regs->status & SR_SUM))) {
 275                 if (fixup_exception(regs))
 276                         return;
 277
 278                 die_kernel_fault("access to user memory without uaccess routines", addr, regs);
 279         }
 280
 281         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
 282
 283         if (cause == EXC_STORE_PAGE_FAULT)
 284                 flags |= FAULT_FLAG_WRITE;
 285         else if (cause == EXC_INST_PAGE_FAULT)
 286                 flags |= FAULT_FLAG_INSTRUCTION;
 287         if (!(flags & FAULT_FLAG_USER))
 288                 goto lock_mmap;
 289
 290         vma = lock_vma_under_rcu(mm, addr);
 291         if (!vma)
 292                 goto lock_mmap;
 293
 294         if (unlikely(access_error(cause, vma))) {
 295                 vma_end_read(vma);
 296                 count_vm_vma_lock_event(VMA_LOCK_SUCCESS);
 297                 tsk->thread.bad_cause = cause;
 298                 bad_area_nosemaphore(regs, SEGV_ACCERR, addr);
 299                 return;
 300         }
 301
 302         fault = handle_mm_fault(vma, addr, flags | FAULT_FLAG_VMA_LOCK, regs);
 303         if (!(fault & (VM_FAULT_RETRY | VM_FAULT_COMPLETED)))
 304                 vma_end_read(vma);
 305
 306         if (!(fault & VM_FAULT_RETRY)) {
 307                 count_vm_vma_lock_event(VMA_LOCK_SUCCESS);
 308                 goto done;
 309         }
 310         count_vm_vma_lock_event(VMA_LOCK_RETRY);
 311         if (fault & VM_FAULT_MAJOR)
 312                 flags |= FAULT_FLAG_TRIED;
 313
 314         if (fault_signal_pending(fault, regs)) {
 315                 if (!user_mode(regs))
 316                         no_context(regs, addr);
 317                 return;
 318         }
 319 lock_mmap:
 320
 321 retry:
 322         vma = lock_mm_and_find_vma(mm, addr, regs);
 323         if (unlikely(!vma)) {
 324                 tsk->thread.bad_cause = cause;
 325                 bad_area_nosemaphore(regs, code, addr);
 326                 return;
 327         }
 328
 329         /*
 330          * Ok, we have a good vm_area for this memory access, so
 331          * we can handle it.
 332          */
 333         code = SEGV_ACCERR;
 334
 335         if (unlikely(access_error(cause, vma))) {
 336                 tsk->thread.bad_cause = cause;
 337                 bad_area(regs, mm, code, addr);
 338                 return;
 339         }
 340
 341         /*
 342          * If for any reason at all we could not handle the fault,
 343          * make sure we exit gracefully rather than endlessly redo
 344          * the fault.
 345          */
 346         fault = handle_mm_fault(vma, addr, flags, regs);
 347
 348         /*
 349          * If we need to retry but a fatal signal is pending, handle the
 350          * signal first. We do not need to release the mmap_lock because it
 351          * would already be released in __lock_page_or_retry in mm/filemap.c.
 352          */
 353         if (fault_signal_pending(fault, regs)) {
 354                 if (!user_mode(regs))
 355                         no_context(regs, addr);
 356                 return;
 357         }
 358
 359         /* The fault is fully completed (including releasing mmap lock) */
 360         if (fault & VM_FAULT_COMPLETED)
 361                 return;
 362
 363         if (unlikely(fault & VM_FAULT_RETRY)) {
 364                 flags |= FAULT_FLAG_TRIED;
 365
 366                 /*
 367                  * No need to mmap_read_unlock(mm) as we would
 368                  * have already released it in __lock_page_or_retry
 369                  * in mm/filemap.c.
 370                  */
 371                 goto retry;
 372         }
 373
 374         mmap_read_unlock(mm);
 375
 376 done:
 377         if (unlikely(fault & VM_FAULT_ERROR)) {
 378                 tsk->thread.bad_cause = cause;
 379                 mm_fault_error(regs, addr, fault);
 380                 return;
 381         }
 382         return;
 383 }