1 // SPDX-License-Identifier: GPL-2.0-or-later
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
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>
17 #include <asm/pgalloc.h>
18 #include <asm/ptrace.h>
19 #include <asm/tlbflush.h>
21 #include "../kernel/head.h"
24 * This routine handles page faults. It determines the address and the
25 * problem, and then passes it off to one of the appropriate routines.
27 asmlinkage void do_page_fault(struct pt_regs *regs)
29 struct task_struct *tsk;
30 struct vm_area_struct *vma;
32 unsigned long addr, cause;
33 unsigned int flags = FAULT_FLAG_DEFAULT;
34 int code = SEGV_MAPERR;
44 * Fault-in kernel-space virtual memory on-demand.
45 * The 'reference' page table is init_mm.pgd.
47 * NOTE! We MUST NOT take any locks for this case. We may
48 * be in an interrupt or a critical region, and should
49 * only copy the information from the master page table,
52 if (unlikely((addr >= VMALLOC_START) && (addr <= VMALLOC_END)))
55 /* Enable interrupts if they were enabled in the parent context. */
56 if (likely(regs->status & SR_PIE))
60 * If we're in an interrupt, have no user context, or are running
61 * in an atomic region, then we must not take the fault.
63 if (unlikely(faulthandler_disabled() || !mm))
67 flags |= FAULT_FLAG_USER;
69 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
72 down_read(&mm->mmap_sem);
73 vma = find_vma(mm, addr);
76 if (likely(vma->vm_start <= addr))
78 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN)))
80 if (unlikely(expand_stack(vma, addr)))
84 * Ok, we have a good vm_area for this memory access, so
91 case EXC_INST_PAGE_FAULT:
92 if (!(vma->vm_flags & VM_EXEC))
95 case EXC_LOAD_PAGE_FAULT:
96 if (!(vma->vm_flags & VM_READ))
99 case EXC_STORE_PAGE_FAULT:
100 if (!(vma->vm_flags & VM_WRITE))
102 flags |= FAULT_FLAG_WRITE;
105 panic("%s: unhandled cause %lu", __func__, cause);
109 * If for any reason at all we could not handle the fault,
110 * make sure we exit gracefully rather than endlessly redo
113 fault = handle_mm_fault(vma, addr, flags);
116 * If we need to retry but a fatal signal is pending, handle the
117 * signal first. We do not need to release the mmap_sem because it
118 * would already be released in __lock_page_or_retry in mm/filemap.c.
120 if (fault_signal_pending(fault, regs))
123 if (unlikely(fault & VM_FAULT_ERROR)) {
124 if (fault & VM_FAULT_OOM)
126 else if (fault & VM_FAULT_SIGBUS)
132 * Major/minor page fault accounting is only done on the
133 * initial attempt. If we go through a retry, it is extremely
134 * likely that the page will be found in page cache at that point.
136 if (flags & FAULT_FLAG_ALLOW_RETRY) {
137 if (fault & VM_FAULT_MAJOR) {
139 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ,
143 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN,
146 if (fault & VM_FAULT_RETRY) {
147 flags |= FAULT_FLAG_TRIED;
150 * No need to up_read(&mm->mmap_sem) as we would
151 * have already released it in __lock_page_or_retry
158 up_read(&mm->mmap_sem);
162 * Something tried to access memory that isn't in our memory map.
163 * Fix it, but check if it's kernel or user first.
166 up_read(&mm->mmap_sem);
167 /* User mode accesses just cause a SIGSEGV */
168 if (user_mode(regs)) {
169 do_trap(regs, SIGSEGV, code, addr);
174 /* Are we prepared to handle this kernel fault? */
175 if (fixup_exception(regs))
179 * Oops. The kernel tried to access some bad page. We'll have to
180 * terminate things with extreme prejudice.
183 pr_alert("Unable to handle kernel %s at virtual address " REG_FMT "\n",
184 (addr < PAGE_SIZE) ? "NULL pointer dereference" :
185 "paging request", addr);
190 * We ran out of memory, call the OOM killer, and return the userspace
191 * (which will retry the fault, or kill us if we got oom-killed).
194 up_read(&mm->mmap_sem);
195 if (!user_mode(regs))
197 pagefault_out_of_memory();
201 up_read(&mm->mmap_sem);
202 /* Kernel mode? Handle exceptions or die */
203 if (!user_mode(regs))
205 do_trap(regs, SIGBUS, BUS_ADRERR, addr);
217 /* User mode accesses just cause a SIGSEGV */
219 return do_trap(regs, SIGSEGV, code, addr);
222 * Synchronize this task's top level page-table
223 * with the 'reference' page table.
225 * Do _not_ use "tsk->active_mm->pgd" here.
226 * We might be inside an interrupt in the middle
229 index = pgd_index(addr);
230 pgd = (pgd_t *)pfn_to_virt(csr_read(CSR_SATP)) + index;
231 pgd_k = init_mm.pgd + index;
233 if (!pgd_present(*pgd_k))
235 set_pgd(pgd, *pgd_k);
237 p4d = p4d_offset(pgd, addr);
238 p4d_k = p4d_offset(pgd_k, addr);
239 if (!p4d_present(*p4d_k))
242 pud = pud_offset(p4d, addr);
243 pud_k = pud_offset(p4d_k, addr);
244 if (!pud_present(*pud_k))
248 * Since the vmalloc area is global, it is unnecessary
249 * to copy individual PTEs
251 pmd = pmd_offset(pud, addr);
252 pmd_k = pmd_offset(pud_k, addr);
253 if (!pmd_present(*pmd_k))
255 set_pmd(pmd, *pmd_k);
258 * Make sure the actual PTE exists as well to
259 * catch kernel vmalloc-area accesses to non-mapped
260 * addresses. If we don't do this, this will just
261 * silently loop forever.
263 pte_k = pte_offset_kernel(pmd_k, addr);
264 if (!pte_present(*pte_k))
268 * The kernel assumes that TLBs don't cache invalid
269 * entries, but in RISC-V, SFENCE.VMA specifies an
270 * ordering constraint, not a cache flush; it is
271 * necessary even after writing invalid entries.
273 local_flush_tlb_page(addr);