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>
16 #include <linux/kprobes.h>
17 #include <linux/kfence.h>
18 #include <linux/entry-common.h>
20 #include <asm/ptrace.h>
21 #include <asm/tlbflush.h>
23 #include "../kernel/head.h"
25 static void die_kernel_fault(const char *msg, unsigned long addr,
30 pr_alert("Unable to handle kernel %s at virtual address " REG_FMT "\n", msg,
35 make_task_dead(SIGKILL);
38 static inline void no_context(struct pt_regs *regs, unsigned long addr)
42 /* Are we prepared to handle this kernel fault? */
43 if (fixup_exception(regs))
47 * Oops. The kernel tried to access some bad page. We'll have to
48 * terminate things with extreme prejudice.
51 msg = "NULL pointer dereference";
53 if (kfence_handle_page_fault(addr, regs->cause == EXC_STORE_PAGE_FAULT, regs))
56 msg = "paging request";
59 die_kernel_fault(msg, addr, regs);
62 static inline void mm_fault_error(struct pt_regs *regs, unsigned long addr, vm_fault_t fault)
64 if (fault & VM_FAULT_OOM) {
66 * We ran out of memory, call the OOM killer, and return the userspace
67 * (which will retry the fault, or kill us if we got oom-killed).
69 if (!user_mode(regs)) {
70 no_context(regs, addr);
73 pagefault_out_of_memory();
75 } else if (fault & (VM_FAULT_SIGBUS | VM_FAULT_HWPOISON | VM_FAULT_HWPOISON_LARGE)) {
76 /* Kernel mode? Handle exceptions or die */
77 if (!user_mode(regs)) {
78 no_context(regs, addr);
81 do_trap(regs, SIGBUS, BUS_ADRERR, addr);
88 bad_area_nosemaphore(struct pt_regs *regs, int code, unsigned long addr)
91 * Something tried to access memory that isn't in our memory map.
92 * Fix it, but check if it's kernel or user first.
94 /* User mode accesses just cause a SIGSEGV */
95 if (user_mode(regs)) {
96 do_trap(regs, SIGSEGV, code, addr);
100 no_context(regs, addr);
104 bad_area(struct pt_regs *regs, struct mm_struct *mm, int code,
107 mmap_read_unlock(mm);
109 bad_area_nosemaphore(regs, code, addr);
112 static inline void vmalloc_fault(struct pt_regs *regs, int code, unsigned long addr)
122 /* User mode accesses just cause a SIGSEGV */
124 return do_trap(regs, SIGSEGV, code, addr);
127 * Synchronize this task's top level page-table
128 * with the 'reference' page table.
130 * Do _not_ use "tsk->active_mm->pgd" here.
131 * We might be inside an interrupt in the middle
134 index = pgd_index(addr);
135 pfn = csr_read(CSR_SATP) & SATP_PPN;
136 pgd = (pgd_t *)pfn_to_virt(pfn) + index;
137 pgd_k = init_mm.pgd + index;
139 if (!pgd_present(*pgd_k)) {
140 no_context(regs, addr);
143 set_pgd(pgd, *pgd_k);
145 p4d_k = p4d_offset(pgd_k, addr);
146 if (!p4d_present(*p4d_k)) {
147 no_context(regs, addr);
151 pud_k = pud_offset(p4d_k, addr);
152 if (!pud_present(*pud_k)) {
153 no_context(regs, addr);
156 if (pud_leaf(*pud_k))
160 * Since the vmalloc area is global, it is unnecessary
161 * to copy individual PTEs
163 pmd_k = pmd_offset(pud_k, addr);
164 if (!pmd_present(*pmd_k)) {
165 no_context(regs, addr);
168 if (pmd_leaf(*pmd_k))
172 * Make sure the actual PTE exists as well to
173 * catch kernel vmalloc-area accesses to non-mapped
174 * addresses. If we don't do this, this will just
175 * silently loop forever.
177 pte_k = pte_offset_kernel(pmd_k, addr);
178 if (!pte_present(*pte_k)) {
179 no_context(regs, addr);
184 * The kernel assumes that TLBs don't cache invalid
185 * entries, but in RISC-V, SFENCE.VMA specifies an
186 * ordering constraint, not a cache flush; it is
187 * necessary even after writing invalid entries.
190 local_flush_tlb_page(addr);
193 static inline bool access_error(unsigned long cause, struct vm_area_struct *vma)
196 case EXC_INST_PAGE_FAULT:
197 if (!(vma->vm_flags & VM_EXEC)) {
201 case EXC_LOAD_PAGE_FAULT:
202 /* Write implies read */
203 if (!(vma->vm_flags & (VM_READ | VM_WRITE))) {
207 case EXC_STORE_PAGE_FAULT:
208 if (!(vma->vm_flags & VM_WRITE)) {
213 panic("%s: unhandled cause %lu", __func__, cause);
219 * This routine handles page faults. It determines the address and the
220 * problem, and then passes it off to one of the appropriate routines.
222 void handle_page_fault(struct pt_regs *regs)
224 struct task_struct *tsk;
225 struct vm_area_struct *vma;
226 struct mm_struct *mm;
227 unsigned long addr, cause;
228 unsigned int flags = FAULT_FLAG_DEFAULT;
229 int code = SEGV_MAPERR;
233 addr = regs->badaddr;
238 if (kprobe_page_fault(regs, cause))
242 * Fault-in kernel-space virtual memory on-demand.
243 * The 'reference' page table is init_mm.pgd.
245 * NOTE! We MUST NOT take any locks for this case. We may
246 * be in an interrupt or a critical region, and should
247 * only copy the information from the master page table,
250 if ((!IS_ENABLED(CONFIG_MMU) || !IS_ENABLED(CONFIG_64BIT)) &&
251 unlikely(addr >= VMALLOC_START && addr < VMALLOC_END)) {
252 vmalloc_fault(regs, code, addr);
256 /* Enable interrupts if they were enabled in the parent context. */
257 if (!regs_irqs_disabled(regs))
261 * If we're in an interrupt, have no user context, or are running
262 * in an atomic region, then we must not take the fault.
264 if (unlikely(faulthandler_disabled() || !mm)) {
265 tsk->thread.bad_cause = cause;
266 no_context(regs, addr);
271 flags |= FAULT_FLAG_USER;
273 if (!user_mode(regs) && addr < TASK_SIZE && unlikely(!(regs->status & SR_SUM))) {
274 if (fixup_exception(regs))
277 die_kernel_fault("access to user memory without uaccess routines", addr, regs);
280 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
282 if (cause == EXC_STORE_PAGE_FAULT)
283 flags |= FAULT_FLAG_WRITE;
284 else if (cause == EXC_INST_PAGE_FAULT)
285 flags |= FAULT_FLAG_INSTRUCTION;
286 if (!(flags & FAULT_FLAG_USER))
289 vma = lock_vma_under_rcu(mm, addr);
293 if (unlikely(access_error(cause, vma))) {
298 fault = handle_mm_fault(vma, addr, flags | FAULT_FLAG_VMA_LOCK, regs);
299 if (!(fault & (VM_FAULT_RETRY | VM_FAULT_COMPLETED)))
302 if (!(fault & VM_FAULT_RETRY)) {
303 count_vm_vma_lock_event(VMA_LOCK_SUCCESS);
306 count_vm_vma_lock_event(VMA_LOCK_RETRY);
307 if (fault & VM_FAULT_MAJOR)
308 flags |= FAULT_FLAG_TRIED;
310 if (fault_signal_pending(fault, regs)) {
311 if (!user_mode(regs))
312 no_context(regs, addr);
318 vma = lock_mm_and_find_vma(mm, addr, regs);
319 if (unlikely(!vma)) {
320 tsk->thread.bad_cause = cause;
321 bad_area_nosemaphore(regs, code, addr);
326 * Ok, we have a good vm_area for this memory access, so
331 if (unlikely(access_error(cause, vma))) {
332 tsk->thread.bad_cause = cause;
333 bad_area(regs, mm, code, addr);
338 * If for any reason at all we could not handle the fault,
339 * make sure we exit gracefully rather than endlessly redo
342 fault = handle_mm_fault(vma, addr, flags, regs);
345 * If we need to retry but a fatal signal is pending, handle the
346 * signal first. We do not need to release the mmap_lock because it
347 * would already be released in __lock_page_or_retry in mm/filemap.c.
349 if (fault_signal_pending(fault, regs)) {
350 if (!user_mode(regs))
351 no_context(regs, addr);
355 /* The fault is fully completed (including releasing mmap lock) */
356 if (fault & VM_FAULT_COMPLETED)
359 if (unlikely(fault & VM_FAULT_RETRY)) {
360 flags |= FAULT_FLAG_TRIED;
363 * No need to mmap_read_unlock(mm) as we would
364 * have already released it in __lock_page_or_retry
370 mmap_read_unlock(mm);
373 if (unlikely(fault & VM_FAULT_ERROR)) {
374 tsk->thread.bad_cause = cause;
375 mm_fault_error(regs, addr, fault);