This commit was manufactured by cvs2svn to create branch 'vserver'.
[linux-2.6.git] / arch / um / kernel / trap.c
diff --git a/arch/um/kernel/trap.c b/arch/um/kernel/trap.c
new file mode 100644 (file)
index 0000000..8c90a63
--- /dev/null
@@ -0,0 +1,250 @@
+/*
+ * Copyright (C) 2000, 2001 Jeff Dike (jdike@karaya.com)
+ * Licensed under the GPL
+ */
+
+#include "linux/kernel.h"
+#include "asm/errno.h"
+#include "linux/sched.h"
+#include "linux/mm.h"
+#include "linux/spinlock.h"
+#include "linux/init.h"
+#include "linux/ptrace.h"
+#include "asm/semaphore.h"
+#include "asm/pgtable.h"
+#include "asm/pgalloc.h"
+#include "asm/tlbflush.h"
+#include "asm/a.out.h"
+#include "asm/current.h"
+#include "asm/irq.h"
+#include "sysdep/sigcontext.h"
+#include "user_util.h"
+#include "kern_util.h"
+#include "kern.h"
+#include "chan_kern.h"
+#include "mconsole_kern.h"
+#include "mem.h"
+#include "mem_kern.h"
+#include "sysdep/sigcontext.h"
+#include "sysdep/ptrace.h"
+#include "os.h"
+#ifdef CONFIG_MODE_SKAS
+#include "skas.h"
+#endif
+#include "os.h"
+
+/* Note this is constrained to return 0, -EFAULT, -EACCESS, -ENOMEM by segv(). */
+int handle_page_fault(unsigned long address, unsigned long ip,
+                     int is_write, int is_user, int *code_out)
+{
+       struct mm_struct *mm = current->mm;
+       struct vm_area_struct *vma;
+       pgd_t *pgd;
+       pud_t *pud;
+       pmd_t *pmd;
+       pte_t *pte;
+       int err = -EFAULT;
+
+       *code_out = SEGV_MAPERR;
+
+       /* If the fault was during atomic operation, don't take the fault, just
+        * fail. */
+       if (in_atomic())
+               goto out_nosemaphore;
+
+       down_read(&mm->mmap_sem);
+       vma = find_vma(mm, address);
+       if(!vma)
+               goto out;
+       else if(vma->vm_start <= address)
+               goto good_area;
+       else if(!(vma->vm_flags & VM_GROWSDOWN))
+               goto out;
+       else if(is_user && !ARCH_IS_STACKGROW(address))
+               goto out;
+       else if(expand_stack(vma, address))
+               goto out;
+
+good_area:
+       *code_out = SEGV_ACCERR;
+       if(is_write && !(vma->vm_flags & VM_WRITE))
+               goto out;
+
+       /* Don't require VM_READ|VM_EXEC for write faults! */
+        if(!is_write && !(vma->vm_flags & (VM_READ | VM_EXEC)))
+                goto out;
+
+       do {
+survive:
+               switch (handle_mm_fault(mm, vma, address, is_write)){
+               case VM_FAULT_MINOR:
+                       current->min_flt++;
+                       break;
+               case VM_FAULT_MAJOR:
+                       current->maj_flt++;
+                       break;
+               case VM_FAULT_SIGBUS:
+                       err = -EACCES;
+                       goto out;
+               case VM_FAULT_OOM:
+                       err = -ENOMEM;
+                       goto out_of_memory;
+               default:
+                       BUG();
+               }
+               pgd = pgd_offset(mm, address);
+               pud = pud_offset(pgd, address);
+               pmd = pmd_offset(pud, address);
+               pte = pte_offset_kernel(pmd, address);
+       } while(!pte_present(*pte));
+       err = 0;
+       /* The below warning was added in place of
+        *      pte_mkyoung(); if (is_write) pte_mkdirty();
+        * If it's triggered, we'd see normally a hang here (a clean pte is
+        * marked read-only to emulate the dirty bit).
+        * However, the generic code can mark a PTE writable but clean on a
+        * concurrent read fault, triggering this harmlessly. So comment it out.
+        */
+#if 0
+       WARN_ON(!pte_young(*pte) || (is_write && !pte_dirty(*pte)));
+#endif
+       flush_tlb_page(vma, address);
+out:
+       up_read(&mm->mmap_sem);
+out_nosemaphore:
+       return(err);
+
+/*
+ * We ran out of memory, or some other thing happened to us that made
+ * us unable to handle the page fault gracefully.
+ */
+out_of_memory:
+       if (current->pid == 1) {
+               up_read(&mm->mmap_sem);
+               yield();
+               down_read(&mm->mmap_sem);
+               goto survive;
+       }
+       goto out;
+}
+
+void segv_handler(int sig, union uml_pt_regs *regs)
+{
+       struct faultinfo * fi = UPT_FAULTINFO(regs);
+
+       if(UPT_IS_USER(regs) && !SEGV_IS_FIXABLE(fi)){
+               bad_segv(*fi, UPT_IP(regs));
+               return;
+       }
+       segv(*fi, UPT_IP(regs), UPT_IS_USER(regs), regs);
+}
+
+struct kern_handlers handlinfo_kern = {
+       .relay_signal = relay_signal,
+       .winch = winch,
+       .bus_handler = relay_signal,
+       .page_fault = segv_handler,
+       .sigio_handler = sigio_handler,
+       .timer_handler = timer_handler
+};
+/*
+ * We give a *copy* of the faultinfo in the regs to segv.
+ * This must be done, since nesting SEGVs could overwrite
+ * the info in the regs. A pointer to the info then would
+ * give us bad data!
+ */
+unsigned long segv(struct faultinfo fi, unsigned long ip, int is_user, void *sc)
+{
+       struct siginfo si;
+       void *catcher;
+       int err;
+        int is_write = FAULT_WRITE(fi);
+        unsigned long address = FAULT_ADDRESS(fi);
+
+        if(!is_user && (address >= start_vm) && (address < end_vm)){
+                flush_tlb_kernel_vm();
+                return(0);
+        }
+       else if(current->mm == NULL)
+               panic("Segfault with no mm");
+
+       if (SEGV_IS_FIXABLE(&fi) || SEGV_MAYBE_FIXABLE(&fi))
+               err = handle_page_fault(address, ip, is_write, is_user, &si.si_code);
+       else {
+               err = -EFAULT;
+               /* A thread accessed NULL, we get a fault, but CR2 is invalid.
+                * This code is used in __do_copy_from_user() of TT mode. */
+               address = 0;
+       }
+
+       catcher = current->thread.fault_catcher;
+       if(!err)
+               return(0);
+       else if(catcher != NULL){
+               current->thread.fault_addr = (void *) address;
+               do_longjmp(catcher, 1);
+       }
+       else if(current->thread.fault_addr != NULL)
+               panic("fault_addr set but no fault catcher");
+        else if(!is_user && arch_fixup(ip, sc))
+               return(0);
+
+       if(!is_user)
+               panic("Kernel mode fault at addr 0x%lx, ip 0x%lx",
+                     address, ip);
+
+       if (err == -EACCES) {
+               si.si_signo = SIGBUS;
+               si.si_errno = 0;
+               si.si_code = BUS_ADRERR;
+               si.si_addr = (void __user *)address;
+                current->thread.arch.faultinfo = fi;
+               force_sig_info(SIGBUS, &si, current);
+       } else if (err == -ENOMEM) {
+               printk("VM: killing process %s\n", current->comm);
+               do_exit(SIGKILL);
+       } else {
+               BUG_ON(err != -EFAULT);
+               si.si_signo = SIGSEGV;
+               si.si_addr = (void __user *) address;
+                current->thread.arch.faultinfo = fi;
+               force_sig_info(SIGSEGV, &si, current);
+       }
+       return(0);
+}
+
+void bad_segv(struct faultinfo fi, unsigned long ip)
+{
+       struct siginfo si;
+
+       si.si_signo = SIGSEGV;
+       si.si_code = SEGV_ACCERR;
+       si.si_addr = (void __user *) FAULT_ADDRESS(fi);
+       current->thread.arch.faultinfo = fi;
+       force_sig_info(SIGSEGV, &si, current);
+}
+
+void relay_signal(int sig, union uml_pt_regs *regs)
+{
+       if(arch_handle_signal(sig, regs)) return;
+       if(!UPT_IS_USER(regs))
+               panic("Kernel mode signal %d", sig);
+        current->thread.arch.faultinfo = *UPT_FAULTINFO(regs);
+       force_sig(sig, current);
+}
+
+void bus_handler(int sig, union uml_pt_regs *regs)
+{
+       if(current->thread.fault_catcher != NULL)
+               do_longjmp(current->thread.fault_catcher, 1);
+       else relay_signal(sig, regs);
+}
+
+void winch(int sig, union uml_pt_regs *regs)
+{
+       do_IRQ(WINCH_IRQ, regs);
+}
+
+void trap_init(void)
+{
+}