Skip to contentSkip to conversation

x86_64 GPF handler (was: [PATCH] remove errornous semicolon)

DISCUSSION Chuck Ebbert 3 participants 3 messages Nov 24, 2004

lore
Share & tools
lore
#0 lore raw
Quoted 11 lines · Jesper Juhl

arch/i386/kernel/traps.c: In function `do_general_protection':
arch/i386/kernel/traps.c:506: warning: empty body in an if-statement

upon inspecting the code I see what looks like a mistakenly placed ";"

        if (!fixup_exception(regs)) {
                if (notify_die(DIE_GPF, "general protection fault", regs,
                                error_code, 13, SIGSEGV) == NOTIFY_STOP);
                        return;
                die("general protection fault", regs, error_code);
        }

Ouch. No matter what the notifier chain returns it will be treated
as if it returned NOTIFY_STOP, and no kernel-mode GPF will ever reach the die().

This bug was introduced 31 Aug 04 by [email protected] during a
kprobes update. The comments say it was ported from x86_64, so I had a look:

        /* kernel gp */
        {
                const struct exception_table_entry *fixup;
                fixup = search_exception_tables(regs->rip);
                if (fixup) {
                        regs->rip = fixup->fixup;
                        return;
                }
                notify_die(DIE_GPF, "general protection fault", regs, error_code,
                           13, SIGSEGV); 
                die("general protection fault", regs, error_code);
        }

x86_64 never checks the result of notify_die() and unconditionally does a die(). I don't know if this is a bug or not...

Andi, if this is not a bug could you explain why not?

--Chuck Ebbert 24-Nov-04 00:23:50

#1 lore raw
Quoted 36 lines · Chuck Ebbert
Quoted 11 lines · Jesper Juhl

arch/i386/kernel/traps.c: In function `do_general_protection':
arch/i386/kernel/traps.c:506: warning: empty body in an if-statement

upon inspecting the code I see what looks like a mistakenly placed ";"

        if (!fixup_exception(regs)) {
                if (notify_die(DIE_GPF, "general protection fault", regs,
                                error_code, 13, SIGSEGV) == NOTIFY_STOP);
                        return;
                die("general protection fault", regs, error_code);
        }

Ouch. No matter what the notifier chain returns it will be treated
as if it returned NOTIFY_STOP, and no kernel-mode GPF will ever reach the die().

This bug was introduced 31 Aug 04 by [email protected] during a
kprobes update. The comments say it was ported from x86_64, so I had a look:

        /* kernel gp */
        {
                const struct exception_table_entry *fixup;
                fixup = search_exception_tables(regs->rip);
                if (fixup) {
                        regs->rip = fixup->fixup;
                        return;
                }
                notify_die(DIE_GPF, "general protection fault", regs, error_code,
                           13, SIGSEGV); 
                die("general protection fault", regs, error_code);
        }
Quoted 4 lines

x86_64 never checks the result of notify_die() and unconditionally does a die(). I don't know if this is a bug or not...

Andi, if this is not a bug could you explain why not?

It depends on what the debugger (or kprobes) wants. These checks are added based on their needs. Perhaps he didn't consider it necessary on x86-64. But why don't you ask Prasanna directly? (cc'ed)

-Andi

#2 lore raw
Quoted 9 lines
Quoted 4 lines

x86_64 never checks the result of notify_die() and unconditionally does a die(). I don't know if this is a bug or not...

Andi, if this is not a bug could you explain why not?

It depends on what the debugger (or kprobes) wants. These checks are added based on their needs. Perhaps he didn't consider it necessary on x86-64. But why don't you ask Prasanna directly? (cc'ed)

I agree with Andi that it depends on the specific debugger. On handling the general protection fault notification, kprobe handler returns NOTIFY_STOP. This check missed out in x86_64 kprobes patch. Below patch should fix this, please let me know if you have any issues.

Thanks
Prasanna

This patch adds the return value check for the exception notifiers at do_general_protection as pointed out by Chuck Ebbert.

---

 linux-2.6.10-rc2-prasanna/arch/x86_64/kernel/traps.c |    5 +++--
 1 files changed, 3 insertions(+), 2 deletions(-)
arch/x86_64/kernel/traps.c renamed from arch/x86_64/kernel/traps.c~notifier-fix +3 -2
diff -puN arch/x86_64/kernel/traps.c~notifier-fix arch/x86_64/kernel/traps.c
@@ -556,8 +556,9 @@ asmlinkage void do_general_protection(st
 			regs->rip = fixup->fixup;
 			return;
 		}
-		notify_die(DIE_GPF, "general protection fault", regs, error_code,
-			   13, SIGSEGV); 
+		if (notify_die(DIE_GPF, "general protection fault", regs,
+					error_code, 13, SIGSEGV) == NOTIFY_STOP)
+			return;
 		die("general protection fault", regs, error_code);
 	}
 }

_

Signature · 5 lines
Prasanna S Panchamukhi
Linux Technology Center
India Software Labs, IBM Bangalore
Ph: 91-80-25044636
<[email protected]>