Skip to contentSkip to conversation

trap exception handling in linux kernel

OpenRISC Vad Rulezz 2 participants 5 messages Jan 6, 2019 – Mar 4, 2019

lore
Share & tools
lore

Vad Rulezz

author
#0 lore raw

Hello

In arch/openrisc/kernel/traps.c we have a function:

asmlinkage void do_trap(struct pt_regs *regs, unsigned long address)

{
	force_sig_fault(SIGTRAP, TRAP_TRACE, (void __user *)address, current);

	regs->pc += 4;
}

Is there a reason for regs->pc increment?

I think typically debuggers wants to resume program execution from the same address which caused a trap (after disabling a breakpoint, and it's a sw breakpoint after replacing an instruction at this address).

Vad Rulezz

#1 lore raw

Hello,

Did you ever get a response for this? I believe this is to allow jumping over the openrisc l.trap instruction. But, as you say GDB/debuggers will replace that instruction with the real instruction, so it shouldn't be needed.

Have you noticed an issue or are you just auditing this? We don't
usually run GDB from within openrisc linux, I haven't run into this issue before.

-Stafford

Quoted 23 lines · Vad Rulezz

Hello

In arch/openrisc/kernel/traps.c we have a function:

asmlinkage void do_trap(struct pt_regs *regs, unsigned long address)

{
        force_sig_fault(SIGTRAP, TRAP_TRACE, (void __user *)address, current);

        regs->pc += 4;
}

Is there a reason for regs->pc increment?

I think typically debuggers wants to resume program execution from the same address which caused a trap (after disabling a breakpoint, and it's a sw breakpoint after replacing an instruction at this address).

Vad Rulezz
_______________________________________________
OpenRISC mailing list
OpenRISC at lists.librecores.org
https://lists.librecores.org/listinfo/openrisc

author
#2 lore raw

I have actually discovered it in a native debugger, ofc it was not hard to remove the line in my copy of kernel and everything ran well after that, but I still decided to ask because I thought that maybe there is a software that depends on existing behavior.

As noone except you replied I think it's safe to remove the increment in the mainline. It likely will simplify things for native GDB too.

Vad Rulezz

Quoted 36 lines · Stafford Horne

Did you ever get a response for this? I believe this is to allow jumping over the openrisc l.trap instruction. But, as you say GDB/debuggers will replace that instruction with the real instruction, so it shouldn't be needed.

Have you noticed an issue or are you just auditing this? We don't
usually run GDB from within openrisc linux, I haven't run into this issue before.

-Stafford

Quoted 23 lines · Vad Rulezz

Hello

In arch/openrisc/kernel/traps.c we have a function:

asmlinkage void do_trap(struct pt_regs *regs, unsigned long address)

{
        force_sig_fault(SIGTRAP, TRAP_TRACE, (void __user *)address, current);

        regs->pc += 4;
}

Is there a reason for regs->pc increment?

I think typically debuggers wants to resume program execution from the same address which caused a trap (after disabling a breakpoint, and it's a sw breakpoint after replacing an instruction at this address).

Vad Rulezz
_______________________________________________
OpenRISC mailing list
OpenRISC at lists.librecores.org
https://lists.librecores.org/listinfo/openrisc

#3 lore raw

Hello,

That sounds good. I'll create a patch.

Btw were you using my gdb Linux support patches?

-Stafford

Quoted 52 lines · Vad Rulezz

I have actually discovered it in a native debugger, ofc it was not hard to remove the line in my copy of kernel and everything ran well after that, but I still decided to ask because I thought that maybe there is a software that depends on existing behavior.

As noone except you replied I think it's safe to remove the increment in the mainline. It likely will simplify things for native GDB too.

Vad Rulezz

Quoted 5 lines · Stafford Horne

Did you ever get a response for this? I believe this is to allow jumping over the openrisc l.trap instruction. But, as you say GDB/debuggers will replace that instruction with the real instruction, so it shouldn't be needed.

Quoted 15 lines

Have you noticed an issue or are you just auditing this? We don't
usually run GDB from within openrisc linux, I haven't run into this issue before.

-Stafford

Quoted 8 lines · Vad Rulezz

Hello

In arch/openrisc/kernel/traps.c we have a function:

asmlinkage void do_trap(struct pt_regs *regs, unsigned long address) {
force_sig_fault(SIGTRAP, TRAP_TRACE, (void __user *)address,

current);

Quoted 7 lines
Quoted 7 lines
        regs->pc += 4;
}

Is there a reason for regs->pc increment?

I think typically debuggers wants to resume program execution from the

same address which caused a trap (after disabling a breakpoint, and it's a sw breakpoint after replacing an instruction at this address).

Quoted 9 lines
Quoted 8 lines

Vad Rulezz
_______________________________________________
OpenRISC mailing list
OpenRISC at lists.librecores.org
https://lists.librecores.org/listinfo/openrisc

Quoted 5 lines

_______________________________________________
OpenRISC mailing list
OpenRISC at lists.librecores.org
https://lists.librecores.org/listinfo/openrisc

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.librecores.org/pipermail/openrisc/attachments/20190304/e12cca70/attachment.html>

author
#4 lore raw
Quoted 1 line

Btw were you using my gdb Linux support patches?

Actually no.
But maybe I'll give them a try.

Vad Rulezz

Quoted 61 lines · Stafford Horne

Hello,

That sounds good. I'll create a patch.

Btw were you using my gdb Linux support patches?

-Stafford

On Mon, Mar 4, 2019, 4:38 AM Vad Rulezz <vad at vr5.epicgamer.org <mailto:[email protected]>> wrote:

    I have actually discovered it in a native debugger, ofc it was not hard to remove the line in my copy of kernel and everything ran well after that,
    but I still decided to ask because I thought that maybe there is a software that depends on existing behavior.

    As noone except you replied I think it's safe to remove the increment in the mainline. It likely will simplify things for native GDB too.


    Vad Rulezz

    On 03.03.2019 18:29, Stafford Horne wrote:
    > Did you ever get a response for this?  I believe this is to allow
    > jumping over the openrisc l.trap instruction.  But, as you say
    > GDB/debuggers will replace that instruction with the real instruction,
    > so it shouldn't be needed.
    >
    > Have you noticed an issue or are you just auditing this?   We don't
    > usually run GDB from within openrisc linux, I haven't run into this
    > issue before.
    >
    > -Stafford
    >
    > On Mon, Jan 7, 2019 at 3:07 AM Vad Rulezz <vad at vr5.epicgamer.org <mailto:[email protected]>> wrote:
    >>
    >> Hello
    >>
    >> In arch/openrisc/kernel/traps.c we have a function:
    >>
    >> asmlinkage void do_trap(struct pt_regs *regs, unsigned long address)
    >> {
    >>         force_sig_fault(SIGTRAP, TRAP_TRACE, (void __user *)address, current);
    >>
    >>         regs->pc += 4;
    >> }
    >>
    >> Is there a reason for regs->pc increment?
    >>
    >> I think typically debuggers wants to resume program execution from the same address which caused a trap (after disabling a breakpoint, and it's a sw breakpoint after replacing an instruction at this address).
    >>
    >>
    >>
    >> Vad Rulezz
    >> _______________________________________________
    >> OpenRISC mailing list
    >> OpenRISC at lists.librecores.org <mailto:[email protected]>
    >> https://lists.librecores.org/listinfo/openrisc
    >

    _______________________________________________
    OpenRISC mailing list
    OpenRISC at lists.librecores.org <mailto:[email protected]>
    https://lists.librecores.org/listinfo/openrisc