Re: [Qemu-discuss] Qemu: AARCH64: Single Step exception does not work

[David Long]

On 01/08/15 00:32, Pratyush Anand wrote:
> Hi All,
>
> Have anyone tried to test single step exception with ARM64 on Qemu? I
> was testing ARM64 uprobe patches[1] with qemu and I noticed that it does
> not generate single step exception. I also tried kprobe[2], which uses
> single step exception and it does not work. However, these code works
> fine with real silicon.
>
> Test case can be summarized as under:
>
> 1. After kernel code is executed, it programs ELR_EL1 with the address
> of instruction which is to be single stepped. Lets say 0x7ffffff004 is
> the address of instruction which is to be single stepped. So, ELR_EL1
> has been programmed with 0x7ffffff004.
>
> 2. MDSCR_EL1.SS is set to 1
>
> 3. ERET has been called to execute instruction to be single stepped.
>
> With Qemu, I always see
> undefined instruction: pc=0000007ffffff008
> Code: bad PC value
>
> It seems that Qemu could not notice MDSCR_EL1.SS = 1 and since, kernel
> had written a single valid instruction at location 0x7ffffff004, so it
> raised an undefined exception while executing next invalid instruction.
>
> My Qemu version is:
>
> QEMU emulator version 2.1.2, Copyright (c) 2003-2008 Fabrice Bellard
>
> You may use code in [3] to test single steping.
>
> Please let me know, if any more input is needed to reproduce it.
>
> ~Pratyush
>
> [1] https://lkml.org/lkml/2014/12/31/151
> [2] https://lkml.org/lkml/2014/11/18/33
> [3] https://github.com/pratyushanand/linux.git:ml_arm64_uprobe_devel_v2

The singlestep support in QEMU is relatively recent.  Make sure you're 
running a fairly recent QEMU.

At one point QEMU was not setting ELR_EL* properly.  I'll forward you an 
email from Peter Maydell that has some relevance.  This was fixed.

Note that you have to both set SS *and* be sure debug exceptions are 
enabled in order to get the single-step exception.  My kprobes patch [2] 
should work in that regard.  Today I plan to post a v4 version of that 
patch, but I think the only thing it fixes relative to single-stepping 
is very intermittent failures due to interrupts not being properly 
disabled (you have to disabke interrupts uring single-stepping or you 
could end up single-stepping an ingterrupt handler).

Note that testing under QEMU will not reveal SMP issues.

-dl