Control: tags -1 - unreproducible

Hi,

> 2026年1月24日 13:23,Tiezhu Yang <[email protected]> 写道:
> 
> On 2026/1/23 下午10:30, Miao Wang wrote:
>> On Wed, 21 Jan 2026 14:47:19 -0500 Sergio Durigan Junior 
>> <[email protected]> wrote:
>>> OK, then let's wait for the loongarch64 builders to be redeployed and
>>> see if that solves the issue.
>>> 
>> Hi, I managed to reproduce the error in qemu emulated virtual machine, which
>> lacks the support of hardware watch point.
> 
> Could you please check whether CONFIG_HAVE_HW_BREAKPOINT is set in your
> kernel config? It is usually set by default via the defconfig.

Apparently yes. The kernel I have used is from Debian. As I've
explained in my pervious reply, the reason why hardware watch points is
not supported in my environment is that qemu does not support this. However,
I guess that gdb fails to set hardware breakpoints might probably be the
failure reason for the buildds, but the root cause can't be the same, since
the kernel used by the buildds is shown in the logs and is also from
Debian ports, 6.17.12+deb14-loong64.

> IMO the problem is related with the Linux kernel config
> CONFIG_HAVE_HW_BREAKPOINT which is set by default in the
> newer kernel used by Debian official ISO:
> 
> https://cdimage.debian.org/cdimage/ports/snapshots/2025-12-06/debian-13.0.0-loong64-NETINST-1.iso
> 
> So in order to avoid the problem, one way is to redeploy the above ISO
> for the builder, the other way is to modify the GDB test case.

Apparently this conclusion is not correct since in the kernel used
by buildds, the above config is enabled, unless the kernel is not
what was in the ports repo with the same name. If it were the kernel
problem, there would be no need to reinstall the whole system but
replacing the kernel would be enough. I suspect that this issue
might be related to the sbuild unshare environment or related to
how the buildd are deployed. The root cause for it should be further
investigated.

> Please send a patch for GDB or let me know that you don't plan to do it
> and I'll do it.
> 
> Thanks,
> Tiezhu
> 

Reply via email to