Re: [PATCH] gcore: Handle unreadable pages within readable memory regions

[Kevin Buettner <kevinb@redhat.com>]

On Tue, 10 Feb 2026 23:28:27 +0530
Abhay Kandpal <abhay@linux.ibm.com> wrote:

> Hi Kevin,
> 
> Thanks for sharing this patch and for the detailed end-to-end explanation — it matches exactly what I’ve been seeing.
> 
> I’ve independently hit the same issue with glibc 2.42 on kernels > 6.3, due to the MADV_GUARD_INSTALL change.
> The behavior is reproducible with both upstream and custom GDB.
> 
> I tested your patch on ppc64le and x86_64, and it fixes the problem on both.
> This looks like a generic GDB issue triggered by the new glibc stack layout rather than an architecture-specific problem.
> 
> The approach of falling back to page-by-page reads on failure looks correct to me.
> Thanks again for the clear analysis and fix.

Hi Abhay,

Thanks for your review!

I ended up posting a v2 patch which does not use the constant
SPARSE_BLOCK_SIZE.  It is used for other purposes in the file.  The
number associated with this constant makes sense, but it seemed better
to instead define a new constant for the size of  the fallback memory reads.

The v2 patch is here:

https://sourceware.org/pipermail/gdb-patches/2026-February/224948.html

Kevin