Re: [musl] Add PAGESIZE definition for LoongArch

[Rich Felker <dalias@libc.org>]

On Tue, May 07, 2024 at 09:45:32AM +0800, lixing wrote:
> 
> 在 2024/5/7 上午9:39, Rich Felker 写道:
> >On Tue, May 07, 2024 at 09:30:13AM +0800, lixing wrote:
> >>在 2024/5/6 下午8:36, Rich Felker 写道:
> >>>On Mon, May 06, 2024 at 05:05:36PM +0800, lixing wrote:
> >>>>Hi, Rich,
> >>>>
> >>>>  arch/loongarch64/bits/limits.h | 1 +
> >>>>  1 file changed, 1 insertion(+)
> >>>>  create mode 100644 arch/loongarch64/bits/limits.h
> >>>>
> >>>>diff --git a/arch/loongarch64/bits/limits.h b/arch/loongarch64/bits/limits.h
> >>>>new file mode 100644
> >>>>index 00000000..5cd9aad6
> >>>>--- /dev/null
> >>>>+++ b/arch/loongarch64/bits/limits.h
> >>>>@@ -0,0 +1 @@
> >>>>+#define PAGESIZE 16384
> >>>Can you explain why you want this change? I would be very hesitant to
> >>>merge it. Defining PAGESIZE for an arch is a contract that the
> >>>application-facing runtime page size (i.e. mmap granularity) can
> >>>*never*, now or in the future, be anything other than the value you
> >>>define this macro as having.
> >>when I debuging "indent" program error, the  caculation of
> >>relro_start and relro_end in function kernel_mapped_dso get wrong
> >>value. It seems the PAGE_SIZE value is 0 and it come from
> >>libc.page_size, but the libc.page_size initialized until __dls3. Is
> >>there any other way to make the PAGE_SIZE correct at the before
> >>__dls3 ?
> >I believe this is a known bug that somehow got skipped over in the
> >last release cycle. I'll get it fixed. It does not mean you need to
> >define a constant PAGESIZE.
> ok,thanks.
> >>Also, if the xmalloc low and high address aparted by the  protect
> >>page in a  loop,  it seems to trigger SIGSEGV when the loop meet the
> >>protect page. Can this situation happen?
> >I don't understand.
> 
> ok, I will do more test to check this happens or not.
> 
> Thanks.

OK, the bug was supposedly already fixed by commit
f47a8cdd250d9163fcfb39bf4e9d813957c0b187, but that commit didn't
actually do anything because of inclusion order issues. A fix for the
fix is now available in commit
6f666231bf51703fadbef10460d462fb573548a1.

Rich