Re: [PATCH] system/physmem: Fix cpu_memory_rw_debug for armv7m MPU

[Richard Henderson <richard.henderson@linaro.org>]

On 11/23/24 22:30, xndcn wrote:
> Thanks. This patch at least guarantees normal read/write access to
> addresses with r/w flags, although there is still a risk of
> misidentifying accessible regions within continuous address spaces.
> 
> Actually, initially I did write a patch with a modified page size as
> an argument, but I soon found that the current implementation of
> armv7a (pmsav7) will return the page size (lg_page_size) as 0 in many
> situations (such as overlapping regions).

0 was arbitrarily chosen as "anything less than TARGET_PAGE_BITS".

> Maybe we can simply make the page size as 2^5 (minimal page size) when
> the returned lg_page_size == 0?

No, don't lie, or make unhelpful guesses at the generic level. The only thing you could do 
for arm m-profile is give accurate information, which for now will be immediately 
discarded as *still* being less than TARGET_PAGE_BITS.  But I do have plans for that.

Anyway, nothing you do with page sizes is helpful along the debug path.  For that, we need 
to swap away from "pages" to "ranges", where some ranges are in fact pages, but others 
aren't.  This means changing the API for cpu_get_phys_<something>_debug.


r~