Re: [PATCH 0/4] arm64: Cleanup ioremap() and support ioremap_prot()

[Kefeng Wang]

On 2022/4/28 18:46, Will Deacon wrote:
> On Wed, Apr 27, 2022 at 08:14:09PM +0800, Kefeng Wang wrote:
> > Let's arm64 use GENERIC_IOREMAP to cleanup code, and
> > support ioremap_prot()/HAVE_IOREMAP_PROT, which could
> > enable generic_access_phys().
> >
> > Kefeng Wang (4):
> >    mm: ioremap: Setup phys_addr of struct vm_struct
> >    mm: ioremap: Add arch_ioremap/iounmap_check()
> >    arm64: mm: Convert to GENERIC_IOREMAP
> >    arm64: Add HAVE_IOREMAP_PROT support
> >
> >   .../features/vm/ioremap_prot/arch-support.txt |  2 +-
> >   arch/arm64/Kconfig                            |  2 +
> >   arch/arm64/include/asm/io.h                   | 14 +--
> >   arch/arm64/include/asm/pgtable.h              | 10 +++
> >   arch/arm64/kernel/acpi.c                      |  2 +-
> >   arch/arm64/mm/hugetlbpage.c                   | 10 ---
> >   arch/arm64/mm/ioremap.c                       | 86 +++----------------
> >   include/asm-generic/io.h                      |  3 +
> >   mm/ioremap.c                                  | 21 ++++-
> >   9 files changed, 56 insertions(+), 94 deletions(-)
> That's not a massively compelling diffstat for a cleanup, in all honesty.
> I looked at generic_access_phys() to try to figure out why we would want
> that on arm64, but it seems like it's related to mmap() of devices in
> userspace. Bearing in mind that CONFIG_STRICT_DEVMEM=y by default, please
> can you justify why this is something worth doing?

The geneirc_access_phys was introduced by

  7ae8ed5053a3 use generic_access_phys for /dev/mem mappings
  28b2ee20c7cb access_process_vm device memory infrastructure

and add into uio.c  by

  7294151d0592 "uio: provide vm access to UIO_MEM_PHYS maps"

It could let user to debug(eg, gdb or ptrace) app via access_process_vm().

also some upstream drivers and out-of-tree modules could use it too
  drivers/fpga/dfl-afu-main.c:    .access = generic_access_phys,
  drivers/pci/mmap.c:     .access = generic_access_phys,

When see the ioremap_prot from generic ioremap, for better debug, I think

we could enabled it on arm64 too.

> Will
> .