Re: [PATCH v4 0/9] memory: assert and define MemoryRegionOps callbacks

[P J P <ppandit@redhat.com>]

[-- Attachment #1: Type: text/plain, Size: 1463 bytes --]

+-- On Sun, 16 Aug 2020, Philippe Mathieu-Daudé wrote --+
| On 8/11/20 1:41 PM, P J P wrote:
| > From: Prasad J Pandit <pjp@fedoraproject.org>
| > * This series asserts that MemoryRegionOps objects define read/write
| >   callback methods. Thus avoids potential NULL pointer dereference.
| >   ex. -> https://git.qemu.org/?p=qemu.git;a=commit;h=bb15013ef34617eb1344f5276292cadd326c21b2
| > 
| > * Also adds various undefined MemoryRegionOps read/write functions
| >   to avoid potential assert failure.
| 
| What about read_with_attrs()/write_with_attrs()? It seems they are part of 
| the same problem.

* read/write_with_attrs function is called if read/write callback is not 
  defined

  ../softmmu/memory.c
    if (mr->ops->write) {
                    ... memory_region_write_accessor, mr,
    } else {
                    ... memory_region_write_with_attrs_accessor,

  So, defining read/write methods may also address read/write_with_attrs 
  issue?

* $ grep -Eri -A 5 -B 5 '(\.read_with_attrs|\.write_with_attrs)' . | fpaste

   -> https://paste.centos.org/view/386c9597

  It doesn't show an occurrence where one of the read/write_with_attrs is 
  missing.

* Nevertheless, if we need to define read/write_with_attrs routines, because 
  memory_region_init_io() would assert(3) for them

  could that be a subsequent patch series please?


Thank you.
--
Prasad J Pandit / Red Hat Product Security Team
8685 545E B54C 486B C6EB 271E E285 8B5A F050 DE8D