[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Qemu-devel] MemoryRegionOps access sizes

From: Peter Maydell
Subject: Re: [Qemu-devel] MemoryRegionOps access sizes
Date: Wed, 20 Feb 2019 15:48:08 +0000

On Fri, 15 Feb 2019 at 09:56, BALATON Zoltan <address@hidden> wrote:
> Hello,
> Could someone please explain how the MemoryRegionOps
> valid.{min,max}_access_size and unaligned and corresponding impl
> constraints work and where are these implemented? And in particular if
> this would work:
> static const MemoryRegionOps ops = {
>      .read = readfn,
>      .write = writefn,
>      .valid.min_access_size = 1,
>      .valid.max_access_size = 4,
>      .impl.min_access_size = 4,
>      .impl.max_access_size = 4,
>      .endianness = DEVICE_LITTLE_ENDIAN,
> };

These are implemented in memory.c. The documentation
is in the comments on the MemoryRegionOps struct in

The .valid parts should specify what the device permits
the guest to do (and where failures result in MemTxErrors
usually meaning bus error/data abort). These ought to be
handled by the memory_region_access_valid() function, but
I can't see how the logic there actually does the right
thing. Maybe Paolo understands that part better than me.

The .impl parts specify what the implementation can
deal with, and access_with_adjusted_size() takes
care of making multiple accesses to the underlying
device if necessary. (The fiddling of the data to
put it in the right place is done at the next level
down by memory_region_shift_read_access()
and memory_region_shift_write_access().)

I think your suggested MemoryRegionOps ought to work.

-- PMM

reply via email to

[Prev in Thread] Current Thread [Next in Thread]