[Top][All Lists]

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

Re: [Qemu-devel] [PATCH 2/3] PPC: Allow little-endian user mode.

From: Peter Maydell
Subject: Re: [Qemu-devel] [PATCH 2/3] PPC: Allow little-endian user mode.
Date: Thu, 8 May 2014 13:39:28 +0100

On 8 May 2014 09:39, Alexander Graf <address@hidden> wrote:
> On 05/08/2014 10:26 AM, Doug Kwan wrote:
>> This all running PPC64 little-endian in user mode if target is configured
>> that way.  In PPC64 LE user mode we set MSR.LE during initialization.
>> Byteswapping logic is reversed also when QEMU is running in that mode.
>> Signed-off-by: Doug Kwan <address@hidden>
> I can't say I'm a huge fan of this patch. It allows for really tricky
> subtile mistakes to happen. Can't we leave the target mode configured on big
> endian?

Unfortunately not if you care about linux-user mode:
linux-user mode uses the target-endian setting to figure
out if it needs to do swapping of data in all the syscall

If you're going to overhaul how PPC deals with endian
dependent loads/stores, I suspect you'll end up with a
cleaner result if you convert to the new "specify endian
setting as part of the memory operation" TCG ops:
So for instance rather than having:

    tcg_gen_qemu_ld16u(arg1, arg2, ctx->mem_idx);
    if (unlikely(ctx->le_mode)) {
        tcg_gen_bswap16_tl(arg1, arg1);

it would be better to do
    TCGMemOp op = MO_UW | (ctx->le_mode ? MO_LE : MO_BE);
    tcg_gen_qemu_ld_i32(arg1, arg2, ctx->mem_idx, op);

This will work regardless of the TARGET_WORD_BIGENDIAN
setting, since we directly ask TCG to do an LE or BE
access, rather than doing a target-endian access and
then swapping. (It's also more efficient if you're
in little-endian mode on a little endian host since
it won't swap at all rather than swapping twice.)

-- PMM

reply via email to

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