[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-devel] [PATCH 2/4] memory: Single byte swap along the I/O path
From: |
Paolo Bonzini |
Subject: |
Re: [Qemu-devel] [PATCH 2/4] memory: Single byte swap along the I/O path |
Date: |
Wed, 17 Jul 2019 12:08:21 +0200 |
User-agent: |
Mozilla/5.0 (X11; Linux x86_64; rv:60.0) Gecko/20100101 Thunderbird/60.7.2 |
On 17/07/19 08:06, address@hidden wrote:
> +
> +static inline MemOp MEMOP(unsigned size)
> +{
> + switch (size) {
> + case 1:
> + return MO_8;
> + case 2:
> + return MO_16;
> + case 4:
> + return MO_32;
> + case 8:
> + return MO_64;
> + default:
> + g_assert_not_reached();
> + }
> +}
FWIW, this is just "ctz(size)".
My main concern is that MO_BE/MO_LE/MO_TE do not really apply to the
memory.c paths. MO_BSWAP is never passed into the MemOp, even if target
endianness != host endianness.
Therefore, you could return MO_TE | MO_{8,16,32,64} from this function,
and change memory_region_endianness_inverted to test
HOST_WORDS_BIGENDIAN instead of TARGET_WORDS_BIGENDIAN. Then the two
MO_BSWAPs (one from MO_TE, one from adjust_endianness because
memory_region_endianness_inverted returns true) cancel out if the
memory region's endianness is the same as the host's but different
from the target's.
Some care is needed in virtio_address_space_write and zpci_write_bar. I
think the latter is okay, while the former could do something like this:
diff --git a/hw/virtio/virtio-pci.c b/hw/virtio/virtio-pci.c
index ce928f2429..61885f020c 100644
--- a/hw/virtio/virtio-pci.c
+++ b/hw/virtio/virtio-pci.c
@@ -541,16 +541,16 @@ void virtio_address_space_write(VirtIOPCIProxy *proxy,
hwaddr addr,
val = pci_get_byte(buf);
break;
case 2:
- val = cpu_to_le16(pci_get_word(buf));
+ val = pci_get_word(buf);
break;
case 4:
- val = cpu_to_le32(pci_get_long(buf));
+ val = pci_get_long(buf);
break;
default:
/* As length is under guest control, handle illegal values. */
return;
}
- memory_region_dispatch_write(mr, addr, val, len, MEMTXATTRS_UNSPECIFIED);
+ memory_region_dispatch_write(mr, addr, val, size_memop(len) & ~MO_BSWAP,
MEMTXATTRS_UNSPECIFIED);
}
static void
Paolo
- [Qemu-devel] [PATCH 0/4] Invert Endian bit in SPARCv9 MMU TTE, tony.nguyen, 2019/07/17
- [Qemu-devel] [PATCH 1/4] tcg: TCGMemOp is now accelerator independent MemOp, tony.nguyen, 2019/07/17
- [Qemu-devel] [PATCH 2/4] memory: Single byte swap along the I/O path, tony.nguyen, 2019/07/17
- [Qemu-devel] [PATCH 4/4] target/sparc: sun4u Invert Endian TTE bit, tony.nguyen, 2019/07/17
- [Qemu-devel] [PATCH 3/4] cputlb: Byte swap memory transaction attribute, tony.nguyen, 2019/07/17
- Re: [Qemu-devel] [PATCH 0/4] Invert Endian bit in SPARCv9 MMU TTE, Paolo Bonzini, 2019/07/17