Am 24.06.2013 18:27, schrieb Paolo Bonzini: > Il 24/06/2013 18:24, Peter Lieven ha scritto: >>>> So far, 4K blocksize is usable if you also specify the same block size >>>> for the guest device. I have posted once the patches to do >>>> read-modify-write, but I never really pursued inclusion of those. >>>> >>>> In any case, the right place to fix this is the block layer; >>>> driver-specific hacks are... hacks. :) >> Where do I find the sector size if not in BDRV_SECTOR_SIZE? >> Is there a dynamic field or is the answer qemu only support 512 Byte >> sector size at the moment? > bdrv_read/write and friends are always in 512-byte units. > > There is bs->buffer_alignment for the guest sector size, but it is not > set at bdrv_open, only later. So I think QEMU support for bigger > sectors is too sparse to be useful. > >> So you would go for fail to open a device if the LUN blocksize is >> not equal to BDRV_SECTOR_SIZE? > Either that, or fails reads/writes that are not aligned. It would be the easiest. Maybe Ronnie can point out what the hack with the acb->buffer_offset in aio_read was meant for. All other functions have the requirement to be properly aligned regarding to the xfersize. I am also not sure if the e.g. reading of 512 byte at sector 0 from a 4K block device does still work with the iovector patches we added earlier.
Peter > > Paolo
