Re: [PULL 03/38] pflash: Only read non-zero parts of backend image

[Kevin Wolf]

Am 07.02.2023 um 10:19 hat Cédric Le Goater geschrieben:
> On 2/7/23 09:38, Kevin Wolf wrote:
> > Am 06.02.2023 um 16:54 hat Cédric Le Goater geschrieben:
> > > On 1/20/23 13:25, Kevin Wolf wrote:
> > > > From: Xiang Zheng <zhengxiang9@huawei.com>
> > > > 
> > > > Currently we fill the VIRT_FLASH memory space with two 64MB NOR images
> > > > when using persistent UEFI variables on virt board. Actually we only use
> > > > a very small(non-zero) part of the memory while the rest significant
> > > > large(zero) part of memory is wasted.
> > > > 
> > > > So this patch checks the block status and only writes the non-zero part
> > > > into memory. This requires pflash devices to use sparse files for
> > > > backends.
> > > > 
> > > > Signed-off-by: Xiang Zheng <zhengxiang9@huawei.com>
> > > > 
> > > > [ kraxel: rebased to latest master ]
> > > > 
> > > > Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
> > > > Message-Id: <20221220084246.1984871-1-kraxel@redhat.com>
> > > > Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
> > > > Reviewed-by: Kevin Wolf <kwolf@redhat.com>
> > > > Signed-off-by: Kevin Wolf <kwolf@redhat.com>
> > > 
> > > This newly merged patch introduces a "regression" when booting an Aspeed
> > > machine. The following extra m25p80 patch (not yet merged) is required
> > > for the issue to show:
> > > 
> > >    
> > > https://lore.kernel.org/qemu-devel/20221115151000.2080833-1-clg@kaod.org/
> > > 
> > > U-Boot fails to find the filesystem in that case.
> > > 
> > > It can be easily reproduced with the witherspoon-bmc machine and seems
> > > to be related to the use of a UBI filesystem. Other Aspeed machines not
> > > using UBI are not impacted.
> > > 
> > > Here is a tentative fix. I don't know enough the block layer to explain
> > > what is happening :/
> > 
> > I was puzzled for a moment, but...
> > 
> > > @@ -39,7 +39,7 @@ static int blk_pread_nonzeroes(BlockBack
> > >               return ret;
> > >           }
> > >           if (!(ret & BDRV_BLOCK_ZERO)) {
> > > -            ret = bdrv_pread(bs->file, offset, bytes,
> > 
> > 'bs->file' rather than 'bs' really looks wrong. I think replacing that
> > would already fix the bug you're seeing.
> > 
> > Just to be sure, how did you configure the block backend? bs->file would
> > happen to work more or less with raw over file-posix (which is probably
> > what Gerd tested), but I think it breaks with anything else.
> 
> The command is  :
> 
>   $ qemu-system-arm -M witherspoon-bmc -net user \
>       -drive file=/path/to/file.mtd,format=raw,if=mtd \
>       -nographic -serial mon:stdio -snapshot
> 
> If I remove '-snapshot', all works fine.

Ok, that makes sense then. -snapshot creates a temporary qcow2 overlay,
and then what your guest sees with bs->file is not the virtual disk
content of the qcow2 image, but the qcow2 file itself.

Kevin