Re: [REGRESSION] commit b520c4eef83d ("block: split bio_alloc_bioset more clearly into a fast and slowpath") causes fsync to fail with pmem

On Fri, Jul 10, 2026 at 12:17:34PM -0600, Keith Busch wrote:
> On Fri, Jul 10, 2026 at 08:07:32PM +0200, Sven-Hendrik Haase wrote:
> > I noticed that starting with Linux 7.1, my pmem volumes inside of QEMU
> > stopped working. My use case is this:
> > 
> > qemu-system-x86_64 ... tons of QEMU flags ...
> > -object memory-backend-file,id=pmem0,share=on,merge=on,discard-data=on,mem-path=/path/pmem0.pmem,size=100G
> > -device virtio-pmem-pci,memdev=pmem0,id=nv0
> > 
> > This used to work but inside the VM I'd get this when running mkfs.ext4
> > /dev/pmem0 with guest kernel 7.1.3:
> 
> This pending fix (which is the same as what you proposed here) should
> take care of it:
> 
> https://lore.kernel.org/linux-block/[email protected]/

Note that it can stil fail with that alone.  The real fix is this:

https://lore.kernel.org/all/[email protected]/