[Qemu-devel] Re: [PATCH 13/13] blockdev: New -blockdev to define a host block device
On 06/10/2010 06:32 PM, Markus Armbruster wrote: I understand why that's the most sensible cache setting. But if the user explicitly asks for something else, I think we better give it to him, or tell him no. Ignoring him silently isn't nice. Ah, it's clearer now... I guess one could use cache=something together with snapshot to do benchmarking. Actually the same changes in behavior (unsafe as default, but observe a non-default value) can be done to -drive ...,snapshot too. Maybe I'll give it a shot. Paolo
[Qemu-devel] Re: [PATCH 13/13] blockdev: New -blockdev to define a host block device
Paolo Bonzini writes: > On 06/02/2010 06:55 PM, Markus Armbruster wrote: >> * Like -drive, -blockdev ignores cache= silently when snapshot=on. Do >>we really want that? > > Yes, the changes are throw-away by definition. Might as well use > cache=unsafe. I understand why that's the most sensible cache setting. But if the user explicitly asks for something else, I think we better give it to him, or tell him no. Ignoring him silently isn't nice. >> +if (snapshot) { >> +/* always use write-back with snapshot */ >> +/* FIXME ignores explicit cache= *silently*; really want that? */ >> +flags &= ~BDRV_O_CACHE_MASK; >> +flags |= (BDRV_O_SNAPSHOT | BDRV_O_CACHE_WB); >> +flags |= BDRV_O_SNAPSHOT; > > Cut and paste? Pasto indeed, will fix. Thanks!
[Qemu-devel] Re: [PATCH 13/13] blockdev: New -blockdev to define a host block device
On 06/02/2010 06:55 PM, Markus Armbruster wrote: * Like -drive, -blockdev ignores cache= silently when snapshot=on. Do we really want that? Yes, the changes are throw-away by definition. Might as well use cache=unsafe. +if (snapshot) { +/* always use write-back with snapshot */ +/* FIXME ignores explicit cache= *silently*; really want that? */ +flags &= ~BDRV_O_CACHE_MASK; +flags |= (BDRV_O_SNAPSHOT | BDRV_O_CACHE_WB); +flags |= BDRV_O_SNAPSHOT; Cut and paste? Paolo