On Mon, May 20, 2019 at 09:05:57AM -0600, Keith Busch wrote: > On Sat, May 18, 2019 at 09:39:05AM +0200, Klaus Birkelund Jensen wrote: > > `nvme_dma_read_prp` erronously used `qemu_iovec_*to*_buf` instead of > > `qemu_iovec_*from*_buf` when the request involved the controller memory > > buffer. > > > > Signed-off-by: Klaus Birkelund Jensen <[email protected]> > > I was wondering how this mistake got by for so long, and it looks like > the only paths here require an admin command with dev->host transfer > to CMB. That's just not done in any host implementation I'm aware of > since it'd make it more difficult to use for no particular gain AFAICS, > so I'd be curious to hear if you have a legit implementation doing this. >
I'm just trying to get the device to be as compliant as possible, but I don't know why you'd have any reason to do a, say Get Feature, to the CMB.
