On Mon, Feb 17, 2020 at 02:08:03PM +0800, Jason Wang wrote:> I thought you were copied in the patch [1], maybe we can move vhost related > discussion there to avoid confusion. > > [1] https://lwn.net/Articles/811210/Wow, that is .. confusing. So this is supposed to duplicate the uAPI of vhost-user? But it is open coded and duplicated because .. vdpa?> So it's cheaper and simpler to introduce a new bus instead of refactoring a > well known bus and API where brunches of drivers and devices had been > implemented for years.If you reason for this approach is to ease the implementation then you should talk about it in the cover letters/etc Maybe it is reasonable to do this because the rework is too great, I don't know, but to me this whole thing looks rather messy. Remember this stuff is all uAPI as it shows up in sysfs, so you can easilly get stuck with it forever. Jason
On Tue, Feb 18, 2020 at 01:56:12PM +0000, Jason Gunthorpe wrote:> On Mon, Feb 17, 2020 at 02:08:03PM +0800, Jason Wang wrote: > > > I thought you were copied in the patch [1], maybe we can move vhost related > > discussion there to avoid confusion. > > > > [1] https://lwn.net/Articles/811210/ > > Wow, that is .. confusing. > > So this is supposed to duplicate the uAPI of vhost-user? But it is > open coded and duplicated because .. vdpa?Do you mean the vhost-user in DPDK? There is no vhost-user in Linux kernel. Thanks, Tiwei> > > So it's cheaper and simpler to introduce a new bus instead of refactoring a > > well known bus and API where brunches of drivers and devices had been > > implemented for years. > > If you reason for this approach is to ease the implementation then you > should talk about it in the cover letters/etc > > Maybe it is reasonable to do this because the rework is too great, I > don't know, but to me this whole thing looks rather messy. > > Remember this stuff is all uAPI as it shows up in sysfs, so you can > easilly get stuck with it forever. > > Jason
On 2020/2/18 ??9:56, Jason Gunthorpe wrote:> On Mon, Feb 17, 2020 at 02:08:03PM +0800, Jason Wang wrote: > >> I thought you were copied in the patch [1], maybe we can move vhost related >> discussion there to avoid confusion. >> >> [1] https://lwn.net/Articles/811210/ > Wow, that is .. confusing. > > So this is supposed to duplicate the uAPI of vhost-user?It tries to reuse the uAPI of vhost with some extension.> But it is > open coded and duplicated because .. vdpa?I'm not sure I get here, vhost module is reused for vhost-vdpa and all current vhost device (e.g net) uses their own char device.> >> So it's cheaper and simpler to introduce a new bus instead of refactoring a >> well known bus and API where brunches of drivers and devices had been >> implemented for years. > If you reason for this approach is to ease the implementation then you > should talk about it in the cover letters/etcI will add more rationale in both cover letter and this patch. Thanks> > Maybe it is reasonable to do this because the rework is too great, I > don't know, but to me this whole thing looks rather messy. > > Remember this stuff is all uAPI as it shows up in sysfs, so you can > easilly get stuck with it forever. > > Jason >
On Wed, Feb 19, 2020 at 01:35:25PM +0800, Jason Wang wrote:> > But it is > > open coded and duplicated because .. vdpa? > > > I'm not sure I get here, vhost module is reused for vhost-vdpa and all > current vhost device (e.g net) uses their own char device.I mean there shouldn't be two fops implementing the same uAPI Jason