>-----Original Message----- >From: Cédric Le Goater <c...@redhat.com> >Subject: Re: [PATCH 1/2] backends/iommufd: Remove check on number of >backend users > >Hello Zhenzhong, > >On 1/3/24 02:40, Duan, Zhenzhong wrote: >> Hi Cédric, >> >>> -----Original Message----- >>> From: Cédric Le Goater <c...@redhat.com> >>> Sent: Tuesday, January 2, 2024 8:32 PM >>> To: qemu-devel@nongnu.org >>> Cc: Liu, Yi L <yi.l....@intel.com>; Eric Auger <eric.au...@redhat.com>; >Duan, >>> Zhenzhong <zhenzhong.d...@intel.com>; Cédric Le Goater >>> <c...@redhat.com> >>> Subject: [PATCH 1/2] backends/iommufd: Remove check on number of >>> backend users >>> >>> QOM already has a ref count on objects and it will assert much >>> earlier, when INT_MAX is reached. >>> >>> Signed-off-by: Cédric Le Goater <c...@redhat.com> >> >> IIUC, /dev/iommu is opened on demand, be->users is used to catch >underflow >> or overflow due to mismatched iommufd_backend_connect/disconnect >> pairs. >> >> It looks different from object ref count in purpose, but I agree >> a correctly written code will never trigger such overflow/underflow. > >Well, we could limit the number of devices sharing the same iommufd >backend but UINT32_MAX seems really too large and the object refcount >will fail earlier anyhow. The max open files per process limit will >also be reached before, since vfio opens a /dev/vfio/devices/vfiox >file per device.
Clear, thanks Cédric. > >So, this check didn't seem necessary after all. > >> So, for the series: >> Reviewed-by: Zhenzhong Duan <zhenzhong.d...@intel.com> > >You should reply to the cover letter with your R-b tag so that >it applies on the whole series. Sure. BRs. Zhenzhong
