>-----Original Message----- >From: Joao Martins <joao.m.mart...@oracle.com> >Sent: Monday, September 11, 2023 6:13 PM >Subject: Re: [PATCH v1] vfio/common: Separate vfio-pci ranges > >On 11/09/2023 10:48, Duan, Zhenzhong wrote: >>> -----Original Message----- >>> From: Joao Martins <joao.m.mart...@oracle.com> >>> Sent: Monday, September 11, 2023 5:07 PM >>> Subject: Re: [PATCH v1] vfio/common: Separate vfio-pci ranges ... >> I have another question, previously I think vfio pci bars are device states >> and >> save/restored through VFIO migration protocol, so we don't need to dirty >> tracking them. Do I understand wrong? > >The general thinking of device dirty tracking is to track all addressable >IOVAs. >But you do raise a good question. My understanding is that migrating the bars >*as is* might be device migration specific (not a guarantee?); the save file >and >precopy interface are the only places we transfer from/to the data and it's >just >opaque data, not bars or anything formatted specifically -- so if we migrate >bars it is hidden in what device f/w wants to move. Might be that BARs aren't >even needed as they are sort of scratch space from h/w side. Ultimately, the >dirty tracker is the one reporting the values, and the device h/w chooses to >not >report those IOVAs as dirty then nothing changes.
Understood, thanks Joao.