[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <BN9PR11MB5276974ABA5981A7361953708C979@BN9PR11MB5276.namprd11.prod.outlook.com>
Date: Wed, 27 Jul 2022 03:20:25 +0000
From: "Tian, Kevin" <kevin.tian@...el.com>
To: Jason Gunthorpe <jgg@...dia.com>,
Baolu Lu <baolu.lu@...ux.intel.com>
CC: Joerg Roedel <joro@...tes.org>,
Christoph Hellwig <hch@...radead.org>,
"Raj, Ashok" <ashok.raj@...el.com>, Will Deacon <will@...nel.org>,
"Robin Murphy" <robin.murphy@....com>,
Jean-Philippe Brucker <jean-philippe@...aro.com>,
"Jiang, Dave" <dave.jiang@...el.com>,
Vinod Koul <vkoul@...nel.org>,
Eric Auger <eric.auger@...hat.com>,
"Liu, Yi L" <yi.l.liu@...el.com>,
"Pan, Jacob jun" <jacob.jun.pan@...el.com>,
"Zhangfei Gao" <zhangfei.gao@...aro.org>,
"Zhu, Tony" <tony.zhu@...el.com>,
"iommu@...ts.linux.dev" <iommu@...ts.linux.dev>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
"Jean-Philippe Brucker" <jean-philippe@...aro.org>
Subject: RE: [PATCH v10 04/12] iommu: Add attach/detach_dev_pasid iommu
interface
> From: Jason Gunthorpe <jgg@...dia.com>
> Sent: Tuesday, July 26, 2022 9:57 PM
>
> On Tue, Jul 26, 2022 at 02:23:26PM +0800, Baolu Lu wrote:
> > On 2022/7/25 22:40, Jason Gunthorpe wrote:
> > > On Sun, Jul 24, 2022 at 03:03:16PM +0800, Baolu Lu wrote:
> > >
> > > > How about rephrasing this part of commit message like below:
> > > >
> > > > Some buses, like PCI, route packets without considering the PASID value.
> > > > Thus a DMA target address with PASID might be treated as P2P if the
> > > > address falls into the MMIO BAR of other devices in the group. To make
> > > > things simple, these interfaces only apply to devices belonging to the
> > > > singleton groups.
> > >
> > > > Considering that the PCI bus supports hot-plug, even a device boots
> with
> > > > a singleton group, a later hot-added device is still possible to share
> > > > the group, which breaks the singleton group assumption. In order to
> > > > avoid this situation, this interface requires that the ACS is enabled on
> > > > all devices on the path from the device to the host-PCI bridge.
> > >
> > > But ACS directly fixes the routing issue above
> > >
> > > This entire explanation can be recast as saying we block PASID
> > > attachment in all cases where the PCI fabric is routing based on
> > > address. ACS disables that.
> > >
> > > Not sure it even has anything to do with hotplug or singleton??
> >
> > Yes, agreed. I polished this patch like below. Does it look good to you?
> >
> > iommu: Add attach/detach_dev_pasid iommu interface
> >
> > Attaching an IOMMU domain to a PASID of a device is a generic operation
> > for modern IOMMU drivers which support PASID-granular DMA address
> > translation. Currently visible usage scenarios include (but not limited):
> >
> > - SVA (Shared Virtual Address)
> > - kernel DMA with PASID
> > - hardware-assist mediated device
> >
> > This adds a pair of domain ops for this purpose and adds the interfaces
> > for device drivers to attach/detach a domain to/from a {device,
> > PASID}.
>
> > The PCI bus routes packets without considering the PASID value.
>
> More like:
>
> Some configurations of the PCI fabric will route device originated TLP
> packets based on memory address, and these configurations are
> incompatible with PASID as the PASID packets form a distinct address
> space. For instance any configuration where switches are present
> without ACS is incompatible with PASID.
This description reads like ACS enables PASID-based routing...
In reality PCI fabric always route TLP based on memory address.
ACS just provides a way to redirect the packet to RC, with or
without PASID.
So it's simply that PASID requires such redirection hence ACS
because only RC/IOMMU understands PASID and related
address space.
>
> > + /*
> > + * Block PASID attachment in all cases where the PCI fabric is
> > + * routing based on address. ACS disables it.
> > + */
> > + if (dev_is_pci(dev) &&
> > + !pci_acs_path_enabled(to_pci_dev(dev), NULL, REQ_ACS_FLAGS))
> > + return -ENODEV;
>
> I would probably still put this in a function just to be clear, and
> probably even a PCI layer funcion 'pci_is_pasid_supported' that
> clearly indicates that the fabric path can route a PASID packet
> without mis-routing it.
But there is no single line in above check related to PASID...
>
> If the fabric routes PASID properly then groups are not an issue - all
> agree on this?
>
IMHO if the fabric can route PASID properly, and according to
above once such redirect is available it applies to both non-PASID
and PASID TLP, then the group will be singleton in the first place.
Is there a real-world example where the fabric can route PASID
properly for a multi-devices group?
Thanks
Kevin
Powered by blists - more mailing lists