Re: [PATCH] xen/virtio: Handle PCI devices which Host controller is described in DT

[Stefano Stabellini <sstabellini@xxxxxxxxxx>]

On Thu, 6 Oct 2022, Oleksandr Tyshchenko wrote:
> From: Oleksandr Tyshchenko <oleksandr_tyshchenko@xxxxxxxx>
> 
> Use the same "xen-grant-dma" device concept (based on generic IOMMU
> device-tree bindings) for the PCI devices behind device-tree based
> PCI Host controller.
> 
> Signed-off-by: Oleksandr Tyshchenko <oleksandr_tyshchenko@xxxxxxxx>
> ---
> Slightly RFC. This is needed to support Xen grant mappings for virtio-pci 
> devices
> on Arm at some point in the future. The Xen toolstack side is not published 
> yet.
> Here, for PCI devices we use the same way to pass backend domid to the guest 
> as for
> platform devices.
> 
> Depends on Juergen's series:
> https://lore.kernel.org/xen-devel/20221006071500.15689-1-jgross@xxxxxxxx/
> ---
>  drivers/xen/grant-dma-ops.c | 51 +++++++++++++++++++++++++++++--------
>  1 file changed, 41 insertions(+), 10 deletions(-)
> 
> diff --git a/drivers/xen/grant-dma-ops.c b/drivers/xen/grant-dma-ops.c
> index ff9be3aff87e..79d13122ec08 100644
> --- a/drivers/xen/grant-dma-ops.c
> +++ b/drivers/xen/grant-dma-ops.c
> @@ -10,6 +10,7 @@
>  #include <linux/module.h>
>  #include <linux/dma-map-ops.h>
>  #include <linux/of.h>
> +#include <linux/pci.h>
>  #include <linux/pfn.h>
>  #include <linux/xarray.h>
>  #include <linux/virtio_anchor.h>
> @@ -273,12 +274,28 @@ static const struct dma_map_ops xen_grant_dma_ops = {
>       .dma_supported = xen_grant_dma_supported,
>  };
>  
> -static bool xen_is_dt_grant_dma_device(struct device *dev)
> +static struct device_node *xen_dt_get_node(struct device *dev)
> +{
> +     if (dev_is_pci(dev)) {
> +             struct pci_dev *pdev = to_pci_dev(dev);
> +             struct pci_bus *bus = pdev->bus;
> +
> +             /* Walk up to the root bus to look for PCI Host controller */
> +             while (!pci_is_root_bus(bus))
> +                     bus = bus->parent;
> +
> +             return of_node_get(bus->bridge->parent->of_node);
> +     }

Is it possible to have multiple virtio devices under a single virtio-pci
root complex? What if virtio-net has the backend in dom0 and
virtio-block has the backend in dom1?

Or each virtio PCI device shows up under a different PCI root complex?

If we can have multiple virtio PCI devices under a single PCI root
complex, then I think it would be better to check for a per-device
property, rather than a single property at the PCI root complex level.

The first thing that comes to mind is to describe each PCI device under
the root complex in device tree. Although it is uncommon (usually only
the PCI root complex is described in device tree), it is possible to
also describe in device tree all the individual PCI devices under the
root complex.

Given that the domU device tree is generated by Xen and/or the Xen
toolstack, it would be easy to arrange for it to happen.

That would solve the issue as far as I can tell, but I worry it might
not be a good idea because if we rely on the per-device device tree node
to be present then it becomes harder to implement virtio hotplug
(Virtio hotplug is important to add dom0less support.)

Let's say that we create a dom0less domU with an emulated PCI root
complex without any devices under it, then after Dom0 is fully booted,
we add a virtio-net emulated device. How do we tell the guest what is
the backend domain id?

Device tree and other firmware tables are not relevant anymore.

We could reuse a PCI config space register to expose the backend id.
However this solution requires a backend change (QEMU) to expose the
backend id via an emulated register for each emulated device.

To avoid having to introduce a special config space register in all
emulated PCI devices (virtio-net, virtio-block, etc) I wonder if we
could add a special PCI config space register at the emulated PCI Root
Complex level.

Basically the workflow would be as follow:

- Linux recognizes the PCI Root Complex as a Xen PCI Root Complex
- Linux writes to special PCI config space register of the Xen PCI Root
  Complex the PCI device id (basically the BDF)
- The Xen PCI Root Complex emulated by Xen answers by writing back to
  the same location the backend id (domid of the backend)
- Linux reads back the same PCI config space register of the Xen PCI
  Root Complex and learn the relevant domid

What do you think?

Other ideas welcome!



> +     return of_node_get(dev->of_node);
> +}
> +
> +static bool xen_is_dt_grant_dma_device(struct device_node *np)
>  {
>       struct device_node *iommu_np;
>       bool has_iommu;
>  
> -     iommu_np = of_parse_phandle(dev->of_node, "iommus", 0);
> +     iommu_np = of_parse_phandle(np, "iommus", 0);
>       has_iommu = iommu_np &&
>                   of_device_is_compatible(iommu_np, "xen,grant-dma");
>       of_node_put(iommu_np);
> @@ -288,9 +305,17 @@ static bool xen_is_dt_grant_dma_device(struct device 
> *dev)
>  
>  bool xen_is_grant_dma_device(struct device *dev)
>  {
> +     struct device_node *np;
> +
>       /* XXX Handle only DT devices for now */
> -     if (dev->of_node)
> -             return xen_is_dt_grant_dma_device(dev);
> +     np = xen_dt_get_node(dev);
> +     if (np) {
> +             bool ret;
> +
> +             ret = xen_is_dt_grant_dma_device(np);
> +             of_node_put(np);
> +             return ret;
> +     }
>  
>       return false;
>  }
> @@ -303,20 +328,20 @@ bool xen_virtio_mem_acc(struct virtio_device *dev)
>       return xen_is_grant_dma_device(dev->dev.parent);
>  }
>  
> -static int xen_dt_grant_setup_dma_ops(struct device *dev,
> +static int xen_dt_grant_setup_dma_ops(struct device_node *np,
>                                      struct xen_grant_dma_data *data)
>  {
>       struct of_phandle_args iommu_spec;
>  
> -     if (of_parse_phandle_with_args(dev->of_node, "iommus", "#iommu-cells",
> +     if (of_parse_phandle_with_args(np, "iommus", "#iommu-cells",
>                       0, &iommu_spec)) {
> -             dev_err(dev, "Cannot parse iommus property\n");
> +             pr_err("%s: Cannot parse iommus property\n", np->name);
>               return -ESRCH;
>       }
>  
>       if (!of_device_is_compatible(iommu_spec.np, "xen,grant-dma") ||
>                       iommu_spec.args_count != 1) {
> -             dev_err(dev, "Incompatible IOMMU node\n");
> +             pr_err("%s: Incompatible IOMMU node\n", iommu_spec.np->name);
>               of_node_put(iommu_spec.np);
>               return -ESRCH;
>       }
> @@ -335,6 +360,7 @@ static int xen_dt_grant_setup_dma_ops(struct device *dev,
>  void xen_grant_setup_dma_ops(struct device *dev)
>  {
>       struct xen_grant_dma_data *data;
> +     struct device_node *np;
>  
>       data = find_xen_grant_dma_data(dev);
>       if (data) {
> @@ -346,8 +372,13 @@ void xen_grant_setup_dma_ops(struct device *dev)
>       if (!data)
>               goto err;
>  
> -     if (dev->of_node) {
> -             if (xen_dt_grant_setup_dma_ops(dev, data))
> +     np = xen_dt_get_node(dev);
> +     if (np) {
> +             int ret;
> +
> +             ret = xen_dt_grant_setup_dma_ops(np, data);
> +             of_node_put(np);
> +             if (ret)
>                       goto err;
>       } else if (IS_ENABLED(CONFIG_XEN_VIRTIO_FORCE_GRANT)) {
>               dev_info(dev, "Using dom0 as backend\n");
> -- 
> 2.25.1
>