Re: [KERNEL PATCH v9 3/3] xen/privcmd: Add new syscall to get gsi from dev

[Juergen Gross <jgross@xxxxxxxx>]

On 23.09.24 09:58, Juergen Gross wrote:
> On 23.09.24 07:49, Chen, Jiqian wrote:
> > On 2024/9/21 05:17, Stefano Stabellini wrote:
> > > On Fri, 20 Sep 2024, Chen, Jiqian wrote:
> > > > On 2024/9/19 06:49, Stefano Stabellini wrote:
> > > > > On Thu, 12 Sep 2024, Jiqian Chen wrote:
> > > > > > On PVH dom0, when passthrough a device to domU, QEMU and xl tools
> > > > > > want to use gsi number to do pirq mapping, see QEMU code
> > > > > > xen_pt_realize->xc_physdev_map_pirq, and xl code
> > > > > > pci_add_dm_done->xc_physdev_map_pirq, but in current codes, the gsi
> > > > > > number is got from file /sys/bus/pci/devices/<sbdf>/irq, that is
> > > > > > wrong, because irq is not equal with gsi, they are in different
> > > > > > spaces, so pirq mapping fails.
> > > > > > And in current linux codes, there is no method to get gsi
> > > > > > for userspace.
> > > > > >
> > > > > > For above purpose, record gsi of pcistub devices when init
> > > > > > pcistub and add a new syscall into privcmd to let userspace
> > > > > > can get gsi when they have a need.
> > > > > >
> > > > > > Signed-off-by: Jiqian Chen <Jiqian.Chen@xxxxxxx>
> > > > > > Signed-off-by: Huang Rui <ray.huang@xxxxxxx>
> > > > > > Signed-off-by: Jiqian Chen <Jiqian.Chen@xxxxxxx>
> > > > > > ---
> > > > > > v8->v9 changes:
> > > > > > Changed the syscall name from "IOCTL_PRIVCMD_GSI_FROM_DEV" to 
> > > > > > "IOCTL_PRIVCMD_PCIDEV_GET_GSI". Also changed the other functions name.
> > > > > > Changed the macro wrapping "pcistub_get_gsi_from_sbdf" from 
> > > > > > "CONFIG_XEN_ACPI" to "CONFIG_XEN_PCIDEV_BACKEND" to fix compile errors 
> > > > > > reported by CI robot.
> > > > > > Changed the parameter gsi of struct privcmd_pcidev_get_gsi from int to u32.
> > > > > >
> > > > > > v7->v8 changes:
> > > > > > In function privcmd_ioctl_gsi_from_dev, return -EINVAL when not confige 
> > > > > > CONFIG_XEN_ACPI.
> > > > > > Used PCI_BUS_NUM PCI_SLOT PCI_FUNC instead of open coding.
> > > > > >
> > > > > > v6->v7 changes:
> > > > > > Changed implementation to add a new parameter "gsi" to struct 
> > > > > > pcistub_device and set gsi when pcistub initialize device. Then when 
> > > > > > userspace wants to get gsi and pass sbdf, we can return that gsi.
> > > > > > v5->v6 changes:
> > > > > > Changed implementation to add a new syscall to translate irq to gsi, 
> > > > > > instead adding a new gsi sysfs node, because the pci Maintainer didn't 
> > > > > > allow to add that sysfs node.
> > > > > > v3->v5 changes:
> > > > > > No.
> > > > > >
> > > > > > v2->v3 changes:
> > > > > > Suggested by Roger: Abandoned previous implementations that added new 
> > > > > > syscall to get gsi from irq and changed to add a new sysfs node for gsi, 
> > > > > > then userspace can get gsi number from sysfs node.
> > > > > > ---
> > > > > > | Reported-by: kernel test robot <lkp@xxxxxxxxx>
> > > > > > | Closes: 
> > > > > > https://lore.kernel.org/oe-kbuild-all/202406090826.whl6Cb7R-lkp@xxxxxxxxx/
> > > > > > ---
> > > > > > | Reported-by: kernel test robot <lkp@xxxxxxxxx>
> > > > > > | Closes: 
> > > > > > https://lore.kernel.org/oe-kbuild-all/202405171113.T431PC8O-lkp@xxxxxxxxx/
> > > > > > ---
> > > > > >   drivers/xen/privcmd.c              | 30 +++++++++++++++++++++++
> > > > > >   drivers/xen/xen-pciback/pci_stub.c | 38 +++++++++++++++++++++++++++---
> > > > > >   include/uapi/xen/privcmd.h         |  7 ++++++
> > > > > >   include/xen/acpi.h                 |  9 +++++++
> > > > > >   4 files changed, 81 insertions(+), 3 deletions(-)
> > > > > >
> > > > > > diff --git a/drivers/xen/privcmd.c b/drivers/xen/privcmd.c
> > > > > > index 9563650dfbaf..1ed612d21543 100644
> > > > > > --- a/drivers/xen/privcmd.c
> > > > > > +++ b/drivers/xen/privcmd.c
> > > > > > @@ -46,6 +46,9 @@
> > > > > >   #include <xen/page.h>
> > > > > >   #include <xen/xen-ops.h>
> > > > > >   #include <xen/balloon.h>
> > > > > > +#ifdef CONFIG_XEN_ACPI
> > > > > > +#include <xen/acpi.h>
> > > > > > +#endif
> > > > > >   #include "privcmd.h"
> > > > > > @@ -844,6 +847,29 @@ static long privcmd_ioctl_mmap_resource(struct file 
> > > > > > *file,
> > > > > >       return rc;
> > > > > >   }
> > > > > > +static long privcmd_ioctl_pcidev_get_gsi(struct file *file, void __user 
> > > > > > *udata)
> > > > > > +{
> > > > > > +#ifdef CONFIG_XEN_ACPI
> > > > > > +    int rc;
> > > > > > +    struct privcmd_pcidev_get_gsi kdata;
> > > > > > +
> > > > > > +    if (copy_from_user(&kdata, udata, sizeof(kdata)))
> > > > > > +        return -EFAULT;
> > > > > > +
> > > > > > +    rc = pcistub_get_gsi_from_sbdf(kdata.sbdf);
> > > > > > +    if (rc < 0)
> > > > > > +        return rc;
> > > > > > +
> > > > > > +    kdata.gsi = rc;
> > > > > > +    if (copy_to_user(udata, &kdata, sizeof(kdata)))
> > > > > > +        return -EFAULT;
> > > > > > +
> > > > > > +    return 0;
> > > > > > +#else
> > > > > > +    return -EINVAL;
> > > > > > +#endif
> > > > > > +}
> > > > > > +
> > > > > >   #ifdef CONFIG_XEN_PRIVCMD_EVENTFD
> > > > > >   /* Irqfd support */
> > > > > >   static struct workqueue_struct *irqfd_cleanup_wq;
> > > > > > @@ -1543,6 +1569,10 @@ static long privcmd_ioctl(struct file *file,
> > > > > >           ret = privcmd_ioctl_ioeventfd(file, udata);
> > > > > >           break;
> > > > > > +    case IOCTL_PRIVCMD_PCIDEV_GET_GSI:
> > > > > > +        ret = privcmd_ioctl_pcidev_get_gsi(file, udata);
> > > > > > +        break;
> > > > > > +
> > > > > >       default:
> > > > > >           break;
> > > > > >       }
> > > > > > diff --git a/drivers/xen/xen-pciback/pci_stub.c 
> > > > > > b/drivers/xen/xen-pciback/pci_stub.c
> > > > > > index 8ce27333f54b..2ea8e4075adc 100644
> > > > > > --- a/drivers/xen/xen-pciback/pci_stub.c
> > > > > > +++ b/drivers/xen/xen-pciback/pci_stub.c
> > > > > > @@ -56,6 +56,9 @@ struct pcistub_device {
> > > > > >       struct pci_dev *dev;
> > > > > >       struct xen_pcibk_device *pdev;/* non-NULL if struct pci_dev is in 
> > > > > > use */
> > > > > > +#ifdef CONFIG_XEN_ACPI
> > > > > > +    int gsi;
> > > > > > +#endif
> > > > > >   };
> > > > > >   /* Access to pcistub_devices & seized_devices lists and the 
> > > > > > initialize_devices
> > > > > > @@ -88,6 +91,9 @@ static struct pcistub_device 
> > > > > > *pcistub_device_alloc(struct pci_dev *dev)
> > > > > >       kref_init(&psdev->kref);
> > > > > >       spin_lock_init(&psdev->lock);
> > > > > > +#ifdef CONFIG_XEN_ACPI
> > > > > > +    psdev->gsi = -1;
> > > > > > +#endif
> > > > > >       return psdev;
> > > > > >   }
> > > > > > @@ -220,6 +226,25 @@ static struct pci_dev 
> > > > > > *pcistub_device_get_pci_dev(struct xen_pcibk_device *pdev,
> > > > > >       return pci_dev;
> > > > > >   }
> > > > > > +#ifdef CONFIG_XEN_PCIDEV_BACKEND
> > > > >
> > > > > This breaks configurations without CONFIG_ACPI and with
> > > > > CONFIG_XEN_PCIDEV_BACKEND.
> > > > >
> > > > > Also there should be no dependency between PCIDEV_BACKEND and
> > > > > pcistub_get_gsi_from_sbdf.
> > > > >
> > > > > I think we should solve the build issues this way:
> > > > >
> > > > > - privcmd_ioctl_pcidev_get_gsi should have:
> > > > > #if defined(CONFIG_XEN_ACPI) && defined(CONFIG_XEN_PCI_STUB)
> > > > >
> > > > > - here we should have #ifdef CONFIG_XEN_ACPI as you had before
> > > > >
> > > > >
> > > > > As far as I can tell the above should be able to address all valid
> > > > > combinations.
> > > > This can't pass a combination that:
> > > > CONFIG_XEN_ACPI=y
> > > > CONFIG_XEN_PCI_STUB=y
> > > > CONFIG_XEN_PCIDEV_BACKEND=m
> > > > Reported by robot 
> > > > https://lore.kernel.org/oe-kbuild-all/202406090826.whl6Cb7R-lkp@xxxxxxxxx/
> > > > At this combination, privcmd_ioctl_pcidev_get_gsi can call 
> > > > pcistub_get_gsi_from_sbdf because CONFIG_XEN_ACPI and CONFIG_XEN_PCI_STUB 
> > > > are both "y".
> > > > But when the compiler tries to find the implementation of 
> > > > pcistub_get_gsi_from_sbdf, it fails (ld: vmlinux.o: in function 
> > > > `privcmd_ioctl_pcidev_get_gsi':
> > > > /home/cjq/code/upstream/kernel_test_robot/linux_xen/build_dir/../drivers/xen/privcmd.c:859: 
> > > > undefined reference to `pcistub_get_gsi_from_sbdf'), because the value of CONFIG_XEN_PCIDEV_BACKEND 
> > > > is "m" (file drivers/xen/xen-pciback/Makefile shows 
> > > > "obj-$(CONFIG_XEN_PCIDEV_BACKEND) += xen-pciback.o"), so that xen-pciback is built as a 
> > > > module.
> > >
> > > The attached .config, with the appended changes on top of your patch
> > > work for me. I have CONFIG_XEN_PCIDEV_BACKEND=m.
> > Using your attached .config, it also works for me.
> > The difference between your .config and the config reported by Robot 
> > (https://download.01.org/0day-ci/archive/20240609/202406090826.whl6Cb7R-lkp@xxxxxxxxx/config) is that your CONFIG_XEN_PRIVCMD=m, but Robot's CONFIG_XEN_PRIVCMD=y.
> > It seems that, with my patch changes, privcmd uses the function implemented by 
> > xen-pciback, so if the configuration is "CONFIG_XEN_PRIVCMD=m & 
> > CONFIG_XEN_PCIDEV_BACKEND=m" or "CONFIG_XEN_PRIVCMD=y & 
> > CONFIG_XEN_PCIDEV_BACKEND=y", the compilation is OK, but if the configuration 
> > is "CONFIG_XEN_PRIVCMD=y & CONFIG_XEN_PCIDEV_BACKEND=m", issue happens.
> > You can change CONFIG_XEN_PRIVCMD from "m" to "y" in your .config, then you 
> > will fail at:
> > ld: vmlinux.o: in function `privcmd_ioctl':
> > privcmd.c:(.text+0x665e42): undefined reference to `pcistub_get_gsi_from_sbdf'
> You can do:
>
>      if (IS_REACHABLE(CONFIG_XEN_PCIDEV_BACKEND)) {
>          rc = pcistub_get_gsi_from_sbdf(kdata.sbdf);
>          if (rc < 0)
>              return rc;
>      }
>
> See Documentation/kbuild/kconfig-language.rst in the kernel source tree.
Oh, I think this will need:

        imply CONFIG_XEN_PCIDEV_BACKEND

in the CONFIG_XEN_PRIVCMD Kconfig definition.


Juergen

Attachment: OpenPGP_0xB0DE9DD628BF132F.asc
Description: OpenPGP public key

Attachment: OpenPGP_signature.asc
Description: OpenPGP digital signature