Am Samstag, 8. April 2023, 01:58:29 CEST schrieb Bjorn Helgaas:
> [+cc Korneliusz, Alexander]
>
> On Tue, Mar 14, 2023 at 01:37:33AM +0800, You-Sheng Yang (vicamo) wrote:
> > VMD may fail to create sysfs entries while `pci_rescan_bus()` called in
> >
> > some other drivers like t7xx wwan driver:
> > sysfs: cannot create duplicate filename '/devices/.../resource0'
> >
> > Call Trace:
> > <TASK>
> > sysfs_warn_dup.cold+0x17/0x34
> > sysfs_add_bin_file_mode_ns+0xc0/0xf0
> > sysfs_create_bin_file+0x6d/0xb0
> > pci_create_attr+0x117/0x260
> > pci_create_resource_files+0x6b/0x150
> > pci_create_sysfs_dev_files+0x18/0x30
> > pci_bus_add_device+0x30/0x80
> > pci_bus_add_devices+0x31/0x80
> > pci_bus_add_devices+0x5b/0x80
> > vmd_enable_domain.constprop.0+0x6b7/0x880 [vmd]
> > vmd_probe+0x16d/0x193 [vmd]
>
> This is a long-standing issue, and I would *love* to nail it, but this
> doesn't feel like the right solution to me. What's unique about vmd
> here?
>
> I guess maybe it's similar to the situation Korneliusz and Alexander
> ran into at [1]?
What's the other path creating those sysfs files? Also vmd seems to be built
as a module in this case. So that's a different issue. pci-imx6 is racing
against an initcall when being built into the kernel. The probe_type
PROBE_PREFER_ASYNCHRONOUS is also important.
Being built as a module this race is essentially impossible.
Best regards,
Alexander
> And why is t7xx called out specifically here? This is a pretty
> generic sysfs attribute issue and it doesn't *seem* like t7xx should
> be special in that respect. Oooh, maybe it's the fact that there's a
> t7xx patch [2] coming that adds pci_rescan_bus() there?
>
> Krzysztof has converted a lot of the sysfs files to static attributes,
> e.g.,
>
> 506140f9c06b ("PCI/sysfs: Convert "index", "acpi_index", "label" to static
> attributes") d93f8399053d ("PCI/sysfs: Convert "vpd" to static attribute")
> f42c35ea3b13 ("PCI/sysfs: Convert "reset" to static attribute")
> 527139d738d7 ("PCI/sysfs: Convert "rom" to static attribute")
> e1d3f3268b0e ("PCI/sysfs: Convert "config" to static attribute")
>
> I think that's the *best* approach since the sysfs infrastructure
> already prevents races here. But these last few files are more
> difficult to convert, so we've been kind of stalled on them.
>
> Bjorn
>
> [1]
> https://lore.kernel.org/r/20230316091540.494366-1-alexander.stein@xxxxx-gro
> up.com [2]
> https://lore.kernel.org/linux-pci/20230124204543.550d88e3@xxxxxxxxxx/
> > Fixes: 185a383ada2e ("x86/PCI: Add driver for Intel Volume Management
> > Device (VMD)") Signed-off-by: You-Sheng Yang (vicamo)
> > <vicamo.yang@xxxxxxxxxxxxx> ---
> >
> > drivers/pci/controller/vmd.c | 6 ++++++
> > 1 file changed, 6 insertions(+)
> >
> > diff --git a/drivers/pci/controller/vmd.c b/drivers/pci/controller/vmd.c
> > index 769eedeb8802..f050991bd1e9 100644
> > --- a/drivers/pci/controller/vmd.c
> > +++ b/drivers/pci/controller/vmd.c
> > @@ -838,9 +838,13 @@ static int vmd_enable_domain(struct vmd_dev *vmd,
> > unsigned long features)>
> > pci_add_resource_offset(&resources, &vmd->resources[1], offset[0]);
> > pci_add_resource_offset(&resources, &vmd->resources[2], offset[1]);
> >
> > + pci_lock_rescan_remove();
> > +
> >
> > vmd->bus = pci_create_root_bus(&vmd->dev->dev, vmd->busn_start,
> >
> > &vmd_ops, sd, &resources);
> >
> > if (!vmd->bus) {
> >
> > + pci_unlock_rescan_remove();
> > +
> >
> > pci_free_resource_list(&resources);
> > vmd_remove_irq_domain(vmd);
> > return -ENODEV;
> >
> > @@ -893,6 +897,8 @@ static int vmd_enable_domain(struct vmd_dev *vmd,
> > unsigned long features)>
> > vmd_acpi_end();
> >
> > + pci_unlock_rescan_remove();
> > +
> >
> > WARN(sysfs_create_link(&vmd->dev->dev.kobj, &vmd->bus->dev.kobj,
> >
> > "domain"), "Can't create symlink to
domain\n");
> >
> > return 0;
--
TQ-Systems GmbH | Mühlstraße 2, Gut Delling | 82229 Seefeld, Germany
Amtsgericht München, HRB 105018
Geschäftsführer: Detlef Schneider, Rüdiger Stahl, Stefan Schneider
http://www.tq-group.com/
