[PATCH v1] arm64:pci: fix the IOV device enabled crash issue in designware

Hi Bjorn,

Best regards,
Liu Po

 -----Original Message-----
 From: Bjorn Helgaas [mailto:helgaas at kernel.org]
 Sent: Friday, August 26, 2016 2:10 AM
 To: Po Liu
 Cc: S?ren Brinkmann; Roy Zang; Arnd Bergmann; linux-pci at vger.kernel.org;
 Pratyush Anand; Stuart Yoder; Yang-Leo Li; Mingkai Hu; Jingoo Han; Bjorn
 Helgaas; Ley Foon Tan; Michal Simek; linux-arm-
 kernel at lists.infradead.org
 Subject: Re: [PATCH v1] arm64:pci: fix the IOV device enabled crash
 issue in designware
 
 On Thu, Aug 25, 2016 at 04:53:19AM +0000, Po Liu wrote:
 > >  -----Original Message-----
 > >  From: Bjorn Helgaas [mailto:helgaas at kernel.org]
 > >  Sent: Thursday, August 25, 2016 4:51 AM
 > >  To: Po Liu
 > >  Cc: linux-pci at vger.kernel.org; Roy Zang; Arnd Bergmann; Stuart
 > > Yoder;  Yang-Leo Li; linux-arm-kernel at lists.infradead.org; Bjorn
 > > Helgaas;  Mingkai Hu; Ley Foon Tan; Michal Simek; S?ren Brinkmann;
 > > Jingoo Han;  Pratyush Anand
 > >  Subject: Re: [PATCH v1] arm64:pci: fix the IOV device enabled crash
 > > issue in designware
 > >
 > >  [+cc Jingoo, Pratyush, Michal, S?ren, Ley]
 > >
 > >  On Tue, Aug 23, 2016 at 02:01:12PM +0800, Po Liu wrote:
 > >  > When echo a number to /sys/bus/pci/devices/xxx/sriov_numvfs to
 > > enable  > the VF devices. A crash log occurred. This found to be
 > > access the IOV  > devices config space failure issue.
 > >
 > >  What was the actual crash?  The mere fact that we made a config
 > > read  fail should not cause a crash.  We might erroneously prevent
 > > access to  VF devices, but it shouldn't crash.  So maybe there's
 > > another bug  elsewhere that we should fix first.
 >
 > I built with CONFIG_PCI_IOV=y and notice a crash when I use it:
 >
 > centqds-60 cd /sys/class/net/
 > centqds-61 ls
 > enP1p1s0@ enP2p1s0f0@ enP2p1s0f1@ lo@ sit0@
 > centqds-62 cd enP2p1s0f1/device
 > centqds-63 ls
 > broken_parity_status driver_override msi_irqs/ sriov_numvfs class
 > enable net/ sriov_totalvfs config iommu_group@ power/ subsystem@
 > consistent_dma_mask_bits irq remove subsystem_device device
 > local_cpulist rescan subsystem_vendor devspec local_cpus reset uevent
 > dma_mask_bits modalias resource vendor driver@ msi_bus rom
 > centqds-64 zcat /proc/config.gz | grep _IOV CONFIG_PCI_IOV=y
 > centqds-65 sudo su
 > [root at centqds 0002:01:00.1]# echo 2 > sriov_numvfs [ 317.604543] ixgbe
 > 0002:01:00.1 enP2p1s0f1: SR-IOV enabled with 2 VFs [ 317.714431]
 > (null): of_irq_parse_pci() failed with rc=134
 
 
 > [ 317.719906] -----------[ cut here ]----------- [ 317.724525]
 > WARNING: CPU: 6 PID: 3179 at drivers/pci/probe.c:1555
 > pci_device_add+0x144/0x148() [ 317.733123] Modules linked in:
 > [ 317.736175] CPU: 6 PID: 3179 Comm: bash Not tainted
 > 4.1.8-00024-g0a32d65-dirty #32 [ 317.743731] Hardware name: Freescale
 > Layerscape 2088a QDS Board (DT) [ 317.750077] Call trace:
 > [ 317.752516] [<fffffe0000096cf8>] dump_backtrace+0x0/0x12c Message
 > from[ 317.757916] [<fffffe0000096e34>] show_stack+0x10/0x1c
 > syslogd at centqds[ 317.764341] [<fffffe00008937fc>] dump_stack+0x84/0xd4
 > at Jul 26 15:51[ 317.770770] [<fffffe00000be590>]
 > warn_slowpath_common+0x94/0xcc
 > :10 ...
 > kerne[ 317.778067] [<fffffe00000be68c>] warn_slowpath_null+0x14/0x20
 > l:Call trace:
 > [ 317.785192] [<fffffe00003c5fa4>] pci_device_add+0x140/0x148 [
 > 317.792133] [<fffffe00003de784>] pci_enable_sriov+0x470/0x7a0 [
 > 317.797873] [<fffffe00005fa9dc>] ixgbe_pci_sriov_configure+0x8c/0x148
 > [ 317.804302] [<fffffe00003cf354>] sriov_numvfs_store+0x78/0x11c [
 > 317.810129] [<fffffe0000443d80>] dev_attr_store+0x14/0x28 [
 > 317.815521] [<fffffe0000200164>] sysfs_kf_write+0x40/0x4c [
 > 317.820908] [<fffffe00001ff5ec>] kernfs_fop_write+0xb8/0x180 [
 > 317.826561] [<fffffe000019a568>] __vfs_write+0x28/0x10c [ 317.831775]
 > [<fffffe000019ad00>] vfs_write+0x90/0x1a0 [ 317.836819]
 > [<fffffe000019b584>] SyS_write+0x40/0xa0 [ 317.841772] --[ end trace
 > 83725a9784fd702a ]-- [ 317.846393] BUG: failure at
 > fs/sysfs/file.c:481/sysfs_create_bin_file()!
 > [ 317.853081] Kernel panic - not syncing: BUG!
 > [ 317.857339] CPU: 6 PID: 3179 Comm: bash Tainted: G W
 > 4.1.8-00024-g0a32d65-dirty #32 [ 317.866110] Hardware name: Freescale
 > Layerscape 2088a QDS Board (DT) [ 317.872451] Call trace:
 > [ 317.874887] [<fffffe0000096cf8>] dump_backtrace+0x0/0x12c [
 > 317.880274] [<fffffe0000096e34>] show_stack+0x10/0x1c [ 317.885315]
 > [<fffffe00008937fc>] dump_stack+0x84/0xd4 [ 317.890354]
 > [<fffffe00008909e8>] panic+0xe4/0x21c [ 317.895047]
 > [<fffffe0000200988>] sysfs_create_bin_file+0x60/0x64 [ 317.901041]
 > [<fffffe00003cf52c>] pci_create_sysfs_dev_files+0x48/0x2a8
 > [ 317.907556] [<fffffe00003c4238>] pci_bus_add_device+0x20/0x6c
 >
 > The code process is that: "echo 2 > sriov_numvf" makes driver
 load .sriov_configure. At last to load pci_enable_sriov().
 > The first time vf device operate the config space in the
 pci_setup_device() (this function was load in the virtfn_add()) is
 pci_read_config_byte(dev, PCI_HEADER_TYPE, &hdr_type) return failure. So
 the virtfn didn't initialized proper.
 >
 > This found to be "bus->primary == pp->root_bus_nr && dev > 0" then
 return failure in host controller. The dev came from devfn must not
 zero(is about 0x10).
 >
 > then read config space failure. This makes the dev->bus is NULL. Lead
 to upper crash.
 
 I think the crash (BUG: failure at
 fs/sysfs/file.c:481/sysfs_create_bin_file())
 happens in this path:
 
   sriov_numvfs_store
     ixgbe_pci_sriov_configure
       ixgbe_pci_sriov_enable
         pci_enable_sriov
           sriov_enable
             pci_iov_add_virtfn
               virtfn = pci_alloc_dev()
               pci_setup_device(virtfn)
                 if (pci_read_config_byte(dev, PCI_HEADER_TYPE,
 &hdr_type))
                   return -EIO
               pci_bus_add_device(virtfn)
                 pci_create_sysfs_dev_files
                   sysfs_create_bin_file
                     BUG_ON(!kobj)
 
 If the config read of PCI_HEADER_TYPE fails, pci_setup_device() returns
 -EIO, but pci_iov_add_virtfn() doesn't check it.  Can you update
 pci_iov_add_virtfn() so it checks that return value?  That should fix
 the crash, even without your designware patch.
 
 Obviously, it won't make SR-IOV work, so we still need both patches.

Thanks, I'll provide patch in V2.

 
 Bjorn