On Mon, Oct 14, 2019 at 4:56 AM Kevin Hao <haoke...@gmail.com> wrote:
>
> In the current code, the API pci_get_domain_bus_and_slot() is used to
> get the root bus. But we would get the below call trace on rt kernel
> due to the spin_lock is used in the klist.
> BUG: sleeping function called from invalid context at
> kernel/locking/rtmutex.c:968
> in_atomic(): 1, irqs_disabled(): 128, pid: 1, name: swapper/0
> 3 locks held by swapper/0/1:
> #0: (____ptrval____) (per_cpu_ptr(&bh_lock.lock, cpu)){....}, at:
> __device_driver_lock+0x34/0x60
> #1: (____ptrval____) (pci_lock){....}, at:
> pci_bus_write_config_word+0x40/0xa8
> #2: (____ptrval____) (&k->k_lock){+.+.}, at: klist_next+0x24/0x110
> irq event stamp: 1293008
> hardirqs last enabled at (1293007): [<ffff000010a3dd48>]
> _raw_spin_unlock_irqrestore+0xa8/0xb8
> hardirqs last disabled at (1293008): [<ffff000010a3daa0>]
> _raw_spin_lock_irqsave+0x38/0x80
> softirqs last enabled at (1032902): [<ffff00001090bd74>]
> raw_unhash_sk+0x74/0xb0
> softirqs last disabled at (1032891): [<ffff00001090bd00>]
> raw_unhash_sk+0x0/0xb0
> Preemption disabled at:
> [<ffff00001059a198>] pci_bus_write_config_word+0x40/0xa8
> CPU: 8 PID: 1 Comm: swapper/0 Not tainted 5.2.20-rt9-yocto-preempt-rt+ #12
> Hardware name: Marvell OcteonTX CN96XX board (DT)
> Call trace:
> dump_backtrace+0x0/0x140
> show_stack+0x24/0x30
> dump_stack+0xbc/0x104
> ___might_sleep+0x16c/0x1d0
> rt_spin_lock+0x64/0x78
> klist_next+0x24/0x110
> bus_find_device+0x70/0xe0
> pci_get_dev_by_id+0x60/0x88
> pci_get_domain_bus_and_slot+0x5c/0xb0
> octeontx2_pem_config_write+0x84/0x3e0
> pci_bus_write_config_word+0x64/0xa8
> pci_write_config_word+0x44/0x68
> pci_setup_device+0x214/0x718
> pci_scan_single_device+0xb4/0xf8
> pci_scan_slot+0x44/0x108
> pci_scan_child_bus_extend+0x58/0x290
> pci_scan_bridge_extend+0x350/0x4f8
> pci_scan_child_bus_extend+0x1e4/0x290
> pci_scan_root_bus_bridge+0x60/0xd0
> pci_host_probe+0x20/0xb8
> pci_host_common_probe+0xf4/0x1f0
> octeontx2_pem_probe+0x2c/0x38
> platform_drv_probe+0x58/0xa8
> really_probe+0xe0/0x288
> driver_probe_device+0x5c/0xf0
> device_driver_attach+0x74/0x80
> __driver_attach+0x64/0xe0
> bus_for_each_dev+0x84/0xd8
> driver_attach+0x30/0x40
> bus_add_driver+0x190/0x1f0
> driver_register+0x64/0x110
> __platform_driver_register+0x54/0x60
> octeontx2_pem_driver_init+0x20/0x28
> do_one_initcall+0xa8/0x468
> kernel_init_freeable+0x514/0x5b8
> kernel_init+0x18/0x108
> ret_from_fork+0x10/0x1c
>
> Actually we can use a more lightweight API to get the rootbus without
> acquiring any lock, then fix this issue.
>
> Signed-off-by: Kevin Hao <haoke...@gmail.com>
> ---
>
> Hi Bruce,
>
> Please help merge this to the following branches:
>
> linux-yocto:
> v5.2/standard/cn96xx
> v5.2/standard/preempt-rt/cn96xx
merged
>
> linux-yocto-dev:
> standard/cn96xx
.. and merged. But I'm about a day away from replacing linux-yocto-dev
with v5.4-rc3, so be prepared that this will change shortly.
Bruce
>
> drivers/pci/controller/pci-octeontx2-pem.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/pci/controller/pci-octeontx2-pem.c
> b/drivers/pci/controller/pci-octeontx2-pem.c
> index 90b5537e8cdb..1e14dc065322 100644
> --- a/drivers/pci/controller/pci-octeontx2-pem.c
> +++ b/drivers/pci/controller/pci-octeontx2-pem.c
> @@ -335,10 +335,10 @@ static void octeontx2_be_workaround_init(struct pci_bus
> *bus)
> static void octeontx2_be_workaround(struct pci_bus *bus, int where,
> int size, u32 val)
> {
> - struct pci_dev *rc;
> + struct pci_host_bridge *rc;
> u32 reg, be = 0;
>
> - rc = pci_get_domain_bus_and_slot(pci_domain_nr(bus), 0, 0);
> + rc = pci_find_host_bridge(bus);
>
> /* Setup RAS to inject one error */
> octeontx2_be_workaround_init(rc->bus);
> --
> 2.14.4
>
--
- Thou shalt not follow the NULL pointer, for chaos and madness await
thee at its end
- "Use the force Harry" - Gandalf, Star Trek II
--
_______________________________________________
linux-yocto mailing list
linux-yocto@yoctoproject.org
https://lists.yoctoproject.org/listinfo/linux-yocto
