Re: [PATCH] wifi: wilc1000: unregister wiphy only after netdev registration

[Kalle Valo <kvalo@kernel.org>]

Alexis Lothoré <alexis.lothore@bootlin.com> writes:

> wiphy_unregister/wiphy_free has been recently decoupled from
> wilc_netdev_cleanup to fix a faulty error path in sdio/spi probe
> functions. However this change introduced a new failure when simply
> loading then unloading the driver:
>   $ modprobe wilc1000-sdio; modprobe -r wilc1000-sdio
>   WARNING: CPU: 0 PID: 115 at net/wireless/core.c:1145 wiphy_unregister+0x904/0xc40 [cfg80211]
>   Modules linked in: wilc1000_sdio(-) wilc1000 cfg80211 bluetooth ecdh_generic ecc
>   CPU: 0 UID: 0 PID: 115 Comm: modprobe Not tainted 6.13.0-rc6+ #45
>   Hardware name: Atmel SAMA5
>   Call trace:
>    unwind_backtrace from show_stack+0x18/0x1c
>    show_stack from dump_stack_lvl+0x44/0x70
>    dump_stack_lvl from __warn+0x118/0x27c
>    __warn from warn_slowpath_fmt+0xcc/0x140
>    warn_slowpath_fmt from wiphy_unregister+0x904/0xc40 [cfg80211]
>    wiphy_unregister [cfg80211] from wilc_sdio_remove+0xb0/0x15c [wilc1000_sdio]
>    wilc_sdio_remove [wilc1000_sdio] from sdio_bus_remove+0x104/0x3f0
>    sdio_bus_remove from device_release_driver_internal+0x424/0x5dc
>    device_release_driver_internal from driver_detach+0x120/0x224
>    driver_detach from bus_remove_driver+0x17c/0x314
>    bus_remove_driver from sys_delete_module+0x310/0x46c
>    sys_delete_module from ret_fast_syscall+0x0/0x1c
>   Exception stack(0xd0acbfa8 to 0xd0acbff0)
>   bfa0:                   0044b210 0044b210 0044b24c 00000800 00000000 00000000
>   bfc0: 0044b210 0044b210 00000000 00000081 00000000 0044b210 00000000 00000000
>   bfe0: 00448e24 b6af99c4 0043ea0d aea2e12c
>   irq event stamp: 0
>   hardirqs last  enabled at (0): [<00000000>] 0x0
>   hardirqs last disabled at (0): [<c01588f0>] copy_process+0x1c4c/0x7bec
>   softirqs last  enabled at (0): [<c0158944>] copy_process+0x1ca0/0x7bec
>   softirqs last disabled at (0): [<00000000>] 0x0
>
> The warning is triggered by the fact that there is still a
> wireless_device linked to the wiphy we are unregistering, due to
> wiphy_unregister now being called after net device unregister (performed
> in wilc_netdev_cleanup). Fix this warning by moving wiphy_unregister
> after wilc_netdev_cleanup is nominal paths (ie: driver removal).
> wilc_netdev_cleanup ordering is left untouched in error paths in probe
> function because net device is not registered in those paths (so the
> warning can not trigger), yet the wiphy can still be registered, and we
> still some cleanup steps from wilc_netdev_cleanup.
>
> Signed-off-by: Alexis Lothoré <alexis.lothore@bootlin.com>
> ---
> I clearly overlooked this simple scenario/misunderstood expected
> unregistration order when fixing some spi probe error path, my bad (see
> commit 89a7616e1715 ("ARM: dts: at91-sama5d27_wlsom1: update bluetooth
> chip description") in wireless-next)

No worries, bugs are business as usual.

> @Kalle: 89a7616e1715 (the faulty commit) is only in wireless-next for
> now IIUC, so I did not provide any Fixes: tag to prevent any faulty SHA1
> if those commits end up being picked in stable tree (however, the faulty
> commit _has_ a Fixes tag). Please let me know if we should proceed
> differently.

Hmm, I don't really follow you here. I feel that always adding the Fixes
tag is the safest option, that way it's clear for everyone what commit
we are fixing. So if it's ok for you, I would like to add the Fixes tag
but I can't find commit 89a7616e1715 anywhere.

-- 
https://patchwork.kernel.org/project/linux-wireless/list/

https://wireless.wiki.kernel.org/en/developers/documentation/submittingpatches