Re: request_irq() with local bh disabled

From: Eric Dumazet
Date: Thu Mar 06 2025 - 08:46:11 EST


On Thu, Mar 6, 2025 at 1:24 PM Borislav Petkov <bp@xxxxxxxxx> wrote:
>
> Hi,
>
> this is latest Linus/master + tip/master on a 32-bit x86:
>
> tglx says one cannot request_irq() with local bh disabled.
>
> [ 17.354927] cfg80211: Loading compiled-in X.509 certificates for regulatory database
> [ 17.906996] Loaded X.509 cert 'sforshee: 00b28ddf47aef9cea7'
> [ 17.950874] Loaded X.509 cert 'wens: 61c038651aabdcf94bd0ac7ff06c7248db18c600'
> [ 18.034884] platform regulatory.0: Direct firmware load for regulatory.db failed with error -2
> [ 18.038920] cfg80211: failed to load regulatory.db
>
> [ 18.726638] =============================
> [ 18.726638] [ BUG: Invalid wait context ]
> [ 18.726638] 6.14.0-rc5+ #1 Not tainted
> [ 18.726638] -----------------------------
> [ 18.726638] ip/991 is trying to lock:
> [ 18.726638] c36a2d64 (&desc->request_mutex){+.+.}-{4:4}, at: __setup_irq+0x98/0x6cc
> [ 18.726638] other info that might help us debug this:
> [ 18.746793] context-{5:5}
> [ 18.746793] 2 locks held by ip/991:
> [ 18.746793] #0: c261f920 (rtnl_mutex){+.+.}-{4:4}, at: rtnl_newlink+0x336/0x9fc
> [ 18.746793] #1: c1f56ea0 (local_bh){.+.+}-{1:3}, at: dev_set_rx_mode+0x5/0x80

Hmmm.. not sure why local_bh is considered held..

void dev_set_rx_mode(struct net_device *dev)
{
netif_addr_lock_bh(dev);
__dev_set_rx_mode(dev); // does not call ndo_open() (rtl_open in
your trace)
netif_addr_unlock_bh(dev);
}


Note that netif_addr_lock_bh() is using:

local_bh_disable();
spin_lock_nested(&dev->addr_list_lock, nest_level);

Would the following "debug" patch change anything for your tree ?
I know this looks strange, but I do not have anything more sensible to suggest.

diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index ab550a89b9bfaa5682e65f1dcc7f5f99ce90eb94..52e10d738bf0e35296a876011828cd0b40580678
100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -4751,7 +4751,8 @@ static inline void netif_addr_unlock(struct
net_device *dev)

static inline void netif_addr_unlock_bh(struct net_device *dev)
{
- spin_unlock_bh(&dev->addr_list_lock);
+ spin_unlock(&dev->addr_list_lock);
+ local_bh_enable();
}

/*



> [ 18.746793] stack backtrace:
> [ 18.746793] CPU: 1 UID: 0 PID: 991 Comm: ip Not tainted 6.14.0-rc5+ #1
> [ 18.746793] Hardware name: Acer AOA150/, BIOS v0.3309 10/06/2008
> [ 18.746793] Call Trace:
> [ 18.746793] dump_stack_lvl+0x94/0x10c
> [ 18.746793] dump_stack+0x13/0x18
> [ 18.746793] __lock_acquire+0xa1c/0x2500
> [ 18.746793] lock_acquire+0xc3/0x2ac
> [ 18.746793] ? __setup_irq+0x98/0x6cc
> [ 18.746793] ? debug_smp_processor_id+0x12/0x14
> [ 18.746793] ? __mutex_lock+0x54/0xcb8
> [ 18.746793] ? __mutex_lock+0x54/0xcb8
> [ 18.746793] ? trace_preempt_off+0x2e/0xb4
> [ 18.746793] ? __might_sleep+0x35/0x6c
> [ 18.746793] ? __mutex_lock+0x54/0xcb8
> [ 18.746793] ? preempt_count_add+0x6c/0xd4
> [ 18.746793] __mutex_lock+0x82/0xcb8
> [ 18.746793] ? __setup_irq+0x98/0x6cc
> [ 18.746793] mutex_lock_nested+0x27/0x2c
> [ 18.746793] ? __setup_irq+0x98/0x6cc
> [ 18.746793] __setup_irq+0x98/0x6cc
> [ 18.746793] ? __kmalloc_cache_noprof+0x1b1/0x2d0
> [ 18.746793] ? request_threaded_irq+0x84/0x188
> [ 18.746793] request_threaded_irq+0xc2/0x188
> [ 18.746793] rtl_open+0x33b/0x5e4 [r8169]
> [ 18.746793] ? raw_notifier_call_chain+0x20/0x24
> [ 18.746793] __dev_open+0xce/0x17c
> [ 18.746793] ? dev_set_rx_mode+0x74/0x80
> [ 18.746793] __dev_change_flags+0x176/0x1cc
> [ 18.746793] dev_change_flags+0x29/0x6c
> [ 18.746793] do_setlink.isra.0+0x28f/0x1180
> [ 18.746793] ? __mutex_lock+0x107/0xcb8
> [ 18.746793] ? __mutex_lock+0x107/0xcb8
> [ 18.746793] ? trace_preempt_on+0x2e/0xac
> [ 18.746793] ? __mutex_lock+0x107/0xcb8
> [ 18.746793] ? preempt_count_sub+0xb1/0x100
> [ 18.746793] ? debug_smp_processor_id+0x12/0x14
> [ 18.746793] ? __mutex_lock+0x107/0xcb8
> [ 18.746793] ? rtnl_newlink+0x336/0x9fc
> [ 18.746793] ? __kmalloc_cache_noprof+0x1b1/0x2d0
> [ 18.746793] ? do_alloc_pages+0x64/0xbc
> [ 18.746793] rtnl_newlink+0x762/0x9fc
> [ 18.746793] ? __this_cpu_preempt_check+0xf/0x20
> [ 18.746793] ? do_alloc_pages+0x64/0xbc
> [ 18.746793] ? do_setlink.isra.0+0x1180/0x1180
> [ 18.746793] rtnetlink_rcv_msg+0x3fd/0x584
> [ 18.746793] ? rtnetlink_rcv_msg+0x58/0x584
> [ 18.746793] ? netlink_deliver_tap.constprop.0+0xe5/0x4ac
> [ 18.746793] ? local_clock_noinstr+0x68/0x1c0
> [ 18.746793] ? rtnl_fdb_dump+0x370/0x370
> [ 18.746793] netlink_rcv_skb+0x42/0xdc
> [ 18.746793] ? do_alloc_pages+0x64/0xbc
> [ 18.746793] rtnetlink_rcv+0x12/0x14
> [ 18.746793] netlink_unicast+0x198/0x2a8
> [ 18.746793] netlink_sendmsg+0x1bb/0x3ec
> [ 18.746793] ? netlink_unicast+0x2a8/0x2a8
> [ 18.746793] ____sys_sendmsg+0x233/0x280
> [ 18.746793] ? netlink_unicast+0x2a8/0x2a8
> [ 18.746793] ? do_alloc_pages+0x64/0xbc
> [ 18.746793] ___sys_sendmsg+0x66/0x9c
> [ 18.746793] ? __might_fault+0x3b/0x84
> [ 18.746793] ? __might_fault+0x3b/0x84
> [ 18.746793] ? local_clock_noinstr+0x68/0x1c0
> [ 18.746793] ? do_alloc_pages+0x64/0xbc
> [ 18.746793] __sys_sendmsg+0x52/0x88
> [ 18.746793] ? _copy_from_user+0x51/0x60
> [ 18.746793] __ia32_sys_socketcall+0x30b/0x320
> [ 18.746793] ? __might_fault+0x7d/0x84
> [ 18.746793] ia32_sys_call+0x2695/0x284c
> [ 18.746793] __do_fast_syscall_32+0x67/0xf0
> [ 18.746793] do_fast_syscall_32+0x29/0x5c
> [ 18.746793] do_SYSENTER_32+0x15/0x18
> [ 18.746793] entry_SYSENTER_32+0x98/0xf9
> [ 18.746793] EIP: 0xb7edd579
> [ 18.746793] Code: b8 01 10 06 03 74 b4 01 10 07 03 74 b0 01 10 08 03 74 d8 01 00 00 00 00 00 00 00 00 00 00 00 00 00 51 52 55 89 e5 0f 34 cd 80 <5d> 5a 59 c3 90 90 90 90 8d 76 00 58 b8 77 00 00 00 cd 80 90 8d 76
> [ 18.746793] EAX: ffffffda EBX: 00000010 ECX: bff4ab70 EDX: 00000000
> [ 18.746793] ESI: b7e5f000 EDI: 005692a0 EBP: 00000010 ESP: bff4ab60
> [ 18.746793] DS: 007b ES: 007b FS: 0000 GS: 0033 SS: 007b EFLAGS: 00000282
> [ 18.874284] hpet: Lost 9 RTC interrupts
> [ 18.874867] RTL8201CP Ethernet r8169-0-200:00: attached PHY driver (mii_bus:phy_addr=r8169-0-200:00, irq=MAC)
> [ 18.932445] r8169 0000:02:00.0 eth0: Link is Down
> [ 19.264574] ath5k 0000:03:00.0: can't disable ASPM; OS doesn't have ASPM control
> [ 19.269295] ath5k 0000:03:00.0: registered as 'phy0'
> [ 19.827560] ath: EEPROM regdomain: 0x65
> [ 19.829929] ath: EEPROM indicates we should expect a direct regpair map
> [ 19.832273] ath: Country alpha2 being used: 00
> [ 19.834396] ath: Regpair used: 0x65
> [ 19.837790] ieee80211 phy0: Selected rate control algorithm 'minstrel_ht'
> [ 19.854976] ath5k: phy0: Atheros AR2425 chip found (MAC: 0xe2, PHY: 0x70)
> [ 20.566012] r8169 0000:02:00.0 eth0: Link is Up - 100Mbps/Full - flow control rx/tx
> [ 20.966578] NET: Registered PF_INET6 protocol family
>
> --
> Regards/Gruss,
> Boris.
>
> https://people.kernel.org/tglx/notes-about-netiquette