netdev - Re: [PATCH net-next] net: Add lockdep asserts to ____napi

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <YjRZY6CXOJPCAoNK@linutronix.de>
Date:   Fri, 18 Mar 2022 11:05:23 +0100
From:   Sebastian Andrzej Siewior <bigeasy@...utronix.de>
To:     Saeed Mahameed <saeed@...nel.org>
Cc:     netdev@...r.kernel.org, "David S. Miller" <davem@...emloft.net>,
        Jakub Kicinski <kuba@...nel.org>,
        Eric Dumazet <edumazet@...gle.com>,
        Thomas Gleixner <tglx@...utronix.de>,
        Peter Zijlstra <peterz@...radead.org>
Subject: Re: [PATCH net-next] net: Add lockdep asserts to ____napi_schedule().

On 2022-03-17 12:21:45 [-0700], Saeed Mahameed wrote:
> On 11 Mar 16:03, Sebastian Andrzej Siewior wrote:
> > ____napi_schedule() needs to be invoked with disabled interrupts due to
> > __raise_softirq_irqoff (in order not to corrupt the per-CPU list).
> 
> This patch is likely causing the following call trace when RPS is enabled:
> 
>  [  690.429122] WARNING: CPU: 0 PID: 0 at net/core/dev.c:4268 rps_trigger_softirq+0x21/0xb0
>  [  690.431236] Modules linked in: bonding ib_ipoib ipip tunnel4 geneve ib_umad ip6_gre ip6_tunnel tunnel6 rdma_ucm nf_tables ip_gre gre mlx5_ib ib_uverbs mlx5_core iptable_raw openvswitch nsh xt_conntrack xt_MASQUERADE nf_conntrack_netlink nfnetlink rpcrdma ib_iser xt_addrtype libiscsi scsi_transport_iscsi rdma_cm iw_cm iptable_nat nf_nat ib_cm br_netfilter ib_core overlay fuse [last unloaded: ib_uverbs]
>  [  690.439693] CPU: 0 PID: 0 Comm: swapper/0 Not tainted 5.17.0-rc7_net_next_4303f9c #1
>  [  690.441709] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.13.0-0-gf21b5a4aeb02-prebuilt.qemu.org 04/01/2014
>  [  690.444587] RIP: 0010:rps_trigger_softirq+0x21/0xb0
>  [  690.445971] Code: ff 0f 1f 84 00 00 00 00 00 0f 1f 44 00 00 8b 05 b1 ea 21 01 53 48 89 fb 85 c0 74 1b 65 8b 05 16 56 71 7e a9 00 ff 0f 00 75 02 <0f> 0b 65 8b 05 4e 5f 72 7e 85 c0 74 5b 48 8b 83 e0 01 00 00 f6 c4
>  [  690.450682] RSP: 0018:ffffffff82803e70 EFLAGS: 00010046
>  [  690.452106] RAX: 0000000000000001 RBX: ffff88852ca3d400 RCX: ffff88852ca3d540
>  [  690.453972] RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88852ca3d400
>  [  690.455860] RBP: 0000000000000000 R08: ffff88852ca3d400 R09: 0000000000000001
>  [  690.457684] R10: 0000000000000001 R11: 0000000000000000 R12: 0000000000000000
>  [  690.459548] R13: ffff88852ca3d540 R14: ffffffff82829628 R15: 0000000000000000
>  [  690.461429] FS:  0000000000000000(0000) GS:ffff88852ca00000(0000) knlGS:0000000000000000
>  [  690.463653] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>  [  690.465180] CR2: 00007ff718200b98 CR3: 000000013b4de003 CR4: 0000000000370eb0
>  [  690.467022] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
>  [  690.468915] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
>  [  690.470742] Call Trace:
>  [  690.471544]  <TASK>
>  [  690.472242]  flush_smp_call_function_queue+0xe5/0x1e0
>  [  690.473639]  flush_smp_call_function_from_idle+0x5f/0xa0
> 
> 
> For some reason - that i haven't looked into yet -
> net_rps_send_ipi() will eventually ____napi_schedule()
> only after enabling IRQ. see net_rps_action_and_irq_enable()

Perfect. There is a do_softirq() in flush_smp_call_function_from_idle()
it so it fine.
PeterZ any idea in how to shut lockdep here? Playing with the preemption
counter will do the trick… I'm worried that by disabling BH
unconditionally here it will need to be done by the upper caller and
this in turn will force a BH-flush on PREEMPT_RT. While it looks
harmless in the idle case, it looks bad for migration_cpu_stop().

Sebastian