Re: Boot crash @hci_uart_tx_wakeup+0x38/0x148 w/ Linus' HEAD?

[Lukas Wunner]

Hi John,

thanks for reporting this and apologies for the breakage.

On Thu, Nov 16, 2017 at 12:58:27PM -0800, John Stultz wrote:
> On Wed, Nov 15, 2017 at 6:00 PM, John Stultz <john.stultz@xxxxxxxxxx> wrote:
> >    After updating to Linus' HEAD today, I'm seeing the following odd
> > boot time crash with the HiKey board (which uses the serdev driver).
> >
> > [    1.963009] Unable to handle kernel read from unreadable memory at
> > virtual address 406f127000
> > [    1.963012] Mem abort info:
> > [    1.963015]   ESR = 0x96000005
> > [    1.963018]   Exception class = DABT (current EL), IL = 32 bits
> > [    1.963021]   SET = 0, FnV = 0
> > [    1.963023]   EA = 0, S1PTW = 0
> > [    1.963025] Data abort info:
> > [    1.963027]   ISV = 0, ISS = 0x00000005
> > [    1.963030]   CM = 0, WnR = 0
> > [    1.963032] [000000406f127000] user address but active_mm is swapper
> > [    1.963038] Internal error: Oops: 96000005 [#1] PREEMPT SMP
> > [    1.963046] CPU: 1 PID: 1282 Comm: kworker/u17:1 Not tainted
> > 4.14.0-07281-g1b386f4 #666
> > [    1.963050] Hardware name: HiKey Development Board (DT)
> > [    1.963068] Workqueue: hci0 hci_cmd_work
> > [    1.963074] task: ffffffc0753c8000 task.stack: ffffff800b6c0000
> > [    1.963079] pstate: 80400005 (Nzcv daif +PAN -UAO)
> > [    1.963090] pc : hci_uart_tx_wakeup+0x38/0x148
> > [    1.963095] lr : hci_uart_tx_wakeup+0x30/0x148
> > [    1.963098] sp : ffffff800b6c3d10
> > [    1.963101] x29: ffffff800b6c3d10 x28: 0000000000000000
> > [    1.963107] x27: ffffffc074a79c78 x26: ffffffc07504d830
> > [    1.963112] x25: ffffff8008f76a30 x24: ffffffc07510a840
> > [    1.963117] x23: ffffffc07510aa10 x22: 0000000000000001
> > [    1.963122] x21: ffffff8008cd2000 x20: ffffffc0351cf288
> > [    1.963128] x19: ffffffc0351cf218 x18: 0000000000000000
> > [    1.963132] x17: 0000000000000000 x16: 0000000000000000
> > [    1.963137] x15: 0000000000000000 x14: ffffffc005fa1c00
> > [    1.963142] x13: 000000406f127000 x12: 0000000034d5d91d
> > [    1.963147] x11: 0000000000000400 x10: ffffffc077f8c480
> > [    1.963152] x9 : 0000000000000000 x8 : 0000000000000005
> > [    1.963157] x7 : 0000000000000000 x6 : ffffff8008f59e88
> > [    1.963162] x5 : 0000000000000003 x4 : 000000406f127000
> > [    1.963167] x3 : 0000000000000001 x2 : 000000406f127000
> > [    1.963172] x1 : ffffff8008cd2d28 x0 : 0000000000000000
> > [    1.963179] Process kworker/u17:1 (pid: 1282, stack limit =
> > 0xffffff800b6c0000)
> > [    1.963182] Call trace:
> > [    1.963188]  hci_uart_tx_wakeup+0x38/0x148
> > [    1.963193]  hci_uart_send_frame+0x28/0x38
> > [    1.963197]  hci_send_frame+0x64/0xc0
> > [    1.963201]  hci_cmd_work+0x98/0x110
> > [    1.963209]  process_one_work+0x134/0x330
> > [    1.963214]  worker_thread+0x130/0x468
> > [    1.963220]  kthread+0xf8/0x128
> > [    1.963227]  ret_from_fork+0x10/0x18
> > [    1.963234] Code: 9134a2a0 97f3be03 f9402280 d538d082 (b8626801)
> > [    1.963239] ---[ end trace 457a26b9096bec64 ]---
> >
> 
> So I bisected this down and the issue looks like the boot regression is from:
>    67d2f8781b9f ("Bluetooth: hci_ldisc: Allow sleeping while proto
> locks are held")
> 
> Reverting that change makes things work again.

Hm, I notice percpu_init_rwsem() is only ever called in the hci_ldisc
case (from hci_uart_tty_open()) but apparently never in the hci_serdev
case.  Could that be the culprit?  Wondering why it's working in the
rwlock case then, perhaps by luck?

Thanks,

Lukas