Re: [PATCH v2 17/17] irq: remove handle_domain_{irq,nmi}()

[Thomas Gleixner <tglx@xxxxxxxxxxxxx>]

On Tue, May 10 2022 at 15:15, Mark Rutland wrote:
> On Tue, May 10, 2022 at 02:13:20PM +0200, Lukas Wunner wrote:
>> For gpio-dln2.c, I believe it from inspection.
>> 
>> For smsc95xx.c, I'm actually seeing it go wrong in practice,
>> unedited dmesg splat is included below FWIW.
>
> Thanks; having the trace makes this much easier to analyse.

which confirmes what I talked about before:

>> WARNING: CPU: 3 PID: 75 at kernel/irq/irqdesc.c:702 generic_handle_domain_irq+0x88/0x94
>>  generic_handle_domain_irq from smsc95xx_status+0x54/0xb0
>>  smsc95xx_status from intr_complete+0x80/0x84
>>  intr_complete from __usb_hcd_giveback_urb+0xa4/0x12c
>>  __usb_hcd_giveback_urb from usb_hcd_giveback_urb+0x118/0x11c
>>  usb_hcd_giveback_urb from completion_tasklet_func+0x7c/0xc8
>>  completion_tasklet_func from tasklet_callback+0x20/0x24
>>  tasklet_callback from tasklet_action_common.constprop.0+0x148/0x220
>>  tasklet_action_common.constprop.0 from tasklet_hi_action+0x28/0x30
>>  tasklet_hi_action from __do_softirq+0x154/0x3e8
>>  __do_softirq from __local_bh_enable_ip+0x12c/0x1a8
>>  __local_bh_enable_ip from irq_forced_thread_fn+0x7c/0xac
>>  irq_forced_thread_fn from irq_thread+0x16c/0x228
>>  irq_thread from kthread+0x100/0x140

So what happens here:

 interrupt
    -> wakeup threaded handler

 threaded handler runs
    local_bh_disable();
    ....
    schedules tasklet
    ...
    local_bh_enable()
      do_softirq()
        run_tasklet()
          urb_completion()
            smsc95xx_status()
              generic_handle_domain_irq()

That interrupt in question is an interrupt, which is not handled by the
primary CPU interrupt chips. It's a synthetic interrupt which is
generated from the received USB packet.

+	/* USB interrupts are received in softirq (tasklet) context.
+	 * Switch to hardirq context to make genirq code happy.
+	 */
+	local_irq_save(flags);
+	__irq_enter_raw();
+
 	if (intdata & INT_ENP_PHY_INT_)
-		;
+		generic_handle_domain_irq(pdata->irqdomain, PHY_HWIRQ);

This __irq_enter_raw() is really wrong. This is _not_ running in hard
interrupt context. Pretending so creates more problems than it
solves. It breaks context tracking, confuses lockdep ...

We also have demultiplexed interrupts which are nested in a threaded
interrupt handler and share the thread context. No, we are not going to
pretend that they run in hard interrupt context either.

So we need a clear distinction between interrupts which really happen in
hard interrupt context and those which are synthetic and can be invoked
from pretty much any context.

Anything else is just a recipe for disaster and endless supply of half
baken hacks.

Thanks,

        tglx