Re: [PATCH] usb: gadget: udc: renesas_usb3: Fix RZ/V2M {modprobe,bind} error

[Laurent Pinchart]

On Mon, May 29, 2023 at 09:39:50AM +0000, Biju Das wrote:
> > Subject: Re: [PATCH] usb: gadget: udc: renesas_usb3: Fix RZ/V2M
> > {modprobe,bind} error
> > 
> > On Mon, 29 May 2023 09:42:34 +0100, Biju Das wrote:
> > >
> > > > Subject: Re: [PATCH] usb: gadget: udc: renesas_usb3: Fix RZ/V2M
> > > > {modprobe,bind} error
> > > >
> > > > On Fri, May 26, 2023 at 03:36:15PM +0100, Biju Das wrote:
> > > > > Currently {modprobe, bind} after {rmmod, unbind} results in probe failure.
> > > > >
> > > > > genirq: Flags mismatch irq 22. 00000004 (85070400.usb3drd) vs.
> > > > > 00000004 (85070400.usb3drd)
> > > > > renesas_usb3: probe of 85070000.usb3peri failed with error -16
> > > > >
> > > > > Fix this issue by replacing "parent dev"->"dev" as the irq
> > > > > resource is managed by this driver.
> > > >
> > > > If the dev pointer passed to devm_request_irq() is not the correct
> > > > one, how does it work the first time the driver is loaded ?
> > >
> > > + Marc/ Kernel.org to give some feedback on this issue
> > >
> > > I believe there may be a bug in the genirq (kernel/irq) driver.
> > > first time it works ok. Maybe this driver is caching on unload with
> > > null value and comparing with actual one (irq 22) during reload??
> > >
> > > Maybe genirq expert can comment what went wrong here??
> > 
> > You get shouted at because you are registering an interrupt handler for
> > the same IRQ twice,
> 
> This not true. It is registering only one IRQ, but with parent device handle.

It uses devm_request_irq() with the parent device, so the interrupt
handler won't be unregistered when the usb3-peri device is unbound. The
next probe will register the same interrupt handler a second time. This
has nothing to do with genirq, it's related to devm_*.

> > and the interrupt is not configured with the SHARED
> > flag.
> 
> I haven't added SHARED flag as there is only one IRQ registration.
> 
>  If, as I understand it, you only have a single device using this
> > interrupt, then it means your driver is not freeing its interrupt on
> > unload.
> 
> You mean devm_request_irq(ddata->dev..)  doesn't free the resource as
> we have unloaded only child device rather than parent.
> 
> But while parent is active, why genirq is giving error during reload?
> It should show same behaviour like initial probe.
> 
> > And that's probably because the device object used when requesting the
> > interrupt isn't the one you load/unload, as indicated by the message.
> > On the first load of "usb3peri", you register an interrupt with
> > "usb3drd" as the requester device. You then unload "usb3peri", which of
> > course has no effect whatsoever on the interrupt.
> > 
> > You could simply have done a "cat /proc/interrupt" and see that
> > interrupt was still there after unload.
> 
> Yes, interrupt still there after unload.
> 
> With devm_request_irq(ddata->dev..), after unload
> =================================================
> 
> root@rzv2m:~# cat /proc/interrupts | grep usb
>  22:          0     GICv2 274 Level     85070400.usb3drd
>  28:          0     GICv2 278 Level     85070000.usb3peri
> root@rzv2m:~# lsmod
> Module                  Size  Used by
> hd3ss3220              12288  0
> typec                  73728  1 hd3ss3220
> renesas_usb3           32768  1
> i2c_rzv2m              12288  0
> crct10dif_ce           12288  1
> ipv6                  450560  16
> root@rzv2m:~# rmmod hd3ss3220
> root@rzv2m:~# rmmod renesas_usb3
> root@rzv2m:~# cat /proc/interrupts | grep usb
>  22:          0     GICv2 274 Level     85070400.usb3drd
> root@rzv2m:~#
> 
> With devm_request_irq(&pdev->dev..), after unload
> ================================================
> 
> root@rzv2m:~# cat /proc/interrupts | grep usb
>  22:          0     GICv2 274 Level     85070400.usb3drd
>  28:          0     GICv2 278 Level     85070000.usb3peri
> root@rzv2m:~# lsmod
> Module                  Size  Used by
> hd3ss3220              12288  0
> typec                  73728  1 hd3ss3220
> renesas_usb3           32768  1
> crct10dif_ce           12288  1
> i2c_rzv2m              12288  0
> ipv6                  450560  16
> root@rzv2m:~# rmmod hd3ss3220
> root@rzv2m:~# rmmod renesas_usb3
> root@rzv2m:~# cat /proc/interrupts | grep usb
> root@rzv2m:~#
> 
> > So the only bug here is in the handling of the interrupt request. And
> > that bug firmly lies in your code. My "expert" advise is to debug the
> > problem rather than suspecting some random failure modes.
> 
> With devm_request_irq(&pdev->dev..) the above issue is fixed.
> 
> Or
> 
> the correct way is passing SHARED flag with devm_request_irq(ddata ->dev..), 
> as the resource is owned by the parent??

No you shouldn't pass the SHARED flag. This patch is a step in the right
direction, but the proper fix would be to register the interrupt handler
in the usb3drd driver.

-- 
Regards,

Laurent Pinchart