Re: BUG Found, was: USB-Serial FT4232H unusable with kernel 5.9.x

Re: BUG Found, was: USB-Serial FT4232H unusable with kernel 5.9.x

[Andy Shevchenko]

On Thu, Oct 29, 2020 at 05:44:56PM +0100, alberto.vignani@fastwebnet.it wrote:
> By bisecting the regression I found that the FT4232 was only an innocent bystander. The real culprit was the PCIe-USB adapter that stopped sending interrupts (or sent wrong interrupts) in kernel 5.9. I have just removed the whole patch and my system is working again as in 5.8.I hope that someone will revise this patch considering that the Pericom PI7C9X440SL is affected by it.

Thank you for the report!

Can you share output of

	% lspci -vvv -nk -s <XXX> # where <XXX> is a slot of PCIe USB

for working and non-working cases?


> Here's the commit responsible:

> commit 306c54d0edb6 usb: hcd: Try MSI interrupts on PCI devices

This commit per se should not affect devices which rightfully announce MSI
support. It is possible that I have missed something or above mentioned device
is not MSI compatible.

Btw, can this PCIe-USB be connected to completely different motherboard /
machine for a testing purposes? It might be that PCI host bridge is not
friendly with the device or other way around.

...

> With 5.9 I noticed that these delays are dependent on how the ports are mapped t o linux devices:- if I remove the JTAG
>  quirk and allow all ports to be connected to ttyUSBs, there seems to 
> be no delay before the first message: 33 usec as in linux 5.8.  Of course I cannot then configure the first two ports for JTAG (FTOpen fails).
> - conversely, if I 
> remove the ftdi_sio module (with all ttyUSB devices) the initial delay 
> is even longer (7.6 seconds), and then the (slow) JTAG configuration s ucceeds.
>  Can someone confirm and investigate? Does the same happen with ST Micro Connect Lite?
> And how can I debug this one?

Can you also monitor /proc/interrupts for working and non-working cases?
Simply share its contents as well.

-- 
With Best Regards,
Andy Shevchenko

Re: BUG Found, was: USB-Serial FT4232H unusable with kernel 5.9.x

[Andy Shevchenko]

On Fri, Oct 30, 2020 at 03:48:26PM +0200, Andy Shevchenko wrote:
> On Thu, Oct 29, 2020 at 05:44:56PM +0100, alberto.vignani@fastwebnet.it wrote:
> > By bisecting the regression I found that the FT4232 was only an innocent bystander. The real culprit was the PCIe-USB adapter that stopped sending interrupts (or sent wrong interrupts) in kernel 5.9. I have just removed the whole patch and my system is working again as in 5.8.I hope that someone will revise this patch considering that the Pericom PI7C9X440SL is affected by it.
> 
> Thank you for the report!
> 
> Can you share output of
> 
> 	% lspci -vvv -nk -s <XXX> # where <XXX> is a slot of PCIe USB
> 
> for working and non-working cases?
> 
> 
> > Here's the commit responsible:
> 
> > commit 306c54d0edb6 usb: hcd: Try MSI interrupts on PCI devices
> 
> This commit per se should not affect devices which rightfully announce MSI
> support. It is possible that I have missed something or above mentioned device
> is not MSI compatible.
> 
> Btw, can this PCIe-USB be connected to completely different motherboard /
> machine for a testing purposes? It might be that PCI host bridge is not
> friendly with the device or other way around.
> 
> ...
> 
> > With 5.9 I noticed that these delays are dependent on how the ports are mapped t o linux devices:- if I remove the JTAG
> >  quirk and allow all ports to be connected to ttyUSBs, there seems to 
> > be no delay before the first message: 33 usec as in linux 5.8.  Of course I cannot then configure the first two ports for JTAG (FTOpen fails).
> > - conversely, if I 
> > remove the ftdi_sio module (with all ttyUSB devices) the initial delay 
> > is even longer (7.6 seconds), and then the (slow) JTAG configuration s ucceeds.
> >  Can someone confirm and investigate? Does the same happen with ST Micro Connect Lite?
> > And how can I debug this one?
> 
> Can you also monitor /proc/interrupts for working and non-working cases?
> Simply share its contents as well.

Also, can you add 'ignore_loglevel' to the kernel command line and add dmesg
output of working and non-working cases?

(You may share the archive or set of files via some share service and give a
 link here)

-- 
With Best Regards,
Andy Shevchenko