Re: [PATCH] serial: 8250: 8250_omap: Fix possible interrupt storm

[Vignesh Raghavendra]

On 7/13/21 1:57 AM, andy@surfacebook.localdomain wrote:
> Tue, Jun 22, 2021 at 11:53:38AM +0530, Vignesh Raghavendra kirjoitti:
>> On 6/22/21 11:45 AM, Jan Kiszka wrote:
>>>>> Vignesh, it seems this quirk needs some soc specific flag added to
>>>>> it maybe? Or maybe UART_OMAP_RX_LVL register is not available for
>>>>> all the SoCs?
>>>>>
>>>> Yes indeed :(
>>>>
>>>>> I think it's best to drop this patch until the issues are resolved,
>>>>> also there are some open comments above that might be answered by
>>>>> limiting this quirk to a specific range of SoCs :)
>>>>>
>>>> Oops, I did test patch AM33xx assuming its equivalent to OMAP3, but UART
>>>> IP is quite different. I will respin the patch making sure, workaround
>>>> applies only to AM65x and K3 SoCs.
>>>>
>>>> Regards
>>>> Vignesh
>>>>
>>> What's the status here for AM65x? The issue remains present on that
>>> platform, and I was hoping to see a quick follow up that limit the fix
>>> to that target.
>>
>> Sorry for the delay, I am trying to find which other TI SoCs are
>> affected by this issue. But that exercise will need a bit more time.
>> Will send a fix to address K3 SoCs like AM65x today/tomo.
> 
> This all reminds me the very similar issue one found on Intel integrated
> (Synopsys DesignWare based) UARTs:
> 

Hmm, yes, seems like common problem with some 8250 UARTs although not
all TI SoCs show this behavior even though they all claim 8250 compatible.

> https://lore.kernel.org/linux-serial/20170206233000.3021-1-dianders@xxxxxxxxxxxx/

I am not sure if reading UART_LSR is a good idea in the above patch.
Some flags in LSR register are cleared on read (at least that's the case
for UARTs on TI SoCs) and thus can result in loss of error/FIFO status
information.

> https://lore.kernel.org/linux-serial/1440015124-28393-1-git-send-email-california.l.sullivan@xxxxxxxxx/
> 

Looks like this never made it.

Given the quirks associated with 8250 UARTs, workarounds would need to
be tied to specific variants, so I don't know if its possible to
implement the fix in 8250 core IRQ handler.

PS: v2 of $patch is already merged.

Regards
Vignesh