Re: [PATCH] V2 pci uart - better cope with UART being temporarily unavailable

[Tomasz Wroblewski <tomasz.wroblewski@citrix.com>]

On 08/27/2013 03:59 PM, Jan Beulich wrote:
>>>> On 27.08.13 at 15:44, Tomasz Wroblewski<tomasz.wroblewski@citrix.com>  wrote:
>> On 08/27/2013 02:26 PM, Jan Beulich wrote:
>>>>>> On 27.08.13 at 12:15, Tomasz Wroblewski<tomasz.wroblewski@citrix.com>   wrote:
>>>> @@ -102,12 +107,17 @@ static void __ns16550_poll(struct cpu_user_regs *regs)
>>>>        if ( uart->intr_works )
>>>>            return; /* Interrupts work - no more polling */
>>>>
>>>> -    while ( ns_read_reg(uart, UART_LSR)&   UART_LSR_DR )
>>>> -        serial_rx_interrupt(port, regs);
>>>> +     while ( ns_read_reg(uart, UART_LSR)&   UART_LSR_DR )
>>>> +     {
>>>> +         serial_rx_interrupt(port, regs);
>>>> +         if ( ns16550_ioport_invalid(uart) )
>>>> +             goto out;
>>>> +     }
>>>>
>>>>        if ( ns_read_reg(uart, UART_LSR)&   UART_LSR_THRE )
>>>>            serial_tx_interrupt(port, regs);
>>>>
>>>> +out:
>>> So serial_rx_interrupt() gets run once in that case, but
>>> serial_tx_interrupt() not at all? That not only inconsistent, I also
>>> can't see why anything would need to be done here at all in this
>>> case. Plus doing the check before the loop would shrink patch
>>> size.
>> So I presume it is impossible for dom0 code to run on another cpu whilst
>> xen is executing the poll routine, I was not sure at all about this?
>> Would like to avoid the possibility of dom0 disabling device whilst this
>> loop is running.
> Of course can a Dom0 vCPU run on another pCPU.
>
> And yes, I hence appreciate reducing the risk window. But then
> please do the check before calling serial_rx_interrupt().
>
Ah sure, will repost v4 soon then