Re: [PATCH v1 2/2] xen/console: unify printout behavior for UART emulators

[Jan Beulich <jbeulich@xxxxxxxx>]

On 05.06.2025 02:57, dmkhn@xxxxxxxxx wrote:
> On Wed, Jun 04, 2025 at 12:48:05PM +0200, Jan Beulich wrote:
>> On 31.05.2025 02:04, dmkhn@xxxxxxxxx wrote:
>>> --- a/xen/arch/x86/hvm/hvm.c
>>> +++ b/xen/arch/x86/hvm/hvm.c
>>> @@ -577,7 +577,7 @@ static int cf_check hvm_print_line(
>>>      if ( (cd->pbuf_idx == (DOMAIN_PBUF_SIZE - 1)) || (c == '\n') )
>>>      {
>>>          cd->pbuf[cd->pbuf_idx] = '\0';
>>> -        guest_printk(cd, XENLOG_G_DEBUG "%s\n", cd->pbuf);
>>> +        guest_printk(cd, "%s\n", cd->pbuf);
>>>          cd->pbuf_idx = 0;
>>>      }
>>
>> Why this and ...
>>
>>> @@ -755,7 +765,7 @@ static long 
>>> guest_console_write(XEN_GUEST_HANDLE_PARAM(char) buffer,
>>>              else
>>>              {
>>>                  cd->pbuf[cd->pbuf_idx] = '\0';
>>> -                guest_printk(cd, XENLOG_G_DEBUG "%s%s\n", cd->pbuf, kbuf);
>>> +                guest_printk(cd, "%s%s\n", cd->pbuf, kbuf);
>>>                  cd->pbuf_idx = 0;
>>>              }
>>
>> ... this change? There's no compensation for it ...
>>
>>> @@ -1013,12 +1023,21 @@ void printk(const char *fmt, ...)
>>>      va_end(args);
>>>  }
>>>
>>> +/*
>>> + * Print message from the guest on the diagnostic console.
>>> + * Prefixes all messages w/ "(dX)" if domain X does not own physical 
>>> console
>>> + * focus.
>>> + */
>>>  void guest_printk(const struct domain *d, const char *fmt, ...)
>>>  {
>>>      va_list args;
>>> -    char prefix[16];
>>> +    char prefix[16] = "";
>>> +    struct domain *consd;
>>>
>>> -    snprintf(prefix, sizeof(prefix), "(d%d) ", d->domain_id);
>>> +    consd = console_get_domain();
>>> +    if ( consd != d )
>>> +        snprintf(prefix, sizeof(prefix), "(d%d) ", d->domain_id);
>>> +    console_put_domain(consd);
>>>
>>>      va_start(args, fmt);
>>>      vprintk_common(fmt, args, prefix);
>>
>> ... here afaics, so it looks like you're undermining rate-limiting of
>> those messages.
> 
> I droppped behavior change for I/O debug port on x86 and HYPERVISOR_console_io
> hypercall.
> 
> But my understanding is that all guest debugging facilities, if enabled, 
> should
> not be rate-limited.

I certainly disagree there. How much rate limiting to apply to guest output is a
matter of the guest_loglvl= command line option. Its default settings are the 
way
they are for a reason.

Jan