Re: [PATCH] kvm/ia64: Add printk support for kvm-intel modules.

[Avi Kivity <avi@qumranet.com>]

Zhang, Xiantao wrote:
>
>>> +static void vcpu_print_vmm_log(void)
>>> +{
>>> +	unsigned int slot;
>>> +
>>> +	spin_lock(&vmm_log->log_lock);
>>>
>>>       
>> You're going to impact scalability with this.  Are per-vcpu logs
>> workable? 
>>     
>
> OK, I will change it to per-vcpu style to avoid this possible
> scalability issue. 
>
>   

Actually, per-vcpu logs have a deficiency where log lines become unordered.

So I suggest a per-vcpu flag that says "there may be something in the
log", but keep a single log buffer.  Since printk()s are rare (and
slow), it's enough that we make the case where the log is empty fast.


>> I suspect this will start breaking when people start using the new
>> printk("%pBLAH") functionality, which will require linking additional
>> files. 
>>     
>
>  If the format string works with vsnprintf, it should be covered. 
>
>   

vsnprintf() may start to be linked with other stuff.  Well, we'll deal
with that when it happens.

>> I can't think of a way on x86, but maybe ia64 varargs are different.
>>
>> (worst case you can limit the number of arguments and just copy a
>> bunch  of stack).
>>     
>
> That maybe infeasible, since some args may be transferred by pointer,
> and this pointer can't be reached in host side.
>   

Okay.

-- 
Do not meddle in the internals of kernels, for they are subtle and quick to panic.