Re: [PATCH] x86/fred: Optimize the FRED entry by prioritizing high-probability event dispatching

["H. Peter Anvin" <hpa@zytor.com>]

On 1/16/25 21:54, Xin Li wrote:
> On 1/16/2025 9:18 PM, Ethan Zhao wrote:
>>>
>>> Just swap the 2 arguments, and it should be:
>>> +    switch_likely (etype, EVENT_TYPE_OTHER) {
>>>
>>>
>> after swapped the parameters as following:
>> +#define switch_likely(v,l) \
>> + switch((__typeof__(v))__builtin_expect((v),(l)))
>> +
>>   __visible noinstr void fred_entry_from_user(struct pt_regs *regs)
>>   {
>>          unsigned long error_code = regs->orig_ax;
>> +       unsigned short etype = regs->fred_ss.type & 0xf;
>>
>>          /* Invalidate orig_ax so that syscall_get_nr() works 
>> correctly */
>>          regs->orig_ax = -1;
>>
>> -       switch (regs->fred_ss.type) {
>> +       switch_likely (etype, (EVENT_TYPE_EXTINT == etype || 
>> EVENT_TYPE_OTHER == etype)) {
> 
> This is not what I suggested, the (l) argument should be only one
> constant; __builtin_expect() doesn't allow 2 different constants.
> 

The (l) argument is not a boolean expression! It is the *expected value* 
of (v).

	-hpa