[ofa-general] Re: [PATCH] ibim/ibsim.c Fixed ibsim console so it doesn't eat up all CPU when creading from a pipe

Nicolas Morey-Chaisemartin devel at morey-chaisemartin.com
Fri Mar 20 09:04:51 PDT 2009


Sasha Khapyorsky a écrit :
> On 13:59 Fri 20 Mar     , Nicolas Morey Chaisemartin wrote:
>> @@ -618,7 +619,8 @@ static int sim_run(int con_fd)
>>  	for (;;) {
>>  		FD_ZERO(&rfds);
>>  		FD_SET(simctl, &rfds);
>> -		FD_SET(con_fd, &rfds);
>> +		if(console)
>> +			FD_SET(con_fd, &rfds);
>>  		for (i = 0; i < IBSIM_MAX_CLIENTS; i++)
>>  			if (clients[i].pid)
>>  				FD_SET(clients[i].fd, &rfds);
>> @@ -634,7 +636,8 @@ static int sim_run(int con_fd)
>>  				sim_read_pkt(clients[i].fd, i);
>>  
>>  		if (FD_ISSET(con_fd, &rfds))
>> -			sim_run_console(con_fd);
>> +			if(sim_run_console(con_fd)<=0)
>> +				console = 0;
> 
> Then one occasional failure will disable the console forever.
> Wouldn't it be better to analyze HUP state (similar to as it is done
> with OpenSM console)?
> 
> Sasha

I guess so but in which case would it fail?
I'm not really good at poll/select and such so I did it my way ;)
We are running hundred thousand of tests using ibsim and it's much faster when not using a whole CPU for itself !

I look for a cleaner way to do this tommorow.

Nicolas




More information about the general mailing list