Hi.... Please don't do top posting... 2009/6/25 Kostya B <bkostya@xxxxxxxxxxx>: > > Mulyadi, thank you for the detailed explanation. Thus > > (a) is a source for interrupt latency and (b) actually fixes it. > > -- > Kostya > > > ---------------------------------------- >> Date: Thu, 25 Jun 2009 02:26:25 +0700 >> Subject: Re: preemption >> From: mulyadi.santosa@xxxxxxxxx >> To: bkostya@xxxxxxxxxxx >> CC: kernelnewbies@xxxxxxxxxxxx >> >> Hi... >> >> 2009/6/24 Kostya B : >>> >>> >>> Dear list, >>> >>> Please help me to analyze the following scenario: >>> >>> By having a userspace process 'A' executing a system call. >>> Hardware interrupt happens. Do you think on exit from interrupt (irq_exit) >>> the context switch could take place? >> Yes it could, depending on the preemption model you pick...see below.... >> >>> Let's assume no softirq work is pending. >>> >>> (a) kernel is not preemptive >> >> then it would stay in the current kernel code path, which is executing >> the syscall on behalf of running task. >> >>> (b) kernel is preemptive >> >> What I am quite sure here is, in full preemptive kernel, after >> returning from interrupt handling, scheduler is allowed to pick new >> task to be executed. So, let's say scheduler has pick new task >> according to certain rules, context switch will happen. >> >> For voluntary preemption, perhaps you need to check the code by your >> own. In my point of view, after reading http://lwn.net/Articles/95334/ >> (probably not so relevant anymore), I have impression rescheduling >> also happen after returning from interrupt handler. >> >>> (c) with RT patch >> >> I have no hands on experience about RT patch. Also you need to >> specificly define whether you refer to hard or soft RT. >> >> I think in both RT style, the preemption might happen after returning >> from the interrupt. By doing that, scheduler might be able to pick the >> job that almost meet its deadline. >> >> regards, >> >> Mulyadi. What do you mean by "interrupt latency"? I thought we're talking about task scheduling latency. CMIIW regards, Mulyadi -- To unsubscribe from this list: send an email with "unsubscribe kernelnewbies" to ecartis@xxxxxxxxxxxx Please read the FAQ at http://kernelnewbies.org/FAQ
