* Fwd: is it possible to liimit the softirq re entrance times [not found] <CAOVJa8FfHWGVf=ccXFCyTUD5YHrMAAeQ9Tkg1ipHA=J_CyVAHQ@mail.gmail.com> @ 2015-08-20 17:12 ` vichy 2015-08-20 17:36 ` Valdis.Kletnieks at vt.edu 0 siblings, 1 reply; 2+ messages in thread From: vichy @ 2015-08-20 17:12 UTC (permalink / raw) To: kernelnewbies after I tracing softirq flow in ARM platform, I have some colusion like below: 1. softirq is running at the end of any irq context, when irq is eabled. 2. as long as any Hard irq happen, softirq can be interrupt and reentrant My question are: 1. in __do_sofiq, there is a jiffies setting to make sure the whole softirq will not took to long. But in my case is there any many hard irq happen like below trace_softirq_entry(vec_nr); ------a h->action(h); trace_softirq_exit(vec_nr); ---------b from lttng, I see there 100 interrupt happen from entry to exit. is there kernel config or module parameter to let me down grade the hard irq count from 100 --> 80 or 90 and event can controlled me. I don't mean I control the count of hard irq, but I found this softirq is interrupt by 80 times. then I will not enable_irq until I finished the job. if no, isn't it possible to make starvation in above a and b? really appreciate ur kind help, ^ permalink raw reply [flat|nested] 2+ messages in thread
* Fwd: is it possible to liimit the softirq re entrance times 2015-08-20 17:12 ` Fwd: is it possible to liimit the softirq re entrance times vichy @ 2015-08-20 17:36 ` Valdis.Kletnieks at vt.edu 0 siblings, 0 replies; 2+ messages in thread From: Valdis.Kletnieks at vt.edu @ 2015-08-20 17:36 UTC (permalink / raw) To: kernelnewbies On Fri, 21 Aug 2015 01:12:10 +0800, vichy said: > > if no, isn't it possible to make starvation in above a and b? Operating systems have to make design choices. You run with IRQs disabled for too long, you can starve other IRQs waiting for service. You run with them mostly enabled, the one handler can get starved by a screaming interrupt (or a large number of devices all generating interrupts). Choosing wisely is required - some devices are low-speed with large buffers, so don't care if it takes a while to service them. Meanwhile, others are extremely latency-sensitive. So in general, each driver gets designed to fit the device's tolerance for delay. -------------- next part -------------- A non-text attachment was scrubbed... Name: not available Type: application/pgp-signature Size: 848 bytes Desc: not available Url : http://lists.kernelnewbies.org/pipermail/kernelnewbies/attachments/20150820/39191d26/attachment.bin ^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2015-08-20 17:36 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
[not found] <CAOVJa8FfHWGVf=ccXFCyTUD5YHrMAAeQ9Tkg1ipHA=J_CyVAHQ@mail.gmail.com>
2015-08-20 17:12 ` Fwd: is it possible to liimit the softirq re entrance times vichy
2015-08-20 17:36 ` Valdis.Kletnieks at vt.edu
This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox; as well as URLs for NNTP newsgroup(s).