On Mon 07 Mar 14:13 PST 2022, Andy Shevchenko wrote: > On Mon, Mar 07, 2022 at 01:04:50PM -0800, Bjorn Andersson wrote: > > On Mon 07 Mar 08:13 PST 2022, Andy Shevchenko wrote: > > > On Mon, Mar 07, 2022 at 06:48:25AM -0800, Bjorn Andersson wrote: > > > > On Mon 07 Mar 02:16 PST 2022, Andy Shevchenko wrote: > > > > > On Sun, Mar 06, 2022 at 07:40:40PM -0800, Bjorn Andersson wrote: > > ... > > > > > > > + /* 15us to allow the SBU switch to turn off */ > > > > > > + usleep_range(15, 1000); > > > > > > > > > > This is quite unusual range. > > > > > > > > > > If you are fine with the long delay, why to stress the system on it? > > > > > Otherwise the use of 1000 is unclear. > > > > > > > > > > That said, I would expect one of the below: > > > > > > > > > > usleep_range(15, 30); > > > > > usleep_range(500, 1000); > > > > > > > > Glad you asked about that, as you say the typical form is to keep the > > > > range within 2x of the lower value, or perhaps lower + 5. > > > > > > > > But if the purpose is to specify a minimum time and then give a max to > > > > give the system some flexibility in it's decision of when to wake up. > > > > And in situations such as this, we're talking about someone connecting a > > > > cable, so we're in "no rush" and I picked the completely arbitrary 1ms > > > > as the max. > > > > > > > > Do you see any drawback of this much higher number? (Other than it > > > > looking "wrong") > > > > > > I see the drawback of low number. > > > > 15us is based on the data sheet and if the kernel is ready to serve us > > after 15us then let's do that. > > > > > The 1000 makes not much sense to me with the minimum 66x times less. > > > If there is no rush, use some reasonable values, > > > what about > > > > > > usleep_range(100, 1000); > > > > > > ? 10x is way better than 66x. > > > > I don't agree, and in particular putting 100 here because it's 1/10 of > > the number I just made up doesn't sounds like a good reason. The > > datasheet says 15us, so that is at least based on something real. > > > > In https://www.kernel.org/doc/Documentation/timers/timers-howto.txt > > I find the following: > > > > With the introduction of a range, the scheduler is > > free to coalesce your wakeup with any other wakeup > > that may have happened for other reasons, or at the > > worst case, fire an interrupt for your upper bound. > > > > The larger a range you supply, the greater a chance > > that you will not trigger an interrupt; this should > > be balanced with what is an acceptable upper bound on > > delay / performance for your specific code path. Exact > > tolerances here are very situation specific, thus it > > is left to the caller to determine a reasonable range. > > > > Which to me says that the wider range is perfectly reasonable. In > > particular 15, 30 (which seems to be quite common) makes the available > > range to the scheduler unnecessarily narrow. > > > > And it's clear that whatever the upper bound it's going to be some > > arbitrary number, but 1ms should ensure that there are other hrtimer > > interrupts to piggy back on. > > Okay, I have grepped for usleep_range(x[x], yyyy) and there are 9 modules > use it. A few commit messages call 1000 as "reasonable upper limit". > Right, we usually see a much more narrow range, as you say 2x or perhaps 10x, and this why I said I was glad you asked. I have been wondering about this in a few different cases... Thanks, Bjorn
