From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jacek Anaszewski Subject: Re: Reading /sys with side effects (was Re: [PATCH 1/2] Documentation: leds: Add description of LED Flash class extension) Date: Mon, 02 Feb 2015 15:51:53 +0100 Message-ID: <54CF8F09.8020203@samsung.com> References: <1422346028-16739-1-git-send-email-j.anaszewski@samsung.com> <20150127221958.GA18993@amd> <54C8A130.8000807@samsung.com> <20150129211420.GA21140@amd> <54CB4702.1090508@samsung.com> <20150130164027.GA25830@kroah.com> <54CF3E36.9000108@samsung.com> <20150202094435.GA6686@amd> <54CF65CB.3060207@samsung.com> <20150202135103.GB26985@amd> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Return-path: In-reply-to: <20150202135103.GB26985@amd> Sender: linux-leds-owner@vger.kernel.org To: Pavel Machek Cc: Greg KH , kernel list , linux-leds@vger.kernel.org, devicetree@vger.kernel.org, kyungmin.park@samsung.com, b.zolnierkie@samsung.com, cooloney@gmail.com, rpurdie@rpsys.net, sakari.ailus@iki.fi, s.nawrocki@samsung.com List-Id: devicetree@vger.kernel.org Hi Pavel, On 02/02/2015 02:51 PM, Pavel Machek wrote: > Hi! > >>> [Actually, you could _always_ do two reads on those devices, discard >>> first result, and return the second. But I'm not sure how hardware >>> will like that.] >> >> This would be the most sensible option. >> >> >> However, let's analyze the typical use cases for flash strobing: >> >> >> ------------------------------------------------------------- >> >> >>>>>>>>>> Version without faults caching: >> >> ============ >> Driver side: >> ============ >> >> read_faults() >> faults = read_i2c(); //read faults >> if faults >> write_i2c(); //clear faults, only for some devices >> faults = read_i2c(); //read faults >> return faults >> >> ================ >> User space side: >> ================ >> >> 1. faults = `cat flash_faults` //read_faults() >> 2. if faults then >> print "Unable to strobe the flash LED due to faults" >> else >> echo 1 > flash_strobe >> >> >>>>>>>>>> Version with faults caching: >> >> ============ >> Driver side: >> ============ >> >> read_faults() >> faults |= read_i2c(); //read faults >> >> clear_faults() >> write_i2c(); //clear faults >> faults = 0; >> >> >> ================ >> User space side: >> ================ >> >> 1. faults = `cat flash_faults` //read_faults() >> 2. if faults then >> echo 0 > flash_faults //clear_faults() >> faults = `cat flash_faults` //read_faults() >> 3, if !faults >> echo 1 > flash_strobe >> else >> print "Unable to strobe the flash LED due to faults" >> >> >> ------------------------------------------------------------- >> >> From the above it seems that version with clearing faults on read >> results in the simpler flash strobing procedure on userspace side, >> by the cost of additional bus access on the driver side. > > I like caching version more (as it will allow by-hand debugging of > "why did not flash fire? Aha, lets see in the file, there was fault), > but both should be acceptable. > >> we don't need additional attribute, just writing the flash_faults >> attribute can do the clearing. > > Yes, writing flash_faults to clear is acceptable. I've been just inspired with another approach: Faults register is read in the strobe_set callback, right after sending flash strobe command to the device. The userspace can read the cached faults through flash_faults attribute. This way, we avoid reading sysfs attribute with side effect and gain the possibility of giving immediate feedback to the user. Exemplary use case: 1. echo 1 > flash_strobe write_i2c(); //strobe flash faults = read_i2c(); //read faults if faults return -EINVAL; return 0; 2. cat flash_faults return faults; -- Best Regards, Jacek Anaszewski