We were discussing introducing mutexes in those critical spots for a long time now (in my team). Stefan, if you find time, you are most welcome to offer your solution with mutexes.
Slawek -----Original Message----- From: Stefan Assmann <sassm...@kpanic.de> Sent: Wednesday, March 17, 2021 8:50 AM To: Brandeburg, Jesse <jesse.brandeb...@intel.com>; Jakub Kicinski <k...@kernel.org> Cc: intel-wired-...@lists.osuosl.org; netdev@vger.kernel.org; Nguyen, Anthony L <anthony.l.ngu...@intel.com>; Yang, Lihong <lihong.y...@intel.com>; Laba, SlawomirX <slawomirx.l...@intel.com>; Nunley, Nicholas D <nicholas.d.nun...@intel.com> Subject: Re: [PATCH] iavf: fix locking of critical sections On 16.03.21 23:02, Jesse Brandeburg wrote: > Jakub Kicinski wrote: >>>> I personally think that the overuse of flags in Intel drivers >>>> brings nothing but trouble. At which point does it make sense to >>>> just add a lock / semaphore here rather than open code all this >>>> with no clear semantics? No code seems to just test the >>>> __IAVF_IN_CRITICAL_TASK flag, all the uses look like poor man's >>>> locking at a quick grep. What am I missing? >>> >>> I agree with you that the locking could be done with other locking >>> mechanisms just as good. I didn't invent the current method so I'll >>> let Intel comment on that part, but I'd like to point out that what >>> I'm making use of is fixing what is currently in the driver. >> >> Right, I should have made it clear that I don't blame you for the >> current state of things. Would you mind sending a patch on top of >> this one to do a conversion to a semaphore? Sure, I'm happy to help working on the conversion once the current issue is resolved. >> Intel folks any opinions? > > I know Slawomir has been working closely with Stefan on figuring out > the right ways to fix this code. Hopefully he can speak for himself, > but I know he's on Europe time. > > As for conversion to mutexes I'm a big fan, and as long as we don't > have too many collisions with the RTNL lock I think it's a reasonable > improvement to do, and if Stefan doesn't want to work on it, we can > look into whether Slawomir or his team can. I'd appreciate to be involved. Thanks! Stefan
