[dpdk-dev] [EXT] [PATCH v3 6/6] spinlock: ticket based to improve fairness

[Jerin Jacob Kollanukkaran]

On Thu, 2018-12-27 at 15:41 -0800, Stephen Hemminger wrote:
> On Thu, 27 Dec 2018 12:08:26 +0000
> Jerin Jacob Kollanukkaran <jerinj at marvell.com> wrote:
> 
> > On Thu, 2018-12-27 at 10:05 +0000, Gavin Hu (Arm Technology China)
> > wrote:
> > > > -----Original Message-----
> > > > From: Jerin Jacob Kollanukkaran <jerinj at marvell.com>
> > > > Sent: Thursday, December 27, 2018 2:58 PM
> > > > To: Gavin Hu (Arm Technology China) <Gavin.Hu at arm.com>; 
> > > > dev at dpdk.org
> > > > Cc: david.marchand at redhat.com; chaozhu at linux.vnet.ibm.com; nd
> > > > <nd at arm.com>; bruce.richardson at intel.com; thomas at monjalon.net;
> > > > Joyce
> > > > Kong (Arm Technology China) <Joyce.Kong at arm.com>;
> > > > hemant.agrawal at nxp.com; stephen at networkplumber.org; Honnappa
> > > > Nagarahalli <Honnappa.Nagarahalli at arm.com>
> > > > Subject: Re: [EXT] [PATCH v3 6/6] spinlock: ticket based to
> > > > improve
> > > > fairness
> > > > 
> > > > On Thu, 2018-12-27 at 12:13 +0800, Gavin Hu wrote:  
> > > > > -----------------------------------------------------------
> > > > > ----
> > > > > ----
> > > > > ---
> > > > > From: Joyce Kong <joyce.kong at arm.com>
> > > > > 
> > > > > The old implementation is unfair, some threads may take locks
> > > > > aggressively  
> > > > 
> > > > I think, one issue here is x86 and ppc follows traditional
> > > > spinlock
> > > > and
> > > > arm64 will be following ticket lock for spinlock
> > > > implementation.
> > > > This would change application behaviour on arm64 compared to
> > > > x86
> > > > and
> > > > ppc.
> > > > 
> > > > How about having a separate API for ticket lock? That would
> > > > give,
> > > > # application choice to use the locking strategy
> > > > # application behaviour will be same across all arch.  
> > > 
> > > Ok, will do in v4 to have a new named rte_ticket_spinlock API.  
> > 
> > I would prefer rte_ticketlock_[lock/unlock/trylock/is_locked] name
> > instead of rte_ticket_spinlock_lock etc to reduce the length of the
> > API.
> 
> NAK to adding new API for this.
> 
> I want the best possible locks for all applications and all
> architectures.
> These should be called spinlock so there is no requirement for
> application
> to change to get better performance. Why not just implement the best
> algorithm
> across the board. Yes, this means collaboration or working on the
> other guys
> architecture.

Then 6/6 patch needs to put on hold if every arch needs to make ticket
lock as default spinlock lock strategy.

How about following to make forward progress:
1) Introduce rte_ticketlock_[lock/unlock/trylock/is_locked] API now as
experimental with default implementation
2) Provide a time line to switch every arch for optimized ticketlock
implementation if needed.
3) Switch rte_ticketlock_ as rte_spinlock_ API.
4) Keep old version of spinlock as new API if some application does not
need fairness between threads at the cost of light weight spinlock
implementation.

I don't want arm64 to behave differently than other arch(s).