Hi, Paul,
SFB can improve the memory bandwidth as much as 30%, and we are planning to enable SFB by default. So, we want to control cpu_relax() under CONFIG_CPU_LOONGSON3, not under CONFIG_LOONGSON3_ENHANCEMENT.
Huacai
------------------ Original ------------------
From: "Paul Burton"<paul.burton@xxxxxxxx>;
Date: Fri, Jul 20, 2018 05:15 AM
To: "Huacai Chen"<chenhc@xxxxxxxxxx>;
Cc: "Ralf Baechle"<ralf@xxxxxxxxxxxxxx>; "James Hogan"<jhogan@xxxxxxxxxx>; "linux-mips"<linux-mips@xxxxxxxxxxxxxx>; "Fuxin Zhang"<zhangfx@xxxxxxxxxx>; "wuzhangjin"<wuzhangjin@xxxxxxxxx>; "stable"<stable@xxxxxxxxxxxxxxx>; "Alan Stern"<stern@xxxxxxxxxxxxxxxxxxx>; "Andrea Parri"<andrea.parri@xxxxxxxxxxxxxxxxxxxx>; "Will Deacon"<will.deacon@xxxxxxx>; "Peter Zijlstra"<peterz@xxxxxxxxxxxxx>; "Boqun Feng"<boqun.feng@xxxxxxxxx>; "Nicholas Piggin"<npiggin@xxxxxxxxx>; "David Howells"<dhowells@xxxxxxxxxx>; "Jade Alglave"<j.alglave@xxxxxxxxx>; "Luc Maranget"<luc.maranget@xxxxxxxx>; "Paul E. McKenney"<paulmck@xxxxxxxxxxxxxxxxxx>; "Akira Yokosawa"<akiyks@xxxxxxxxx>; "LKML"<linux-kernel@xxxxxxxxxxxxxxx>;
Subject: Re: [PATCH] MIPS: Change definition of cpu_relax() for Loongson-3
Hi Huacai,
On Wed, Jul 18, 2018 at 09:15:46AM +0800, Huacai Chen wrote:
> >> diff --git a/arch/mips/include/asm/processor.h b/arch/mips/include/asm/processor.h
> >> index af34afb..a8c4a3a 100644
> >> --- a/arch/mips/include/asm/processor.h
> >> +++ b/arch/mips/include/asm/processor.h
> >> @@ -386,7 +386,17 @@ unsigned long get_wchan(struct task_struct *p);
> >> #define KSTK_ESP(tsk) (task_pt_regs(tsk)->regs[29])
> >> #define KSTK_STATUS(tsk) (task_pt_regs(tsk)->cp0_status)
> >>
> >> +#ifdef CONFIG_CPU_LOONGSON3
> >> +/*
> >> + * Loongson-3's SFB (Store-Fill-Buffer) may get starved when stuck in a read
> >> + * loop. Since spin loops of any kind should have a cpu_relax() in them, force
> >> + * a Store-Fill-Buffer flush from cpu_relax() such that any pending writes will
> >> + * become available as expected.
> >> + */
> >
> > I think "may starve writes" or "may queue writes indefinitely" would be
> > clearer than "may get starved".
>
> Need I change the comment and resend? Or you change the comment and get merged?
I'm happy to fix up the comment - but have a couple more questions.
Looking into the history, would it be fair to say that this is only a
problem after commit 1e820da3c9af ("MIPS: Loongson-3: Introduce
CONFIG_LOONGSON3_ENHANCEMENT") when CONFIG_LOONGSON3_ENHANCEMENT=y,
which adds code to enable the SFB?
If so would it make sense to use CONFIG_LOONGSON3_ENHANCEMENT to select
the use of smp_mb()?
How much does performance gain does enabling the SFB give you? Would it
be reasonable to just disable it, rather than using this workaround?
Thanks,
Paul
