Re: [POC][RFC][PATCH v2] sched: Extended Scheduler Time Slice

From: Steven Rostedt <rostedt@goodmis.org>
To: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>,
	Peter Zijlstra <peterz@infradead.org>,
	LKML <linux-kernel@vger.kernel.org>,
	Thomas Gleixner <tglx@linutronix.de>,
	Ankur Arora <ankur.a.arora@oracle.com>,
	linux-mm@kvack.org, x86@kernel.org, akpm@linux-foundation.org,
	luto@kernel.org, bp@alien8.de, dave.hansen@linux.intel.com,
	hpa@zytor.com, mingo@redhat.com, juri.lelli@redhat.com,
	vincent.guittot@linaro.org, willy@infradead.org, mgorman@suse.de,
	jon.grimm@amd.com, bharata@amd.com, raghavendra.kt@amd.com,
	boris.ostrovsky@oracle.com, konrad.wilk@oracle.com,
	jgross@suse.com, andrew.cooper3@citrix.com,
	Joel Fernandes <joel@joelfernandes.org>,
	Youssef Esmat <youssefesmat@chromium.org>,
	Vineeth Pillai <vineethrp@google.com>,
	Suleiman Souhlal <suleiman@google.com>,
	Ingo Molnar <mingo@kernel.org>,
	Daniel Bristot de Oliveira <bristot@kernel.org>
Subject: Re: [POC][RFC][PATCH v2] sched: Extended Scheduler Time Slice
Date: Mon, 30 Oct 2023 14:39:04 -0400	[thread overview]
Message-ID: <20231030143904.5db873b9@gandalf.local.home> (raw)
In-Reply-To: <c573ef1a-b269-469c-a63f-6ec9c137d789@efficios.com>

On Mon, 30 Oct 2023 14:27:10 -0400
Mathieu Desnoyers <mathieu.desnoyers@efficios.com> wrote:

> > So I just made every unlock disable the extended time slot. I need to go
> > back and enable both a counter and an on/off as I now realize that the spin
> > locks (called within the lwlock) will disable the extend time before the
> > lwlock is released. This should work if I have the spinlocks inc and dec
> > (they are straight forward and all locks are associated with an easily
> > found unlock), and have the lwlock use bit 31 as an on/off switch.  
> 
> This extra on/off switch appears to be working around userspace issues.

Yep!

But that doesn't mean there's not a legitimate use case for it. I don't
want to limit the feature for that. It's unlikely bit 31 would ever be hit
by a counter anyway, for which it could be used as an on/off switch the
same way the NEED_RESCHED bit is used as an on/off switch for preempt_count
in the kernel.

> 
> > Anyway, I would let user space decide what it wants to do, and giving it 31
> > bits to say "I'm extended" and let user space come up with how it handles
> > those 31 bits.  
> 
> If this makes it into the RSEQ uapi, RSEQ should state how userspace
> should collaborate wrt those bits (e.g. nesting counter protocol), even
> though it's not a kernel ABI per se. Otherwise we'll just push this to
> libc to specify this, which is odd.

I agree that user space should have the usage specified. Hell, that bit
could just be used for testing purposes. I think having it reserved is a
good thing than not specifying it and limiting its usage later.

-- Steve