linux-kernel - Re: [PATCH] nvme: Check for readiness more quickly, to speed up boot time

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20200301191501.GA235404@localhost>
Date:   Sun, 1 Mar 2020 11:15:01 -0800
From:   Josh Triplett <josh@...htriplett.org>
To:     Keith Busch <kbusch@...nel.org>
Cc:     Jens Axboe <axboe@...com>, Christoph Hellwig <hch@....de>,
        Sagi Grimberg <sagi@...mberg.me>,
        linux-nvme@...ts.infradead.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH] nvme: Check for readiness more quickly, to speed up boot
 time

On Sun, Mar 01, 2020 at 10:32:31AM -0800, Keith Busch wrote:
> On Fri, Feb 28, 2020 at 06:52:28PM -0800, Josh Triplett wrote:
> > @@ -2074,7 +2074,7 @@ static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
> >  		if ((csts & NVME_CSTS_RDY) == bit)
> >  			break;
> >  
> > -		msleep(100);
> > +		usleep_range(1000, 2000);
> >  		if (fatal_signal_pending(current))
> >  			return -EINTR;
> >  		if (time_after(jiffies, timeout)) {
> 
> The key being this sleep schedules the task unlike udelay.

Right; I don't think it's reasonable to busyloop here, just sleep for
less time.

> It's neat you can boot where 100ms is considered a long time.

It's been fun. This was one of the longest single delays in a ~1s boot.

> This clearly helps when you've one nvme that becomes ready quickly, but
> what happens with many nvme's that are slow to ready? This change will
> end up polling the status of those 1000's of times, I wonder if there's
> a point where this frequent sleep/wake cycle initializing a whole lot
> of nvme devices in parallel may interfere with other init tasks.

usleep_range allows the kernel to consolidate those wakeups, so if you
have multiple NVMe devices, the kernel should in theory just wake up
once, check them all for readiness, and go back to sleep.

> I doubt there's really an issue there, but thought it's worth considering
> what happens at the other end of the specturm.
> 
> Anyway, the patch looks fine to me.
> 
> Reviewed-by: Keith Busch <kbusch@...nel.org>

Thank you!

Does this seem reasonable to enqueue for 5.7?

- Josh Triplett