linux-kernel - Re: [PATCH V2] arm64: psci: Reduce waiting time of cpu_psci_cpu

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]

Date:   Thu, 17 Oct 2019 14:54:16 +0100
From:   Sudeep Holla <sudeep.holla@....com>
To:     Yunfeng Ye <yeyunfeng@...wei.com>
Cc:     David Laight <David.Laight@...LAB.COM>,
        "catalin.marinas@....com" <catalin.marinas@....com>,
        "will@...nel.org" <will@...nel.org>,
        "kstewart@...uxfoundation.org" <kstewart@...uxfoundation.org>,
        "gregkh@...uxfoundation.org" <gregkh@...uxfoundation.org>,
        "ard.biesheuvel@...aro.org" <ard.biesheuvel@...aro.org>,
        "tglx@...utronix.de" <tglx@...utronix.de>,
        "wuyun.wu@...wei.com" <wuyun.wu@...wei.com>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
        "linux-arm-kernel@...ts.infradead.org" 
        <linux-arm-kernel@...ts.infradead.org>, hushiyuan@...wei.com,
        linfeilong@...wei.com, Sudeep Holla <sudeep.holla@....com>
Subject: Re: [PATCH V2] arm64: psci: Reduce waiting time of
 cpu_psci_cpu_kill()

On Thu, Oct 17, 2019 at 09:26:15PM +0800, Yunfeng Ye wrote:
>
>
> On 2019/10/16 23:32, Sudeep Holla wrote:
> > On Wed, Oct 09, 2019 at 12:45:16PM +0800, Yunfeng Ye wrote:
> >> If psci_ops.affinity_info() fails, it will sleep 10ms, which will not
> >> take so long in the right case. Use usleep_range() instead of msleep(),
> >> reduce the waiting time, and give a chance to busy wait before sleep.
> >>
> >> Signed-off-by: Yunfeng Ye <yeyunfeng@...wei.com>
> >> ---
> >> V1->V2:
> >> - use usleep_range() instead of udelay() after waiting for a while
> >>
> >>  arch/arm64/kernel/psci.c | 17 +++++++++++++----
> >>  1 file changed, 13 insertions(+), 4 deletions(-)
> >>
> >> diff --git a/arch/arm64/kernel/psci.c b/arch/arm64/kernel/psci.c
> >> index c9f72b2..99b3122 100644
> >> --- a/arch/arm64/kernel/psci.c
> >> +++ b/arch/arm64/kernel/psci.c
> >> @@ -82,6 +82,7 @@ static void cpu_psci_cpu_die(unsigned int cpu)
> >>  static int cpu_psci_cpu_kill(unsigned int cpu)
> >>  {
> >>  	int err, i;
> >> +	unsigned long timeout;
> >>
> >>  	if (!psci_ops.affinity_info)
> >>  		return 0;
> >> @@ -91,16 +92,24 @@ static int cpu_psci_cpu_kill(unsigned int cpu)
> >>  	 * while it is dying. So, try again a few times.
> >>  	 */
> >>
> >> -	for (i = 0; i < 10; i++) {
> >> +	i = 0;
> >> +	timeout = jiffies + msecs_to_jiffies(100);
> >> +	do {
> >>  		err = psci_ops.affinity_info(cpu_logical_map(cpu), 0);
> >>  		if (err == PSCI_0_2_AFFINITY_LEVEL_OFF) {
> >>  			pr_info("CPU%d killed.\n", cpu);
> >>  			return 0;
> >>  		}
> >>
> >> -		msleep(10);
> >> -		pr_info("Retrying again to check for CPU kill\n");
> >
> > You dropped this message, any particular reason ?
> >
> When reduce the time interval to 1ms, the print message maybe increase 10
> times. on the other hand, cpu_psci_cpu_kill() will print message on success
> or failure, which this retry log is not very necessary. of cource, I think
> use pr_info_once() instead of pr_info() is better.
>

Yes changing it to pr_info_once is better than dropping it as it gives
some indication to the firmware if there's scope for improvement.

> >> -	}
> >> +		/* busy-wait max 1ms */
> >> +		if (i++ < 100) {
> >> +			cond_resched();
> >> +			udelay(10);
> >> +			continue;
> >
> > Why can't it be simple like loop of 100 * msleep(1) instead of loop of
> > 10 * msleep(10). The above initial busy wait for 1 ms looks too much
> > optimised for your setup where it takes 50-500us, what if it take just
> > over 1 ms ?
> >
> msleep() is implemented by jiffies. when HZ=100 or HZ=250, msleep(1) is not
> accurate. so I think usleep_range() is better. 1 ms looks simple and good, but how
> about 100us is better? I refer a function sunxi_mc_smp_cpu_kill(), it use
> usleep_range(50, 100).
>

Again that's specific to sunxi platforms and may work well. While I agree
msleep(1) may not be accurate, I am still inclined to have a max value
of 1000(i.e. 1ms) for usleep_range.

--
Regards,
Sudeep