netdev - Re: [PATCH net-next v3 08/11] net: ethernet: ti: cpts: move rx timestamp processing to ptp worker only

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: <13dd9d58-7417-2f39-aa7d-dceae946482c@ti.com>
Date:   Tue, 24 Mar 2020 17:34:34 +0200
From:   Grygorii Strashko <grygorii.strashko@...com>
To:     Richard Cochran <richardcochran@...il.com>
CC:     "David S . Miller" <davem@...emloft.net>,
        Lokesh Vutla <lokeshvutla@...com>,
        Tony Lindgren <tony@...mide.com>, Sekhar Nori <nsekhar@...com>,
        Murali Karicheri <m-karicheri2@...com>,
        netdev <netdev@...r.kernel.org>, <linux-omap@...r.kernel.org>,
        <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH net-next v3 08/11] net: ethernet: ti: cpts: move rx
 timestamp processing to ptp worker only



On 24/03/2020 15:43, Richard Cochran wrote:
> On Fri, Mar 20, 2020 at 09:42:41PM +0200, Grygorii Strashko wrote:
>> Once CPTS IRQ will be enabled the CPTS irq handler may compete with netif
>> RX sofirq path and so RX timestamp might not be ready at the moment packet
>> is processed. As result, packet has to be deferred and processed later.
> 
> This change is not necessary.  The Rx path can simply take a spinlock,
> check the event list and the HW queue.
>   
>> This patch moves RX timestamp processing tx timestamp processing to PTP
>> worker always the same way as it's been done for TX timestamps.
> 
> There is no advantage to delaying Rx time stamp delivery.  In fact, it
> can degrade synchronization performance.  The only reason the
> implementation delays Tx time stamps delivery is because there is no
> other way.

I tested both ways and kept this version as i'v not seen any degradation,
but, of course, i'll redo the test (or may be you can advise what test to run).

My thoughts were - network stack might not immediately deliver packet to the application
and PTP worker can be tuned (pri and smp_affinity), resulted code will be more structured,
less locks and less time spent in softirq context.

I also can drop it.

-- 
Best regards,
grygorii