linux-kernel - Re: [PATCH 1/4] spi: lantiq: Synchronize interrupt handlers and transfers

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: <28f6511e-fe85-a834-1652-fd70def9ca88@linux.intel.com>
Date:   Mon, 4 May 2020 18:15:47 +0800
From:   Dilip Kota <eswara.kota@...ux.intel.com>
To:     Mark Brown <broonie@...nel.org>
Cc:     Daniel Schwierzeck <daniel.schwierzeck@...il.com>, robh@...nel.org,
        linux-spi@...r.kernel.org, devicetree@...r.kernel.org,
        linux-kernel@...r.kernel.org, hauke@...ke-m.de,
        andriy.shevchenko@...el.com, cheol.yong.kim@...el.com,
        chuanhua.lei@...ux.intel.com, qi-ming.wu@...el.com
Subject: Re: [PATCH 1/4] spi: lantiq: Synchronize interrupt handlers and
 transfers


On 4/29/2020 8:13 PM, Mark Brown wrote:
> On Wed, Apr 29, 2020 at 04:20:53PM +0800, Dilip Kota wrote:
>> On 4/28/2020 7:10 PM, Daniel Schwierzeck wrote:
>>> actually there is no real bottom half. Reading or writing the FIFOs is
>>> fast and is therefore be done in hard IRQ context. But as the comment
>> Doing FIFO r/w in threaded irqs shouldn't cause any impact on maximum
>> transfer rate i think.
> Have you actually tested this?  Generally adding extra latency is going
> to lead to some opportunity for the hardware to idle and the longer the
> hardware is idle the lower the throughput.
>
>> Also the ISR should be quick enough, doing FIFO r/w in ISR adds up more
>> latency to ISR.
>> Handling the FIFOs r/w in threaded irq will be a better way.
> Consider what happens on a heavily loaded system - the threaded
> interrupt will have to be scheduled along with other tasks.
>
>>> for lantiq_ssc_bussy_work() state, the driver needs some busy-waiting
>>> after the last interrupt. I don't think it's worth to replace this with
>>> threaded interrupts which add more runtime overhead and likely decrease
>>> the maximum transfer speed.
>> Workqueue has a higher chances of causing SPI transfers timedout.
> because...?
I just tried to get the history of removing workqueue in SPI driver, on 
GRX500 (earlier chipset of LGM) the SPI transfers got timedout with 
workqueues during regression testing. Once changed to threaded IRQs 
transfers are working successfully.

Regards,
Dilip