netdev - Re: [PATCH] [net-next] stmmac: use netif_set_real_num_{rx,tx}

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:   Mon, 3 Apr 2017 14:12:10 +0100
From:   Joao Pinto <Joao.Pinto@...opsys.com>
To:     Giuseppe CAVALLARO <peppe.cavallaro@...com>,
        Joao Pinto <Joao.Pinto@...opsys.com>,
        Niklas Cassel <niklas.cassel@...s.com>,
        "Thierry Reding" <treding@...dia.com>,
        Corentin Labbe <clabbe.montjoie@...il.com>
CC:     David Miller <davem@...emloft.net>, <arnd@...db.de>,
        <alexandre.torgue@...com>, <netdev@...r.kernel.org>,
        <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] [net-next] stmmac: use netif_set_real_num_{rx,tx}_queues


Hello Peppe,

Às 2:07 PM de 4/3/2017, Giuseppe CAVALLARO escreveu:
> Hello Joao
> 
> On 3/30/2017 6:42 PM, Joao Pinto wrote:
>> Às 5:35 PM de 3/30/2017, Niklas Cassel escreveu:
>>> On 03/30/2017 04:34 PM, Thierry Reding wrote:
>>>> On Thu, Mar 30, 2017 at 09:45:36AM +0200, Corentin Labbe wrote:
>>>>> On Tue, Mar 28, 2017 at 06:01:05PM -0700, David Miller wrote:
>>>>>> From: Arnd Bergmann <arnd@...db.de>
>>>>>> Date: Tue, 28 Mar 2017 11:48:21 +0200
>>>>>>
>>>>>>> A driver must not access the two fields directly but should instead use
>>>>>>> the helper functions to set the values and keep a consistent internal
>>>>>>> state:
>>>>>>>
>>>>>>> ethernet/stmicro/stmmac/stmmac_main.c: In function 'stmmac_dvr_probe':
>>>>>>> ethernet/stmicro/stmmac/stmmac_main.c:4083:8: error: 'struct net_device'
>>>>>>> has no member named 'real_num_rx_queues'; did you mean 'real_num_tx_queues'?
>>>>>>>
>>>>>>> Fixes: a8f5102af2a7 ("net: stmmac: TX and RX queue priority configuration")
>>>>>>> Signed-off-by: Arnd Bergmann <arnd@...db.de>
>>>>>> Applied.
>>>>> This break my revert patch. (since it patch ("net: stmmac: enable multiple
>>>>> buffers").
>>>>> Since dwmac-sunxi is still broken, what can I do ? send two revert patch ?
>>>>> or adapt the reverting patch.
>>>> Have you tried if the kcalloc() patch I sent on Tuesday fixes things the
>>>> issues introduced by the multiple buffers patch? Niklas reported that it
>>>> restores functionality on his setup.
>>>>
>>>> If it makes things work for you as well, we could maybe avoid the revert
>>>> altogether.
>>> Thierry, I know that you are using DWMAC CORE 4.XX
>>> How many RX queues and how many TX queues have you got?
>>>
>>> I'm also using DWMAC CORE 4.XX
>>> We have 2 TX queues and 1 RX queue.
>>>
>>> I think that Corentin is using DWMAC CORE 3.XX
>>>
>>> I know that Joao is using an IP Prototyping Kit that uses
>>> DWMAC CORE 4.XX (connected via PCIe).
>>> It would be nice if Joao could get an IP Prototyping Kit
>>> based on DWMAC CORE 3.XX.
>>>
>>> Doesn't Synopsys have an IP Prototyping Kit based on
>>> DWMAC CORE 3.XX laying around somewhere? :)
>>>
>> I requested a prototyping platform with MAC 100 or a MAC 1000 in order to make
>> more tests, but I don't have an ETA for it yet.
>>
>> The implication of the multiple buffers patch in 3.xx is some flow change in the
>> configuration of dma op mode or similar. I would recomend Corentin to dump the
>> dma & mac registers in the end of the _open function in order to see if the DMA
>> is really being well configured and is really started.
> 
> Old devices managed as stmmac: mac10/100 platform driver have no multi-queue
> support.
> 
> It was introduced in the new versions managed by mac1000 but in my opinion
> the stmmac driver has to only work with a single queue for tx and rx on ALL
> the 3.x versions.
> 
> In fact, although the latest series have the multi-queue and channels there is
> just one IRQ and this is a very poor implementation.
> The 3.xx is very different from what we have in the 4.XX (QoS) on this part and
> AV/B.
> 
> Nobody has ever asked for having this support in all these years for the previous
> chips.
> 
> Adding this kind of support we'll spend time for having a no useful optimization
> and risking to break the compatibility with a lot platforms that use these chips.
> 
> I image, multi-queue stable on 4.XX and no support on all older versions.
> Let me know if you share that.

Yes older cores do not support multiple queues and I tried to isolate the
features not to affect older versions.

Do you think that functions as "ndev = alloc_etherdev_mqs" has some sort of
influence?

Thanks.

> 
> Regards
> Peppe
> 
>>
>> Thanks.
>>
>> Joao
>>
>>
>