linux-kernel - Re: [PATCH] dmaengine: delete redundant chan_id and chancnt initialization in dma drivers

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, 22 Sep 2011 13:55:54 +0800
From:	Barry Song <21cnbao@...il.com>
To:	Jassi Brar <jassisinghbrar@...il.com>
Cc:	Barry Song <Baohua.Song@....com>,
	Piotr Ziecik <kosmo@...ihalf.com>,
	Jaswinder Singh <jassi.brar@...sung.com>,
	Vinod Koul <vinod.koul@...el.com>,
	Pelagicore AB <info@...agicore.com>,
	Nicolas Ferre <nicolas.ferre@...el.com>,
	linux-kernel@...r.kernel.org, Yong Wang <yong.y.wang@...el.com>,
	dan.j.williams@...el.com, linux-arm-kernel@...ts.infradead.org
Subject: Re: [PATCH] dmaengine: delete redundant chan_id and chancnt
 initialization in dma drivers

2011/9/22 Jassi Brar <jassisinghbrar@...il.com>:
> On Thu, Sep 22, 2011 at 5:38 AM, Barry Song <21cnbao@...il.com> wrote:
>> 2011/9/21 Jassi Brar <jassisinghbrar@...il.com>:
>>> On Fri, Sep 16, 2011 at 3:13 PM, Barry Song <Baohua.Song@....com> wrote:
>>>> dma_async_device_register will re-init chan_id and chancnt,
>>>> so whatever chan_id and chancnt are set in drivers, they will
>>>> be re-written by dma_async_device_register.
>>>>
>>>> Cc: Nicolas Ferre <nicolas.ferre@...el.com>
>>>> Cc: Viresh Kumar <viresh.kumar@...com>
>>>> Cc: Vinod Koul <vinod.koul@...el.com>
>>>> Cc: Piotr Ziecik <kosmo@...ihalf.com>
>>>> Cc: Yong Wang <yong.y.wang@...el.com>
>>>> Cc: Jaswinder Singh <jassi.brar@...sung.com>
>>>> Cc: Pelagicore AB <info@...agicore.com>
>>>> Signed-off-by: Barry Song <Baohua.Song@....com>
>>>> ---
>>>>  drivers/dma/at_hdmac.c      |    5 ++---
>>>>  drivers/dma/dw_dmac.c       |    5 ++---
>>>>  drivers/dma/intel_mid_dma.c |    2 --
>>>>  drivers/dma/mpc512x_dma.c   |    1 -
>>>>  drivers/dma/pch_dma.c       |    2 --
>>>>  drivers/dma/pl330.c         |    2 --
>>>>  drivers/dma/timb_dma.c      |    3 +--
>>>
>>> Apparently ....
>>>
>>> drivers/dma/ppc4xx/adma.c
>>> drivers/dma/ipu/ipu_idmac.c
>>>       still write to chan_id.
>>>
>>> drivers/dma/amba-pl08x.c
>>> drivers/dma/fsldma.c
>>> drivers/dma/ioat/dma_v2.c
>>> drivers/dma/ioat/dma.c
>>> drivers/dma/mpc512x_dma.c
>>> drivers/dma/shdma.c
>>>       still write to chancnt
>>>
> My question was if you really intended to convert only some of
> the drivers and not others?

i intended to convert all. but i lost some....

>
>>> Most of them are simply a matter of removal, but some seem
>>> like really depending upon setting them(?)
>>>
>>> Anyways, even after you modify those as well, chan_id and chancnt
>>> are rendered overstaying guests in dmaengine core. Because chan_id
>>> of each channel would be _precisely_ the order in which the
>>> _dmac-driver_ added the channel to the 'channels' list.
>>>
>>> So if their values are _always_ gonna be just contiguously incrementing
>>> why need variables for that in the dmaengine api?
>>> Dmac drivers could use local variables for that.
>>>
>>> OTOH, why not chan_id be left solely for use by dmac drivers read by the
>>> dmaengine only to create sysfs entries ?
>>
>> actually i don't like the way dmaengine core handles chan_id. in my
>> opinion, every dmac only needs to tell dmaengine core the chan_base.
>> then dmaengine core set the id in a global scale.
>>
>> for example, if you have two dmac in system:
>> dmac0 chan_base 0
>> dmac1 chan_base 16
>> then core will have chan 0~31 by chan_id from 0 to 31 but not
>> dmac0chan0~15, dmac1chan0~15.
> But note the numbering is still decided by the dmac driver - directly
> or indirectly.
> That is, if the dmac driver reorder registering of dmac or adding
> channels to the 'channels' list, the assigned chan_id's would change.
> So why not rather get rid of chan_id assignment from dmaengine
> and do explicitly in dmac drivers ?
> Which I already submitted a patch for https://lkml.org/lkml/2011/7/21/1

i don't agree chan_id is just a property of dmac. i think chan_id is
also a property core wants.
why could not dmac driver decide the chan_id in dmaengine core? i
think that is just what dmac driver wants.dmac just tells the base,
core re-number them one by one.
if dmac can't decide the ID directly or indirectly, this will be
random. how could client driver request the channel it wants?

>
>> client drivers can request a specific chan_id just like it can request a gpio.
> No dear. That's not gonna fly if we are to have common client drivers
> across SoCs.
>
>> now dmac drivers and dmaengine core have repeated chan_id and chancnt,
>> anyway, it is a problem that should be fixed at first.
> Correct, chan_id assignment should be completely either in dmaengine
> or in dmac drivers. But this patch does neither.
>
i like the way gpiolib handles ID more.

-barry
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/