Date:	Thu, 25 Apr 2013 14:43:16 +0200
From:	Linus Walleij <linus.walleij@...aro.org>
To:	Lee Jones <lee.jones@...aro.org>
Cc:	"linux-arm-kernel@...ts.infradead.org" 
	<linux-arm-kernel@...ts.infradead.org>,
	"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
	Arnd Bergmann <arnd@...db.de>,
	Linus WALLEIJ <linus.walleij@...ricsson.com>,
	Vinod Koul <vinod.koul@...el.com>, Dan Williams <djbw@...com>,
	Per Forlin <per.forlin@...ricsson.com>,
	Rabin Vincent <rabin@....in>
Subject: Re: [PATCH 04/32] dmaengine: ste_dma40: Amalgamate DMA source and
 destination channel numbers

On Thu, Apr 25, 2013 at 11:06 AM, Lee Jones <lee.jones@...aro.org> wrote:

>> Are we now sacrificing that ability on the altar of simplification?
>>
>> I actually think not, but that we should do periph-to-periph transfers
>> in some other way, and that the .dir attribute should go away from
>> the struct stedma40_chan_cfg as well but I'm not entirely sure.
>> Someone else?
>
> Although the DMA40 device supports device-to-device transfers, Linux
> does not, so this subject is moot AFAICT.

So while there is no active usecase, Linux surely has the ambition to do
that as can be seen in:

/**
 * enum dma_transfer_direction - dma transfer mode and direction indicator
 * @DMA_MEM_TO_MEM: Async/Memcpy mode
 * @DMA_MEM_TO_DEV: Slave mode & From Memory to Device
 * @DMA_DEV_TO_MEM: Slave mode & From Device to Memory
 * @DMA_DEV_TO_DEV: Slave mode & From Device to Device
 */
enum dma_transfer_direction {
        DMA_MEM_TO_MEM,
        DMA_MEM_TO_DEV,
        DMA_DEV_TO_MEM,
        DMA_DEV_TO_DEV,
        DMA_TRANS_NONE,
};

I think we need a handshake with Vinod on this.

>> If you're doing this change, and after this RX and TX has no semantical
>> meaning for these lists, join these two config lists
>> into one.
>
> I agree. See patch: ARM: ux500: Strip out duplicate USB DMA configuration

Please squash the applicable portions into this patch then, I don't
particularly like fix-later patchstack patterns, it makes series hard to
review.

>> (...)
>> > diff --git a/arch/arm/mach-ux500/usb.c b/arch/arm/mach-ux500/usb.c
>> >  static u32 d40_chan_has_events(struct d40_chan *d40c)
>> > @@ -1744,8 +1740,6 @@ static int d40_validate_conf(struct d40_chan *d40c,
>> >                              struct stedma40_chan_cfg *conf)
>> >  {
>> >         int res = 0;
>> > -       u32 dst_event_group = D40_TYPE_TO_GROUP(conf->dst_dev_type);
>> > -       u32 src_event_group = D40_TYPE_TO_GROUP(conf->src_dev_type);
>>
>> Please explain why this is not important to check anymore, I'm not
>> following.
>>
>> >         if (conf->dir == STEDMA40_MEM_TO_PERIPH &&
>> > -           dst_event_group == STEDMA40_DEV_DST_MEMORY) {
>> > -               chan_err(d40c, "Invalid dst\n");
>> > +           d40c->base->plat_data->dev_tx[conf->dev_type] == 0 &&
>> > +           d40c->runtime_addr == 0) {
>> > +               chan_err(d40c, "Invalid TX channel address (%d)\n",
>> > +                        conf->dev_type);
>>
>> Like here. We are checking for inconsistency between group
>> and channel direction, why is it no longer important to check this?
>
> I'm not entirely sure how this ever worked:
>
>   #define D40_TYPE_TO_GROUP(type) (type / 16)
>   #define STEDMA40_DEV_DST_MEMORY (-1)
>
>   (dev_type / 16) == -1
>
> What number would dev_type have to be for this to be true? -16?

No, since it's u32 it cannot really represent negative numbers.

This is equivalent:
#define D40_TYPE_TO_GROUP(type) (type >> 4)

As -1 is 0xffffffff in u32 it will compare at best
0x0fffffff to 0xfffffff. And that is non-attainable.

So the line checking event group for == DTEDMA40_DEV_DST_MEMORY
should be  removed in a separate patch prior to this one, with
something like the above as commit message.

We cannot really mix that cleanup into this patch...

>> >         if (conf->dir == STEDMA40_PERIPH_TO_MEM &&
>> > -           src_event_group == STEDMA40_DEV_SRC_MEMORY) {
>> > -               chan_err(d40c, "Invalid src\n");
>> > -               res = -EINVAL;
>> > -       }
>
> As above.

And same comment.

>> > -       if (conf->dir == STEDMA40_PERIPH_TO_PERIPH &&
>> > -           (src_event_group != dst_event_group)) {
>> > -               chan_err(d40c, "Invalid event group\n");
>> > +           d40c->base->plat_data->dev_rx[conf->dev_type] == 0 &&
>> > +           d40c->runtime_addr == 0) {
>> > +               chan_err(d40c, "Invalid RX channel address (%d)\n",
>> > +                        conf->dev_type);
>>
>> Same here.
>
> I stopped all 'dev_src/dev_dest' comparisons, as there is only 'dev' now.

It is checking for:
conf->dir == STEDMA40_PERIPH_TO_PERIPH

As we may want to support DEV_TO_DEV at some point.

Then no longer, and that is not related to $SUBJECT.

>> (...)
>> > @@ -2062,7 +2035,7 @@ static int d40_free_dma(struct d40_chan *d40c)
>> >  {
>> >
>> >         int res = 0;
>> > -       u32 event;
>> > +       u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type);
>> >         struct d40_phy_res *phy = d40c->phy_chan;
>> >         bool is_src;
>> >
>> > @@ -2081,13 +2054,11 @@ static int d40_free_dma(struct d40_chan *d40c)
>> >         }
>> >
>> >         if (d40c->dma_cfg.dir == STEDMA40_MEM_TO_PERIPH ||
>> > -           d40c->dma_cfg.dir == STEDMA40_MEM_TO_MEM) {
>> > -               event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dst_dev_type);
>> > +           d40c->dma_cfg.dir == STEDMA40_MEM_TO_MEM)
>>
>> Why did you just stop checking dma_cfg.dir for == STEDMA40_MEM_TO_MEM
>> above?
>
> That's not what this is doing. STEDMA40_MEM_TO_MEM is still there.
>
>> And why is dma_cfg.dir suddenly hardcoded to MEM_TO_MEM??
>
> It's not. Look again. :)

Argh I misread == MEM_TO_MEM for = MEM_TO_MEM ...
comparison to assignment. Sorry.

>> This seems like a totally unrelated change and should it be done
>> it need to be a separate patch with a separate explanation
>> AFAICT.
>>
>> This seems to happen in some other places too,
>
> If you could point those out, I'll re-evaluate, or explain.

I'm after that the change to omit checks for some impossible
type/group configs need to be a separate patch.

>> and I find it
>> very hard to follow the changes here ... can you please consider
>> splitting the changes to groups and types semantics into a separate
>> patch?
>
> Can you read the patch again and reconsider please?

Yes and now I am even more convinced that the patch needs
to be split.

Yours,
Linus Walleij
--
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/

Powered by blists - more mailing lists