Date:	Mon, 18 Nov 2013 23:29:21 +0530
From:	Sourav Poddar <sourav.poddar@...com>
To:	Lee Jones <lee.jones@...aro.org>
CC:	Mark Brown <broonie@...nel.org>, <angus.clark@...com>,
	Linus Walleij <linus.walleij@...aro.org>,
	"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
	"linux-mtd@...ts.infradead.org" <linux-mtd@...ts.infradead.org>,
	David Woodhouse <dwmw2@...radead.org>,
	"linux-arm-kernel@...ts.infradead.org" 
	<linux-arm-kernel@...ts.infradead.org>
Subject: Re: [PATCH 02/10] mtd: st_spi_fsm: Supply all register address and
 bit logic defines

On Monday 18 November 2013 11:02 PM, Lee Jones wrote:
> On Mon, 18 Nov 2013, Mark Brown wrote:
>
>> On Mon, Nov 18, 2013 at 04:02:26PM +0000, Lee Jones wrote:
>>> On Mon, 18 Nov 2013, Mark Brown wrote:
>>>> Like I say I'm suggesting that the bit of the code that understands the
>>>> flash chip is separate to the bit of code that knows the mechanics of
>>>> sending commands and data to the chip.
>>> The issue is that almost the entire driver is controller side. The
>>> only bits that are the same (and not in all cases) are the OPCODEs,
>>> but they are one liners (21 lines out of 1153). Most of the
>>> controllers which use this stuff could reuse quite a bit of the m25p80
>>> driver as they just write the message containing the OPCODE as the
>>> m25p80 driver sets it up, but that's simply not the case with our
>>> controller. We would have to pull the OPCODE out and based on which
>>> one it is, we'd have to build our own message.
>> OK, so then perhaps the abstraction here is simply to export the table
>> with the opcodes from the m25p80 driver so that when someone comes along
>> and adds a new chip they can just add it there and other drivers will
>> get the update too.
> We could do that, although I'd have to insist on extending the current
> framework to add a configuration call-back, as it's the neatest way to
> configure chip specific attributes.
>
This looks like the problem which some other controllers(from ti,
freescale) are facing.
I will just summarise the problem with the ti qspi flash controller 
which I am
working on. There is a set of registers which need to be filled with flash
specific commands. One way to deal it with to provide a device tree bindings
for all the requirements(which is really cumbersome.).

Similarly, for freescale there is a LUT registers which has such flash 
requirements.

So, surely we need a way out from m25p80 driver to handle such cases.

drivers/mtd/devices/m25p80.c


int pass_flash_info () {

    u8 info[6];

      info[0] = DUMMY WR;
      info[1] = RD_OPCODE;
      info[2] - DUMMY BITS;
      ......
     .......
    spi_write(flash, info, 6)
}
Then, somehow parse this information to set up the required info.
  This is just a rough idea, and can be implemented in a better way.
>   I can get a patch out tomorrow if the MTD guys agree. Where are they
> by the way? I haven't seen hide nor hair of them since sending out the
> patch set.
>
>>> Put it this way, if we tried to use the m25p80 our controller driver
>>> would most likely be twice as large and twice as complex as it is
>>> currently, which is exactly the inverse of what we're trying to
>>> achieve here.
>> If we're having to add new flashes to multiple drivers I'd not say we're
>> winning.
> I agree.
>

--
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