[U-Boot] MMC: proposal to support multiple physical partitions

[U-Boot] MMC: proposal to support multiple physical partitions

[Steve Rae]

In addition to using the MMC "user area" (the device's physical partition), I am also using the "first MMC boot partition" and the "second MMC boot partition"...
As a result, I am currently using the following code snippet in a number of places:

err = -1;
if (mmc->part_num != part_num) {
        if (mmc_switch_part(dev_num, part_num)) {
                printf("%s: MMC partition switch to %d failed\n",
                       __func__, part_num);
                err = 0;
        }
}

if (err != 0) {
        err = mmc->block_dev.block_read(dev_num, start, blkcnt, buffer);
}

if (mmc->part_num != part_num) {
        if (mmc_switch_part(dev_num, mmc->part_num)) {
                printf("%s: MMC partition switching back from %d failed\n",
                       __func__, part_num);
        }
}

I have two different proposals:
1) overload the "int dev_num" argument with encoded  "dev_num" and "part_num" fields

-          the dev_num  in the [15:0] bits,

-          the part_num in the [30:16] bits,

-          a flag to indicate this encoding in [31] bit.

-          and modify mmc_bread() to handle this encoded argument, and implement the above code...
2) create a wrapper function to perform the above code, with an added argument "int part_num", possibly named:

-          mmc_block_dev_block_read() -- so that it is similar to the original calling convention [mmc->block_dev.block_read], or

-          mmc_pbread() [PartitionBlockRead] -- so that it is similar to the mmc_bread() [which is the implementation of the callback function]



Also, would implement this solution for mmc->block_dev.block_write() and mmc->block_dev.block_erase() too.

Either proposals would affect:

include/mmc.h

drivers/mmc/mmc.c

drivers/mmc/mmc_write.c



Would either of these proposals be acceptable to upstream?

Thanks in advance, Steve

[U-Boot] MMC: proposal to support multiple physical partitions

[Pantelis Antoniou]

Hi Steve,

On Apr 24, 2014, at 8:50 PM, Steve Rae wrote:

> In addition to using the MMC "user area" (the device's physical partition), I am also using the "first MMC boot partition" and the "second MMC boot partition"...
> As a result, I am currently using the following code snippet in a number of places:
>  
> err = -1;
> if (mmc->part_num != part_num) {
>         if (mmc_switch_part(dev_num, part_num)) {
>                 printf("%s: MMC partition switch to %d failed\n",
>                        __func__, part_num);
>                 err = 0;
>         }
> }
>  
> if (err != 0) {
>         err = mmc->block_dev.block_read(dev_num, start, blkcnt, buffer);
> }
>  
> if (mmc->part_num != part_num) {
>         if (mmc_switch_part(dev_num, mmc->part_num)) {
>                 printf("%s: MMC partition switching back from %d failed\n",
>                        __func__, part_num);
>         }
> }
>  
> I have two different proposals:
> 1) overload the "int dev_num" argument with encoded  "dev_num" and "part_num" fields
> -          the dev_num  in the [15:0] bits,
> -          the part_num in the [30:16] bits,
> -          a flag to indicate this encoding in [31] bit.
> -          and modify mmc_bread() to handle this encoded argument, and implement the above code...
> 2) create a wrapper function to perform the above code, with an added argument "int part_num", possibly named:
> -          mmc_block_dev_block_read() -- so that it is similar to the original calling convention [mmc->block_dev.block_read], or
> -          mmc_pbread() [PartitionBlockRead] -- so that it is similar to the mmc_bread() [which is the implementation of the callback function]
>  


I'd rather go with the wrapper function. Perhaps it's not even needed. The function called is part of the block_dev (block_read/write etc).

Overwrite those with functions that implemented the switching first, and then call the original block* function.

> Also, would implement this solution for mmc->block_dev.block_write() and mmc->block_dev.block_erase() too.
> Either proposals would affect:
> include/mmc.h
> drivers/mmc/mmc.c
> drivers/mmc/mmc_write.c
>  
> Would either of these proposals be acceptable to upstream?
> Thanks in advance, Steve

Anything that cleans things up is acceptable.

Regards

-- Pantelis

[U-Boot] MMC: proposal to support multiple physical partitions

[Steve Rae]

Thanks for the response,

> -----Original Message-----
> From: Pantelis Antoniou [mailto:panto at antoniou-consulting.com]
> Sent: Tuesday, April 29, 2014 10:25
> To: Steve Rae
> Cc: u-boot at lists.denx.de; trini at ti.com; albert.u.boot at aribaud.net
> Subject: Re: MMC: proposal to support multiple physical partitions
> 
> Hi Steve,
> 
> On Apr 24, 2014, at 8:50 PM, Steve Rae wrote:
> 
[... snip ...]

> > I have two different proposals:
> > 1) overload the "int dev_num" argument with encoded  "dev_num" and
> "part_num" fields
> > -          the dev_num  in the [15:0] bits,
> > -          the part_num in the [30:16] bits,
> > -          a flag to indicate this encoding in [31] bit.
> > -          and modify mmc_bread() to handle this encoded argument, and
> implement the above code...
> > 2) create a wrapper function to perform the above code, with an added
> argument "int part_num", possibly named:
> > -          mmc_block_dev_block_read() -- so that it is similar to the original
> calling convention [mmc->block_dev.block_read], or
> > -          mmc_pbread() [PartitionBlockRead] -- so that it is similar to the
> mmc_bread() [which is the implementation of the callback function]
> >
> 
> 
> I'd rather go with the wrapper function. Perhaps it's not even needed. The
> function called is part of the block_dev (block_read/write etc).
> 
> Overwrite those with functions that implemented the switching first, and
> then call the original block* function.
>
The callback function is:
    mmc->block_dev.block_read = mmc_bread
and it accepts four arguments:
    include/part.h: unsigned long   (*block_read)(int dev,
    include/part.h-                               lbaint_t start,
    include/part.h-                               lbaint_t blkcnt,
    include/part.h-                               void *buffer);
Are you suggesting that I should add another argument to this callback?

[...snip...]

> 
> Regards
> 
> -- Pantelis

[U-Boot] MMC: proposal to support multiple physical partitions

[Steve Rae]

Pantelis,

As per comments below, I suspect that adding another argument to the callback function is unacceptable. Please clarify!

Thanks, Steve

On 14-04-29 11:08 AM, Steve Rae wrote:

> Thanks for the response,
>
>> -----Original Message-----
>> From: Pantelis Antoniou [mailto:panto at antoniou-consulting.com]
>> Sent: Tuesday, April 29, 2014 10:25
>> To: Steve Rae
>> Cc: u-boot at lists.denx.de; trini at ti.com; albert.u.boot at aribaud.net
>> Subject: Re: MMC: proposal to support multiple physical partitions
>>
>> Hi Steve,
>>
>> On Apr 24, 2014, at 8:50 PM, Steve Rae wrote:
>>
> [... snip ...]
>
>>> I have two different proposals:
>>> 1) overload the "int dev_num" argument with encoded  "dev_num" and
>> "part_num" fields
>>> -          the dev_num  in the [15:0] bits,
>>> -          the part_num in the [30:16] bits,
>>> -          a flag to indicate this encoding in [31] bit.
>>> -          and modify mmc_bread() to handle this encoded argument, and
>> implement the above code...
>>> 2) create a wrapper function to perform the above code, with an added
>> argument "int part_num", possibly named:
>>> -          mmc_block_dev_block_read() -- so that it is similar to the original
>> calling convention [mmc->block_dev.block_read], or
>>> -          mmc_pbread() [PartitionBlockRead] -- so that it is similar to the
>> mmc_bread() [which is the implementation of the callback function]
>>
>> I'd rather go with the wrapper function. Perhaps it's not even needed. The
>> function called is part of the block_dev (block_read/write etc).
>>
>> Overwrite those with functions that implemented the switching first, and
>> then call the original block* function.
>>
> The callback function is:
>      mmc->block_dev.block_read = mmc_bread
> and it accepts four arguments:
>      include/part.h: unsigned long   (*block_read)(int dev,
>      include/part.h-                               lbaint_t start,
>      include/part.h-                               lbaint_t blkcnt,
>      include/part.h-                               void *buffer);
> Are you suggesting that I should add another argument to this callback?
>
> [...snip...]
>
>> Regards
>>
>> -- Pantelis
> _______________________________________________
> U-Boot mailing list
> U-Boot at lists.denx.de
> http://lists.denx.de/mailman/listinfo/u-boot
>