[U-Boot] [RFC PATCH 2/2] mmc: add support for block device cache

[Eric Nelson]

Hi Stephen,

On 03/17/2016 02:23 PM, Stephen Warren wrote:
> On 03/16/2016 03:40 PM, Eric Nelson wrote:
>> Signed-off-by: Eric Nelson <eric at nelint.com>
> 
> Patch description.
> 
>> ---
>>   drivers/mmc/mmc.c       | 10 +++++++++-
>>   drivers/mmc/mmc_write.c |  7 +++++++
> 
> Presumably it makes sense for the cache to work for IDE, SATA, USB,
> SCSI, ... too. I wonder if it's possible to put this code somewhere more
> central than mmc*.c so it automatically applies to
> dev_desc->block_read() (see include/part.h). Perhaps not since each
> implementation supplies its own block_read function directly, so the
> cache calls do need to be duplicated everywhere.
> 

Yeah. I haven't found a spot that would allow interception of
the various block_read/write functions.

The get_dev_hwpart() routine in disk/part.c is close, but it seems
to be bypassed by cmd_mmc, cmd_sata, et cetera.

I think the best that can be done is to provide a common shim that can
easily be inserted from within the various block driver code.

>> diff --git a/drivers/mmc/mmc.c b/drivers/mmc/mmc.c
>>    *
>>    * SPDX-License-Identifier:    GPL-2.0+
>>    */
>> -
>>   #include <config.h>
> 
> Nit: unrelated change.
> 
> I think there is a missing call to cache_block_invalidate() when the MMC
> device gets re-enumerated/re-initialized. The user would do something to
> trigger this (e.g. mmc rescan) when they'd swapped an SD card out for
> example.
> 

Good catch.

> Do you have any stats on how many operations this saves for typical FS
> operations such as:
> 
> - Partition table type identification (with various types such as
> MBR/DOS, GPT, ...)
> - Partition enumeration
> - Filesystem identification (with various filesystems such as FAT, ext,
> ...)
> - File reads
> 

Not yet, but I'm working something up that will allow this to be
gathered easily. As soon as we implement a cache, it provides a nice
spot for tracing operations.

Regards,


Eric