[U-Boot] [PATCH] blk: Invalidate block cache when switching hwpart

Thu Aug 22 13:58:11 UTC 2019

On 11.07.19 09:10, Weijie Gao wrote:

> Some storage devices have multiple hw partitions and both address from
> zero, for example eMMC.
> However currently block cache invalidation only applies to block
> write/erase.
> This can cause a problem that data of current hw partition is cached
> before switching to another hw partition. And the following read
> operation of the latter hw partition will get wrong data when reading
> from the addresses that have been cached previously.
> 
> To solve this problem, invalidate block cache after a successful
> select_hwpart operation.
> 
> Signed-off-by: Weijie Gao <weijie.gao at mediatek.com>
> ---
This patch breaks correct operation of PDU001 board.

>  drivers/block/blk-uclass.c | 14 ++++++++++++--
>  1 file changed, 12 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/block/blk-uclass.c b/drivers/block/blk-uclass.c
> index baaf431e5e0..c23b6682a6c 100644
> --- a/drivers/block/blk-uclass.c
> +++ b/drivers/block/blk-uclass.c
> @@ -208,7 +208,11 @@ int blk_select_hwpart_devnum(enum if_type if_type, int devnum, int hwpart)
>  	if (ret)
>  		return ret;
>  
> -	return blk_select_hwpart(dev, hwpart);
> +	ret = blk_select_hwpart(dev, hwpart);
> +	if (!ret)
> +		blkcache_invalidate(if_type, devnum);
> +
> +	return ret;
>  }
>  
>  int blk_list_part(enum if_type if_type)
> @@ -348,7 +352,13 @@ int blk_select_hwpart(struct udevice *dev, int hwpart)
>  
>  int blk_dselect_hwpart(struct blk_desc *desc, int hwpart)
>  {
> -	return blk_select_hwpart(desc->bdev, hwpart);
> +	int ret;
> +
> +	ret = blk_select_hwpart(desc->bdev, hwpart);
> +	if (!ret)
> +		blkcache_invalidate(desc->if_type, desc->devnum);
> +
Commenting the invalidation of the block cache on this line results in
proper working of the board.

> +	return ret;
>  }
>  
>  int blk_first_device(int if_type, struct udevice **devp)
> 
With the patch active, files from the boot FAT partition of the SD-card
(mmc device 1) do not load anymore, hence booting fails.
Using the eMMC (mmc device 0) instead works fine, files can bee loaded
and the board boots.

To isolate the problem I have modified the configuration to only load a
10k test file (test.bin) instead of loading the DTB and zImage files
followed by booting LINUX. Furthermore I have enabled debugging for some
parts of the code.

When I boot from the SD-card (which fails) I get the following logged:
===
CPU  : AM335X-GP rev 1.0
Model: EETS,PDU001
DRAM:  256 MiB
mmc_bind: alias ret=-2, devnum=-1
mmc_bind: alias ret=-2, devnum=-1
MMC:   OMAP SD/MMC: 0, OMAP SD/MMC: 1
In:    serial at 481a6000
Out:   serial at 481a6000
Err:   serial at 481a6000
Press SPACE to abort autoboot in 1 seconds
blk_get_devnum_by_typename: if_type=6, devnum=1: mmc at 48060000.blk, 6, 0
blk_get_devnum_by_typename: if_type=6, devnum=1: mmc at 481d8000.blk, 6, 1
clock is disabled (0Hz)
clock is enabled (400000Hz)
clock is enabled (50000000Hz)
miss: start 0, count 1
blk_find_device: if_type=6, devnum=1: mmc at 48060000.blk, 6, 0
blk_find_device: if_type=6, devnum=1: mmc at 481d8000.blk, 6, 1
fill: start 0, count 1
part_init: try 'EFI': ret=-1
hit: start 0, count 1
part_init: try 'DOS': ret=0
blk_get_devnum_by_typename: Device desc 8dfb3a98
miss: start 0, count 1
blk_find_device: if_type=6, devnum=1: mmc at 48060000.blk, 6, 0
blk_find_device: if_type=6, devnum=1: mmc at 481d8000.blk, 6, 1
fill: start 0, count 1
part_init: try 'EFI': ret=-1
hit: start 0, count 1
part_init: try 'DOS': ret=0
hit: start 0, count 1
miss: start 800, count 1
blk_find_device: if_type=6, devnum=1: mmc at 48060000.blk, 6, 0
blk_find_device: if_type=6, devnum=1: mmc at 481d8000.blk, 6, 1
fill: start 800, count 1
hit: start 800, count 1
FAT16, fat_sect: 1, fatlength: 64
Rootdir begins at cluster: 0, sector: 129, offset: 10200
Data begins at: 153
Sector size: 512, cluster size: 4
FAT read(sect=129), clust_size=4, read_size=4, DIRENTSPERBLOCK=16
miss: start 881, count 4
blk_find_device: if_type=6, devnum=1: mmc at 48060000.blk, 6, 0
blk_find_device: if_type=6, devnum=1: mmc at 481d8000.blk, 6, 1
fill: start 881, count 4
miss: start 800, count 1
blk_find_device: if_type=6, devnum=1: mmc at 48060000.blk, 6, 0
blk_find_device: if_type=6, devnum=1: mmc at 481d8000.blk, 6, 1
fill: start 800, count 1
FAT16, fat_sect: 1, fatlength: 64
Rootdir begins at cluster: 0, sector: 129, offset: 10200
Data begins at: 153
Sector size: 512, cluster size: 4
FAT read(sect=129), clust_size=4, read_size=4, DIRENTSPERBLOCK=16
miss: start 881, count 4
blk_find_device: if_type=6, devnum=1: mmc at 48060000.blk, 6, 0
blk_find_device: if_type=6, devnum=1: mmc at 481d8000.blk, 6, 1
fill: start 881, count 4
reading test.bin at pos 0
Filesize: 0 bytes
Read position past EOF: 0
0 bytes read in 94 ms (0 Bytes/s)
===

Obviously the the file test.bin does not get loaded.

Booting from the eMMC (which works) logs the following:
===
CPU  : AM335X-GP rev 1.0
Model: EETS,PDU001
DRAM:  256 MiB
mmc_bind: alias ret=-2, devnum=-1
mmc_bind: alias ret=-2, devnum=-1
MMC:   OMAP SD/MMC: 0, OMAP SD/MMC: 1
In:    serial at 481a6000
Out:   serial at 481a6000
Err:   serial at 481a6000
Press SPACE to abort autoboot in 1 seconds
blk_get_devnum_by_typename: if_type=6, devnum=0: mmc at 48060000.blk, 6, 0
clock is disabled (0Hz)
clock is enabled (400000Hz)
clock is enabled (25000000Hz)
clock is enabled (52000000Hz)
miss: start 0, count 1
blk_find_device: if_type=6, devnum=0: mmc at 48060000.blk, 6, 0
fill: start 0, count 1
part_init: try 'EFI': ret=-1
hit: start 0, count 1
part_init: try 'DOS': ret=0
blk_get_devnum_by_typename: Device desc 8dfb3800
miss: start 0, count 1
blk_find_device: if_type=6, devnum=0: mmc at 48060000.blk, 6, 0
fill: start 0, count 1
part_init: try 'EFI': ret=-1
hit: start 0, count 1
part_init: try 'DOS': ret=0
hit: start 0, count 1
miss: start 1, count 1
blk_find_device: if_type=6, devnum=0: mmc at 48060000.blk, 6, 0
fill: start 1, count 1
hit: start 1, count 1
FAT16, fat_sect: 1, fatlength: 20
Rootdir begins at cluster: 0, sector: 41, offset: 5200
Data begins at: 65
Sector size: 512, cluster size: 4
FAT read(sect=41), clust_size=4, read_size=4, DIRENTSPERBLOCK=16
miss: start 2a, count 4
blk_find_device: if_type=6, devnum=0: mmc at 48060000.blk, 6, 0
fill: start 2a, count 4
miss: start 1, count 1
blk_find_device: if_type=6, devnum=0: mmc at 48060000.blk, 6, 0
fill: start 1, count 1
FAT16, fat_sect: 1, fatlength: 20
Rootdir begins at cluster: 0, sector: 41, offset: 5200
Data begins at: 65
Sector size: 512, cluster size: 4
FAT read(sect=41), clust_size=4, read_size=4, DIRENTSPERBLOCK=16
miss: start 2a, count 4
blk_find_device: if_type=6, devnum=0: mmc at 48060000.blk, 6, 0
fill: start 2a, count 4
reading test.bin at pos 0
Filesize: 10240 bytes
10240 bytes
FAT16: entry: 0x000000a3 = 163, offset: 0x00a3 = 163
miss: start 2, count 6
blk_find_device: if_type=6, devnum=0: mmc at 48060000.blk, 6, 0
fill: start 2, count 6
FAT16: ret: 0x000000a4, entry: 0x000000a3, offset: 0x00a3
FAT16: entry: 0x000000a4 = 164, offset: 0x00a4 = 164
FAT16: ret: 0x000000a5, entry: 0x000000a4, offset: 0x00a4
FAT16: entry: 0x000000a5 = 165, offset: 0x00a5 = 165
FAT16: ret: 0x000000a6, entry: 0x000000a5, offset: 0x00a5
FAT16: entry: 0x000000a6 = 166, offset: 0x00a6 = 166
FAT16: ret: 0x000000a7, entry: 0x000000a6, offset: 0x00a6
gc - clustnum: 163, startsect: 717
miss: start 2ce, count 20
blk_find_device: if_type=6, devnum=0: mmc at 48060000.blk, 6, 0
10240 bytes read in 135 ms (73.2 KiB/s)
===

This time the file test.bin gets loaded correctly.

To be honest I don't really see why things are going wrong. It looks
like the block cache gets filled but then again invalidated before it's
data is used. I also feel that this problem is related to the SD-card
not being the first (number 0) but the second device (number 1).
Can anybody shed some light on this and give me a hint?

many thanks, Felix