[U-Boot] [RFC PATCH] dfu: allow get_medium_size() to return bigger medium sizes than 2GiB

Lukasz Majewski l.majewski at samsung.com
Fri Jan 29 15:13:12 CET 2016


Hi Heiko,

> Hello Lukasz,
> 
> Am 28.01.2016 um 15:56 schrieb Lukasz Majewski:
> > Hi Heiko,
> >
> >> change the get_medium_size() function from
> >> -       long (*get_medium_size)(struct dfu_entity *dfu);
> >> +       int (*get_medium_size)(struct dfu_entity *dfu, u64 *size);
> >>
> >> so it can return bigger medium sizes than 2GiB, and the return
> >> value is seperate from the size.
> >>
> >> Signed-off-by: Heiko Schocher <hs at denx.de>
> >>
> >> ---
> >> I just have a DDP nand with a size of 4GiB, and the
> >> mtd partition layout is:
> >> device nand2 <omap2-nand.0>, # parts = 9
> >>   #: name                size            offset          mask_flags
> >>   0: spl                 0x00080000      0x00000000      0
> >>   1: spl.backup1         0x00080000      0x00080000      0
> >>   2: spl.backup2         0x00080000      0x00100000      0
> >>   3: spl.backup3         0x00080000      0x00180000      0
> >>   4: u-boot              0x00780000      0x00200000      0
> >>   5: u-boot.env0         0x00200000      0x00980000      0
> >>   6: u-boot.env1         0x00200000      0x00b80000      0
> >>   7: mtdoops             0x00200000      0x00d80000      0
> >>   8: rootfs              0xff080000      0x00f80000      0
> >>
> >> so the last partition is to big for returning the size in a long.
> >>
> >>   drivers/dfu/dfu.c      | 8 ++++----
> >>   drivers/dfu/dfu_mmc.c  | 8 +++++---
> >>   drivers/dfu/dfu_nand.c | 5 +++--
> >>   drivers/dfu/dfu_ram.c  | 5 +++--
> >>   drivers/dfu/dfu_sf.c   | 5 +++--
> >>   include/dfu.h          | 4 ++--
> >>   6 files changed, 20 insertions(+), 15 deletions(-)
> >>
> >> diff --git a/drivers/dfu/dfu.c b/drivers/dfu/dfu.c
> >> index 8f5915e..daa2eb9 100644
> >> --- a/drivers/dfu/dfu.c
> >> +++ b/drivers/dfu/dfu.c
> >> @@ -335,11 +335,11 @@ int dfu_read(struct dfu_entity *dfu, void
> >> *buf, int size, int blk_seq_num) if (dfu->i_buf_start == NULL)
> >>   			return -ENOMEM;
> >>
> >> -		dfu->r_left = dfu->get_medium_size(dfu);
> >> -		if (dfu->r_left < 0)
> >> -			return dfu->r_left;
> >> +		ret = dfu->get_medium_size(dfu, &dfu->r_left);
> >> +		if (ret < 0)
> >> +			return ret;
> >>
> >> -		debug("%s: %s %ld [B]\n", __func__, dfu->name,
> >> dfu->r_left);
> >> +		debug("%s: %s %lld [B]\n", __func__, dfu->name,
> >> dfu->r_left);
> >>   		dfu->i_blk_seq_num = 0;
> >>   		dfu->crc = 0;
> >> diff --git a/drivers/dfu/dfu_mmc.c b/drivers/dfu/dfu_mmc.c
> >> index 395d472..5c1c1d1 100644
> >> --- a/drivers/dfu/dfu_mmc.c
> >> +++ b/drivers/dfu/dfu_mmc.c
> >> @@ -205,14 +205,15 @@ int dfu_flush_medium_mmc(struct dfu_entity
> >> *dfu) return ret;
> >>   }
> >>
> >> -long dfu_get_medium_size_mmc(struct dfu_entity *dfu)
> >> +int dfu_get_medium_size_mmc(struct dfu_entity *dfu, u64 *size)
> >
> > The idea to use the return value to get error code and separate
> > pointer for passing the size is the way to go in my opinion.
> 
> Thats, why I posted it as an RFC, I was unsure too...
> 
> > The problem is in details. Please find my comments below.
> >
> >>   {
> >>   	int ret;
> >>   	long len;
> >>
> >>   	switch (dfu->layout) {
> >>   	case DFU_RAW_ADDR:
> >> -		return dfu->data.mmc.lba_size *
> >> dfu->data.mmc.lba_blk_size;
> >> +		*size = dfu->data.mmc.lba_size *
> >> dfu->data.mmc.lba_blk_size;
> >> +		return 0;
> >>   	case DFU_FS_FAT:
> >>   	case DFU_FS_EXT4:
> >>   		dfu_file_buf_filled = -1;
> >> @@ -221,7 +222,8 @@ long dfu_get_medium_size_mmc(struct dfu_entity
> >> *dfu) return ret;
> >>   		if (len > CONFIG_SYS_DFU_MAX_FILE_SIZE)
> >>   			return -1;
> >> -		return len;
> >> +		*size = len;
> >> +		return 0;
> >>   	default:
> >>   		printf("%s: Layout (%s) not (yet) supported!\n",
> >> __func__, dfu_get_layout(dfu->layout));
> >> diff --git a/drivers/dfu/dfu_nand.c b/drivers/dfu/dfu_nand.c
> >> index a975492..4612e09 100644
> >> --- a/drivers/dfu/dfu_nand.c
> >> +++ b/drivers/dfu/dfu_nand.c
> >> @@ -114,9 +114,10 @@ static int dfu_write_medium_nand(struct
> >> dfu_entity *dfu, return ret;
> >>   }
> >>
> >> -long dfu_get_medium_size_nand(struct dfu_entity *dfu)
> >> +int dfu_get_medium_size_nand(struct dfu_entity *dfu, u64 *size)
> >>   {
> >> -	return dfu->data.nand.size;
> >> +	*size = dfu->data.nand.size;
> >> +	return 0;
> >>   }
> >>
> >>   static int dfu_read_medium_nand(struct dfu_entity *dfu, u64
> >> offset, void *buf, diff --git a/drivers/dfu/dfu_ram.c
> >> b/drivers/dfu/dfu_ram.c index e094a94..466759d 100644
> >> --- a/drivers/dfu/dfu_ram.c
> >> +++ b/drivers/dfu/dfu_ram.c
> >> @@ -41,9 +41,10 @@ static int dfu_write_medium_ram(struct
> >> dfu_entity *dfu, u64 offset, return
> >> dfu_transfer_medium_ram(DFU_OP_WRITE, dfu, offset, buf, len); }
> >>
> >> -long dfu_get_medium_size_ram(struct dfu_entity *dfu)
> >> +int dfu_get_medium_size_ram(struct dfu_entity *dfu, u64 *size)
> >>   {
> >> -	return dfu->data.ram.size;
> >> +	*size = dfu->data.ram.size;
> >> +	return 0;
> >>   }
> >>
> >>   static int dfu_read_medium_ram(struct dfu_entity *dfu, u64
> >> offset, diff --git a/drivers/dfu/dfu_sf.c b/drivers/dfu/dfu_sf.c
> >> index 9702eee..35c5fa1 100644
> >> --- a/drivers/dfu/dfu_sf.c
> >> +++ b/drivers/dfu/dfu_sf.c
> >> @@ -12,9 +12,10 @@
> >>   #include <spi.h>
> >>   #include <spi_flash.h>
> >>
> >> -static long dfu_get_medium_size_sf(struct dfu_entity *dfu)
> >> +int dfu_get_medium_size_sf(struct dfu_entity *dfu, u64 *size)
> >
> > Originally the dfu_get_medium_size returns int. It is not suitable
> > for sizes > 2GiB.
> >
> > I don't like the u64 for *size, since we use that size to perform
> > some arithmetic operation and comparison (>) in the dfu.c file. I
> > would prefer to have long long here.
> 
> Ok.
> 
> >>   {
> >> -	return dfu->data.sf.size;
> >> +	*size = dfu->data.sf.size;
> >> +	return 0;
> >>   }
> >>
> >>   static int dfu_read_medium_sf(struct dfu_entity *dfu, u64 offset,
> >> void *buf, diff --git a/include/dfu.h b/include/dfu.h
> >> index 6118dc2..323b032 100644
> >> --- a/include/dfu.h
> >> +++ b/include/dfu.h
> >> @@ -110,7 +110,7 @@ struct dfu_entity {
> >>   		struct sf_internal_data sf;
> >>   	} data;
> >>
> >> -	long (*get_medium_size)(struct dfu_entity *dfu);
> >> +	int (*get_medium_size)(struct dfu_entity *dfu, u64 *size);
> >>
> >>   	int (*read_medium)(struct dfu_entity *dfu,
> >>   			u64 offset, void *buf, long *len);
> >> @@ -132,7 +132,7 @@ struct dfu_entity {
> >>   	u8 *i_buf;
> >>   	u8 *i_buf_start;
> >>   	u8 *i_buf_end;
> >> -	long r_left;
> >> +	u64 r_left;
> >
> > This patch changes r_left to be unsigned, but leaves b_left as
> > signed (long).
> >
> > I think that both b_left and r_left should be promoted to long long.
> 
> Ok, I change it to long long ... but, I have detected another problem:
> 
> drivers/dfu/dfu_nand.c:
> 
> static int nand_block_op(enum dfu_op op, struct dfu_entity *dfu,
>                          u64 offset, void *buf, long *len)
> {
>          loff_t start, lim;
>          size_t count, actual;
>          int ret;
>          nand_info_t *nand;
> 
>          /* if buf == NULL return total size of the area */
>          if (buf == NULL) {
>                  *len = dfu->data.nand.size;
>                  return 0;
>          }
> 
> returns also a long only. nand_block_op gets called in the end from
> 
> static int dfu_write_medium_nand(struct dfu_entity *dfu,
>                  u64 offset, void *buf, long *len)
> 
> and
> 
> static int dfu_read_medium_nand(struct dfu_entity *dfu, u64 offset,
> void *buf, long *len)
> {
> 
> So I have to change also this functions ... Why returns the read/write
> operation  the length of the complete area, if buf is NULL ? Is this
> a DFU feature?

Git blame hints that this code is a reminiscent of first code, which
added NAND support to dfu (SHA1: c6631764, year 2013).

It seems like in the past the nand_block_op() has been called with buf
= NULL parameter to read the NAND area size.

However, Stephen Warren has added the get_medium_size_nand() function a
year later (in 2014) to read the medium size.

Hence it seems to me, that the buf == NULL code in nand_block_op could
be safely removed. 

As I've check - in contemporary u-boot - there is no call to this
function with buf NULL parameter.


> 
> bye,
> Heiko
> >
> >>   	long b_left;
> >>
> >>   	u32 bad_skip;	/* for nand use */
> >
> >
> >
> 



-- 
Best regards,

Lukasz Majewski

Samsung R&D Institute Poland (SRPOL) | Linux Platform Group


More information about the U-Boot mailing list