mtd filesystem across different sector sizes in the same part

mtd filesystem across different sector sizes in the same part

[Cam Mayor]

After doing a flash storage map for one of our devices and based on reading a 
recent question to this list, i thought of another question.

In my flash part (intel 28f320c3) there are 8 x 8kB blocks and 63 x 64 kB 
blocks.  I believe that [1] MTD can span the two block sizes effectively for 
it to be recognized as a single MTD device.  JFFS2, IIRC, uses 5 flash blocks 
for flash filesystem overhead.  Do those blocks have to be any specific size 
relative to the rest of the filesystem blocks?  Will it use the first five 
blocks, or does it move around for wear levelling?  Will JFFS2 get confused 
by different block sizes, or is that already handled in the MTD layer?  

If the 5 blocks move around, is there any way to predict how much space will 
be available in the filesystem?  (eg. one moment you have 5 overhead blocks 
in the 8kB area and have 70kB free.  the next moment you have 4 overhead 
blocks in the 8kB area, and one in the 64kB area, leaving you with 14kB free 
space)

thanks,
cam

[1] please correct me if i'm wrong


-- 
Cameron Mayor
Iders Incorporated

Re: mtd filesystem across different sector sizes in the same part

[David Woodhouse]

On Mon, 2003-10-06 at 14:33 -0500, Cam Mayor wrote:
> After doing a flash storage map for one of our devices and based on reading a 
> recent question to this list, i thought of another question.
> 
> In my flash part (intel 28f320c3) there are 8 x 8kB blocks and 63 x 64 kB 
> blocks.  I believe that [1] MTD can span the two block sizes effectively for 
> it to be recognized as a single MTD device.  JFFS2, IIRC, uses 5 flash blocks 
> for flash filesystem overhead.  Do those blocks have to be any specific size 
> relative to the rest of the filesystem blocks?  Will it use the first five 
> blocks, or does it move around for wear levelling?  Will JFFS2 get confused 
> by different block sizes, or is that already handled in the MTD layer?  

JFFS2 will use the 'major' erase size, and treat your 8 * 8KiB blocks as
a single 64KiB block.

-- 
dwmw2

Re: mtd filesystem across different sector sizes in the same part

[Cam Mayor]

On Monday 06 October 2003 14:36, David Woodhouse wrote:
> On Mon, 2003-10-06 at 14:33 -0500, Cam Mayor wrote:
> > In my flash part (intel 28f320c3) there are 8 x 8kB blocks and 63 x 64 kB
> > blocks.  I believe that [1] MTD can span the two block sizes effectively
> > for it to be recognized as a single MTD device.  JFFS2, IIRC, uses 5
> > flash blocks for flash filesystem overhead.  Do those blocks have to be
> > any specific size relative to the rest of the filesystem blocks?  Will it
> > use the first five blocks, or does it move around for wear levelling? 
> > Will JFFS2 get confused by different block sizes, or is that already
> > handled in the MTD layer?
>
> JFFS2 will use the 'major' erase size, and treat your 8 * 8KiB blocks as
> a single 64KiB block.

In my particular usage, the first two (of eight) "small" blocks are used by 
the bootloader, leaving a free area of 6 x 8kiB blocks + 6 x 64 kiB blocks.   
Just now i tried specifying this oddly shaped area on the kernel command 
line, with the result:
	mtd: partition "free" doesn't start on an erase block boundary -- force 
read-only

So, it sounds like JFFS2 won't handle this specific situation, as mtd itself 
is unwilling to deal with it.  (the partition "free" does start on an erase 
block boundary - but it is one of the smaller block boundaries, not the major 
erase size)

I don't have a multi-flash system, so i can't test this, but i wonder how 
this would work for a system concatenated across two flash chips with 
different erase block sizes where the numbers don't allow them to align 
nicely.

Here's another thing i just thought of but haven't noticed in my mtd research 
in the past (and couldn't see anything relevant in Configure.help): is there 
a way to get two non-continuous mtd partitions to host a jffs2 filesystem?  
eg. 
mtd5: 00400000 00020000 "userarea1"
mtd6: 00400000 00020000 "something_else"
mtd7: 00400000 00020000 "userarea2"
is there a way to use the disjoint mtd5+mtd7 together to host a continuous 
filesystem?

thanks for the info.

-- 
Cameron Mayor
Iders Incorporated

Re: mtd filesystem across different sector sizes in the same part

[David Woodhouse]

On Mon, 2003-10-06 at 15:45 -0500, Cam Mayor wrote:
> In my particular usage, the first two (of eight) "small" blocks are used by 
> the bootloader, leaving a free area of 6 x 8kiB blocks + 6 x 64 kiB blocks.   
> Just now i tried specifying this oddly shaped area on the kernel command 
> line, with the result:
> 	mtd: partition "free" doesn't start on an erase block boundary -- force 
> read-only

Yep. There exists no 'major' erasesize. All real hardware devices have
one, we don't try to work around it.

> is there a way to use the disjoint mtd5+mtd7 together to host a continuous 
> filesystem?

See mtdconcat.

-- 
dwmw2