flushing buffer cache on mtd device: is the verify in flashcp actually reading back from the flash chip or the cache?

[David Woodhouse]

On Mon, 2006-10-16 at 16:02 -0400, Clem Taylor wrote:
> 
> Recently a question came up as to where we were really verifying that
> the data was correctly being written to the flash. We are using
> flashcp from a 2006.05.04 mtd snapshot to program the flash. Looking
> at the code, after the write, it just does a lseek() back to the
> beginning of the device and then starts reading. It is opening the
> device with O_SYNC, but I thought that only effects writes, not reads.

s/effect/affect/

> Without doing something special, wouldn't this data just come from the
> buffer cache instead of the physical device?
> 
> Is there something else doing on under the hood that would force a
> read from physical device after a flash write? 

Access to character devices doesn't go through the buffer cache.

-- 
dwmw2