Bartlomiej Zolnierkiewicz wrote:

>>Vanilla 2.6.6:
>>
>>rene@7ixe4:~/cache$ grep wcache settings-2.6.6
>>wcache                  1               0               1               rw
>>
>>2.6.6 with the de{flush,spin}ification patches:
>>
>>rene@7ixe4:~/cache$ grep wcache settings-2.6.6-hackedup
>>wcache                  0               0               1               rw
>>
>>Hrmpf. 0?
> 
> Very interesting, can you try vanilla 2.6.5 (or some other 2.6.x)?

2.4.26		: wcache = 0, 0, 1
2.6.6-rc3-mm2	: wcache = 0, 0, 1
2.6.6		: wcache = 1, 0, 1
2.6.6-hackedup	: wcache = 0, 0, 1

(on none of these hdparm -W0/1 has any effect, by the way)

> We are getting close to the real problem.

Indeed. I have attached two "tiobench" seq. write results, one for 
vanilla 2.6.6, and one for 2.6.6-hackedup (your changes). I remember the 
result for the latter are the normal ones I got with previous kernels as 
well. The differences are very interesting.

2.6.6-vanilla (ie, the one showing wcache=1): +/- 10MB/s
2.6.6-hackedup (ie, shows wcache=0)	    : +/- 46MB/s

As said, that 46 is the expected value. Is this a logic inversion bug in 
either driver (but then both 2.4.26 and 2.6) or drive firmware?

Rene.