On Tue, 25 Jan 2000, Tomasz Motylewski wrote:
> On Tue, 25 Jan 2000, Fabrice Bellard wrote:
>
> > I am currently working with a Millenium disk-on-chip under Linux with the
> > driver provided by M-systems & IGEL, and I noticed a very bad thing in the
> > open source wrapper (fldrvlnx.c): in the block request function
> > (fl_request), the irq are never activated. They are off when we enter the
> > function, and just in case they would miraclously get enabled (!!), a
> > cli() is put at the beginning of the function.
>
> Is there a busy-wait loop there, or may be schedule() is called?
> schedule() contains sti().
> Tomek
>
According to what I saw, schedule() is not called, and when I make big
disk accesses, a few ticks are lost. I discovered this problem when I made
some read / write benchmarks (with the time unix command) which were
surprisingly too good !
It turns out that on my hardware, the time to read one page (4K) from the
disk on chip is just longer than one tick (10 ms). This is why the ticks
are lost !
Fabrice.
--
To unsubscribe from this list, send a message to [EMAIL PROTECTED]
with the command "unsubscribe linux-embedded" in the message body.
For more information, see <http://waste.org/mail/linux-embedded>.
