[prev in list] [next in list] [prev in thread] [next in thread] 

List:       linux-msdos
Subject:    Re: stability patches for dosemu
From:       Hans Lermen <lermen () fgan ! de>
Date:       2001-03-18 20:37:52
[Download RAW message or body]

On Sun, 18 Mar 2001, Stas Sergeev wrote:

> The reason is, IMHO, here:
>  t_time = nmUStoTICK(s_time->td % 900000000);
>                              ^^^^^^^^^^^^^^^^^^^^^^^^^^

Ok, but this is one of the odds in the old timer code, see below.

> But anyway, non-monoton timing is broken. 

It always has been so 'by principal', that's why we have the 'new'
tmonoton code.

> Don't you think that it should be repaired?

No, we have monoton timing since 96/12/30 (dosemu-0.64.2.1) and it is
the default since 97/05/24 (dosemu-0.67.3.3) and since then all
distributions are using it.

The old timer code has some realy nasty odds (intermediate counter
freezes, jumping back and forth in time at will), which in fact are
fixed by the 'new' (monoton) one. The reasons why we left in the old code
as an option are no longer valid and we will erase it in the near future.

BTW: if your testing of applications is based on the old timer code,
you should redo it with tmonoton on.

>   >> dpmi_tim.diff
>   >> 4. run_dpmi() must not be called from timer_int_engine(), because
>   >This one breaks my 'doom testsuite' (Locks at the wellcome screen)
> Yes, maybe, because this was a workaround and it still stays a workaround
> still unknown bug.

Your (much smaller) new patch looks a lot better and doesn't lock any
more and doesn't seem to do harm ... went into 1.0.2.5

> Strange... pic_sched() is called *only* from timer_int_engine(), so
> it can lose only timer ints. And how can you pass timer int via
> $_irqpassing?

I did not say that;-) The driver's IRQ handler passes time consuming
stuff to an 'offline' handler triggered by the timer. This one locks
waiting for events which normally would have been scheduled.
Something in your patch changed the dynamic behave.

> C version works for me some years already. But OK, I didn't test it on an
> SMP. But this patch also fixes 2 bugs in the asm version.
> What do you think about that fixes?

I'm not sure yet, wether they realy are only fixes, have to digg deeper.
As your patch will do essentiel change to the algorithme, this for
sure is no 1.0 material and (if we consider) it has to go into 1.1

Hans
<lermen@fgan.de>


-
To unsubscribe from this list: send the line "unsubscribe linux-msdos" in
the body of a message to majordomo@vger.kernel.org

[prev in list] [next in list] [prev in thread] [next in thread]