Re: ISDN massive packet drops while DVD burn/verify

[Stephan von Krawczynski]

On Sat, 19 Apr 2003 20:50:00 +0200
Christian Vogel <[e-mail]> wrote:

> Hi Stephan,
> 
> On Sat, Apr 19, 2003 at 07:38:48PM +0200, Stephan von Krawczynski wrote:
> > > My best guess would be that IDE blocks IRQs for too long and hisax 
> > > interrupts get lost. You could try whether hdparm -u1 helps, and a 
> > > debugging log from the hisax driver may confirm over/underruns.
> 
> > I don't buy that explanation. Reason is simple: during this all network
> > connections work flawlessly, and they do have quite a lot of interrupts
> > compared to ISDN. ISDN is so slow and has so few interrupts that it is
> > quite unlikely in a SMP-beyond-GHz-limit box that you loose some. The
> > ancient hardware days are long gone ...
> 
> I would. At least try it. Network connections are not subject to smaller
> delays, ISDN on the other hand depends on a synchronous processing of the
> data (it's like a sync serial port after all...). It's not important how
> fast your CPU is if it's just doing nothing while waiting for your
> harddisk...

First: we are not talking about harddisks by any means.

Second: if it is doing _nothing_ while waiting for ide-whatever, then it is
presumably also not updating my mouse position or performing anything else
useful. This is _not_ the case here.

Third: a full-blown (1 channel) ISDN-download has around 8kBytes/sec of data.
The worst chipsets ever will create around 250 interrupts/sec for this (32 byte
fifo per hdlc-chunk). Don't count on the exact number, it may be some few more
than 250, but not a lot. This may have been a problem on very old boxes from
some years back, but for sure not on todays GHz monsters. You can only kill it
by completely braindead interrupt programming, which - I honor Alans' IDE stuff
quite a lot - I do not presume existing somewhere inside the current kernel.

Forth: There are parts in the ISDN code like:

static int
isdn_writebuf_stub(int drvidx, int chan, const u_char * buf, int len,
                   int user)
{
        int ret;
        int hl = dev->drv[drvidx]->interface->hl_hdrlen;
        struct sk_buff *skb = alloc_skb(hl + len, GFP_ATOMIC);

        if (!skb)
                return 0;

Used here:
		lock_kernel();
		...

                while (isdn_writebuf_stub(drvidx, chidx, buf, count, 1) != count)
                        interruptible_sleep_on(&dev->drv[drvidx]->snd_waitq[chidx]);

Which basically means we are looping around for some mem without telling anybody... or not?
This is probably not a very good example of what I mean, but nevertheless ...

> I would recommend doing /sbin/hdparm -u1 -c1 -d1 /dev/hda which makes
> all systems I know more performant and better responding.

You cannot issue that command to a DVD-writer used with ide-scsi.

> 
> 	Chris
> 
> -- 
> The code was willing,
> It considered your request,
> But the chips were weak.
> -- Barry L. Brumitt
> 

Regards,
Stephan