Bug in console driver?

Bug in console driver?

[Michael Schroeder]

Hi!

I suppose there is a bug in the 2.4-kernel console driver 
(../drivers/char/console.c).

It seems that (if there is a slow video device) this driver disables the 
handling of interrupts while writing or switching the consoles. 

Symptoms:

If you use the VESA framebuffer, switching to another console prevents 
updating the internal kernel clock (jiffies, xtime) for a short time. 
This effect can grow up to some seconds:

Load the GNU Midnight Commander (mc) and synchronize kernel clock and 
hardware clock with the command 'hwclock --hctosys'.

Then you can see the the difference with 'hwclock && date'.

  # hwclock && date
  Sun Sep 16 10:31:33 2001  -0,154753 ...
  Sun Sep 10 10:31:32 CEST 2001

Now hold down the keys <Ctrl+o> for a some seconds. (This switches the 
window of the Midnight Commander on/off many times)

If you now compare the clocks you yield a difference of some seconds:

  # hwclock && date
  Sun Sep 16 10:33:17 2001  -0,456384 ...
  Sun Sep 16 10:33:12 CEST 2001

5 seconds - I think this is a bug.


My system:

Intel Pentium II, 512 MB RAM
NVIdia Riva TNT2 M64 with 32 MB
Kernel 2.4.4 and 2.4.7
SuSE Linux 7.2
VESA Framebuffer 


Greetings

Michael

Re: Bug in console driver?

[Andrew Morton]

Michael Schroeder wrote:
> 
> Hi!
> 
> I suppose there is a bug in the 2.4-kernel console driver
> (../drivers/char/console.c).
> 
> It seems that (if there is a slow video device) this driver disables the
> handling of interrupts while writing or switching the consoles.
> 

Yup.  This was fixed in -ac kernels a few months ago.  We're working
on a merge into Linus' kernels at present - it's still a fairly large
patch though.

The latest diff (against 2.4.10-pre10) is at
http://www.uow.edu.au/~andrewm/linux/2.4.10-pre10-no-console-lock-2.patch

Could you please test it?

-