OSDev.org

Hello,

I have one problem with my Kernel. It seems that sometimes a Memory write isn't performed.

I have programmed a class that handles text output to 0xB8000. The class stores the
cursor position and updates the cursor. But: Sometimes If I write a text, only one char
either is missing and there is the old char(a space) or the cursor isn't updated, wich results
in that the char is overwritten, if it is not the last. But they occur only in Bochs and about
2 to 4 chars each screen (80x25).
I think it is a not performed write to memory. Either the char isn't written to 0xB8000, then it
left blank. Or if the cursor isn't updating the new value of the cursor position stored by the
class is not written.

I get no exception and no NMI. And the programm runs always fine. On real hardware I never
saw this phenomen.

Does anyone have an idea what that can be?

cheers
sebihepp

Hi,

Do you have SMP? I wonder if Bochs is showing up a bug that only appears if a lock isn't occurring in a timely manner.

Cheers,
Adam

Well, very good idea.
But no, I have bochs compiled without specifying SMP.

I thought about Interrupts, but I searched many times through the code
and found nothing. Every used register is pushed and popped, ...
I think it can only be an interrupt. But on real Hardware this doesn't appear.

I will post the interrupt code later.

cheers
sebihepp

What AJ was talking about is not the potential bugs in your interrupt handling code, but rather the presence of a race condition, which in software requires a form of multitasking to show up.

From experience, bochs is more likely to show race conditions, since on average there are less instructions between external interventions