Zhao Yakui wrote:
Hi, Alexey
Maybe you misunderstand what I have done in the patch of bug 10724.
Ok, the patch is worthless from both debugging point (you _will_ get less interrupts if you service them longer), so you will
not find anything new with this patch, and it is stupid as a patch intended at production just because you'll spend same time
in interrupt context as before, but now you flush system responsiveness down the tube too.
We should firstly make it clear why there exists EC GPE interrupt storm on some laptops. EC controller
uses the pulse interrupt model to trigger GPE interrupt and EC GPE is registered as Edge mode.
If the EC interrupt pulse is too wide on some broken BIOS, several ACPI interrupts are triggered
although only one EC interrupt pulse is generated. Maybe on some broken BIOS the situation will be more serious.And
when EC GPE handler is served, OS will check the EC status by reading the EC status I/O port. On most laptops the
EC and other controller(Keyboard, UART) are accessed through LPC bus. If the EC GPE interrupt is triggered too many times
( For example: 50000 per second), too much bandwith of LPC bus will be wasted on reading the EC status register.
On the bug 9998 it causes that sometimes the keystrobes are lost. Sometimes it will cause that the info of the
battery/SBS battery related with EC can't be obtained.
If we can add some delays on the EC GPE handler, the number of EC GPE interrupt can be reduced very obviously, which
makes the system work well. For example: the laptop in bug 9998 can work well and the EC working flowchart is also simple.
My patch is only for the broken BIOS. For most laptops no delay is added in the EC GPE handler.
Only when five EC GPE interrupts happens in the same jiffies and the EC status is the same, OS will add some delay in
the EC GPE handler. If the same thing still happens after the delay is added, the delay time will be increased.
But the max delay time is 5 microseconds.
If you care to look at the log I've sent several days ago, you will see that interrupts happen for about 1ms after the command
write. So you will spend this whole millisecond spinning.
At the same time IMO udelay is harmless. It is realized by repeat loop instead of halt.
Well, most people try to make interrupt context as short as possible, just because it's not interruptible,
so you loose interactivity. For example music/video could become jerky.
--
To unsubscribe from this list: send the line "unsubscribe linux-acpi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
