Re: Serial: bug in 8250.c when handling PCI or other level triggers

From: Russell King
Date: Wed Dec 14 2005 - 14:54:17 EST


On Wed, Dec 14, 2005 at 07:08:08PM +0000, Alan Cox wrote:
> On Mer, 2005-12-14 at 16:55 +0000, Russell King wrote:
> > If we trigger this, we can assume that the port is dead anyway, or
> > we're in a situation where the host CPU can not keep up with the
> > data stream.
>
> Not actually true in some cases.
>
> - When your UART has a large FIFO and pretends to be an 8250 you can get
> a 256 byte burst triggered by the box sleeping for a moment or the BIOS
> SMI crap going to chat to the battery

In which case the receive_chars() function gobbles up to 255 characters
from the device before relinquishing to the main interrupt loop. The
main interrupt loop has two exit conditions - no further interrupts
are pending from any device, or we run this loop 256 times.

In the case where further characters are waiting, we will re-run the
receive_chars() function.

Hence, we will check the device up to 256 times and each will potentially
receive 255 characters, which gives about 64K of character reception
before the warning triggers.

Therefore, this scenario is _very_ _very_ unlikely.

> - On a virtualised system this trap can trigger because the emulations
> don't emulate the bit arrival and baud rate.

Again, only if there's more than about 64K of data waiting.

--
Russell King
Linux kernel 2.6 ARM Linux - http://www.arm.linux.org.uk/
maintainer of: 2.6 Serial core
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/