Messages in this thread | | | From | (Alan Cox) | Subject | Re: ping fails under 2.1.39 | Date | Fri, 23 May 1997 19:54:03 +0100 (BST) |
| |
> Also, the method used by the kernel to delay I/O (writing to an unused port) > can (read will) lead to other problems. You should not perform a bus > operation if you intend to allow the bus to settle. It makes more glitches
It isnt waiting for the bus to settle. If you need to wait for bus events you've got FAULTY ISA bus hardware, you just have to give the poor little NS8390 a breather. Similarly for some other I/O chips that are specced for stuff like 2MHz operation.
> compensated for a chip-to-chip select-time bug with the delay, the resulting > bus activity can cause invalid reads/writes during this operation.
The only select bug we have to deal with is the parallel port one, and that is done by writing the byte twice to ensure it remembers to latch it.
> or PCI do not have these bus-settle problems. Further, "modern" 8390 clones > don't have the the chip-to-chip select-time bug. Basically, if you have
Alas not true. The Simply PCI NE2K's have all the standard NS8390 timing problems. So the driver has to treat it like you are supposed to treat an NS8390 - at which point it works beautifully.
The problem with stuff like bus delays now is that the CPU isnt driving the ISA bus on a VLB/ISA or PCI/ISA board. Instead a whole chipset is doing speculative PCI and ISA reads and other clever magic to glue things all together. It really is out of the CPU's hands now.
Alan
| |