Skip to contentSkip to conversation

PCI bus mastering question

DISCUSSION Mateusz Berezecki 2 participants 2 messages Mar 29, 2005

lore
Share & tools
lore
#0 lore raw

Hello dear list readers,

I've been googling on the topic for a couple of minutes and I got some question. If PCI bus mastering means the device gets the control over the bus and does all device voodoo is it possible to achieve the same effects without using mastering?
To be more specific: if the device wrote some - let's say register for example - to a
memory location, would the result be identical if a kernel read the same register from a device and stored it somewhere in memory? Or does mastering trigger some special
stuff in a device so it acts different?

kind regards
/mb

#1 lore raw
Quoted 16 lines · Mateusz Berezecki

Hello dear list readers,

I've been googling on the topic for a couple of minutes and I got some question.
If PCI bus mastering means the device gets the control over the bus and does all
device voodoo is it possible to achieve the same effects without using mastering?
To be more specific: if the device wrote some - let's say register for example - to a
memory location, would the result be identical if a kernel read the same register from a device and stored it somewhere in memory? Or does mastering trigger some special stuff in a device so it acts different?

Quoted 2 lines

kind regards
/mb

Bus-mastering is a DMA operation. Devices that can perform bus-mastering can read or write data to or from RAM while the CPU is doing something else.

Actually, it isn't all that "free", though, because during a bus-master operation, the CPU is locked off the bus. The bus-mastering operation is usually broken up into bursts so the CPU isn't locked off the bus for very long.

With many PCI devices, there isn't any to get data into or out of the device without using bus-mastering so you would need to properly program the device to perform the operation.

This comes about because the internal RAM, inside the device, isn't accessible from the PCI/Bus. The only path is through the bus-mastering controller on the board. Otherwise, there isn't any difference between the CPU reading data from the board and putting it into memory, or the bus-mastering controller doing it for you --except, when the CPU does it, it's busy 100% of the time until done, it the case of bus-mastering the CPU is busy only a small fraction of the total transfer time.

Wonderful thing about bus-mastering controllers. When they are programmed correctly, the work like a charm. If you screw up the setup, though, they can overwrite the whole kernel, with no way out but the power switch!

Fun! Do lots of back-ups, keep important stuff on external USB or Firewire drives so a gigantic crash won't kill a week's work. The world's "best" documentation might not tell you that the scatter-list address needs to be big-endian or something like that! It's easy to get things wrong and you don't know until you throw it off the cliff and see if it flies! If it works, it works. If it doesn't it tries to get even by destroying your work!

Nevertheless, don't be frightened. Just be prepared. The best performing software you will ever write involves bus-mastering devices. It's well worth the effort.

Cheers,
Dick Johnson
Penguin : Linux version 2.6.11 on an i686 machine (5537.79 BogoMips).
Notice : All mail here is now cached for review by Dictator Bush.
98.36% of all statistics are fiction.