KMDF DMA questions?

Hello,
1. Can I have two DMA transactions active at same time?

2. If not, can I create two DMA enabler and have two DMA transactions active
for each enabler at same time?

3. Can KMDF handle as big as 2M Bytes DMA in one transaction? Because I pass
a 2M bytes data buffer to request, from SgList->Elements[0].Length, I only
got SgList->Elements[0].Length= 0x7000, dose it mean I can only transfer
0x7000 bytes in one transfer? Since my DMA mode is WdfDmaProfilePacket.

Thanks very much,

Xinshou

Stephan
Sat Oct 20 07:15:02 PDT 2007

On Oct 19, 7:49 pm, Xinshou2 <Xinsh...@discussions.microsoft.com>
wrote:
> 1. Can I have two DMA transactions active at same time?

Large DMA transfers are usually done in bursts of (multiples of) the
cache line size. A busmaster DMA device acquires the respective bus
(e.g. PCI), transfers data, and releases the bus [this is somewhat
simplified, but I am not a real hardware expert.]

If two devices want to transfer data via DMA on the same bus, (at
least) one device needs to wait until the bus is "free".

So in this respect, the answer is no, there cannot be two simultaneous
DMA transactions. Much like there cannot be two simultaneous running
threads on one processor. Sure they run "at the same time" from a user
perspective, but they actually can only use the same CPU, one after
the other.

A DMA busmaster device can do DMA transfers of any length, but IIRC
these are actually broken down into smaller pieces by the hardware.
This is completely transparent to software, that is, the software can
tell a device to do a transfer of any length but this will actually be
handled in smaller pieces in hardware.

But all this has nothing to do with how many scatter/gather lists can
be there at the same time, or how long one scatter/gather element can
be.

Keep in mind that the DMA hardware needs to use physical addresses. So
if you have, say, a 4 MB buffer in software, the undelying physical
pages can be anywhere in memory and in any order, so you can end up
with many scatter/gather elements of different lenghts for the 4 MB
block.

HTH, Stephan

Doron
Mon Oct 22 13:27:46 PDT 2007

note that what stephen described is a limit of PCI's architecture, not a
limit created by KMDF. anything you can do in WDM's DMA APIs you can do in
KMDF's APIs.

--
Please do not send e-mail directly to this alias. this alias is for
newsgroup purposes only.
This posting is provided "AS IS" with no warranties, and confers no rights.


"Stephan Wolf [MVP]" <stewo68@hotmail.com> wrote in message
news:1192889702.728111.6700@z24g2000prh.googlegroups.com...
> On Oct 19, 7:49 pm, Xinshou2 <Xinsh...@discussions.microsoft.com>
> wrote:
>> 1. Can I have two DMA transactions active at same time?
>
> Large DMA transfers are usually done in bursts of (multiples of) the
> cache line size. A busmaster DMA device acquires the respective bus
> (e.g. PCI), transfers data, and releases the bus [this is somewhat
> simplified, but I am not a real hardware expert.]
>
> If two devices want to transfer data via DMA on the same bus, (at
> least) one device needs to wait until the bus is "free".
>
> So in this respect, the answer is no, there cannot be two simultaneous
> DMA transactions. Much like there cannot be two simultaneous running
> threads on one processor. Sure they run "at the same time" from a user
> perspective, but they actually can only use the same CPU, one after
> the other.
>
> A DMA busmaster device can do DMA transfers of any length, but IIRC
> these are actually broken down into smaller pieces by the hardware.
> This is completely transparent to software, that is, the software can
> tell a device to do a transfer of any length but this will actually be
> handled in smaller pieces in hardware.
>
> But all this has nothing to do with how many scatter/gather lists can
> be there at the same time, or how long one scatter/gather element can
> be.
>
> Keep in mind that the DMA hardware needs to use physical addresses. So
> if you have, say, a 4 MB buffer in software, the undelying physical
> pages can be anywhere in memory and in any order, so you can end up
> with many scatter/gather elements of different lenghts for the 4 MB
> block.
>
> HTH, Stephan
>