* RE: Problem using more than 2 GB with my driver
2003-09-30 14:13 Problem using more than 2 GB with my driver Patrice Martinez
@ 2003-09-30 15:45 ` Luck, Tony
2003-09-30 17:49 ` Grant Grundler
` (2 subsequent siblings)
3 siblings, 0 replies; 5+ messages in thread
From: Luck, Tony @ 2003-09-30 15:45 UTC (permalink / raw)
To: linux-ia64
> I've got a little problem:
> I wrote a driver using the physical addresses of a buffer computed by
> the "ia64_tpa" function (but no using DMA) that works only on machines
> with 2Gb of RAM at most. With more RAM, the machine crashes very badly
> after a while, and even KDB is not usable...
> If anybody has an a idea ;-)
What does your driver do with the physical address (if not for DMA, why
does it need the physical address?)
-Tony
^ permalink raw reply [flat|nested] 5+ messages in thread* Re: Problem using more than 2 GB with my driver
2003-09-30 14:13 Problem using more than 2 GB with my driver Patrice Martinez
2003-09-30 15:45 ` Luck, Tony
@ 2003-09-30 17:49 ` Grant Grundler
2003-09-30 18:06 ` Alex Williamson
2003-10-01 11:05 ` Patrice Martinez
3 siblings, 0 replies; 5+ messages in thread
From: Grant Grundler @ 2003-09-30 17:49 UTC (permalink / raw)
To: linux-ia64
On Tue, Sep 30, 2003 at 04:13:30PM +0200, Patrice Martinez wrote:
> I wrote a driver using the physical addresses of a buffer computed by
> the "ia64_tpa" function (but no using DMA) that works only on machines
> with 2Gb of RAM at most. With more RAM, the machine crashes very badly
> after a while, and even KDB is not usable...
> If anybody has an a idea ;-)
Which machine is this on?
rx2600/rx5670 has a memory hole from 2-4GB because of IO "stuff".
ie memory > 2G ends up getting placed above 4GB physical.
Sounds like your code is using one or more of:
1) unsigned int
2) signed int -> signed long (automatic conversion for math).
3) linear addressing (not aware of memory holes)
And I'm curious as well what your driver does with physical addresses...
It's obviously not DMA since that wouldn't work on most of the ia64 machines
out there today.
grant
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: Problem using more than 2 GB with my driver
2003-09-30 14:13 Problem using more than 2 GB with my driver Patrice Martinez
2003-09-30 15:45 ` Luck, Tony
2003-09-30 17:49 ` Grant Grundler
@ 2003-09-30 18:06 ` Alex Williamson
2003-10-01 11:05 ` Patrice Martinez
3 siblings, 0 replies; 5+ messages in thread
From: Alex Williamson @ 2003-09-30 18:06 UTC (permalink / raw)
To: linux-ia64
On Tue, 2003-09-30 at 11:49, Grant Grundler wrote:
> On Tue, Sep 30, 2003 at 04:13:30PM +0200, Patrice Martinez wrote:
> > I wrote a driver using the physical addresses of a buffer computed by
> > the "ia64_tpa" function (but no using DMA) that works only on machines
> > with 2Gb of RAM at most. With more RAM, the machine crashes very badly
> > after a while, and even KDB is not usable...
> > If anybody has an a idea ;-)
>
> Which machine is this on?
> rx2600/rx5670 has a memory hole from 2-4GB because of IO "stuff".
> ie memory > 2G ends up getting placed above 4GB physical.
>
Minor correction, on zx1, memory in the 1-4GB range is relocated. So
there's actually only 1GB of memory below 4GB.
Alex
--
Alex Williamson HP Linux & Open Source Lab
^ permalink raw reply [flat|nested] 5+ messages in thread
* RE: Problem using more than 2 GB with my driver
2003-09-30 14:13 Problem using more than 2 GB with my driver Patrice Martinez
` (2 preceding siblings ...)
2003-09-30 18:06 ` Alex Williamson
@ 2003-10-01 11:05 ` Patrice Martinez
3 siblings, 0 replies; 5+ messages in thread
From: Patrice Martinez @ 2003-10-01 11:05 UTC (permalink / raw)
To: linux-ia64
Hello,
In fact, Tony, I'm working on a driver that can copy data directly
between user spaces using physical addresses for MPI applications.
And thank you, Grant, there was an integer expression lying between long
ones in an address computation which causes losses of high order bits in
addresses...
So it works far better now!
Bye ;-)
^ permalink raw reply [flat|nested] 5+ messages in thread