I suspect this points to the issue (from the file pinfo.mk):

define PINFO
PINFO DESCRIPTION=Intel 854 non-BIOS PCI server
endef

But then again, I have been confused by the variations in this project before.

I finally understand what is happening.  I am not sure how to resolve it.

So, the PCI server allocates a block of memory for all busses from 0xff900000 to 0xffdfffff.  There are a number of 
devices that are on Bus 1 that start allocating that memory in fairly small chunks starting at 0xffdfffff and working 
down.  

Next the 82574 driver grabs 0x24000 in memory starting at 0xff900000.  It grabs a full 1 MB of memory even though it 
only needs a little more than 128kB. Both requests are given an alignment value of 0x80000.

Then the driver for the Intel 82580 in e1000 starts grabbing memory.  It grabs two buffers for each ethernet port in the
 52850. the first is 0x80000 bytes and the second is 0x4000.  The problem is that for some reason *both* allocation 
requests are given an alignment value of 0x80000.  That means we only need 16kB for that small buffer but we chew up an 
entire 1 MB of memory for the two buffers.

By the time it allocates buffers for the first 4 ethernet ports, we have chewed up 4 MB of memory when the devices 
arguably only need 2 MB plus another 64 kB.

Now that 5th ethernet port comes along and *should* be able to grab 1/2 MB for its big buffer at 0xffd00000 and another 
16 kB at 0xffd80000 for its small buffer.  However, since the request sets a 0x80000 alignment, the request fails.

So, is there some reason that the e1000 driver needs to have those memory buffers on 1 MB boundaries?  Is there memory 
available at 0xff800000 that would give us the space we need for one more buffer?

Thanks.

The PCI server determines the size of the memory/I/O aperture by writing
all F¹s to the register and then reading it back. You can see the size of
the apertures by starting the PCI server with Œ-vvv¹ and looking at the
sloginfo output. There you should see something like ³MEM 0xffc00000
0x80000². This is where the PCI server has determined the size of the
aperture. You could maybe expand the find_all() function in the PCI server
to dump the base address registers to see what the BIOS assigns to them.



On 2014/4/14, 6:17 PM, "Ray Mack" <community-noreply@qnx.com> wrote:

>I finally understand what is happening.  I am not sure how to resolve it.
>
>So, the PCI server allocates a block of memory for all busses from
>0xff900000 to 0xffdfffff.  There are a number of devices that are on Bus
>1 that start allocating that memory in fairly small chunks starting at
>0xffdfffff and working down.
>
>Next the 82574 driver grabs 0x24000 in memory starting at 0xff900000.  It
>grabs a full 1 MB of memory even though it only needs a little more than
>128kB. Both requests are given an alignment value of 0x80000.
>
>Then the driver for the Intel 82580 in e1000 starts grabbing memory.  It
>grabs two buffers for each ethernet port in the 52850. the first is
>0x80000 bytes and the second is 0x4000.  The problem is that for some
>reason *both* allocation requests are given an alignment value of
>0x80000.  That means we only need 16kB for that small buffer but we chew
>up an entire 1 MB of memory for the two buffers.
>
>By the time it allocates buffers for the first 4 ethernet ports, we have
>chewed up 4 MB of memory when the devices arguably only need 2 MB plus
>another 64 kB.
>
>Now that 5th ethernet port comes along and *should* be able to grab 1/2
>MB for its big buffer at 0xffd00000 and another 16 kB at 0xffd80000 for
>its small buffer.  However, since the request sets a 0x80000 alignment,
>the request fails.
>
>So, is there some reason that the e1000 driver needs to have those memory
>buffers on 1 MB boundaries?  Is there memory available at 0xff800000 that
>would give us the space we need for one more buffer?
>
>Thanks.
>
>
>
>_______________________________________________
>
>Networking Drivers
>http://community.qnx.com/sf/go/post109915
>To cancel your subscription to this discussion, please e-mail
>drivers-networking-unsubscribe@community.qnx.com

A quick test to see why my posts are not making it to the forum.  I have tried at least 5 messages recently that have 
not made it to the forum.

I know what is happening just not why.  devb-mmcsd is starting libcam.so which gets 0x100 bytes from ethernet memory and
 cam-disk.so is getting allocated 0x400 bytes from ethernet memory.

If this post works, I'll try again with appropriate information.