On Monday, 10/23/2017 at 12:38 GMT, Michael MacIsaac
wrote:
> The wrapper function tries to trap the error for buffer overflow and if
> found, resend the vmcp command with the required size. So I'm guessing
the
> vast majority of vmcp calls made will have output that will fit in 32K.
>
> Does t
Hi,
I thought of a way to work around this issue. Then I looked at my vmcp
wrapper function (below) and it seems I already did (Homer slapping head -
D'oh).
But I still had the default buffer size for the vmcp command set to 1M.
Given the previous comments, I moved that down to 32K in zoom and zu
Christian wrote:
> I am aware of at least one case where we fixed that as such in an IBM
> product, but I can not remember which one it was.
GDPS xDR was one of them. Some of us have a lot of DASD addresses.
Marcy
As I said, In the future vmcp will make use of the CMA allocator, which
should make sure that the allocation will succeed. The advantage of that
allocator is that this memory is not kept free, but it can be used by movable
pages. So if vmcp needs that memory, CMA will be able to move that memory a
Would it be possible for the vmcp kernel driver to have an option to
pre-allocate some fixed size of contiguous memory and hold onto it? This would
prevent fragmentation which can hit you at any time. I remember using vmcp from
a Linux based installer which typically started by downloading large
An enhancement to DIAG 8 to use a buffer list like DIAG 250 does for disk
I/O would circumvent the contiguous memory requirement. The Ry register
already contains a flag byte that could be extended to indicate Rx+1 being
a buffer list rather than the buffer.
On 10/18/17, 2:13 PM, "Linux on 390 Por
Yes, this is memory fragmentation.
Heiko recently did some changes that should improve the situation.
This upstream change
https://git.kernel.org/pub/scm/linux/kernel/git/s390/linux.git/commit/?h=features&id=cd4386a931b6310b05559d2e28efda04d30ab593
will change the allocation from below 2GB to wher
Fragmentation. Vmcp needs contiguous real storage. So if there is not a
slab of 1MB in size hanging around then it will complain.
On 10/18/17, 1:44 PM, "Linux on 390 Port on behalf of Michael MacIsaac"
wrote:
>Hello list,
>
>We use vmcp extensively to issue CP commands. More and more I'm seeing
Hello list,
We use vmcp extensively to issue CP commands. More and more I'm seeing
this error:
Error: Could not issue CP command: Cannot allocate memory
I always use the --buffer=1M flag to be able to get the largest output. I've
bumped up the VM size from 2 to 3 to 4GB, but still see it. We ha