I've now tried a number of things, and still get failures on the tests. I
also get "inconclusive" tests, and I'm wondering what this means.
Most recently, I cloned the git repo and checked out the next branch (13
Dec 2014), compiled it and did make func in several ways.
Because results had been bad at my usual 12288 2MiB-pages, I followed
guidance in the HOWTO document from git. This gave 3 tests at 32-bit
killed by signal and 2 bad configurations. On investigation, the lines
with the bad configurations offered the notation that at least 32 pages
were required.
So, I retried with 32 pages. The bad configurations disappeared, but the 3
tests were still killed. They show up as tests with a caption, but neither
PASS nor FAIL. Grepping for this, I find
HUGETLB_VERBOSE=0 xB.linkhuge_nofd (2M: 32):
HUGETLB_VERBOSE=0 HUGETLB_MINIMAL_COPY=no xB.linkhuge_nofd (2M: 32):
HUGETLB_VERBOSE=0 HUGETLB_ELFMAP=no xB.linkhuge_nofd (2M: 32):
The word "inconclusive" also shows up on 6 tests:
slbpacaflush (2M: 32): PASS (inconclusive)
slbpacaflush (2M: 64): PASS (inconclusive)
misaligned_offset (2M: 32): PASS (inconclusive)
misaligned_offset (2M: 64): PASS (inconclusive)
LD_PRELOAD=libheapshrink.so HUGETLB_MORECORE_SHRINK=yes
HUGETLB_MORECORE=yes heapshrink (2M: 32): PASS (inconclusive)
LD_PRELOAD=libheapshrink.so HUGETLB_MORECORE_SHRINK=yes
HUGETLB_MORECORE=yes heapshrink (2M: 64): PASS (inconclusive)
Forgive me for the obvious, but I recommend changing the HOWTO document to
recommend at least 32 pages, not 25. It would also be nice to know how to
make all tests conclusive, or to know why that's not possible.
As for the killed tests, I would appreciate some information about what
features are being reported as broken. If I ever use huge pages in a
32-bit process, I'd like to avoid them.
I'd also be glad to help diagnosing and fixing the problem.
Additional info:
uname -a gives
Linux treat 3.13.0-40-generic #69-Ubuntu SMP Thu Nov 13 17:53:56 UTC 2014
x86_64 x86_64 x86_64 GNU/Linux
There's a full 32 GB of RAM. I intend to run with 12288 huge pages, which
is precisely 24 GB, figuring that this machine was running these tasks just
fine with 8GB, and I added 24GB more just to make use of a huge cache on my
pet hobby project, so reserving all 24 should be no problem.
I'd make it a single page if I could. That would sure help with TLB misses.
--
Kevin O'Gorman
#define QUESTION ((bb) || (!bb)) /* Shakespeare */
Please consider the environment before printing this email.
------------------------------------------------------------------------------
Download BIRT iHub F-Type - The Free Enterprise-Grade BIRT Server
from Actuate! Instantly Supercharge Your Business Reports and Dashboards
with Interactivity, Sharing, Native Excel Exports, App Integration & more
Get technology previously reserved for billion-dollar corporations, FREE
http://pubads.g.doubleclick.net/gampad/clk?id=164703151&iu=/4140/ostg.clktrk
_______________________________________________
Libhugetlbfs-devel mailing list
Libhugetlbfs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/libhugetlbfs-devel
