At 2011-03-28 10:00:27，"David Gibson" <[email protected]> wrote:
>On Sun, Mar 27, 2011 at 10:25:23PM +0800, rocky wrote:
>> Hey :)
>>
>> I am using libhugetlbfs to test ARM hugetlb support , I am a bit of
>> confused of some test cases in libhugetlbfs.
>
>> take alloc-instantiate-race for exmapple,
>> for shared mapping, why does we employ fork to produce 2 process do the
>> racing?
>> while for private mapping, we actually use pthread do it .
>> I don't get it the reason of this testing theme,
>
>alloc-instantiate-race is designed to test for a specific kernel bug
>we had at one point. This bug occurred when two processes or threads
>raced to instantiate the last available huge page in the system. The
>buggy case only occurred when the two threads/processes were
>attempting to instantiate the *same* huge page - that is, that if the
>two faults did not race, but occurred sequentially the second fault
>would simply use the hugepage already instantiated by the first
>fault. The bug was that the second fault when racing would fail to
>allocate a new page (because none were available) and fail the fault,
>even though if it waited the necessary page would be supplied by the
>first fault.
>
>There are two situations where this can occur. The first is in a
>MAP_SHARED mapping, shared between two processes. Here the race it to
>instantiate the same page in the shared mapping's radix tree. We
>could theoretically test this with threads instead of processes, but
>in that case we wouldn't really be testing the proper synchronization
>between different processes that share a mapping.
>
>The second case is a MAP_PRIVATE mapping, with the race being between
>two threads sharing that private mapping (since a private mapping is
>not shared between processes). In this case the race is to
>instantiate the same page in the mm's page tables which are shared
>between the threads.
>
>--
>David Gibson | I'll have my music baroque, and my code
>david AT gibson.dropbear.id.au | minimalist, thank you. NOT _the_ _other_
> | _way_ _around_!
>http://www.ozlabs.org/~dgibson
You are the guy who wrote libhugetlbfs, nice to reach you in mail.
Yes, I understand now, thanks for your help:)
And a couple of question if you don't mind, I do the homework and I'm
struggling to make it clear.
platform : ARM(32bit) cortex A9 SMP platform with 16MB hugepage, linux-2.6.34
1:mremap-expand-slice-collision
this test FAIL both on X86_64/ARM , it's about mremap on normal 4K page,not
hugepage
mmap SLICE_BOUNDARY-0x400 ~ SLICE_BOUNDARY OK
mremap SLICE_BOUNDARY-0x400 ~ SLICE_BOUNDARY+0x400 OK
readback SLICE_BOUNDARY-0x400 ~ SLICE_BOUNDARY OK
readback SLICE_BOUNDARY ~ SLICE_BOUNDARY+0X400 BUS ERROR
attachment is mremap test code.
2: icache-hygiene
test rationale says:
'This test will never trigger (obviously) on machines with coherent icache
and dcache (including x86 and POWER5). '
Is this test case applicable to arm ?
3:slbpacaflush
Is this test case applicable to arm ?
4:truncate_above_4GB
This test failed on mmap64, I dig the code, it fails here:
fs/hugetlbfs/inode.c/hugetlbfs_file_mmap:
if (vma->vm_pgoff & ~(huge_page_mask(h) >> PAGE_SHIFT))
return -EINVAL;
I don't know the FOURGIG number is OK for 32bit arm?
thanks for you time.
rocky
/*
* libhugetlbfs - Easy use of Linux hugepages
* Copyright (C) 2009 David Gibson, IBM Corporation.
*
* This library is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public License
* as published by the Free Software Foundation; either version 2.1 of
* the License, or (at your option) any later version.
*
* This library is distributed in the hope that it will be useful, but
* WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with this library; if not, write to the Free Software
* Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
*/
#define _GNU_SOURCE
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <sys/mman.h>
#include <errno.h>
#define RANDOM_CONSTANT 0x1234ABCD
#ifdef __LP64__
#define SLICE_BOUNDARY 0x20000000000
#else
//#define SLICE_BOUNDARY 0xe0000000
#define SLICE_BOUNDARY 0x80000000
#endif
long hpage_size, page_size;
void do_readback(void *p, size_t size, const char *stage)
{
unsigned int *q = p;
int i;
printf("do_readback(%p, 0x%lx, \"%s\")\n", p,
(unsigned long)size, stage);
for (i = 0; i < (size / sizeof(*q)); i++) {
q[i] = RANDOM_CONSTANT ^ i;
}
for (i = 0; i < (size / sizeof(*q)); i++) {
if (q[i] != (RANDOM_CONSTANT ^ i))
printf("Stage \"%s\": Mismatch at offset 0x%x: 0x%x
instead of 0x%x",
stage, i, q[i], RANDOM_CONSTANT ^ i);
}
}
int main(int argc, char *argv[])
{
int fd, rc;
void *p, *q, *r;
page_size = getpagesize();
q = mmap((void *)(SLICE_BOUNDARY - page_size), page_size,
PROT_READ | PROT_WRITE,
MAP_SHARED | MAP_ANONYMOUS | MAP_FIXED, -1, 0);
if (q == MAP_FAILED)
printf("mmap(normal below): %s", strerror(errno));
do_readback(q, page_size, "normal below");
printf("Attempting to remap...");
r = mremap(q, page_size, 2*page_size, 0);
if (r == MAP_FAILED) {
printf("disallowed\n");
rc = munmap(q, page_size);
if (rc != 0)
printf("munmap(normal below): %s", strerror(errno));
} else {
if (r != q)
printf("mremap() moved without MREMAP_MAYMOVE!?");
printf("testing...");
do_readback(q, 2*page_size, "normal below expanded");
rc = munmap(q, 2*page_size);
if (rc != 0)
printf("munmap(normal below expanded): %s",
strerror(errno));
}
}
------------------------------------------------------------------------------
Enable your software for Intel(R) Active Management Technology to meet the
growing manageability and security demands of your customers. Businesses
are taking advantage of Intel(R) vPro (TM) technology - will your software
be a part of the solution? Download the Intel(R) Manageability Checker
today! http://p.sf.net/sfu/intel-dev2devmar
_______________________________________________
Libhugetlbfs-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/libhugetlbfs-devel
