Re: [Bug 56881] New: MAP_HUGETLB mmap fails for certain sizes

["Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>]

Naoya Horiguchi <n-horiguchi@ah.jp.nec.com> writes:

> On Wed, Apr 24, 2013 at 11:39:51AM -0400, Johannes Weiner wrote:
>> On Wed, Apr 24, 2013 at 11:16:39AM -0400, Naoya Horiguchi wrote:
>> > On Wed, Apr 24, 2013 at 04:14:54AM -0400, Johannes Weiner wrote:
>> > > @@ -491,10 +491,13 @@ static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
>> > >  
>> > >  	sprintf (name, "SYSV%08x", key);
>> > >  	if (shmflg & SHM_HUGETLB) {
>> > > +		unsigned int hugesize;
>> > > +
>> > >  		/* hugetlb_file_setup applies strict accounting */
>> > >  		if (shmflg & SHM_NORESERVE)
>> > >  			acctflag = VM_NORESERVE;
>> > > -		file = hugetlb_file_setup(name, 0, size, acctflag,
>> > > +		hugesize = ALIGN(size, huge_page_size(&default_hstate));
>> > > +		file = hugetlb_file_setup(name, hugesize, acctflag,
>> > >  				  &shp->mlock_user, HUGETLB_SHMFS_INODE,
>> > >  				(shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
>> > >  	} else {
>> > 
>> > Would it be better to find proper hstate instead of using default_hstate?
>> 
>> You are probably right, I guess we can't assume default_hstate anymore
>> after page_size_log can be passed in.
>> 
>> Can we have hugetlb_file_setup() return an adjusted length, or an
>> alignment requirement?
>
> Yes, it's possible if callers pass the pointer of size (length) to
> hugetlb_file_setup() and make it adjusted inside the function.
> And as for alignment, I think it's not a hugetlb_file_setup's job,
> so we don't have to do it in this function.
>
>> Or pull the hstate lookup into the callsites (since they pass in
>> page_size_log to begin with)?
>
> This is also a possible solution, where we might need to define and
> export a function converting hugepage order to hstate.
>
> I like the former one, so wrote a patch like below.
> # I added your Signed-off-by: because this's based on your draft patch.
> # if you don't like it, please let me know.
>
> Thanks,
> Naoya Horiguchi
> ---
> From: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
> Date: Wed, 24 Apr 2013 16:44:19 -0400
> Subject: [PATCH] hugetlbfs: fix mmap failure in unaligned size request
>
> As reported in https://bugzilla.kernel.org/show_bug.cgi?id=56881, current
> kernel returns -EINVAL unless a given mmap length is "almost" hugepage
> aligned. This is because in sys_mmap_pgoff() the given length is passed to
> vm_mmap_pgoff() as it is without being aligned with hugepage boundary.
>
> This is a regression introduced in commit 40716e29243d "hugetlbfs: fix
> alignment of huge page requests", where alignment code is pushed into
> hugetlb_file_setup() and the variable len in caller side is not changed.
>
> To fix this, this patch partially reverts that commit, and changes
> the type of parameter size from size_t to (size_t *) in order to
> align the size in caller side.

After the change af73e4d9506d3b797509f3c030e7dcd554f7d9c4 we have
alignment related failures in libhugetlbfs test suite. misalign test
fails with 3.10-rc5, while it works with 3.9.

-aneesh

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>