[dpdk-dev,2/2] memalloc: keep in mind a failed MAP_FIXED mmap may still perform an unmap
Checks
Commit Message
This isn't documented in the manuals, but a failed
mmap(..., MAP_FIXED) may still unmap overlapping
regions. In such case, we need to remap these regions
back into our address space to ensure mem contiguity.
We do it unconditionally now on mmap failure just to
be safe.
Verified on Linux 4.9.0-4-amd64. I was getting
ENOMEM when trying to map in hugetlbfs with no space
left, but the previous anonymous mapping was still
being removed.
Signed-off-by: Dariusz Stojaczyk <dariuszx.stojaczyk@intel.com>
---
lib/librte_eal/linuxapp/eal/eal_memalloc.c | 9 ++++++++-
1 file changed, 8 insertions(+), 1 deletion(-)
Comments
On 01-Jun-18 1:51 PM, Dariusz Stojaczyk wrote:
> This isn't documented in the manuals, but a failed
> mmap(..., MAP_FIXED) may still unmap overlapping
> regions. In such case, we need to remap these regions
> back into our address space to ensure mem contiguity.
> We do it unconditionally now on mmap failure just to
> be safe.
>
> Verified on Linux 4.9.0-4-amd64. I was getting
> ENOMEM when trying to map in hugetlbfs with no space
> left, but the previous anonymous mapping was still
> being removed.
>
> Signed-off-by: Dariusz Stojaczyk <dariuszx.stojaczyk@intel.com>
> ---
Does this also happen with other error values?
> -----Original Message-----
> From: Burakov, Anatoly
> Sent: Friday, June 1, 2018 1:09 PM
> To: Stojaczyk, DariuszX <dariuszx.stojaczyk@intel.com>; dev@dpdk.org
> Subject: Re: [PATCH 2/2] memalloc: keep in mind a failed MAP_FIXED
> mmap may still perform an unmap
>
> On 01-Jun-18 1:51 PM, Dariusz Stojaczyk wrote:
> > This isn't documented in the manuals, but a failed mmap(...,
> > MAP_FIXED) may still unmap overlapping regions. In such case, we need
> > to remap these regions back into our address space to ensure mem
> > contiguity.
> > We do it unconditionally now on mmap failure just to be safe.
> >
> > Verified on Linux 4.9.0-4-amd64. I was getting ENOMEM when trying to
> > map in hugetlbfs with no space left, but the previous anonymous
> > mapping was still being removed.
> >
> > Signed-off-by: Dariusz Stojaczyk <dariuszx.stojaczyk@intel.com>
> > ---
>
> Does this also happen with other error values?
The man pages aren't clear on this. It could. I know this patch only remaps the region if errrno == ENOMEM, but we could remap it unconditionally after each failed mmap just as well. It won't hurt if we remap a region unnecessarily. This was my original intent, but apparently I didn't fully commit my changes before publishing patches here. Uhh, sorry.
D.
>
> --
> Thanks,
> Anatoly
@@ -527,7 +527,13 @@ alloc_seg(struct rte_memseg *ms, void *addr, int socket_id,
if (va == MAP_FAILED) {
RTE_LOG(DEBUG, EAL, "%s(): mmap() failed: %s\n", __func__,
strerror(errno));
- goto resized;
+ if (errno == ENOMEM) {
+ /* mmap failed, but the previous region might have
+ * been unmapped anyway; try to remap it */
+ goto unmapped;
+ } else {
+ goto resized;
+ }
}
if (va != addr) {
RTE_LOG(DEBUG, EAL, "%s(): wrong mmap() address\n", __func__);
@@ -588,6 +594,7 @@ alloc_seg(struct rte_memseg *ms, void *addr, int socket_id,
mapped:
munmap(addr, alloc_sz);
+unmapped:
flags = MAP_FIXED;
#ifdef RTE_ARCH_PPC_64
flags |= MAP_HUGETLB;