diff mbox series

mem: Fix anonymous mapping on Power9.

Message ID 20181114204051.9580-1-dwilder@us.ibm.com (mailing list archive)
State Superseded, archived
Delegated to: Thomas Monjalon
Headers
Series mem: Fix anonymous mapping on Power9. |

Checks

Context Check Description
ci/checkpatch success coding style OK
ci/mellanox-Performance-Testing success Performance Testing PASS
ci/intel-Performance-Testing success Performance Testing PASS
ci/Intel-compilation success Compilation OK

Commit Message

David Wilder Nov. 14, 2018, 8:40 p.m. UTC 
  Removed the use of MAP_HUGETLB for anonymous mapping on ppc64.  The
MAP_HUGETLB had previously been added to workaround issues on IBM Power8
systems when mapping /dev/zero (see: commit
284ae3e9ff9a92575c28c858efd2c85c8de6d440).  In the current code the
MAP_HUGETLB flag will cause the anonymous mapping to fail on Power9.
Note, Power8 is currently failing to correctly mmap Hugepages, with and
without this change.

Signed-off-by: David Wilder <dwilder@us.ibm.com>
Reviewed-by: Pradeep Satyanarayana <pradeep@us.ibm.com>
---
This patch corrects problems with virtual address mapping on IBM Power9 system.
I have tested with primary and secondary process tests. IBM power8 systems
continue to fail with and without this change. I am perusing a fix for power8
and will post a patch when its working. As Power9 is our primary platform I
wanted to get this patch out. I will follow up with a power8 fix when its ready.

Please include this patch in 18.11 as well, this patch is not suitable
for 17.11.

 lib/librte_eal/linuxapp/eal/eal_memalloc.c | 3 ---
 lib/librte_eal/linuxapp/eal/eal_memory.c   | 4 ----
 2 files changed, 7 deletions(-)

Comments

Anatoly Burakov Nov. 15, 2018, 10:49 a.m. UTC | #1 
On 14-Nov-18 8:40 PM, David Wilder wrote:
> Removed the use of MAP_HUGETLB for anonymous mapping on ppc64.  The
> MAP_HUGETLB had previously been added to workaround issues on IBM Power8
> systems when mapping /dev/zero (see: commit
> 284ae3e9ff9a92575c28c858efd2c85c8de6d440).  In the current code the
> MAP_HUGETLB flag will cause the anonymous mapping to fail on Power9.
> Note, Power8 is currently failing to correctly mmap Hugepages, with and
> without this change.
> 
> Signed-off-by: David Wilder <dwilder@us.ibm.com>
> Reviewed-by: Pradeep Satyanarayana <pradeep@us.ibm.com>
> ---

Can't test it (obviously), but LGTM

Acked-by: Anatoly Burakov <anatoly.burakov@intel.com>
Anatoly Burakov Nov. 15, 2018, 3:11 p.m. UTC | #2 
On 14-Nov-18 8:40 PM, David Wilder wrote:

> Please include this patch in 18.11 as well, this patch is not suitable
> for 17.11.
> 
Small note: while this patch is not suitable for 17.11 as is, it can be 
made suitable if you drop the memalloc part. I copied this flag from old 
memory allocation code, so if 17.11 has this flag there, you can fix it 
in 17.11 as well.
Thomas Monjalon Nov. 15, 2018, 6:17 p.m. UTC | #3 
Hi,

14/11/2018 21:40, David Wilder:
> Removed the use of MAP_HUGETLB for anonymous mapping on ppc64.  The
> MAP_HUGETLB had previously been added to workaround issues on IBM Power8
> systems when mapping /dev/zero (see: commit
> 284ae3e9ff9a92575c28c858efd2c85c8de6d440).  In the current code the
> MAP_HUGETLB flag will cause the anonymous mapping to fail on Power9.
> Note, Power8 is currently failing to correctly mmap Hugepages, with and
> without this change.
> 
> Signed-off-by: David Wilder <dwilder@us.ibm.com>
> Reviewed-by: Pradeep Satyanarayana <pradeep@us.ibm.com>
> ---
> This patch corrects problems with virtual address mapping on IBM Power9 system.
> I have tested with primary and secondary process tests. IBM power8 systems
> continue to fail with and without this change. I am perusing a fix for power8
> and will post a patch when its working. As Power9 is our primary platform I
> wanted to get this patch out. I will follow up with a power8 fix when its ready.

We need to document that Power8 is not supported in DPDK 18.11.
Please send a patch to explain it in "known issues":
	http://doc.dpdk.org/guides/rel_notes/release_18_11.html#known-issues

We may also note that IBM POWER support is in alpha stage:
	http://git.dpdk.org/dpdk/tree/MAINTAINERS#n236

Thanks in advance
David Wilder Nov. 16, 2018, 11:56 p.m. UTC | #4 
On 2018-11-15 07:11, Burakov, Anatoly wrote:
> On 14-Nov-18 8:40 PM, David Wilder wrote:
> 
>> Please include this patch in 18.11 as well, this patch is not suitable
>> for 17.11.
>> Small note: while this patch is not suitable for 17.11 as is, it can 
>> be made suitable if you drop the memalloc part. I copied this flag 
>> from old memory allocation code, so if 17.11 has this flag there, you 
>> can fix it in 17.11 as well.

Thank you

I am working on a 17.11 fix as well,  the issue is a little different on 
17.11 but removing the flag is indeed part of the fix.
diff mbox series

Patch

diff --git a/lib/librte_eal/linuxapp/eal/eal_memalloc.c b/lib/librte_eal/linuxapp/eal/eal_memalloc.c
index 48b9c7360..784939566 100644
--- a/lib/librte_eal/linuxapp/eal/eal_memalloc.c
+++ b/lib/librte_eal/linuxapp/eal/eal_memalloc.c
@@ -753,9 +753,6 @@  alloc_seg(struct rte_memseg *ms, void *addr, int socket_id,
 	munmap(addr, alloc_sz);
 unmapped:
 	flags = MAP_FIXED;
-#ifdef RTE_ARCH_PPC_64
-	flags |= MAP_HUGETLB;
-#endif
 	new_addr = eal_get_virtual_area(addr, &alloc_sz, alloc_sz, 0, flags);
 	if (new_addr != addr) {
 		if (new_addr != NULL)
diff --git a/lib/librte_eal/linuxapp/eal/eal_memory.c b/lib/librte_eal/linuxapp/eal/eal_memory.c
index 48b23ce19..6f94621d4 100644
--- a/lib/librte_eal/linuxapp/eal/eal_memory.c
+++ b/lib/librte_eal/linuxapp/eal/eal_memory.c
@@ -847,10 +847,6 @@  alloc_va_space(struct rte_memseg_list *msl)
 	void *addr;
 	int flags = 0;
 
-#ifdef RTE_ARCH_PPC_64
-	flags |= MAP_HUGETLB;
-#endif
-
 	page_sz = msl->page_sz;
 	mem_sz = page_sz * msl->memseg_arr.len;