mem: mark pages as not accessed when reserving VA
authorDavid Marchand <david.marchand@redhat.com>
Mon, 9 Mar 2020 14:54:42 +0000 (15:54 +0100)
committerDavid Marchand <david.marchand@redhat.com>
Fri, 13 Mar 2020 12:53:12 +0000 (13:53 +0100)
When the memory allocator reserves virtual addresses, it still does not
know what they will be used for.
Besides, huge areas are reserved for memory hotplug in multiprocess
setups. But most of the pages are unused in the whole life of the
processes.

Change protection mode to PROT_NONE when only reserving VA.
The memory allocator already switches to the right mode when making use
of it.

It also has the nice effect of getting those pages skipped by the kernel
when calling mlockall() or when a coredump gets generated.

Cc: stable@dpdk.org
Suggested-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: David Marchand <david.marchand@redhat.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Acked-by: Aaron Conole <aconole@redhat.com>
Acked-by: Anatoly Burakov <anatoly.burakov@intel.com>
lib/librte_eal/common/eal_common_memory.c

index 4a9cc1f..cc7d54e 100644 (file)
@@ -97,7 +97,7 @@ eal_get_virtual_area(void *requested_addr, size_t *size,
                        return NULL;
                }
 
-               mapped_addr = mmap(requested_addr, (size_t)map_sz, PROT_READ,
+               mapped_addr = mmap(requested_addr, (size_t)map_sz, PROT_NONE,
                                mmap_flags, -1, 0);
                if (mapped_addr == MAP_FAILED && allow_shrink)
                        *size -= page_sz;