mm/shmem.c: thp, shmem: fix conflict of above-47bit hint address and PMD alignment
authorKirill A. Shutemov <kirill@shutemov.name>
Tue, 14 Jan 2020 00:29:13 +0000 (16:29 -0800)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Thu, 23 Jan 2020 07:21:30 +0000 (08:21 +0100)
commit 991589974d9c9ecb24ee3799ec8c415c730598a2 upstream.

Shmem/tmpfs tries to provide THP-friendly mappings if huge pages are
enabled.  But it doesn't work well with above-47bit hint address.

Normally, the kernel doesn't create userspace mappings above 47-bit,
even if the machine allows this (such as with 5-level paging on x86-64).
Not all user space is ready to handle wide addresses.  It's known that
at least some JIT compilers use higher bits in pointers to encode their
information.

Userspace can ask for allocation from full address space by specifying
hint address (with or without MAP_FIXED) above 47-bits.  If the
application doesn't need a particular address, but wants to allocate
from whole address space it can specify -1 as a hint address.

Unfortunately, this trick breaks THP alignment in shmem/tmp:
shmem_get_unmapped_area() would not try to allocate PMD-aligned area if
*any* hint address specified.

This can be fixed by requesting the aligned area if the we failed to
allocated at user-specified hint address.  The request with inflated
length will also take the user-specified hint address.  This way we will
not lose an allocation request from the full address space.

[kirill@shutemov.name: fold in a fixup]
Link: http://lkml.kernel.org/r/20191223231309.t6bh5hkbmokihpfu@box
Link: http://lkml.kernel.org/r/20191220142548.7118-3-kirill.shutemov@linux.intel.com
Fixes: b569bab78d8d ("x86/mm: Prepare to expose larger address space to userspace")
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: "Willhalm, Thomas" <thomas.willhalm@intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: "Bruggeman, Otto G" <otto.g.bruggeman@intel.com>
Cc: "Aneesh Kumar K . V" <aneesh.kumar@linux.vnet.ibm.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
mm/shmem.c

index 650ef56..a6fc82a 100644 (file)
@@ -2072,9 +2072,10 @@ unsigned long shmem_get_unmapped_area(struct file *file,
        /*
         * Our priority is to support MAP_SHARED mapped hugely;
         * and support MAP_PRIVATE mapped hugely too, until it is COWed.
-        * But if caller specified an address hint, respect that as before.
+        * But if caller specified an address hint and we allocated area there
+        * successfully, respect that as before.
         */
-       if (uaddr)
+       if (uaddr == addr)
                return addr;
 
        if (shmem_huge != SHMEM_HUGE_FORCE) {
@@ -2108,7 +2109,7 @@ unsigned long shmem_get_unmapped_area(struct file *file,
        if (inflated_len < len)
                return addr;
 
-       inflated_addr = get_area(NULL, 0, inflated_len, 0, flags);
+       inflated_addr = get_area(NULL, uaddr, inflated_len, 0, flags);
        if (IS_ERR_VALUE(inflated_addr))
                return addr;
        if (inflated_addr & ~PAGE_MASK)