[PATCH v2] xen/gntdev: Use mempolicy instead of VM_IO flag to avoid NUMA balancing

[Boris Ostrovsky]

Commit 9c17d96500f7 ("xen/gntdev: Grant maps should not be subject to
NUMA balancing") set VM_IO flag to prevent grant maps from being
subjected to NUMA balancing.

It was discovered recently that this flag causes get_user_pages() to
always fail with -EFAULT.

check_vma_flags
__get_user_pages
__get_user_pages_locked
__get_user_pages_unlocked
get_user_pages_fast
iov_iter_get_pages
dio_refill_pages
do_direct_IO
do_blockdev_direct_IO
do_blockdev_direct_IO
ext4_direct_IO_read
generic_file_read_iter
aio_run_iocb

(which can happen if guest's vdisk has direct-io-safe option).

To avoid this, instead of setting VM_IO use mempolicy that prohibits page
migration (i.e. clear policy's MPOL_F_MOF|MPOL_F_MORON)

Reported-by: Olaf Hering <olaf@xxxxxxxxx>
Signed-off-by: Boris Ostrovsky <boris.ostrovsky@xxxxxxxxxx>
Reviewed-by: David Vrabel <david.vrabel@xxxxxxxxxx>
---
Changes in v2:
* Corrected the commit message

Unfortunately I haven't been able to trigger NUMA balancing
so while I tested this in general I am not sure I actually
exercised the code path.

 drivers/xen/gntdev.c | 15 ++++++++++++++-
 1 file changed, 14 insertions(+), 1 deletion(-)

diff --git a/drivers/xen/gntdev.c b/drivers/xen/gntdev.c
index bb95212..743e6f0 100644
--- a/drivers/xen/gntdev.c
+++ b/drivers/xen/gntdev.c
@@ -35,6 +35,7 @@
 #include <linux/spinlock.h>
 #include <linux/slab.h>
 #include <linux/highmem.h>
+#include <linux/mempolicy.h>
 
 #include <xen/xen.h>
 #include <xen/grant_table.h>
@@ -1007,8 +1008,20 @@ static int gntdev_mmap(struct file *flip, struct vm_area_struct *vma)
 
 	vma->vm_ops = &gntdev_vmops;
 
-	vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP | VM_IO;
+	vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
 
+#ifdef CONFIG_NUMA
+	/* Prevent NUMA balancing */
+	if (vma->vm_policy)
+		vma->vm_policy->flags &= ~(MPOL_F_MOF | MPOL_F_MORON);
+	else {
+		struct mempolicy *pol = get_task_policy(current);
+
+		vma->vm_policy = mpol_dup(pol);
+		if (vma->vm_policy)
+			vma->vm_policy->flags &= ~(MPOL_F_MOF | MPOL_F_MORON);
+	}
+#endif
 	if (use_ptemod)
 		vma->vm_flags |= VM_DONTCOPY;
 
-- 
2.5.5