[PATCH QEMU HCI-8.1 18/21] VZ: physmem: align fd-backed RAM like anonymous RAM #VSTOR-103860

[Andrey Drobyshev <andrey.drobyshev@virtuozzo.com>]

file_ram_alloc() aligns the mapping of an FD-backed RAM block to 4K for
memfd and tmpfs.  Anon RAM by qemu_anon_ram_alloc() is aligned to
QEMU_VMALLOC_ALIGN instead, which is 2M on x86_64.  Currently, for
FD-backed RAM that's only done on s390x KVM.

Without such alignment, shared memory can't be mapped with 2M
transparent huge pages.  That makes "memory-backend-memfd" guest RAM
backed by 4K pages only, even with 'advice' set to
/sys/kernel/mm/transparent_hugepage/shmem_enabled.

This matters for VMs with VFIO devices and cpr-exec.  For them, anon RAM
pages preserved via MADV_DOEXEC get pinned by VFIO_IOMMU_MAP_DMA and
then replaced by copies on first guest write after exec(), while the
device keeps doing DMA to the old pinned pages.  Therefore we're going
to use "memory-backend-memfd" RAM backend for such VMs, and we don't
want to lose THP.

So align FD-backed RAM blocks to QEMU_VMALLOC_ALIGN on all hosts.
Nothing changes for hugetlbfs-backed RAM, whose page size is at least
2M already.

Measured on a 2G guest with a vfio-pci NIC, memory-backend-memfd with
share=on, and shmem_enabled=advise:

                        before      after
  mapping offset        0x1ff000    0x0
  ShmemPmdMapped        0 kB        2097152 kB

Signed-off-by: Andrey Drobyshev <andrey.drobyshev@virtuozzo.com>
---
 system/physmem.c | 13 ++++++++-----
 1 file changed, 8 insertions(+), 5 deletions(-)

diff --git a/system/physmem.c b/system/physmem.c
index 6f89232d257..ae5aa534c60 100644
--- a/system/physmem.c
+++ b/system/physmem.c
@@ -1476,11 +1476,14 @@ static void *file_ram_alloc(RAMBlock *block,
         return NULL;
     }
     block->mr->align = MAX(block->page_size, block->mr->align);
-#if defined(__s390x__)
-    if (kvm_enabled()) {
-        block->mr->align = MAX(block->mr->align, QEMU_VMALLOC_ALIGN);
-    }
-#endif
+
+    /*
+     * Align like anonymous RAM (qemu_anon_ram_alloc), so that shmem-backed
+     * RAM (memfd, tmpfs files) is eligible for THP.
+     *
+     * On s390x KVM this alignment is required.
+     */
+    block->mr->align = MAX(block->mr->align, QEMU_VMALLOC_ALIGN);
 
     if (memory < block->page_size) {
         error_setg(errp, "memory size 0x" RAM_ADDR_FMT " must be equal to "

-- 
2.47.1