When setting up the DAX window during the virtiofs driver probe inside the guest, the Linux arch-specific function arch_add_memory is called, which on ppc tries to do a cache flush [1] of the recently added memory. At this point the window is mmap'ed PROT_NONE by QEMU, which causes the following:
<snip> [ 10.136703] virtio_fs virtio0: Cache len: 0x80000000 @ 0x210000000000 [ 10.165106] radix-mmu: Mapped 0xc000210000000000-0xc000210080000000 with 1.00 GiB pages error: kvm run failed Bad address NIP c000000000072350 LR c000000000072304 CTR 0000000001000000 XER 0000000020040000 CPU#0 MSR 8000000002009033 HID0 0000000000000000 HF 8000000000000000 iidx 3 didx 3 TB 00000000 00000000 DECR 0 GPR00 c000000000072304 c0000000fa383100 c000000001190300 0000000000000000 GPR04 0000000000000001 0000000000000000 c0000000fa383208 0000000000000080 GPR08 c000210000000000 000000008000007f 0000000001000000 6874697720303030 <snip> The problem is the same for the memory device removal path (e.g. during filesystem unmount). Since powerpc expects the memory to be accessible during device addition/removal, this patch makes the DAX window readable at creation and after virtio-fs unmap. 1 - https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=fb5924fd Signed-off-by: Fabiano Rosas <faro...@linux.ibm.com> --- hw/virtio/vhost-user-fs.c | 17 ++++++++++++++--- 1 file changed, 14 insertions(+), 3 deletions(-) diff --git a/hw/virtio/vhost-user-fs.c b/hw/virtio/vhost-user-fs.c index 455e97beea..92958797d0 100644 --- a/hw/virtio/vhost-user-fs.c +++ b/hw/virtio/vhost-user-fs.c @@ -24,6 +24,16 @@ #include "exec/address-spaces.h" #include "trace.h" +/* + * The powerpc kernel code expects the memory to be accessible during + * addition/removal. + */ +#if defined(TARGET_PPC64) && defined(CONFIG_LINUX) +#define DAX_WINDOW_PROT PROT_READ +#else +#define DAX_WINDOW_PROT PROT_NONE +#endif + uint64_t vhost_user_fs_slave_map(struct vhost_dev *dev, VhostUserFSSlaveMsg *sm, int fd) { @@ -133,8 +143,8 @@ uint64_t vhost_user_fs_slave_unmap(struct vhost_dev *dev, VhostUserFSSlaveMsg *s continue; } - ptr = mmap(cache_host + sm->c_offset[i], sm->len[i], - PROT_NONE, MAP_PRIVATE | MAP_ANONYMOUS | MAP_FIXED, -1, 0); + ptr = mmap(cache_host + sm->c_offset[i], sm->len[i], DAX_WINDOW_PROT, + MAP_PRIVATE | MAP_ANONYMOUS | MAP_FIXED, -1, 0); if (ptr != (cache_host + sm->c_offset[i])) { res = -errno; fprintf(stderr, "%s: mmap failed (%s) [%d] %" @@ -485,8 +495,9 @@ static void vuf_device_realize(DeviceState *dev, Error **errp) if (fs->conf.cache_size) { /* Anonymous, private memory is not counted as overcommit */ - cache_ptr = mmap(NULL, fs->conf.cache_size, PROT_NONE, + cache_ptr = mmap(NULL, fs->conf.cache_size, DAX_WINDOW_PROT, MAP_ANONYMOUS | MAP_PRIVATE, -1, 0); + if (cache_ptr == MAP_FAILED) { error_setg(errp, "Unable to mmap blank cache"); return; -- 2.23.0
