Xen devel - Mar 2012 - [PATCH] x86/mm/paging: Prevent the guest from faulting on the same gfn when dying

xen/arch/x86/mm/p2m.c |  2 ++
 1 files changed, 2 insertions(+), 0 deletions(-)


A crashing guest due to paging may hit an "endless" loop faulting
repeatedly on
the current paged-out gfn, until the toolstack comes around to killing the
domain.

Unfortunately domain_crash at this point may not pause the vcpu due to the
shutting_down flag, so we pause it explicitly when detecting the condition.

Signed-off-by: Andres Lagar-Cavilla <andres@lagarcavilla.org>

diff -r dfbcb092aa66 -r 0fe4b3326167 xen/arch/x86/mm/p2m.c
--- a/xen/arch/x86/mm/p2m.c
+++ b/xen/arch/x86/mm/p2m.c
@@ -988,6 +988,8 @@ void p2m_mem_paging_populate(struct doma
     {
         gdprintk(XENLOG_ERR, "Domain %hu paging gfn %lx yet no ring "
                              "in place\n", d->domain_id, gfn);
+        /* Prevent the vcpu from faulting repeatedly on the same gfn */
+        vcpu_pause_nosync(v);
         domain_crash(d);
         return;
     }

xen/arch/x86/mm/p2m.c |  3 +++
 1 files changed, 3 insertions(+), 0 deletions(-)


Please ignore previous post from a minute ago. Use this instead.

A crashing guest due to paging may hit an "endless" loop faulting
repeatedly on
the current paged-out gfn, until the toolstack comes around to killing the
domain.

Unfortunately domain_crash at this point may not pause the vcpu due to the
shutting_down flag, so we pause it explicitly when detecting the condition.

Signed-off-by: Andres Lagar-Cavilla <andres@lagarcavilla.org>

diff -r dfbcb092aa66 -r 88584a17d2da xen/arch/x86/mm/p2m.c
--- a/xen/arch/x86/mm/p2m.c
+++ b/xen/arch/x86/mm/p2m.c
@@ -988,6 +988,9 @@ void p2m_mem_paging_populate(struct doma
     {
         gdprintk(XENLOG_ERR, "Domain %hu paging gfn %lx yet no ring "
                              "in place\n", d->domain_id, gfn);
+        /* Prevent the vcpu from faulting repeatedly on the same gfn */
+        if ( v->domain == d )
+            vcpu_pause_nosync(v);
         domain_crash(d);
         return;
     }

At 11:42 -0400 on 12 Mar (1331552563), Andres Lagar-Cavilla
wrote:
> A crashing guest due to paging may hit an "endless" loop faulting
repeatedly on
> the current paged-out gfn, until the toolstack comes around to killing the
> domain.
> 
> Unfortunately domain_crash at this point may not pause the vcpu due to the
> shutting_down flag, so we pause it explicitly when detecting the condition.

Applied, thanks.

Tim.