From mboxrd@z Thu Jan 1 00:00:00 1970 From: Julien Grall Subject: Re: [PATCH v2] xen: arm: avoid reusing incorrect mappings when walking the p2m. Date: Fri, 18 Jul 2014 21:29:29 +0100 Message-ID: <53C983A9.5030103@linaro.org> References: <1405690440-32211-1-git-send-email-ian.campbell@citrix.com> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii"; Format="flowed" Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <1405690440-32211-1-git-send-email-ian.campbell@citrix.com> List-Unsubscribe: , List-Post: List-Help: List-Subscribe: , Sender: xen-devel-bounces@lists.xen.org Errors-To: xen-devel-bounces@lists.xen.org To: Ian Campbell , xen-devel@lists.xen.org Cc: tim@xen.org, stefano.stabellini@eu.citrix.com List-Id: xen-devel@lists.xenproject.org Hi Ian, On 18/07/14 14:33, Ian Campbell wrote: > When we change which PT page we are mapping at a given level then we need to > invalidate any cached mappings further down the tree, otherwise we risk using > them because their offset might match but be based on a different offset > further up the table. > > e.g. when remapping first then cur_first_offset and cur_second_offset (which > indicate the currently mapped second and third tables respectively) both become > invalid > > Signed-off-by: Ian Campbell Acked-by: Julien Grall Regards, > --- > v2: Also invalidate cur_first_offset when changing cur_first_page. > Corrected/clarified commit message, this issue isn't really specific to > superpages, they just happen to expose it. > --- > xen/arch/arm/p2m.c | 4 ++++ > 1 file changed, 4 insertions(+) > > diff --git a/xen/arch/arm/p2m.c b/xen/arch/arm/p2m.c > index 2f855b5..8ffddac 100644 > --- a/xen/arch/arm/p2m.c > +++ b/xen/arch/arm/p2m.c > @@ -743,6 +743,8 @@ static int apply_p2m_changes(struct domain *d, > goto out; > } > cur_first_page = p2m_first_level_index(addr); > + /* Any mapping further down is now invalid */ > + cur_first_offset = cur_second_offset = ~0; > } > > /* We only use a 3 level p2m at the moment, so no level 0, > @@ -765,6 +767,8 @@ static int apply_p2m_changes(struct domain *d, > if (second) unmap_domain_page(second); > second = map_domain_page(first[first_table_offset(addr)].p2m.base); > cur_first_offset = first_table_offset(addr); > + /* Any mapping further down is now invalid */ > + cur_second_offset = ~0; > } > /* else: second already valid */ > > -- Julien Grall