From mboxrd@z Thu Jan 1 00:00:00 1970 From: George Dunlap Subject: Re: [PATCH] pvh: change epte_get_entry_emt() for pvh mem types Date: Fri, 22 Nov 2013 11:03:58 +0000 Message-ID: <528F3A1E.4010301@eu.citrix.com> References: <20131121170051.424971fd@mantra.us.oracle.com> <528F3765.8080806@eu.citrix.com> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii"; Format="flowed" Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <528F3765.8080806@eu.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: Mukesh Rathor , "Xen-devel@lists.xensource.com" Cc: Keir Fraser , eddie.dong@intel.com, "Nakajima, Jun" , Jan Beulich , Roger Pau Monne List-Id: xen-devel@lists.xenproject.org On 22/11/13 10:52, George Dunlap wrote: > On 22/11/13 01:00, Mukesh Rathor wrote: >> For pvh guests, epte_get_entry_emt() is incorrectly returning WB for >> all mem types because of the following check: >> if ( !v->domain->arch.hvm_domain.params[HVM_PARAM_IDENT_PT] ) >> Skip the check for pvh guests. >> >> Also note, MTRR ranges are not maintained for pvh, and a solution is >> being contrived using PAT. >> >> Signed-off-by: Mukesh Rathor >> --- >> xen/arch/x86/hvm/mtrr.c | 7 ++++++- >> 1 files changed, 6 insertions(+), 1 deletions(-) >> >> diff --git a/xen/arch/x86/hvm/mtrr.c b/xen/arch/x86/hvm/mtrr.c >> index 4ff1e55..5427e1c 100644 >> --- a/xen/arch/x86/hvm/mtrr.c >> +++ b/xen/arch/x86/hvm/mtrr.c >> @@ -693,7 +693,8 @@ uint8_t epte_get_entry_emt(struct domain *d, >> unsigned long gfn, mfn_t mfn, >> ((d->vcpu == NULL) || ((v = d->vcpu[0]) == NULL)) ) >> return MTRR_TYPE_WRBACK; >> - if ( !v->domain->arch.hvm_domain.params[HVM_PARAM_IDENT_PT] ) >> + if ( !is_pvh_vcpu(v) && >> + !v->domain->arch.hvm_domain.params[HVM_PARAM_IDENT_PT] ) >> return MTRR_TYPE_WRBACK; >> if ( !mfn_valid(mfn_x(mfn)) ) >> @@ -717,6 +718,10 @@ uint8_t epte_get_entry_emt(struct domain *d, >> unsigned long gfn, mfn_t mfn, >> return MTRR_TYPE_WRBACK; >> } >> + /* MTRR ranges are not maintained for pvh. */ >> + if ( is_pvh_vcpu(v) ) >> + return MTRR_TYPE_WRBACK; >> + >> gmtrr_mtype = get_mtrr_type(&v->arch.hvm_vcpu.mtrr, (gfn << >> PAGE_SHIFT)); >> hmtrr_mtype = get_mtrr_type(&mtrr_state, (mfn_x(mfn) << >> PAGE_SHIFT)); >> return ((gmtrr_mtype <= hmtrr_mtype) ? gmtrr_mtype : hmtrr_mtype); > > So this will bypass the host mtrr settings, and always return WRBACK, > even if in mtrr_state it was set to something lower. Presumably that > "min(host,guest)" was there for a reason. Are you sure it's safe to > just ignore it? > > This is why I suggested the following instead: > > gmtrr_mtype = is_hvm_domain(v) ? > get_mtrr_type(&v->arch.hvm_vcpu.mtrr, (gfn << PAGE_SHIFT)) : > MTRR_TYPE_WRBACK; Also, while I suppose this is fine for now (since we're starting the code freeze), for the future: is it actually better to "bake in" that MTRRs are completely disabled and always effectively return WRBACK (except for direct_mmio memory), or would it be better to initialize the virtual MTRRs to WRBACK (the way a BIOS would), and allow the guest to change them if it wants? -George