Hi,
On 14/08/2019 11:27, Paul Durrant wrote:
-----Original Message-----
From: Julien Grall <julien.gr...@arm.com>
Sent: 14 August 2019 11:21
To: Paul Durrant <paul.durr...@citrix.com>; 'Jan Beulich' <jbeul...@suse.com>
Cc: xen-devel@lists.xenproject.org; Andrew Cooper <andrew.coop...@citrix.com>;
Roger Pau Monne
<roger....@citrix.com>; Volodymyr Babchuk <volodymyr_babc...@epam.com>; George
Dunlap
<george.dun...@citrix.com>; Ian Jackson <ian.jack...@citrix.com>; Stefano
Stabellini
<sstabell...@kernel.org>; Konrad Rzeszutek Wilk <konrad.w...@oracle.com>; Tim
(Xen.org) <t...@xen.org>;
Wei Liu <w...@xen.org>
Subject: Re: [PATCH 5/6] iommu: tidy up iommu_us_hap_pt() and
need_iommu_pt_sync() macros
Hi Paul,
On 14/08/2019 11:13, Paul Durrant wrote:
--- a/xen/include/xen/iommu.h
+++ b/xen/include/xen/iommu.h
@@ -268,6 +268,13 @@ struct domain_iommu {
#define iommu_set_feature(d, f) set_bit(f, dom_iommu(d)->features)
#define iommu_clear_feature(d, f) clear_bit(f, dom_iommu(d)->features)
+/* Are we using the domain P2M table as its IOMMU pagetable? */
+#define iommu_use_hap_pt(d) \
+ (hap_enabled(d) && is_iommu_enabled(d) && iommu_hap_pt_share)
Does this build for Arm, seeing that there's no hap_enabled()
definition there? Or have I missed its addition earlier in this
series?
It moved to common code sched.h in an earlier patch.
I went through the series and didn't find where hap_enabled() is defined for Arm
in this series. Do you mind pointing the exact patch?
Sorry, I wasn't clear... The change is in my other series, "use stashed domain
create flags", which is a pre-requisite for this series (as called out in the cover
letter). The change is made in patch #2 of that series:
https://lists.xenproject.org/archives/html/xen-devel/2019-07/msg02256.html.
Oh. I understand this adds benefits as the implementation is now common. But the
downside is hap_enabled() will now require evaluation on Arm even it is
evaluates to true... This will prevent the compiler to remove any non-HAP code
paths (assuming there are any in the common code).
Furthermore, 2 parts of the iommu_use_hap_pt() condition will always returning
always true. But as they are non-constant, so they will always be evaluated.
It is also probably going to confuse developer as they may think non-HAP is
supported on Arm. You can't find easily that both hap_enabled(...) and
iommu_hap_pt_share will always evaluate to true.
So aside the common implementation, what is the real gain for Arm?
Cheers,
--
Julien Grall
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-devel
