|
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] Re: [Xen-devel] [PATCH] iommu / p2m: add a page_order parameter to iommu_map/unmap_page()
> -----Original Message-----
> From: Andrew Cooper
> Sent: 17 October 2018 12:20
> To: Paul Durrant <Paul.Durrant@xxxxxxxxxx>; xen-devel@xxxxxxxxxxxxxxxxxxxx
> Cc: Jan Beulich <jbeulich@xxxxxxxx>; Wei Liu <wei.liu2@xxxxxxxxxx>; George
> Dunlap <George.Dunlap@xxxxxxxxxx>; Ian Jackson <Ian.Jackson@xxxxxxxxxx>;
> Julien Grall <julien.grall@xxxxxxx>; Konrad Rzeszutek Wilk
> <konrad.wilk@xxxxxxxxxx>; Stefano Stabellini <sstabellini@xxxxxxxxxx>; Tim
> (Xen.org) <tim@xxxxxxx>; Jun Nakajima <jun.nakajima@xxxxxxxxx>; Kevin Tian
> <kevin.tian@xxxxxxxxx>
> Subject: Re: [PATCH] iommu / p2m: add a page_order parameter to
> iommu_map/unmap_page()
>
> On 17/10/18 09:19, Paul Durrant wrote:
> > diff --git a/xen/arch/x86/mm/p2m-pt.c b/xen/arch/x86/mm/p2m-pt.c
> > index 55df18501e..b264a97bd8 100644
> > --- a/xen/arch/x86/mm/p2m-pt.c
> > +++ b/xen/arch/x86/mm/p2m-pt.c
> > @@ -683,41 +684,13 @@ p2m_pt_set_entry(struct p2m_domain *p2m, gfn_t
> gfn_, mfn_t mfn,
> > {
> > ASSERT(rc == 0);
> >
> > - if ( iommu_use_hap_pt(p2m->domain) )
> > - {
> > - if ( iommu_old_flags )
> > - amd_iommu_flush_pages(p2m->domain, gfn, page_order);
> > - }
> > - else if ( need_iommu_pt_sync(p2m->domain) )
> > - {
> > - dfn_t dfn = _dfn(gfn);
> > -
> > - if ( iommu_pte_flags )
> > - for ( i = 0; i < (1UL << page_order); i++ )
> > - {
> > - rc = iommu_map_page(p2m->domain, dfn_add(dfn, i),
> > - mfn_add(mfn, i),
> iommu_pte_flags);
> > - if ( unlikely(rc) )
> > - {
> > - while ( i-- )
> > - /* If statement to satisfy __must_check. */
> > - if ( iommu_unmap_page(p2m->domain,
> > - dfn_add(dfn, i)) )
> > - continue;
> > -
> > - break;
> > - }
> > - }
> > - else
> > - for ( i = 0; i < (1UL << page_order); i++ )
> > - {
> > - int ret = iommu_unmap_page(p2m->domain,
> > - dfn_add(dfn, i));
> > -
> > - if ( !rc )
> > - rc = ret;
> > - }
> > - }
> > + if ( need_iommu_pt_sync(p2m->domain) )
> > + rc = iommu_pte_flags ?
> > + iommu_map_page(d, _dfn(gfn), mfn, page_order,
> > + iommu_pte_flags) :
> > + iommu_unmap_page(d, _dfn(gfn), page_order);
> > + else if ( iommu_use_hap_pt(d) && iommu_old_flags )
> > + amd_iommu_flush_pages(p2m->domain, gfn, page_order);
>
> This logically reverses the
> iommu_use_hap_pt(d)/need_iommu_pt_sync(p2m->domain) conditions.
Yes it does, but I think this I ok as they will never both be true at the same
time. Doing it this way allowed me to get rid of the nested if.
>
> I'd be tempted confine this change to the else if (
> need_iommu_pt_sync(p2m->domain) ) block.
>
>
> Tangentially related, calling amd_iommu_flush_pages() is a laying
> violation here because this is supposedly common code. In reality, it
> is the NPT code, so might perhaps be better named as p2m-npt.c. George?
>
The boilerplate says:
/******************************************************************************
* arch/x86/mm/p2m-pt.c
*
* Implementation of p2m datastructures as pagetables, for use by
* NPT and shadow-pagetable code
*
so calling AMD IOMMU functions is not really a layering violation.
> > }
> >
> > /*
> > diff --git a/xen/arch/x86/mm/p2m.c b/xen/arch/x86/mm/p2m.c
> > index f1df1debc7..3fa559da01 100644
> > --- a/xen/arch/x86/mm/p2m.c
> > +++ b/xen/arch/x86/mm/p2m.c
> > @@ -718,24 +718,8 @@ p2m_remove_page(struct p2m_domain *p2m, unsigned
> long gfn_l, unsigned long mfn,
> > p2m_access_t a;
> >
> > if ( !paging_mode_translate(p2m->domain) )
> > - {
> > - int rc = 0;
> > -
> > - if ( need_iommu_pt_sync(p2m->domain) )
> > - {
> > - dfn_t dfn = _dfn(mfn);
> > -
> > - for ( i = 0; i < (1 << page_order); i++ )
> > - {
> > - int ret = iommu_unmap_page(p2m->domain, dfn_add(dfn,
> i));
> > -
> > - if ( !rc )
> > - rc = ret;
> > - }
> > - }
> > -
> > - return rc;
> > - }
> > + return need_iommu_pt_sync(p2m->domain) ?
> > + iommu_unmap_page(p2m->domain, _dfn(mfn), page_order) : 0;
>
> TBH, I think this is harder to read than the non ternary alternative.
>
Ok. I'm not fussed either way.
> >
> > ASSERT(gfn_locked_by_me(p2m, gfn));
> > P2M_DEBUG("removing gfn=%#lx mfn=%#lx\n", gfn_l, mfn);
> > diff --git a/xen/drivers/passthrough/iommu.c
> b/xen/drivers/passthrough/iommu.c
> > index 8b438ae4bc..40db9e7849 100644
> > --- a/xen/drivers/passthrough/iommu.c
> > +++ b/xen/drivers/passthrough/iommu.c
> > @@ -305,50 +305,71 @@ void iommu_domain_destroy(struct domain *d)
> > }
> >
> > int iommu_map_page(struct domain *d, dfn_t dfn, mfn_t mfn,
> > - unsigned int flags)
> > + unsigned int page_order, unsigned int flags)
> > {
> > const struct domain_iommu *hd = dom_iommu(d);
> > - int rc;
> > + unsigned long i;
> >
> > if ( !iommu_enabled || !hd->platform_ops )
> > return 0;
> >
> > - rc = hd->platform_ops->map_page(d, dfn, mfn, flags);
> > - if ( unlikely(rc) )
> > + for ( i = 0; i < (1ul << page_order); i++ )
> > {
> > - if ( !d->is_shutting_down && printk_ratelimit() )
> > - printk(XENLOG_ERR
> > - "d%d: IOMMU mapping dfn %"PRI_dfn" to mfn %"PRI_mfn"
> failed: %d\n",
> > - d->domain_id, dfn_x(dfn), mfn_x(mfn), rc);
> > + int ignored, rc = hd->platform_ops->map_page(d, dfn_add(dfn,
> i),
> > + mfn_add(mfn, i),
> > + flags);
> >
> > - if ( !is_hardware_domain(d) )
> > - domain_crash(d);
> > + if ( unlikely(rc) )
> > + {
> > + while (i--)
>
> Spaces, but you're also off-by-one when cleaning up i = 0. Wouldn't it
> be easier to reuse iommu_unmap_page() rather than opencode it?
>
I don't think this is off by one. When unmap_page() is called then i will have
been decremented, so the last iteration of the loop will call unmap_page() with
the base dfn. Calling iommu_unmap_page() comes with baggage I'd rather avoid,
such as the possibility of the domain crash occurring there instead. I'll fix
the indent.
Paul
> ~Andrew
_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel
|
 |
Lists.xenproject.org is hosted with RackSpace, monitoring our |