RE: [PATCH v2] xen: support priv-mapping in an HVM tools domain

[Paul Durrant <Paul.Durrant@citrix.com>]

> -----Original Message-----
> From: Juergen Gross [mailto:jgross@suse.com]
> Sent: 01 November 2017 13:40
> To: Paul Durrant <Paul.Durrant@citrix.com>; x86@kernel.org; xen-
> devel@lists.xenproject.org; linux-kernel@vger.kernel.org
> Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>; Thomas Gleixner
> <tglx@linutronix.de>; Ingo Molnar <mingo@redhat.com>; H. Peter Anvin
> <hpa@zytor.com>
> Subject: Re: [PATCH v2] xen: support priv-mapping in an HVM tools domain
> 
> On 01/11/17 12:31, Paul Durrant wrote:
> > If the domain has XENFEAT_auto_translated_physmap then use of the PV-
> > specific HYPERVISOR_mmu_update hypercall is clearly incorrect.
> >
> > This patch adds checks in xen_remap_domain_gfn_array() and
> > xen_unmap_domain_gfn_array() which call through to the approprate
> > xlate_mmu function if the feature is present.
> >
> > This patch also moves xen_remap_domain_gfn_range() into the PV-only
> MMU
> > code and #ifdefs the (only) calling code in privcmd accordingly.
> >
> > Signed-off-by: Paul Durrant <paul.durrant@citrix.com>
> > ---
> > Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
> > Cc: Juergen Gross <jgross@suse.com>
> > Cc: Thomas Gleixner <tglx@linutronix.de>
> > Cc: Ingo Molnar <mingo@redhat.com>
> > Cc: "H. Peter Anvin" <hpa@zytor.com>
> > ---
> >  arch/x86/xen/mmu.c    | 36 +++++++++++++++++-------------------
> >  arch/x86/xen/mmu_pv.c | 11 +++++++++++
> >  drivers/xen/privcmd.c | 17 +++++++++++++----
> >  include/xen/xen-ops.h |  7 +++++++
> >  4 files changed, 48 insertions(+), 23 deletions(-)
> >
> > diff --git a/arch/x86/xen/mmu.c b/arch/x86/xen/mmu.c
> > index 3e15345abfe7..01837c36e293 100644
> > --- a/arch/x86/xen/mmu.c
> > +++ b/arch/x86/xen/mmu.c
> > @@ -91,12 +91,12 @@ static int remap_area_mfn_pte_fn(pte_t *ptep,
> pgtable_t token,
> >  	return 0;
> >  }
> >
> > -static int do_remap_gfn(struct vm_area_struct *vma,
> > -			unsigned long addr,
> > -			xen_pfn_t *gfn, int nr,
> > -			int *err_ptr, pgprot_t prot,
> > -			unsigned domid,
> > -			struct page **pages)
> > +int xen_remap_gfn(struct vm_area_struct *vma,
> > +		  unsigned long addr,
> > +		  xen_pfn_t *gfn, int nr,
> > +		  int *err_ptr, pgprot_t prot,
> > +		  unsigned int domid,
> > +		  struct page **pages)
> >  {
> >  	int err = 0;
> >  	struct remap_data rmd;
> > @@ -166,36 +166,34 @@ static int do_remap_gfn(struct vm_area_struct
> *vma,
> >  	return err < 0 ? err : mapped;
> >  }
> >
> > -int xen_remap_domain_gfn_range(struct vm_area_struct *vma,
> > -			       unsigned long addr,
> > -			       xen_pfn_t gfn, int nr,
> > -			       pgprot_t prot, unsigned domid,
> > -			       struct page **pages)
> > -{
> > -	return do_remap_gfn(vma, addr, &gfn, nr, NULL, prot, domid,
> pages);
> > -}
> > -EXPORT_SYMBOL_GPL(xen_remap_domain_gfn_range);
> > -
> >  int xen_remap_domain_gfn_array(struct vm_area_struct *vma,
> >  			       unsigned long addr,
> >  			       xen_pfn_t *gfn, int nr,
> >  			       int *err_ptr, pgprot_t prot,
> >  			       unsigned domid, struct page **pages)
> >  {
> > +	if (xen_feature(XENFEAT_auto_translated_physmap))
> > +		return xen_xlate_remap_gfn_array(vma, addr, gfn, nr,
> err_ptr,
> > +						 prot, domid, pages);
> > +
> >  	/* We BUG_ON because it's a programmer error to pass a NULL
> err_ptr,
> >  	 * and the consequences later is quite hard to detect what the actual
> >  	 * cause of "wrong memory was mapped in".
> >  	 */
> >  	BUG_ON(err_ptr == NULL);
> > -	return do_remap_gfn(vma, addr, gfn, nr, err_ptr, prot, domid,
> pages);
> > +	return xen_remap_gfn(vma, addr, gfn, nr, err_ptr, prot, domid,
> > +			     pages);
> >  }
> >  EXPORT_SYMBOL_GPL(xen_remap_domain_gfn_array);
> >
> >  /* Returns: 0 success */
> >  int xen_unmap_domain_gfn_range(struct vm_area_struct *vma,
> > -			       int numpgs, struct page **pages)
> > +			       int nr, struct page **pages)
> >  {
> > -	if (!pages || !xen_feature(XENFEAT_auto_translated_physmap))
> > +	if (xen_feature(XENFEAT_auto_translated_physmap))
> > +		return xen_xlate_unmap_gfn_range(vma, nr, pages);
> > +
> > +	if (!pages)
> >  		return 0;
> >
> >  	return -EINVAL;
> > diff --git a/arch/x86/xen/mmu_pv.c b/arch/x86/xen/mmu_pv.c
> > index 71495f1a86d7..4974d8a6c2b4 100644
> > --- a/arch/x86/xen/mmu_pv.c
> > +++ b/arch/x86/xen/mmu_pv.c
> > @@ -2670,3 +2670,14 @@ phys_addr_t paddr_vmcoreinfo_note(void)
> >  		return __pa(vmcoreinfo_note);
> >  }
> >  #endif /* CONFIG_KEXEC_CORE */
> > +
> > +int xen_remap_domain_gfn_range(struct vm_area_struct *vma,
> > +			       unsigned long addr,
> > +			       xen_pfn_t gfn, int nr,
> > +			       pgprot_t prot, unsigned int domid,
> > +			       struct page **pages)
> > +{
> > +	return xen_remap_gfn(vma, addr, &gfn, nr, NULL, prot, domid,
> > +			     pages);
> > +}
> > +EXPORT_SYMBOL_GPL(xen_remap_domain_gfn_range);
> > diff --git a/drivers/xen/privcmd.c b/drivers/xen/privcmd.c
> > index feca75b07fdd..b58a1719b606 100644
> > --- a/drivers/xen/privcmd.c
> > +++ b/drivers/xen/privcmd.c
> > @@ -215,6 +215,8 @@ static int traverse_pages_block(unsigned nelem,
> size_t size,
> >  	return ret;
> >  }
> >
> > +#ifdef CONFIG_XEN_PV
> > +
> >  struct mmap_gfn_state {
> >  	unsigned long va;
> >  	struct vm_area_struct *vma;
> > @@ -261,10 +263,6 @@ static long privcmd_ioctl_mmap(struct file *file,
> void __user *udata)
> >  	LIST_HEAD(pagelist);
> >  	struct mmap_gfn_state state;
> >
> > -	/* We only support privcmd_ioctl_mmap_batch for auto translated.
> */
> > -	if (xen_feature(XENFEAT_auto_translated_physmap))
> > -		return -ENOSYS;
> > -
> 
> CONFIG_XEN_PV doesn't mean we don't support running as HVM guest.
> 
> I don't think you can remove this test, nor do I think it makes sense
> to put privcmd_ioctl_mmap() inside an #ifdef.

Ok, I can leave the test in, but I only put the function inside the ifdef at Boris's request (unless I misinterpreted what he meant).

> 
> You should rather add something like privcmd_ioctl_mmap_pv() with all
> the PV-specific functionality and call this here. For !CONFIG_XEN_PV
> it can e.g. just BUG().
> 

Personally, I prefer my original patch where all the changes were kept inside mmu.c. It was much neater.

  Paul

> 
> Juergen