[PATCH v8 2/8] kvmppc: Movement of pages between normal and secure memory

[Bharata B Rao]

On Wed, Sep 18, 2019 at 12:42:10PM +0530, Bharata B Rao wrote:
> On Tue, Sep 17, 2019 at 04:31:39PM -0700, Sukadev Bhattiprolu wrote:
> > 
> > Minor: Can this allocation be outside the lock? I guess it would change
> > the order of cleanup at the end of the function.
> 
> Cleanup has bitmap_clear which needs be under spinlock, so this order
> of setup/alloc and cleanup will keep things simple is what I felt.
> 
> > 
> > > +	spin_unlock(&kvmppc_uvmem_pfn_lock);
> > > +
> > > +	*rmap = uvmem_pfn | KVMPPC_RMAP_UVMEM_PFN;
> > > +	pvt->rmap = rmap;
> > > +	pvt->gpa = gpa;
> > > +	pvt->lpid = lpid;
> > > +	dpage->zone_device_data = pvt;
> > > +
> > > +	get_page(dpage);
> > > +	return dpage;
> > > +
> > > +out_unlock:
> > > +	unlock_page(dpage);
> > > +out_clear:
> > > +	bitmap_clear(kvmppc_uvmem_pfn_bitmap, uvmem_pfn - pfn_first, 1);
> > 
> > Reuse variable 'bit'  here?
> 
> Sure.
> 
> > 
> > > +out:
> > > +	spin_unlock(&kvmppc_uvmem_pfn_lock);
> > > +	return NULL;
> > > +}
> > > +
> > > +/*
> > > + * Alloc a PFN from private device memory pool and copy page from normal
> > > + * memory to secure memory using UV_PAGE_IN uvcall.
> > > + */
> > > +static int
> > > +kvmppc_svm_page_in(struct vm_area_struct *vma, unsigned long start,
> > > +		   unsigned long end, unsigned long *rmap,
> > > +		   unsigned long gpa, unsigned int lpid,
> > > +		   unsigned long page_shift)
> > > +{
> > > +	unsigned long src_pfn, dst_pfn = 0;
> > > +	struct migrate_vma mig;
> > > +	struct page *spage;
> > > +	unsigned long pfn;
> > > +	struct page *dpage;
> > > +	int ret = 0;
> > > +
> > > +	memset(&mig, 0, sizeof(mig));
> > > +	mig.vma = vma;
> > > +	mig.start = start;
> > > +	mig.end = end;
> > > +	mig.src = &src_pfn;
> > > +	mig.dst = &dst_pfn;
> > > +
> > > +	ret = migrate_vma_setup(&mig);
> > > +	if (ret)
> > > +		return ret;
> > > +
> > > +	spage = migrate_pfn_to_page(*mig.src);
> > > +	pfn = *mig.src >> MIGRATE_PFN_SHIFT;
> > > +	if (!spage || !(*mig.src & MIGRATE_PFN_MIGRATE)) {
> > > +		ret = 0;
> > 
> > Do we want to return success here (and have caller return H_SUCCESS) if
> > we can't find the source page?
> 
> spage is NULL for zero page. In this case we return success but there is
> no UV_PAGE_IN involved.
> 
> Absence of MIGRATE_PFN_MIGRATE indicates that the requested page
> can't be migrated. I haven't hit this case till now. Similar check
> is also present in the nouveau driver. I am not sure if this is strictly
> needed here.
> 
> Christoph, Jason - do you know if !(*mig.src & MIGRATE_PFN_MIGRATE)
> check is required and if so in which cases will it be true?

Looks like the currently existing check

if (!spage || !(*mig.src & MIGRATE_PFN_MIGRATE)) {
	ret = 0;
	goto out_finalize;
}

will prevent both

1. Zero pages and
2. Pages for which no page table entries exist

from getting migrated to secure (device) side. In both the above cases
!spage is true (and MIGRATE_PFN_MIGRATE is set). In both cases
we needn't copy the page, but migration should complete.

Guess the following comment extract from migrate_vma_setup() is talking about
Case 2 above.

 * For empty entries inside CPU page table (pte_none() or pmd_none() is true) we
 * do set MIGRATE_PFN_MIGRATE flag inside the corresponding source array thus
 * allowing the caller to allocate device memory for those unback virtual
 * address.  For this the caller simply has to allocate device memory and
 * properly set the destination entry like for regular migration.  Note that

Is the above understanding correct? Will the same apply to nouveau driver too?
nouveau_dmem_migrate_copy_one() also seems to bail out after a similar
check.

Regards,
Bharata.