Re: [PATCH for_v23 3/3] x86/sgx: Move reclaim logic out of sgx_free_page()

[Sean Christopherson <sean.j.christopherson@intel.com>]

On Wed, Oct 23, 2019 at 03:42:20PM +0300, Jarkko Sakkinen wrote:
> On Tue, Oct 22, 2019 at 03:49:22PM -0700, Sean Christopherson wrote:
> > Move the reclaim logic out of sgx_free_page() and into a standalone
> > helper to avoid taking sgx_active_page_list_lock when the page is known
> > to be unreclaimable, which is the vast majority of flows that free EPC
> > pages.
> > 
> > Movig reclaim logic to a separate function also eliminates any
> > possibility of silently leaking a page because it is unexpectedly
> > reclaimable (and being actively reclaimed).
> > 
> > Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
> > ---
> > 
> > I really don't like the sgx_unmark_...() name, but couldn't come up with
> > anything better.  Suggestions welcome...
> > 
> >  arch/x86/kernel/cpu/sgx/encl.c    |  3 ++-
> >  arch/x86/kernel/cpu/sgx/main.c    | 32 ++++++++-----------------------
> >  arch/x86/kernel/cpu/sgx/reclaim.c | 32 +++++++++++++++++++++++++++++++
> >  arch/x86/kernel/cpu/sgx/sgx.h     |  3 ++-
> >  4 files changed, 44 insertions(+), 26 deletions(-)
> > 
> > diff --git a/arch/x86/kernel/cpu/sgx/encl.c b/arch/x86/kernel/cpu/sgx/encl.c
> > index 8045f1ddfd62..22186d89042a 100644
> > --- a/arch/x86/kernel/cpu/sgx/encl.c
> > +++ b/arch/x86/kernel/cpu/sgx/encl.c
> > @@ -474,9 +474,10 @@ void sgx_encl_destroy(struct sgx_encl *encl)
> >  			 * The page and its radix tree entry cannot be freed
> >  			 * if the page is being held by the reclaimer.
> >  			 */
> > -			if (sgx_free_page(entry->epc_page))
> > +			if (sgx_unmark_page_reclaimable(entry->epc_page))
> >  				continue;
> >  
> > +			sgx_free_page(entry->epc_page);
> >  			encl->secs_child_cnt--;
> >  			entry->epc_page = NULL;
> >  		}
> > diff --git a/arch/x86/kernel/cpu/sgx/main.c b/arch/x86/kernel/cpu/sgx/main.c
> > index 8e7557d3ff03..cfd8480ef563 100644
> > --- a/arch/x86/kernel/cpu/sgx/main.c
> > +++ b/arch/x86/kernel/cpu/sgx/main.c
> > @@ -108,45 +108,29 @@ struct sgx_epc_page *sgx_alloc_page(void *owner, bool reclaim)
> >   * sgx_free_page() - Free an EPC page
> >   * @page:	pointer a previously allocated EPC page
> >   *
> > - * EREMOVE an EPC page and insert it back to the list of free pages. If the
> > - * page is reclaimable, delete it from the active page list.
> > - *
> > - * Return:
> > - *   0 on success,
> > - *   -EBUSY if a reclaim is in progress
> > + * EREMOVE an EPC page and insert it back to the list of free pages. The page
> > + * must not be reclaimable.
> >   */
> > -int sgx_free_page(struct sgx_epc_page *page)
> > +void sgx_free_page(struct sgx_epc_page *page)
> >  {
> >  	struct sgx_epc_section *section = sgx_epc_section(page);
> >  	int ret;
> >  
> >  	/*
> > -	 * Remove the page from the active list if necessary.  If the page
> > -	 * is actively being reclaimed, i.e. RECLAIMABLE is set but the
> > -	 * page isn't on the active list, return -EBUSY as we can't free
> > -	 * the page at this time since it is "owned" by the reclaimer.
> > +	 * Don't take sgx_active_page_list_lock when asserting the page isn't
> > +	 * reclaimable, missing a WARN in the very rare case is preferable to
> > +	 * unnecessarily taking a global lock in the common case.
> >  	 */
> > -	spin_lock(&sgx_active_page_list_lock);
> > -	if (page->desc & SGX_EPC_PAGE_RECLAIMABLE) {
> > -		if (list_empty(&page->list)) {
> > -			spin_unlock(&sgx_active_page_list_lock);
> > -			return -EBUSY;
> > -		}
> > -		list_del(&page->list);
> > -		page->desc &= ~SGX_EPC_PAGE_RECLAIMABLE;
> > -	}
> > -	spin_unlock(&sgx_active_page_list_lock);
> > +	WARN_ON_ONCE(page->desc & SGX_EPC_PAGE_RECLAIMABLE);
> >  
> >  	ret = __eremove(sgx_epc_addr(page));
> >  	if (WARN_ONCE(ret, "EREMOVE returned %d (0x%x)", ret, ret))
> > -		return -EIO;
> > +		return;
> >  
> >  	spin_lock(&section->lock);
> >  	list_add_tail(&page->list, &section->page_list);
> >  	atomic_inc(&sgx_nr_free_pages);
> >  	spin_unlock(&section->lock);
> > -
> > -	return 0;
> >  }
> >  
> >  static void __init sgx_free_epc_section(struct sgx_epc_section *section)
> > diff --git a/arch/x86/kernel/cpu/sgx/reclaim.c b/arch/x86/kernel/cpu/sgx/reclaim.c
> > index 8067ce1915a4..e64c810883ec 100644
> > --- a/arch/x86/kernel/cpu/sgx/reclaim.c
> > +++ b/arch/x86/kernel/cpu/sgx/reclaim.c
> > @@ -125,6 +125,38 @@ void sgx_mark_page_reclaimable(struct sgx_epc_page *page)
> >  	spin_unlock(&sgx_active_page_list_lock);
> >  }
> >  
> > +/**
> > + * sgx_unmark_page_reclaimable() - Remove a page from the reclaim list
> > + * @page:	EPC page
> > + *
> > + * Clear the reclaimable flag and remove the page from the active page list.
> > + *
> > + * Return:
> > + *   0 on success,
> > + *   -EBUSY if the page is in the process of being reclaimed
> > + */
> > +int sgx_unmark_page_reclaimable(struct sgx_epc_page *page)
> > +{
> > +	/*
> > +	 * Remove the page from the active list if necessary.  If the page
> > +	 * is actively being reclaimed, i.e. RECLAIMABLE is set but the
> > +	 * page isn't on the active list, return -EBUSY as we can't free
> > +	 * the page at this time since it is "owned" by the reclaimer.
> > +	 */
> > +	spin_lock(&sgx_active_page_list_lock);
> > +	if (page->desc & SGX_EPC_PAGE_RECLAIMABLE) {
> > +		if (list_empty(&page->list)) {
> > +			spin_unlock(&sgx_active_page_list_lock);
> > +			return -EBUSY;
> > +		}
> > +		list_del(&page->list);
> > +		page->desc &= ~SGX_EPC_PAGE_RECLAIMABLE;
> > +	}
> > +	spin_unlock(&sgx_active_page_list_lock);
> 
> Would a WARN_ONCE() make sense when SGX_EPC_PAGE_RECLAIMABLE is not set,
> or do we have a legit flow where this can happen?

No, there's a legit case.   sgx_reclaim_pages() clears RECLAIMABLE when
it can't get a reference to the enclave, i.e. the enclave is being
released.  That would incorrectly trigger the WARN when sgx_encl_destroy()
frees all pages as part of sgx_encl_release().