path: root/docs/misc/grant-tables.txt

********************************************************************************
 A Rough Introduction to Using Grant Tables
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
                                              Christopher Clark, March, 2005.

Grant tables are a mechanism for sharing and transferring frames between
domains, without requiring the participating domains to be privileged.

The first mode of use allows domA to grant domB access to a specific frame,
whilst retaining ownership. The block front driver uses this to grant memory
access to the block back driver, so that it may read or write as requested.

 1. domA creates a grant access reference, and transmits the ref id to domB.
 2. domB uses the reference to map the granted frame.
 3. domB performs the memory access.
 4. domB unmaps the granted frame.
 5. domA removes its grant.


The second mode allows domA to accept a transfer of ownership of a frame from
domB. The net front and back driver will use this for packet tx/rx. This
mechanism is still being implemented, though the xen<->guest interface design
is complete.

 1. domA creates an accept transfer grant reference, and transmits it to domB.
 2. domB uses the ref to hand over a frame it owns.
 3. domA accepts the transfer
 4. domA clears the used reference.


********************************************************************************
 Data structures
 ~~~~~~~~~~~~~~~

 The following data structures are used by Xen and the guests to implement
 grant tables:

 1. Shared grant entries
 2. Active grant entries
 3. Map tracking

 These are not the users primary interface to grant tables, but are discussed
 because an understanding of how they work may be useful. Each of these is a
 finite resource.

 Shared grant entries
 ~~~~~~~~~~~~~~~~~~~~

 A set of pages are shared between Xen and a guest, holding the shared grant
 entries. The guest writes into these entries to create grant references. The
 index of the entry is transmitted to the remote domain: this is the
 reference used to activate an entry. Xen will write into a shared entry to
 indicate to a guest that its grant is in use.
  sha->domid : remote domain being granted rights
  sha->frame : machine frame being granted
  sha->flags : allow access, allow transfer, remote is reading/writing, etc.

 Active grant entries
 ~~~~~~~~~~~~~~~~~~~~

 Xen maintains a set of private frames per domain, holding the active grant
 entries for safety, and to reference count mappings.
  act->domid : remote domain being granted rights
  act->frame : machine frame being granted
  act->pin   : used to hold reference counts

 Map tracking
 ~~~~~~~~~~~~

 Every time a frame is mapped, a map track entry is stored in the metadata of
 the mapping domain. The index of this entry is returned from the map call,
 and is used to unmap the frame. Map track entries are also searched whenever a
 page table entry containing a foreign frame number is overwritten: the first
 matching map track entry is then removed, as if unmap had been invoked.
 These are not used by the transfer mechanism.
  map->domid         : owner of the mapped frame
  map->ref_and_flags : grant reference, ro/rw, mapped for host or device access

********************************************************************************

 Granting a foreign domain access to frames
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

 domA [frame]--> domB


 domA:  #include <asm-xen/gnttab.h>
        grant_ref_t gref[BATCH_SIZE];

        for ( i = 0; i < BATCH_SIZE; i++ )
            gref[i] = gnttab_grant_foreign_access( domBid, mfn, (readonly ? 1 : 0) );


 .. gref is then somehow transmitted to domB for use.


 Mapping foreign frames
 ~~~~~~~~~~~~~~~~~~~~~~

 domB:  #include <asm-xen/hypervisor.h>
        unsigned long       mmap_vstart;
        gnttab_op_t         aop[BATCH_SIZE];
        grant_ref_t         mapped_handle[BATCH_SIZE];

        if ( (mmap_vstart = allocate_empty_lowmem_region(BATCH_SIZE)) == 0 )
            BUG();

        for ( i = 0; i < BATCH_SIZE; i++ )
        {
            aop[i].u.map_grant_ref.host_virt_addr =
                                              mmap_vstart + (i * PAGE_SIZE);
            aop[i].u.map_grant_ref.dom      = domAid;
            aop[i].u.map_grant_ref.ref      = gref[i];
            aop[i].u.map_grant_ref.flags    = ( GNTMAP_host_map | GNTMAP_readonly );
        }

        if ( unlikely(HYPERVISOR_grant_table_op(
                        GNTTABOP_map_grant_ref, aop, BATCH_SIZE)))
            BUG();

        for ( i = 0; i < BATCH_SIZE; i++ )
        {
            if ( unlikely(aop[i].u.map_grant_ref.handle < 0) )
            {
                tidyup_all(aop, i);
                goto panic;
            }

            phys_to_machine_mapping[__pa(mmap_vstart + (i * PAGE_SIZE))>>PAGE_SHIFT] =
                FOREIGN_FRAME(aop[i].u.map_grant_ref.dev_bus_addr);

            mapped_handle[i] = aop[i].u.map_grant_ref.handle;
        }



 Unmapping foreign frames
 ~~~~~~~~~~~~~~~~~~~~~~~~

 domB:
        for ( i = 0; i < BATCH_SIZE; i++ )
        {
            aop[i].u.unmap_grant_ref.host_virt_addr = mmap_vstart + (i * PAGE_SIZE);
            aop[i].u.unmap_grant_ref.dev_bus_addr   = 0;
            aop[i].u.unmap_grant_ref.handle         = mapped_handle[i];
        }
        if ( unlikely(HYPERVISOR_grant_table_op(
                        GNTTABOP_unmap_grant_ref, aop, BATCH_SIZE)))
            BUG();


 Ending foreign access
 ~~~~~~~~~~~~~~~~~~~~~

    Note that this only prevents further mappings; it does _not_ revoke access.
    Should _only_ be used when the remote domain has unmapped the frame.
    gnttab_query_foreign_access( gref ) will indicate the state of any mapping.

 domA:
        if ( gnttab_query_foreign_access( gref[i] ) == 0 )
            gnttab_end_foreign_access( gref[i], readonly );

        TODO: readonly yet to be implemented.


********************************************************************************

 Transferring ownership of a frame to another domain
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

 [ XXX: Transfer mechanism is alpha-calibre code, untested, use at own risk XXX ]
 [ XXX: show use of batch operations below, rather than single frame XXX ]
 [ XXX: linux internal interface could/should be wrapped to be tidier XXX ]


 Prepare to accept a frame from a foreign domain
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

  domA:
        if ( (p = alloc_page(GFP_HIGHUSER)) == NULL )
        {
            printk("Cannot alloc a frame to surrender\n");
            break;
        }
        pfn = p - mem_map;
        mfn = phys_to_machine_mapping[pfn];
                                                                                       
        if ( !PageHighMem(p) )
        {
            v = phys_to_virt(pfn << PAGE_SHIFT);
            scrub_pages(v, 1);
            queue_l1_entry_update(get_ptep((unsigned long)v), 0);
        }
                                                                                       
        /* Ensure that ballooned highmem pages don't have cached mappings. */
        kmap_flush_unused();

        /* Flush updates through and flush the TLB. */
        xen_tlb_flush();
                                                                                       
        phys_to_machine_mapping[pfn] = INVALID_P2M_ENTRY;
                                                                                       
        if ( HYPERVISOR_dom_mem_op(
            MEMOP_decrease_reservation, &mfn, 1, 0) != 1 )
        {
            printk("MEMOP_decrease_reservation failed\n");
            /* er... ok. free the page then */
            __free_page(p);
            break;
        }
                                                                                       
        accepting_pfn = pfn;
        ref = gnttab_grant_foreign_transfer( (domid_t) args.arg[0], pfn );
        printk("Accepting dom %lu frame at ref (%d)\n", args.arg[0], ref);
                                                                                       

 Transfer a frame to a foreign domain
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

  domB:
        mmu_update_t            update;
        domid_t                 domid;
        grant_ref_t             gref;
        unsigned long           pfn, mfn, *v;
        struct page            *transfer_page = 0;
                                                                                       
        /* alloc a page and grant access.
         * alloc page returns a page struct. */
        if ( (transfer_page = alloc_page(GFP_HIGHUSER)) == NULL )
            return -ENOMEM;

        pfn = transfer_page - mem_map;
        mfn = phys_to_machine_mapping[pfn];

        /* need to remove all references to this page */
        if ( !PageHighMem(transfer_page) )
        {
            v = phys_to_virt(pfn << PAGE_SHIFT);
            scrub_pages(v, 1);
            sprintf((char *)v, "This page (%lx) was transferred.\n", mfn);
            queue_l1_entry_update(get_ptep((unsigned long)v), 0);
        }
#ifdef CONFIG_XEN_SCRUB_PAGES
        else
        {
            v = kmap(transfer_page);
            scrub_pages(v, 1);
            sprintf((char *)v, "This page (%lx) was transferred.\n", mfn);
            kunmap(transfer_page);
        }
#endif
        /* Delete any cached kmappings */
        kmap_flush_unused();

        /* Flush updates through and flush the TLB */
        xen_tlb_flush();

        /* invalidate in P2M */
        phys_to_machine_mapping[pfn] = INVALID_P2M_ENTRY;

        domid = (domid_t)args.arg[0];
        gref  = (grant_ref_t)args.arg[1];

        update.ptr  = MMU_EXTENDED_COMMAND;
        update.ptr |= ((gref & 0x00FF) << 2);
        update.ptr |= mfn << PAGE_SHIFT;
                                                                                       
        update.val  = MMUEXT_TRANSFER_PAGE;
        update.val |= (domid << 16);
        update.val |= (gref & 0xFF00);
                                                                                       
        ret = HYPERVISOR_mmu_update(&update, 1, NULL);
                                                                                       

 Map a transferred frame
 ~~~~~~~~~~~~~~~~~~~~~~~

 TODO:


 Clear the used transfer reference
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

 TODO:


********************************************************************************

 Using a private reserve of grant references
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Where it is known in advance how many grant references are required, and
failure to allocate them on demand would cause difficulty, a batch can be
allocated and held in a private reserve.

To reserve a private batch:

    /* housekeeping data - treat as opaque: */
    grant_ref_t gref_head, gref_terminal;

    if ( 0 > gnttab_alloc_grant_references( number_to_reserve,
                                            &gref_head, &gref_terminal ))
        return -ENOSPC;


To release a batch back to the shared pool:

    gnttab_free_grant_references( number_reserved, gref_head );


To claim a reserved reference:

    ref = gnttab_claim_grant_reference( &gref_head, gref_terminal );


To release a claimed reference back to the reserve pool:

    gnttab_release_grant_reference( &gref_head, gref );


To use a claimed reference to grant access, use these alternative functions
that take an additional parameter of the grant reference to use:

    gnttab_grant_foreign_access_ref
    gnttab_grant_foreign_transfer_ref