Re: [PATCH v2 07/18] IOMMU/x86: perform PV Dom0 mappings in batches

["Roger Pau Monné" <roger.pau@citrix.com>]

On Fri, Sep 24, 2021 at 11:47:41AM +0200, Jan Beulich wrote:
> For large page mappings to be easily usable (i.e. in particular without
> un-shattering of smaller page mappings) and for mapping operations to
> then also be more efficient, pass batches of Dom0 memory to iommu_map().
> In dom0_construct_pv() and its helpers (covering strict mode) this
> additionally requires establishing the type of those pages (albeit with
> zero type references).
> 
> The earlier establishing of PGT_writable_page | PGT_validated requires
> the existing places where this gets done (through get_page_and_type())
> to be updated: For pages which actually have a mapping, the type
> refcount needs to be 1.
> 
> There is actually a related bug that gets fixed here as a side effect:
> Typically the last L1 table would get marked as such only after
> get_page_and_type(..., PGT_writable_page). While this is fine as far as
> refcounting goes, the page did remain mapped in the IOMMU in this case
> (when "iommu=dom0-strict").
> 
> Signed-off-by: Jan Beulich <jbeulich@suse.com>
> ---
> Subsequently p2m_add_identity_entry() may want to also gain an order
> parameter, for arch_iommu_hwdom_init() to use. While this only affects
> non-RAM regions, systems typically have 2-16Mb of reserved space
> immediately below 4Gb, which hence could be mapped more efficiently.
> 
> The installing of zero-ref writable types has in fact shown (observed
> while putting together the change) that despite the intention by the
> XSA-288 changes (affecting DomU-s only) for Dom0 a number of
> sufficiently ordinary pages (at the very least initrd and P2M ones as
> well as pages that are part of the initial allocation but not part of
> the initial mapping) still have been starting out as PGT_none, meaning
> that they would have gained IOMMU mappings only the first time these
> pages would get mapped writably.
> 
> I didn't think I need to address the bug mentioned in the description in
> a separate (prereq) patch, but if others disagree I could certainly
> break out that part (needing to first use iommu_legacy_unmap() then).
> 
> Note that 4k P2M pages don't get (pre-)mapped in setup_pv_physmap():
> They'll end up mapped via the later get_page_and_type().
> 
> As to the way these refs get installed: I've chosen to avoid the more
> expensive {get,put}_page_and_type(), putting in place the intended type
> directly. I guess I could be convinced to avoid this bypassing of the
> actual logic; I merely think it's unnecessarily expensive.
> 
> --- a/xen/arch/x86/pv/dom0_build.c
> +++ b/xen/arch/x86/pv/dom0_build.c
> @@ -106,11 +106,26 @@ static __init void mark_pv_pt_pages_rdon
>      unmap_domain_page(pl3e);
>  }
>  
> +/*
> + * For IOMMU mappings done while building Dom0 the type of the pages needs to
> + * match (for _get_page_type() to unmap upon type change). Set the pages to
> + * writable with no type ref. NB: This is benign when !need_iommu_pt_sync(d).
> + */
> +static void __init make_pages_writable(struct page_info *page, unsigned long nr)
> +{
> +    for ( ; nr--; ++page )
> +    {
> +        ASSERT(!page->u.inuse.type_info);
> +        page->u.inuse.type_info = PGT_writable_page | PGT_validated;
> +    }
> +}
> +
>  static __init void setup_pv_physmap(struct domain *d, unsigned long pgtbl_pfn,
>                                      unsigned long v_start, unsigned long v_end,
>                                      unsigned long vphysmap_start,
>                                      unsigned long vphysmap_end,
> -                                    unsigned long nr_pages)
> +                                    unsigned long nr_pages,
> +                                    unsigned int *flush_flags)
>  {
>      struct page_info *page = NULL;
>      l4_pgentry_t *pl4e, *l4start = map_domain_page(_mfn(pgtbl_pfn));
> @@ -123,6 +138,8 @@ static __init void setup_pv_physmap(stru
>  
>      while ( vphysmap_start < vphysmap_end )
>      {
> +        int rc = 0;
> +
>          if ( domain_tot_pages(d) +
>               ((round_pgup(vphysmap_end) - vphysmap_start) >> PAGE_SHIFT) +
>               3 > nr_pages )
> @@ -176,7 +193,22 @@ static __init void setup_pv_physmap(stru
>                                               L3_PAGETABLE_SHIFT - PAGE_SHIFT,
>                                               MEMF_no_scrub)) != NULL )
>              {
> -                *pl3e = l3e_from_page(page, L1_PROT|_PAGE_DIRTY|_PAGE_PSE);
> +                mfn_t mfn = page_to_mfn(page);
> +
> +                if ( need_iommu_pt_sync(d) )
> +                    rc = iommu_map(d, _dfn(mfn_x(mfn)), mfn,
> +                                   SUPERPAGE_PAGES * SUPERPAGE_PAGES,
> +                                   IOMMUF_readable | IOMMUF_writable,
> +                                   flush_flags);
> +                if ( !rc )
> +                    make_pages_writable(page,
> +                                        SUPERPAGE_PAGES * SUPERPAGE_PAGES);
> +                else
> +                    printk(XENLOG_ERR
> +                           "pre-mapping P2M 1G-MFN %lx into IOMMU failed: %d\n",
> +                           mfn_x(mfn), rc);
> +
> +                *pl3e = l3e_from_mfn(mfn, L1_PROT|_PAGE_DIRTY|_PAGE_PSE);
>                  vphysmap_start += 1UL << L3_PAGETABLE_SHIFT;
>                  continue;
>              }
> @@ -202,7 +234,20 @@ static __init void setup_pv_physmap(stru
>                                               L2_PAGETABLE_SHIFT - PAGE_SHIFT,
>                                               MEMF_no_scrub)) != NULL )
>              {
> -                *pl2e = l2e_from_page(page, L1_PROT|_PAGE_DIRTY|_PAGE_PSE);
> +                mfn_t mfn = page_to_mfn(page);
> +
> +                if ( need_iommu_pt_sync(d) )
> +                    rc = iommu_map(d, _dfn(mfn_x(mfn)), mfn, SUPERPAGE_PAGES,
> +                                   IOMMUF_readable | IOMMUF_writable,
> +                                   flush_flags);
> +                if ( !rc )
> +                    make_pages_writable(page, SUPERPAGE_PAGES);
> +                else
> +                    printk(XENLOG_ERR
> +                           "pre-mapping P2M 2M-MFN %lx into IOMMU failed: %d\n",
> +                           mfn_x(mfn), rc);
> +
> +                *pl2e = l2e_from_mfn(mfn, L1_PROT|_PAGE_DIRTY|_PAGE_PSE);
>                  vphysmap_start += 1UL << L2_PAGETABLE_SHIFT;
>                  continue;
>              }
> @@ -310,6 +355,7 @@ int __init dom0_construct_pv(struct doma
>      unsigned long initrd_pfn = -1, initrd_mfn = 0;
>      unsigned long count;
>      struct page_info *page = NULL;
> +    unsigned int flush_flags = 0;
>      start_info_t *si;
>      struct vcpu *v = d->vcpu[0];
>      void *image_base = bootstrap_map(image);
> @@ -572,6 +618,18 @@ int __init dom0_construct_pv(struct doma
>                      BUG();
>          }
>          initrd->mod_end = 0;
> +
> +        count = PFN_UP(initrd_len);
> +
> +        if ( need_iommu_pt_sync(d) )
> +            rc = iommu_map(d, _dfn(initrd_mfn), _mfn(initrd_mfn), count,
> +                           IOMMUF_readable | IOMMUF_writable, &flush_flags);
> +        if ( !rc )
> +            make_pages_writable(mfn_to_page(_mfn(initrd_mfn)), count);
> +        else
> +            printk(XENLOG_ERR
> +                   "pre-mapping initrd (MFN %lx) into IOMMU failed: %d\n",
> +                   initrd_mfn, rc);
>      }
>  
>      printk("PHYSICAL MEMORY ARRANGEMENT:\n"
> @@ -605,6 +663,22 @@ int __init dom0_construct_pv(struct doma
>  
>      process_pending_softirqs();
>  
> +    /*
> +     * We map the full range here and then punch a hole for page tables via
> +     * iommu_unmap() further down, once they have got marked as such.
> +     */
> +    if ( need_iommu_pt_sync(d) )
> +        rc = iommu_map(d, _dfn(alloc_spfn), _mfn(alloc_spfn),
> +                       alloc_epfn - alloc_spfn,
> +                       IOMMUF_readable | IOMMUF_writable, &flush_flags);
> +    if ( !rc )
> +        make_pages_writable(mfn_to_page(_mfn(alloc_spfn)),
> +                            alloc_epfn - alloc_spfn);
> +    else
> +        printk(XENLOG_ERR
> +               "pre-mapping MFNs [%lx,%lx) into IOMMU failed: %d\n",
> +               alloc_spfn, alloc_epfn, rc);
> +
>      mpt_alloc = (vpt_start - v_start) + pfn_to_paddr(alloc_spfn);
>      if ( vinitrd_start )
>          mpt_alloc -= PAGE_ALIGN(initrd_len);
> @@ -689,7 +763,8 @@ int __init dom0_construct_pv(struct doma
>          l1tab++;
>  
>          page = mfn_to_page(_mfn(mfn));
> -        if ( !page->u.inuse.type_info &&
> +        if ( (!page->u.inuse.type_info ||
> +              page->u.inuse.type_info == (PGT_writable_page | PGT_validated)) &&

Would it be clearer to get page for all pages that have a 0 count:
!(type_info & PGT_count_mask). Or would that interact badly with page
table pages?

>               !get_page_and_type(page, d, PGT_writable_page) )
>              BUG();
>      }
> @@ -720,6 +795,17 @@ int __init dom0_construct_pv(struct doma
>      /* Pages that are part of page tables must be read only. */
>      mark_pv_pt_pages_rdonly(d, l4start, vpt_start, nr_pt_pages);
>  
> +    /*
> +     * This needs to come after all potentially excess
> +     * get_page_and_type(..., PGT_writable_page) invocations; see the loop a
> +     * few lines further up, where the effect of calling that function in an
> +     * earlier loop iteration may get overwritten by a later one.
> +     */
> +    if ( need_iommu_pt_sync(d) &&
> +         iommu_unmap(d, _dfn(PFN_DOWN(mpt_alloc) - nr_pt_pages), nr_pt_pages,
> +                     &flush_flags) )
> +        BUG();

Wouldn't such unmap better happen as part of changing the types of the
pages that become part of the guest page tables?

>      /* Mask all upcalls... */
>      for ( i = 0; i < XEN_LEGACY_MAX_VCPUS; i++ )
>          shared_info(d, vcpu_info[i].evtchn_upcall_mask) = 1;
> @@ -793,7 +879,7 @@ int __init dom0_construct_pv(struct doma
>      {
>          pfn = pagetable_get_pfn(v->arch.guest_table);
>          setup_pv_physmap(d, pfn, v_start, v_end, vphysmap_start, vphysmap_end,
> -                         nr_pages);
> +                         nr_pages, &flush_flags);
>      }
>  
>      /* Write the phys->machine and machine->phys table entries. */
> @@ -824,7 +910,9 @@ int __init dom0_construct_pv(struct doma
>          if ( get_gpfn_from_mfn(mfn) >= count )
>          {
>              BUG_ON(compat);
> -            if ( !page->u.inuse.type_info &&
> +            if ( (!page->u.inuse.type_info ||
> +                  page->u.inuse.type_info == (PGT_writable_page |
> +                                              PGT_validated)) &&
>                   !get_page_and_type(page, d, PGT_writable_page) )
>                  BUG();
>  
> @@ -840,22 +928,41 @@ int __init dom0_construct_pv(struct doma
>  #endif
>      while ( pfn < nr_pages )
>      {
> -        if ( (page = alloc_chunk(d, nr_pages - domain_tot_pages(d))) == NULL )
> +        count = domain_tot_pages(d);
> +        if ( (page = alloc_chunk(d, nr_pages - count)) == NULL )
>              panic("Not enough RAM for DOM0 reservation\n");
> +        mfn = mfn_x(page_to_mfn(page));
> +
> +        if ( need_iommu_pt_sync(d) )
> +        {
> +            rc = iommu_map(d, _dfn(mfn), _mfn(mfn), domain_tot_pages(d) - count,
> +                           IOMMUF_readable | IOMMUF_writable, &flush_flags);
> +            if ( rc )
> +                printk(XENLOG_ERR
> +                       "pre-mapping MFN %lx (PFN %lx) into IOMMU failed: %d\n",
> +                       mfn, pfn, rc);
> +        }
> +
>          while ( pfn < domain_tot_pages(d) )
>          {
> -            mfn = mfn_x(page_to_mfn(page));
> +            if ( !rc )
> +                make_pages_writable(page, 1);

There's quite a lot of repetition of the pattern: allocate, iommu_map,
set as writable. Would it be possible to abstract this into some
kind of helper?

I've realized some of the allocations use alloc_chunk while others use
alloc_domheap_pages, so it might require some work.

> +
>  #ifndef NDEBUG
>  #define pfn (nr_pages - 1 - (pfn - (alloc_epfn - alloc_spfn)))
>  #endif
>              dom0_update_physmap(compat, pfn, mfn, vphysmap_start);
>  #undef pfn
> -            page++; pfn++;
> +            page++; mfn++; pfn++;
>              if ( !(pfn & 0xfffff) )
>                  process_pending_softirqs();
>          }
>      }
>  
> +    /* Use while() to avoid compiler warning. */
> +    while ( iommu_iotlb_flush_all(d, flush_flags) )
> +        break;

Might be worth to print a message here in case of error?

> +
>      if ( initrd_len != 0 )
>      {
>          si->mod_start = vinitrd_start ?: initrd_pfn;
> --- a/xen/drivers/passthrough/x86/iommu.c
> +++ b/xen/drivers/passthrough/x86/iommu.c
> @@ -325,8 +325,8 @@ static unsigned int __hwdom_init hwdom_i
>  
>  void __hwdom_init arch_iommu_hwdom_init(struct domain *d)
>  {
> -    unsigned long i, top, max_pfn;
> -    unsigned int flush_flags = 0;
> +    unsigned long i, top, max_pfn, start, count;
> +    unsigned int flush_flags = 0, start_perms = 0;
>  
>      BUG_ON(!is_hardware_domain(d));
>  
> @@ -357,9 +357,9 @@ void __hwdom_init arch_iommu_hwdom_init(
>       * First Mb will get mapped in one go by pvh_populate_p2m(). Avoid
>       * setting up potentially conflicting mappings here.
>       */
> -    i = paging_mode_translate(d) ? PFN_DOWN(MB(1)) : 0;
> +    start = paging_mode_translate(d) ? PFN_DOWN(MB(1)) : 0;
>  
> -    for ( ; i < top; i++ )
> +    for ( i = start, count = 0; i < top; )
>      {
>          unsigned long pfn = pdx_to_pfn(i);
>          unsigned int perms = hwdom_iommu_map(d, pfn, max_pfn);
> @@ -372,16 +372,30 @@ void __hwdom_init arch_iommu_hwdom_init(
>                                          perms & IOMMUF_writable ? p2m_access_rw
>                                                                  : p2m_access_r,
>                                          0);
> +        else if ( pfn != start + count || perms != start_perms )
> +        {
> +        commit:
> +            rc = iommu_map(d, _dfn(start), _mfn(start), count,
> +                           start_perms, &flush_flags);
> +            SWAP(start, pfn);
> +            start_perms = perms;
> +            count = 1;
> +        }
>          else
> -            rc = iommu_map(d, _dfn(pfn), _mfn(pfn), 1ul << PAGE_ORDER_4K,
> -                           perms, &flush_flags);
> +        {
> +            ++count;
> +            rc = 0;
> +        }
>  
>          if ( rc )
>              printk(XENLOG_WARNING "%pd: identity %smapping of %lx failed: %d\n",
>                     d, !paging_mode_translate(d) ? "IOMMU " : "", pfn, rc);

Would be nice to print the count (or end pfn) in case it's a range.

While not something that you have to fix here, the logic here is
becoming overly complicated IMO. It might be easier to just put all
the ram and reserved regions (or everything < 4G) into a rangeset and
then punch holes on it for non guest mappable regions, and finally use
rangeset_consume_ranges to iterate and map those. That's likely faster
than having to iterate over all pfns on the system, and easier to
understand from a logic PoV.

Thanks, Roger.