Re: [PATCH RFCv2 14/15] xen/arm: mm: Rework setup_xenheap_mappings()

[Stefano Stabellini <sstabellini@kernel.org>]

On Sun, 25 Apr 2021, Julien Grall wrote:
> From: Julien Grall <julien.grall@arm.com>
> 
> A few issues have been reported with setup_xenheap_mappings() over the
> last couple of years. The main ones are:
>     - It will break on platform supporting more than 512GB of RAM
>       because the memory allocated by the boot allocator is not yet
>       mapped.
>     - Aligning all the regions to 1GB may lead to unexpected result
>       because we may alias non-cacheable region (such as device or reserved
>       regions).
> 
> map_pages_to_xen() was recently reworked to allow superpage mappings and
> deal with the use of page-tables before they are mapped.
> 
> Most of the code in setup_xenheap_mappings() is now replaced with a
> single call to map_pages_to_xen().
> 
> This also require to re-order the steps setup_mm() so the regions are
> given to the boot allocator first and then we setup the xenheap
> mappings.

I know this is paranoia but doesn't this introduce a requirement on the
size of the first bank in bootinfo.mem.bank[] ?

It should be at least 8KB?

I know it is unlikely but it is theoretically possible to have a first
bank of just 1KB.

I think we should write the requirement down with an in-code comment?
Or better maybe we should write a message about it in the panic below?


> Note that the 1GB alignment is not yet removed.
> 
> Signed-off-by: Julien Grall <julien.grall@arm.com>
> Signed-off-by: Julien Grall <jgrall@amazon.com>
> 
> ---
>     Changes in v2:
>         - New patch
> 
>     TODO:
>         - Remove the 1GB alignment
>         - Add support for setting the contiguous bit
> ---
>  xen/arch/arm/mm.c    | 60 ++++----------------------------------------
>  xen/arch/arm/setup.c | 10 ++++++--

I love it!


>  2 files changed, 13 insertions(+), 57 deletions(-)
> 
> diff --git a/xen/arch/arm/mm.c b/xen/arch/arm/mm.c
> index f5768f2d4a81..c49403b687f5 100644
> --- a/xen/arch/arm/mm.c
> +++ b/xen/arch/arm/mm.c
> @@ -143,17 +143,6 @@ static DEFINE_PAGE_TABLE(cpu0_pgtable);
>  static DEFINE_PAGE_TABLES(cpu0_dommap, DOMHEAP_SECOND_PAGES);
>  #endif
>  
> -#ifdef CONFIG_ARM_64
> -/* The first page of the first level mapping of the xenheap. The
> - * subsequent xenheap first level pages are dynamically allocated, but
> - * we need this one to bootstrap ourselves. */
> -static DEFINE_PAGE_TABLE(xenheap_first_first);
> -/* The zeroeth level slot which uses xenheap_first_first. Used because
> - * setup_xenheap_mappings otherwise relies on mfn_to_virt which isn't
> - * valid for a non-xenheap mapping. */
> -static __initdata int xenheap_first_first_slot = -1;
> -#endif
> -
>  /* Common pagetable leaves */
>  /* Second level page tables.
>   *
> @@ -818,9 +807,9 @@ void __init setup_xenheap_mappings(unsigned long base_mfn,
>  void __init setup_xenheap_mappings(unsigned long base_mfn,
>                                     unsigned long nr_mfns)
>  {
> -    lpae_t *first, pte;
>      unsigned long mfn, end_mfn;
>      vaddr_t vaddr;
> +    int rc;
>  
>      /* Align to previous 1GB boundary */
>      mfn = base_mfn & ~((FIRST_SIZE>>PAGE_SHIFT)-1);
> @@ -846,49 +835,10 @@ void __init setup_xenheap_mappings(unsigned long base_mfn,
>       */
>      vaddr = (vaddr_t)__mfn_to_virt(base_mfn) & FIRST_MASK;
>  
> -    while ( mfn < end_mfn )
> -    {
> -        int slot = zeroeth_table_offset(vaddr);
> -        lpae_t *p = &xen_pgtable[slot];
> -
> -        if ( p->pt.valid )
> -        {
> -            /* mfn_to_virt is not valid on the 1st 1st mfn, since it
> -             * is not within the xenheap. */
> -            first = slot == xenheap_first_first_slot ?
> -                xenheap_first_first : mfn_to_virt(lpae_get_mfn(*p));
> -        }
> -        else if ( xenheap_first_first_slot == -1)
> -        {
> -            /* Use xenheap_first_first to bootstrap the mappings */
> -            first = xenheap_first_first;
> -
> -            pte = pte_of_xenaddr((vaddr_t)xenheap_first_first);
> -            pte.pt.table = 1;
> -            write_pte(p, pte);
> -
> -            xenheap_first_first_slot = slot;
> -        }
> -        else
> -        {
> -            mfn_t first_mfn = alloc_boot_pages(1, 1);
> -
> -            clear_page(mfn_to_virt(first_mfn));
> -            pte = mfn_to_xen_entry(first_mfn, MT_NORMAL);
> -            pte.pt.table = 1;
> -            write_pte(p, pte);
> -            first = mfn_to_virt(first_mfn);
> -        }
> -
> -        pte = mfn_to_xen_entry(_mfn(mfn), MT_NORMAL);
> -        /* TODO: Set pte.pt.contig when appropriate. */
> -        write_pte(&first[first_table_offset(vaddr)], pte);
> -
> -        mfn += FIRST_SIZE>>PAGE_SHIFT;
> -        vaddr += FIRST_SIZE;
> -    }
> -
> -    flush_xen_tlb_local();
> +    rc = map_pages_to_xen(vaddr, _mfn(mfn), end_mfn - mfn,
> +                          PAGE_HYPERVISOR_RW | _PAGE_BLOCK);
> +    if ( rc )
> +        panic("Unable to setup the xenheap mappings.\n");

This is the panic I was talking about


>  }
>  #endif
>  
> diff --git a/xen/arch/arm/setup.c b/xen/arch/arm/setup.c
> index 00aad1c194b9..0993a4bb52d4 100644
> --- a/xen/arch/arm/setup.c
> +++ b/xen/arch/arm/setup.c
> @@ -761,8 +761,11 @@ static void __init setup_mm(void)
>          ram_start = min(ram_start,bank_start);
>          ram_end = max(ram_end,bank_end);
>  
> -        setup_xenheap_mappings(bank_start>>PAGE_SHIFT, bank_size>>PAGE_SHIFT);
> -
> +        /*
> +         * Add the region to the boot allocator first, so we can use
> +         * some to allocate page-tables for setting up the xenheap
> +         * mappings.
> +         */
>          s = bank_start;
>          while ( s < bank_end )
>          {
> @@ -781,6 +784,9 @@ static void __init setup_mm(void)
>              fw_unreserved_regions(s, e, init_boot_pages, 0);
>              s = n;
>          }
> +
> +        setup_xenheap_mappings(bank_start >> PAGE_SHIFT,
> +                               bank_size >> PAGE_SHIFT);
>      }
>  
>      total_pages += ram_size >> PAGE_SHIFT;
> -- 
> 2.17.1
>