[PATCH v4 14/19] KVM: arm/arm64: Move HYP IO VAs to the "idmap" range

[Christoffer Dall]

On Thu, Jan 04, 2018 at 06:43:29PM +0000, Marc Zyngier wrote:
> We so far mapped our HYP IO (which is essencially the GICv2 control
> registers) using the same method as for memory. It recently appeared
> that is a bit unsafe:
> 
> We compute the HYP VA using the kern_hyp_va helper, but that helper
> is only designed to deal with kernel VAs coming from the linear map,
> and not from the vmalloc region... This could in turn cause some bad
> aliasing between the two, amplified by the upcoming VA randomisation.
> 
> A solution is to come up with our very own basic VA allocator for
> MMIO. Since half of the HYP address space only contains a single
> page (the idmap), we have plenty to borrow from. Let's use the idmap
> as a base, and allocate downwards from it. GICv2 now lives on the
> other side of the great VA barrier.
> 
> Signed-off-by: Marc Zyngier <marc.zyngier at arm.com>
> ---
>  virt/kvm/arm/mmu.c | 56 +++++++++++++++++++++++++++++++++++++++++-------------
>  1 file changed, 43 insertions(+), 13 deletions(-)
> 
> diff --git a/virt/kvm/arm/mmu.c b/virt/kvm/arm/mmu.c
> index 6192d45d1e1a..14c5e5534f2f 100644
> --- a/virt/kvm/arm/mmu.c
> +++ b/virt/kvm/arm/mmu.c
> @@ -43,6 +43,9 @@ static unsigned long hyp_idmap_start;
>  static unsigned long hyp_idmap_end;
>  static phys_addr_t hyp_idmap_vector;
>  
> +static DEFINE_MUTEX(io_map_lock);
> +static unsigned long io_map_base;
> +
>  #define S2_PGD_SIZE	(PTRS_PER_S2_PGD * sizeof(pgd_t))
>  #define hyp_pgd_order get_order(PTRS_PER_PGD * sizeof(pgd_t))
>  
> @@ -502,27 +505,31 @@ static void unmap_hyp_range(pgd_t *pgdp, phys_addr_t start, u64 size)
>   *
>   * Assumes hyp_pgd is a page table used strictly in Hyp-mode and
>   * therefore contains either mappings in the kernel memory area (above
> - * PAGE_OFFSET), or device mappings in the vmalloc range (from
> - * VMALLOC_START to VMALLOC_END).
> + * PAGE_OFFSET), or device mappings in the idmap range.
>   *
> - * boot_hyp_pgd should only map two pages for the init code.
> + * boot_hyp_pgd should only map the idmap range, and is only used in
> + * the extended idmap case.
>   */
>  void free_hyp_pgds(void)
>  {
> +	pgd_t *id_pgd;
> +
>  	mutex_lock(&kvm_hyp_pgd_mutex);
>  
> +	id_pgd = boot_hyp_pgd ? boot_hyp_pgd : hyp_pgd;
> +
> +	if (id_pgd)
> +		unmap_hyp_range(id_pgd, io_map_base,
> +				hyp_idmap_start + PAGE_SIZE - io_map_base);
> +
>  	if (boot_hyp_pgd) {
> -		unmap_hyp_range(boot_hyp_pgd, hyp_idmap_start, PAGE_SIZE);
>  		free_pages((unsigned long)boot_hyp_pgd, hyp_pgd_order);
>  		boot_hyp_pgd = NULL;
>  	}
>  
>  	if (hyp_pgd) {
> -		unmap_hyp_range(hyp_pgd, hyp_idmap_start, PAGE_SIZE);
>  		unmap_hyp_range(hyp_pgd, kern_hyp_va(PAGE_OFFSET),
>  				(uintptr_t)high_memory - PAGE_OFFSET);
> -		unmap_hyp_range(hyp_pgd, kern_hyp_va(VMALLOC_START),
> -				VMALLOC_END - VMALLOC_START);
>  
>  		free_pages((unsigned long)hyp_pgd, hyp_pgd_order);
>  		hyp_pgd = NULL;
> @@ -721,7 +728,8 @@ int create_hyp_io_mappings(phys_addr_t phys_addr, size_t size,
>  			   void __iomem **kaddr,
>  			   void __iomem **haddr)
>  {
> -	unsigned long start, end;
> +	pgd_t *pgd = hyp_pgd;
> +	unsigned long base;
>  	int ret;
>  
>  	*kaddr = ioremap(phys_addr, size);
> @@ -733,17 +741,38 @@ int create_hyp_io_mappings(phys_addr_t phys_addr, size_t size,
>  		return 0;
>  	}
>  
> +	mutex_lock(&io_map_lock);
> +
> +	base = io_map_base - size;

are we guaranteed that hyp_idmap_start (and therefore io_map_base) is
sufficiently greater than 0 ?  I suppose that even if RAM starts at 0,
and the kernel was loaded at 0, the idmap page for Hyp would be at some
reasonable offset from the start of the kernel image?

> +	base &= ~(size - 1);

I'm not sure I understand this line.  Wouldn't it make more sense to use
PAGE_SIZE here?

>  
> -	start = kern_hyp_va((unsigned long)*kaddr);
> -	end = kern_hyp_va((unsigned long)*kaddr + size);
> -	ret = __create_hyp_mappings(hyp_pgd, start, end,
> +	/*
> +	 * Verify that BIT(VA_BITS - 1) hasn't been flipped by
> +	 * allocating the new area, as it would indicate we've
> +	 * overflowed the idmap/IO address range.
> +	 */
> +	if ((base ^ io_map_base) & BIT(VA_BITS - 1)) {
> +		ret = -ENOMEM;
> +		goto out;
> +	}
> +
> +	if (__kvm_cpu_uses_extended_idmap())
> +		pgd = boot_hyp_pgd;
> +

I don't understand why you can change the pgd used for mappings here,
perhaps a patch splitting issue?

> +	ret = __create_hyp_mappings(pgd, base, base + size,
>  				     __phys_to_pfn(phys_addr), PAGE_HYP_DEVICE);
>  
> +	if (!ret) {
> +		*haddr = (void __iomem *)base;
> +		io_map_base = base;
> +	}
> +
> +out:
> +	mutex_unlock(&io_map_lock);
> +
>  	if (ret) {
>  		iounmap(*kaddr);
>  		*kaddr = NULL;
> -	} else {
> -		*haddr = (void __iomem *)start;

ah, it gets reworked here, so never mind the comment on the previous patch.

>  	}
>  
>  	return ret;
> @@ -1826,6 +1855,7 @@ int kvm_mmu_init(void)
>  			goto out;
>  	}
>  
> +	io_map_base = hyp_idmap_start;
>  	return 0;
>  out:
>  	free_hyp_pgds();
> -- 
> 2.14.2
> 

Thanks,
-Christoffer