Re: [PATCH for-4.22 v5] dom0less: Prevent division by zero in handle_passthrough_prop()

[Dmytro Prokopchuk1 <dmytro_prokopchuk1@xxxxxxxx>]

On 7/10/26 10:24, Orzel, Michal wrote:
>
>
> On 09-Jul-26 14:31, Dmytro Prokopchuk1 wrote:
>> A malformed partial DTB specifying both '#address-cells = <0>' and
>> '#size-cells = <0>' causes '(address_cells * 2 + size_cells)' to
>> evaluate to 0. This sum is subsequently used as a divisor when
>> calculating the number of regions in the 'xen,reg' property inside
>> handle_passthrough_prop():
>>
>>      len = fdt32_to_cpu(xen_reg->len) / ((address_cells * 2 + size_cells) *
>>                                          sizeof(uint32_t));
>>
>> This leads to a division by zero exception in the Xen hypervisor during
>> boot, causing a hypervisor panic/crash.
>>
>> Fix this by validating that both 'address_cells' and 'size_cells'
>> are within the valid range of [1, 2] at the read side in scan_pfdt_node()
>> immediately after they are parsed. Any invalid cell size combination is
>> safely rejected early with an error message and return -EINVAL.
>>
>> Fixes: 9ce974c47588 ("xen/arm: assign devices to boot domains")
>> Signed-off-by: Dmytro Prokopchuk <dmytro_prokopchuk1@xxxxxxxx>
>> ---
>> Changes in v5:
>> - fixed mistakes around dprintk()
>> Test CI pipeline:
>> https://gitlab.com/xen-project/people/dimaprkp4k/xen/-/pipelines/2664600678
>> ---
>>   xen/common/device-tree/dom0less-build.c | 8 ++++++++
>>   1 file changed, 8 insertions(+)
>>
>> diff --git a/xen/common/device-tree/dom0less-build.c 
>> b/xen/common/device-tree/dom0less-build.c
>> index eacfd93087..c054ea4e2f 100644
>> --- a/xen/common/device-tree/dom0less-build.c
>> +++ b/xen/common/device-tree/dom0less-build.c
>> @@ -341,6 +341,14 @@ static int __init scan_pfdt_node(struct kernel_info 
>> *kinfo, const void *pfdt,
>>       size_cells = device_tree_get_u32(pfdt, nodeoff, "#size-cells",
>>                                        DT_ROOT_NODE_SIZE_CELLS_DEFAULT);
>>
>> +    if ( (address_cells < 1) || (address_cells > 2) ||
>> +         (size_cells < 1) || (size_cells > 2) )
> Too many discussions yesterday... Today, I looked at the complete flow once
> again and I'm sorry to say that my previous comment to move the check right at
> the read side was incorrect. #address/size-cells define the number of cells 
> for
> their children and these don't need to have xen,reg. Even our passthrough
> example (docs/misc/arm/passthrough.txt) defines a ethernet node with 
> #size-cells
> = <0> for its PHY child with a 1 cell reg but no xen,reg. Here, for our sanity
> check, we only care that number of cells for xen,reg is either 1 or 2. 0 is a
> legitimate value for nodes without xen,reg (clock nodes, SPI, GPIO, etc.).
> Values above 2 i.e. 3 is PCI only which does not go through this path but 
> still,
> we should only care about nodes with xen,reg.
>
> Therefore, please move this check at the top of handle_passthrough_prop() 
> with a
> comment below (the number of discussions about this subject is a clear
> indication that a comment is needed):
>
> /*
>   * xen,reg holds flat host/guest physical addresses and sizes, so the
>   * inherited #address-cells/#size-cells must each be 1 or 2. This also
>   * guards the len division below against a zero or wrapped divisor.
>   */
>
> ~Michal
>

Ack. I will do.

Thanks, Dmytro.