[rds-devel] general protection fault in __rds_rdma_map

Santosh Shilimkar santosh.shilimkar at oracle.com
Wed Jan 31 08:32:40 PST 2018


On 1/31/2018 5:11 AM, Dmitry Vyukov wrote:
> On Tue, Jan 30, 2018 at 10:52 PM, Eric Biggers <ebiggers3 at gmail.com> wrote:
>> Hi Or,

[...]

>>> I wondered how rds-rdma comes into play on these tests? do you have
>>> some rdma device on the VM
>>> where the syz bot runs? is it backed up by real HW card probed to the
>>> VM or it's software based (e.g
>>> some sort of emulator or software-RoCE (RXE driver))
>>
>> I don't think there is any rdma device at all; it's just testing a kernel with
>> tons of config options enabled, including CONFIG_RDS and CONFIG_RDS_TCP.  And it
>> has definitions for AF_RDS:
>> https://github.com/google/syzkaller/blob/master/sys/linux/socket_rds.txt
> 

[...]

> 
> 
> What Eric said. This is a VM with no special setup for RDS.
> If there something reasonably simple that we can do to improve RDS
> coverage, then we can look into it. E.g. we use tun to emulate remote
> network activity, and vcan devices to test AF_CAN.

Some of your tests using RDS TCP and that doesn't need in any special
hardware. With SOFT ROCE driver getting mature, one can use that
device as RDMA device and run RDS RDMA over it.

Regards,
Santosh



More information about the rds-devel mailing list