On 06/06/19, 6:31 PM, "Jason Gunthorpe" jgg@mellanox.com wrote:
On Thu, Jun 06, 2019 at 10:25:23AM +0000, Ajay Kaher wrote:
I think in this kernel the mm handling code for IB is in three different drivers, it probably needs to be fixed too?
Thanks Jason for pointing this. Crossed checked the locking of mmap_sem in IB driver code of 4.9 to 4.14 with >5.0 and found it requires to handle at following locations of 4.9 and 4.14: mlx4_ib_disassociate_ucontext() of drivers/infiniband/hw/mlx5/main.c: mlx5_ib_disassociate_ucontext() of drivers/infiniband/hw/mlx4/main.c
To fix at above location, would you like me to modify the original patch or submit in another patch.
I think it is a backporting thing, so you should put the new work in this patch? I'm not sure.
Ok, I would like to submit new patch for new work in 4.14, and then backport to 4.9, soon I will submit for review.
Re-submitted the this patch again after correcting the upstream commit, as: Sub: [PATCH v2 1/1] [v4.9.y] coredump: fix race condition between mmget_not_zero()/get_task_mm() and core dumping
Please review.
I'm also surprised hns isn't in the above list of drivers, but maybe hns didn't have this support in these kernels..
IB->hns doesn't have any instance of mmap_sem checked for 4.9-4.14.
Jason