On Wed, Feb 20, 2019 at 04:20:38PM -0800, Matthew Wilcox wrote:
> The lock protecting the data structure does not need to be an rwlock.
> The only read access to the lock is in an error path, and if that's
> limiting your scalability, you have bigger performance problems.
>
> Eliminate mlx5_mkey_table in favour of using the xarray directly.
> Continue to use GFP_ATOMIC for allocating XArray nodes as we may be
> called in interrupt context.
>
> Signed-off-by: Matthew Wilcox <willy@xxxxxxxxxxxxx>
> drivers/infiniband/hw/mlx5/cq.c | 4 ++--
> drivers/infiniband/hw/mlx5/mr.c | 10 +++++-----
> 2 files changed, 7 insertions(+), 7 deletions(-)
This patch doesn't compile..
drivers/infiniband/hw/mlx5/cq.c: In function ‘mlx5_poll_one’:
./include/linux/xarray.h:444:39: error: ‘struct mlx5_mkey_table’ has no member named ‘xa_lock’; did you mean ‘lock’?
#define xa_lock(xa) spin_lock(&(xa)->xa_lock)
^~~~~~~
drivers/infiniband/hw/mlx5/cq.c:525:3: note: in expansion of macro ‘xa_lock’
xa_lock(&dev->mdev->priv.mkey_table);
^~~~~~~
./include/linux/xarray.h:445:43: error: ‘struct mlx5_mkey_table’ has no member named ‘xa_lock’; did you mean ‘lock’?
#define xa_unlock(xa) spin_unlock(&(xa)->xa_lock)
^~~~~~~
drivers/infiniband/hw/mlx5/cq.c:540:3: note: in expansion of macro ‘xa_unlock’
xa_unlock(&dev->mdev->priv.mkey_table);
^~~~~~~~~
scripts/Makefile.build:276: recipe for target 'drivers/infiniband/hw/mlx5/cq.o' failed
make[4]: *** [drivers/infiniband/hw/mlx5/cq.o] Error 1
make[4]: *** Waiting for unfinished jobs....
drivers/infiniband/hw/mlx5/mr.c: In function ‘reg_mr_callback’:
drivers/infiniband/hw/mlx5/mr.c:133:25: error: initialization of ‘struct xarray *’ from incompatible pointer type ‘struct mlx5_mkey_table *’ [-Werror=incompatible-pointer-types]
struct xarray *mkeys = &dev->mdev->priv.mkey_table;
^
cc1: some warnings being treated as errors
The change to the header file is missing. I think we will find other
readers than just MLX5_CQE_SIG_ERR when the patch is made complete.
Jason
