On Thu, Apr 08, 2021 at 08:42:57PM +0530, Devesh Sharma wrote: > On Thu, Apr 8, 2021 at 5:14 PM Leon Romanovsky <leon@xxxxxxxxxx> wrote: > > > > On Thu, Apr 08, 2021 at 05:06:24PM +0530, Devesh Sharma wrote: > > > On Sat, Apr 3, 2021 at 5:12 PM Leon Romanovsky <leon@xxxxxxxxxx> wrote: > > > > > > > > On Sat, Apr 03, 2021 at 03:52:13PM +0530, Devesh Sharma wrote: > > > > > On Thu, Apr 1, 2021 at 12:27 PM Leon Romanovsky <leon@xxxxxxxxxx> wrote: > > > > > > > > > > > > From: Leon Romanovsky <leonro@xxxxxxxxxx> > > > > > > > > > > > > Changelog: > > > > > > v2: > > > > > > * kbuild spotted that I didn't delete all code in patch #5, so deleted > > > > > > even more ulp_ops derefences. > > > > > > v1: https://lore.kernel.org/linux-rdma/20210329085212.257771-1-leon@xxxxxxxxxx > > > > > > * Go much deeper and removed useless ULP indirection > > > > > > v0: https://lore.kernel.org/linux-rdma/20210324142524.1135319-1-leon@xxxxxxxxxx > > > > > > ----------------------------------------------------------------------- > > > > > > > > > > > > The following series fixes issue spotted in [1], where bnxt_re driver > > > > > > messed with module reference counting in order to implement symbol > > > > > > dependency of bnxt_re and bnxt modules. All of this is done, when in > > > > > > upstream we have only one ULP user of that bnxt module. The simple > > > > > > declaration of exported symbol would do the trick. > > > > > > > > > > > > This series removes that custom module_get/_put, which is not supposed > > > > > > to be in the driver from the beginning and get rid of nasty indirection > > > > > > logic that isn't relevant for the upstream code. > > > > > > > > > > > > Such small changes allow us to simplify the bnxt code and my hope that > > > > > > Devesh will continue where I stopped and remove struct bnxt_ulp_ops too. > > > > > > > > > > > > Thanks > > > > > > > > > > > > [1] https://lore.kernel.org/linux-rdma/20210324142524.1135319-1-leon@xxxxxxxxxx > > > > > > > > > > > > Leon Romanovsky (5): > > > > > > RDMA/bnxt_re: Depend on bnxt ethernet driver and not blindly select it > > > > > > RDMA/bnxt_re: Create direct symbolic link between bnxt modules > > > > > > RDMA/bnxt_re: Get rid of custom module reference counting > > > > > > net/bnxt: Remove useless check of non-existent ULP id > > > > > > net/bnxt: Use direct API instead of useless indirection > > > > > > > > > > > > drivers/infiniband/hw/bnxt_re/Kconfig | 4 +- > > > > > > drivers/infiniband/hw/bnxt_re/main.c | 93 ++----- > > > > > > drivers/net/ethernet/broadcom/bnxt/bnxt.c | 4 +- > > > > > > drivers/net/ethernet/broadcom/bnxt/bnxt.h | 1 - > > > > > > drivers/net/ethernet/broadcom/bnxt/bnxt_ulp.c | 245 +++++++----------- > > > > > > drivers/net/ethernet/broadcom/bnxt/bnxt_ulp.h | 32 +-- > > > > > > 6 files changed, 119 insertions(+), 260 deletions(-) > > > > > > > > > > Hi Leon, > > > > > > > > > > After a couple of internal discussions we reached a conclusion to > > > > > implement the Auxbus driver interface and fix the problem once and for > > > > > all. > > > > > > > > Thanks Devesh, > > > > > > > > Jason, it looks like we can proceed with this patchset, because in > > > > auxbus mode this module refcount and ULP indirection logics will be > > > > removed anyway. > > > > > > > > Thanks > > > Hi Leon, > > > > > > In my internal testing, I am seeing a crash using the 3rd patch. I am > > > spending a few cycles on debugging it. expect my input in a day or so. > > > > Can you please post the kernel crash report here? > > I don't see how function rename in patch #3 can cause to the crash. > Hey, unfortunately my kdump service config is giving me tough time on > my host. I will share if I get it. Any news here? > > > > Thanks > > > > > > > > > > > > > > > > > > -- > > > > > > 2.30.2 > > > > > > > > > > > > > > > > > > > > > -- > > > > > -Regards > > > > > Devesh > > > > > > > > > > > > > > > > > -- > > > -Regards > > > Devesh > > > > > > > -- > -Regards > Devesh
