| Message ID | 20200803061941.1139994-1-haakon.bugge@oracle.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | Add CM packets missing and harden the proxying | expand |
A friendly reminder. Thxs, Håkon > On 3 Aug 2020, at 08:19, Håkon Bugge <haakon.bugge@oracle.com> wrote: > > A high number of MAD packet drops are observed in the mlx4 MAD proxy > system. These are fixed by separating the parameters for the tunnel > vs. wire QPs and by introducing a separate worker-thread for the wire > QPs. > > Support for MRA and REJ with its reason being timeout is also added. > > Dynamic debug prints adjusted and amended. > > v1->v2: > * Added commit ("Adjust delayed work when a dup is observed") > * Minor adjustments in some of the commits > > Håkon Bugge (6): > IB/mlx4: Add and improve logging > IB/mlx4: Add support for MRA > IB/mlx4: Separate tunnel and wire bufs parameters > IB/mlx4: Fix starvation in paravirt mux/demux > IB/mlx4: Add support for REJ due to timeout > IB/mlx4: Adjust delayed work when a dup is observed > > drivers/infiniband/hw/mlx4/cm.c | 148 ++++++++++++++++++++++++- > drivers/infiniband/hw/mlx4/mad.c | 158 +++++++++++++++------------ > drivers/infiniband/hw/mlx4/mlx4_ib.h | 8 +- > 3 files changed, 241 insertions(+), 73 deletions(-) > > -- > 2.20.1 >
On Mon, Aug 10, 2020 at 01:20:43PM +0200, Håkon Bugge wrote: > A friendly reminder. We are in merge window. BTW, the patches are in our regression all that time and everything works as expected. Thanks > > > Thxs, Håkon > > > > On 3 Aug 2020, at 08:19, Håkon Bugge <haakon.bugge@oracle.com> wrote: > > > > A high number of MAD packet drops are observed in the mlx4 MAD proxy > > system. These are fixed by separating the parameters for the tunnel > > vs. wire QPs and by introducing a separate worker-thread for the wire > > QPs. > > > > Support for MRA and REJ with its reason being timeout is also added. > > > > Dynamic debug prints adjusted and amended. > > > > v1->v2: > > * Added commit ("Adjust delayed work when a dup is observed") > > * Minor adjustments in some of the commits > > > > Håkon Bugge (6): > > IB/mlx4: Add and improve logging > > IB/mlx4: Add support for MRA > > IB/mlx4: Separate tunnel and wire bufs parameters > > IB/mlx4: Fix starvation in paravirt mux/demux > > IB/mlx4: Add support for REJ due to timeout > > IB/mlx4: Adjust delayed work when a dup is observed > > > > drivers/infiniband/hw/mlx4/cm.c | 148 ++++++++++++++++++++++++- > > drivers/infiniband/hw/mlx4/mad.c | 158 +++++++++++++++------------ > > drivers/infiniband/hw/mlx4/mlx4_ib.h | 8 +- > > 3 files changed, 241 insertions(+), 73 deletions(-) > > > > -- > > 2.20.1 > > >
On 10/08/2020 14:46, Leon Romanovsky wrote: > On Mon, Aug 10, 2020 at 01:20:43PM +0200, Håkon Bugge wrote: >> A friendly reminder. > > We are in merge window. The merge window shouldn't affect bug fixes submissions, no?
On Mon, Aug 10, 2020 at 05:10:44PM +0300, Gal Pressman wrote: > On 10/08/2020 14:46, Leon Romanovsky wrote: > > On Mon, Aug 10, 2020 at 01:20:43PM +0200, Håkon Bugge wrote: > >> A friendly reminder. > > > > We are in merge window. > > The merge window shouldn't affect bug fixes submissions, no? It is hard to call bug fixes, according to Fixes line and description the code is broken from day one. There is no urgency to merge it now. Thanks
On Mon, Aug 03, 2020 at 08:19:35AM +0200, Håkon Bugge wrote: > A high number of MAD packet drops are observed in the mlx4 MAD proxy > system. These are fixed by separating the parameters for the tunnel > vs. wire QPs and by introducing a separate worker-thread for the wire > QPs. > > Support for MRA and REJ with its reason being timeout is also added. > > Dynamic debug prints adjusted and amended. > > v1->v2: > * Added commit ("Adjust delayed work when a dup is observed") > * Minor adjustments in some of the commits > > Håkon Bugge (6): > IB/mlx4: Add and improve logging > IB/mlx4: Add support for MRA > IB/mlx4: Separate tunnel and wire bufs parameters > IB/mlx4: Fix starvation in paravirt mux/demux > IB/mlx4: Add support for REJ due to timeout > IB/mlx4: Adjust delayed work when a dup is observed Applied to for-next, this does not look like -rc material Thanks, Jason
A high number of MAD packet drops are observed in the mlx4 MAD proxy system. These are fixed by separating the parameters for the tunnel vs. wire QPs and by introducing a separate worker-thread for the wire QPs. Support for MRA and REJ with its reason being timeout is also added. Dynamic debug prints adjusted and amended. v1->v2: * Added commit ("Adjust delayed work when a dup is observed") * Minor adjustments in some of the commits Håkon Bugge (6): IB/mlx4: Add and improve logging IB/mlx4: Add support for MRA IB/mlx4: Separate tunnel and wire bufs parameters IB/mlx4: Fix starvation in paravirt mux/demux IB/mlx4: Add support for REJ due to timeout IB/mlx4: Adjust delayed work when a dup is observed drivers/infiniband/hw/mlx4/cm.c | 148 ++++++++++++++++++++++++- drivers/infiniband/hw/mlx4/mad.c | 158 +++++++++++++++------------ drivers/infiniband/hw/mlx4/mlx4_ib.h | 8 +- 3 files changed, 241 insertions(+), 73 deletions(-) -- 2.20.1