Re: [PATCH net-next 2/2] net: hns3: add vf fault process in hns3 ras

[Leon Romanovsky <leon@kernel.org>]

On Thu, Feb 02, 2023 at 09:08:37PM +0800, wangjie (L) wrote:
> 
> 
> On 2023/1/31 21:24, Leon Romanovsky wrote:
> > On Tue, Jan 31, 2023 at 08:04:14PM +0800, wangjie (L) wrote:
> > > 
> > > 
> > > On 2023/1/21 1:12, Leon Romanovsky wrote:
> > > > On Wed, Jan 18, 2023 at 08:34:03PM +0800, wangjie (L) wrote:
> > > > > 
> > > > > 
> > > > > On 2023/1/17 19:21, Leon Romanovsky wrote:
> > > > > > On Tue, Jan 17, 2023 at 03:04:15PM +0800, wangjie (L) wrote:
> > > > > > > 
> > > > > > > 
> > > > > > > On 2023/1/13 14:51, Leon Romanovsky wrote:
> > > > > > > > On Fri, Jan 13, 2023 at 10:08:29AM +0800, Hao Lan wrote:
> > > > > > > > > From: Jie Wang <wangjie125@huawei.com>
> > > > > > > > > 
> > > > > > > > > Currently hns3 driver supports vf fault detect feature. Several ras caused
> > > > > > > > > by VF resources don't need to do PF function reset for recovery. The driver
> > > > > > > > > only needs to reset the specified VF.
> > > > > > > > > 
> > > > > > > > > So this patch adds process in ras module. New process will get detailed
> > > > > > > > > information about ras and do the most correct measures based on these
> > > > > > > > > accurate information.
> > > > > > > > > 
> > > > > > > > > Signed-off-by: Jie Wang <wangjie125@huawei.com>
> > > > > > > > > Signed-off-by: Hao Lan <lanhao@huawei.com>
> > > > > > > > > ---
> > > > > > > > >  drivers/net/ethernet/hisilicon/hns3/hnae3.h   |   1 +
> > > > > > > > >  .../hns3/hns3_common/hclge_comm_cmd.h         |   1 +
> > > > > > > > >  .../hisilicon/hns3/hns3pf/hclge_err.c         | 113 +++++++++++++++++-
> > > > > > > > >  .../hisilicon/hns3/hns3pf/hclge_err.h         |   2 +
> > > > > > > > >  .../hisilicon/hns3/hns3pf/hclge_main.c        |   3 +-
> > > > > > > > >  .../hisilicon/hns3/hns3pf/hclge_main.h        |   1 +
> > > > > > > > >  6 files changed, 115 insertions(+), 6 deletions(-)
> > > > > > > > 
> > > > > > > > Why is it good idea to reset VF from PF?
> > > > > > > > What will happen with driver bound to this VF?
> > > > > > > > Shouldn't PCI recovery handle it?
> > > > > > > > 
> > > > > > > > Thanks
> > > > > > > > .
> > > > > > > PF doesn't reset VF directly. These VF faults are detected by hardware,
> > > > > > > and only reported to PF. PF get the VF id from firmware, then notify the VF
> > > > > > > that it needs reset. VF will do reset after receive the request.
> > > > > > 
> > > > > > This description isn't aligned with the code. You are issuing
> > > > > > hclge_func_reset_cmd() command which will reset VF, while notification
> > > > > > are handled by hclge_func_reset_notify_vf().
> > > > > > 
> > > > > > It also doesn't make any sense to send notification event to VF through
> > > > > > FW while the goal is to recover from stuck FW in that VF.
> > > > > > 
> > > > > Yes, I misunderstand the hclge_func_reset_notify_vf and
> > > > > hclge_func_reset_cmd. It should use hclge_func_reset_notify_vf to inform
> > > > > the VF for recovery. I will fix and retest it in V2.
> > > > > 
> > > > > This patch is used to recover specific vf hardware errors, for example the
> > > > > tx queue configuration exceptions. It make sense in these cases for the
> > > > > firmware is still working properly and can do the recovery rightly.
> > > > 
> > > > If FW is operational and knows about failure, why can't FW do recovery
> > > > internally to that VF without PF involvement?
> > > I'm sorry to reply so late because I took a vacation. If firmware reset VF
> > > hardware directly without notify the running VF driver, it will cause VF
> > > driver works abnormal.
> > 
> > mlx5 health recovery code proves that it is possible to do.
> > Even in your case, FW can notify VF without PF in the middle.
> > 
> > Thanks
> > 
> These faults only report to PF in hns3 devices, even if devlink health is
> used in hns3 driver, these faults also need to report to PF.

I don't think that it is correct, but let's see your v2 without VF reset
and with PF notifications.

Thanks

> 
> Thanks
> > > 
> > > Thanks
> > > > 
> > > > Thanks
> > > > .
> > > > 
> > .
> >