[PATCH net] nfp: always unmask aux interrupts at init

[PATCH net] nfp: always unmask aux interrupts at init

[Jakub Kicinski]

The link state and exception interrupts may be masked when we probe.
The firmware should in theory prevent sending (and automasking) those
interrupts if the device is disabled, but if my reading of the FW code
is correct there are firmwares out there with race conditions in this
area.  The interrupt may also be masked if previous driver which used
the device was malfunctioning and we didn't load the FW (there is no
other good way to comprehensively reset the PF).

Note that FW unmasks the data interrupts by itself when vNIC is
enabled, such helpful operation is not performed for LSC/EXN interrupts.

Always unmask the auxiliary interrupts after request_irq().  On the
remove path add missing PCI write flush before free_irq().

Fixes: 4c3523623dc0 ("net: add driver for Netronome NFP4000/NFP6000 NIC VFs")
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
---
 drivers/net/ethernet/netronome/nfp/nfp_net_common.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c b/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
index 1a603fdd9e80..99b0487b6d82 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
@@ -568,6 +568,7 @@ nfp_net_aux_irq_request(struct nfp_net *nn, u32 ctrl_offset,
 		return err;
 	}
 	nn_writeb(nn, ctrl_offset, entry->entry);
+	nfp_net_irq_unmask(nn, entry->entry);
 
 	return 0;
 }
@@ -582,6 +583,7 @@ static void nfp_net_aux_irq_free(struct nfp_net *nn, u32 ctrl_offset,
 				 unsigned int vector_idx)
 {
 	nn_writeb(nn, ctrl_offset, 0xff);
+	nn_pci_flush(nn);
 	free_irq(nn->irq_entries[vector_idx].vector, nn);
 }
 
-- 
2.15.1

Re: [PATCH net] nfp: always unmask aux interrupts at init

[David Miller]

From: Jakub Kicinski <jakub.kicinski@netronome.com>
Date: Tue,  9 Jan 2018 18:14:28 -0800

> The link state and exception interrupts may be masked when we probe.
> The firmware should in theory prevent sending (and automasking) those
> interrupts if the device is disabled, but if my reading of the FW code
> is correct there are firmwares out there with race conditions in this
> area.  The interrupt may also be masked if previous driver which used
> the device was malfunctioning and we didn't load the FW (there is no
> other good way to comprehensively reset the PF).
> 
> Note that FW unmasks the data interrupts by itself when vNIC is
> enabled, such helpful operation is not performed for LSC/EXN interrupts.
> 
> Always unmask the auxiliary interrupts after request_irq().  On the
> remove path add missing PCI write flush before free_irq().
> 
> Fixes: 4c3523623dc0 ("net: add driver for Netronome NFP4000/NFP6000 NIC VFs")
> Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
> Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>

Applied and queued up for -stable, thanks.