Re: [PATCH] net: mana: Fix MANA VF unload when host is unresponsive

[Simon Horman]

On Wed, Jun 21, 2023 at 03:29:01AM -0700, souradeep chakrabarti wrote:
> From: Souradeep Chakrabarti <schakrabarti@xxxxxxxxxxxxxxxxxxx>
> 
> This patch addresses  the VF unload issue, where mana_dealloc_queues()
> gets stuck in infinite while loop, because of host unresponsiveness.
> It adds a timeout in the while loop, to fix it.
> 
> Also this patch adds a new attribute in mana_context, which gets set when
> mana_hwc_send_request() hits a timeout because of host unresponsiveness.
> This flag then helps to avoid the timeouts in successive calls.
> 
> Signed-off-by: Souradeep Chakrabarti <schakrabarti@xxxxxxxxxxxxxxxxxxx>

Hi Souradeep,

thanks for your patch.
Some minor feedback from my aide.

> diff --git a/drivers/net/ethernet/microsoft/mana/gdma_main.c b/drivers/net/ethernet/microsoft/mana/gdma_main.c
> index 8f3f78b68592..5cc43ae78334 100644
> --- a/drivers/net/ethernet/microsoft/mana/gdma_main.c
> +++ b/drivers/net/ethernet/microsoft/mana/gdma_main.c
> @@ -946,10 +946,12 @@ int mana_gd_deregister_device(struct gdma_dev *gd)
>  	struct gdma_context *gc = gd->gdma_context;
>  	struct gdma_general_resp resp = {};
>  	struct gdma_general_req req = {};
> +	struct mana_context *ac;
>  	int err;
>  
>  	if (gd->pdid == INVALID_PDID)
>  		return -EINVAL;
> +	ac = (struct mana_context *)gd->driver_data;

drive_data is a void *.
There is no need to cast it to another type of pointer.

...

> diff --git a/drivers/net/ethernet/microsoft/mana/hw_channel.c b/drivers/net/ethernet/microsoft/mana/hw_channel.c
> index 9d1507eba5b9..557b890ad0ae 100644
> --- a/drivers/net/ethernet/microsoft/mana/hw_channel.c
> +++ b/drivers/net/ethernet/microsoft/mana/hw_channel.c

...

> @@ -786,12 +788,19 @@ int mana_hwc_send_request(struct hw_channel_context *hwc, u32 req_len,
>  	struct hwc_wq *txq = hwc->txq;
>  	struct gdma_req_hdr *req_msg;
>  	struct hwc_caller_ctx *ctx;
> +	struct mana_context *ac;
>  	u32 dest_vrcq = 0;
>  	u32 dest_vrq = 0;
>  	u16 msg_id;
>  	int err;
>  
>  	mana_hwc_get_msg_index(hwc, &msg_id);
> +	ac = (struct mana_context *)hwc->gdma_dev->driver_data;

Ditto.

...

> diff --git a/drivers/net/ethernet/microsoft/mana/mana_en.c b/drivers/net/ethernet/microsoft/mana/mana_en.c
> index d907727c7b7a..24f5508d2979 100644
> --- a/drivers/net/ethernet/microsoft/mana/mana_en.c
> +++ b/drivers/net/ethernet/microsoft/mana/mana_en.c
> @@ -2330,7 +2330,10 @@ static int mana_dealloc_queues(struct net_device *ndev)
>  	struct mana_port_context *apc = netdev_priv(ndev);
>  	struct gdma_dev *gd = apc->ac->gdma_dev;
>  	struct mana_txq *txq;
> +	struct sk_buff *skb;
> +	struct mana_cq *cq;
>  	int i, err;
> +	unsigned long timeout;

Please use reverse xmas tree - longest line to shortest - for
local variable declarations in Networking code.

...

> diff --git a/include/net/mana/mana.h b/include/net/mana/mana.h
> index 9eef19972845..34f5d8e06ede 100644
> --- a/include/net/mana/mana.h
> +++ b/include/net/mana/mana.h
> @@ -361,6 +361,8 @@ struct mana_context {
>  	struct mana_eq *eqs;
>  
>  	struct net_device *ports[MAX_PORTS_IN_MANA_DEV];
> +
> +	bool vf_unload_timeout;

Perhaps it is not important, but on x86_54 there is a 6 byte hole in the first
cacheline after num_ports where this could go.

pahole reports:

struct mana_context {
	struct gdma_dev *          gdma_dev;             /*     0     8 */
	u16                        num_ports;            /*     8     2 */

	/* XXX 6 bytes hole, try to pack */

	struct mana_eq *           eqs;                  /*    16     8 */
	struct net_device *        ports[256];           /*    24  2048 */
	/* --- cacheline 32 boundary (2048 bytes) was 24 bytes ago --- */
	bool                       vf_unload_timeout;    /*  2072     1 */

	/* size: 2080, cachelines: 33, members: 5 */
	/* sum members: 2067, holes: 1, sum holes: 6 */
	/* padding: 7 */
	/* last cacheline: 32 bytes */
};

-- 
pw-bot: changes-requested