[PATCH net] net: libwx: fix firmware mailbox abnormal return

[PATCH net] net: libwx: fix firmware mailbox abnormal return

[Jiawen Wu]

Firmware writes back 'firmware ready' and 'unknown command' in the mailbox
message if there is an unknown command sent by driver. It tends to happen
with the use of custom firmware. So move the check for 'unknown command'
out of the poll timeout for 'firmware ready'. And adjust the debug log so
that mailbox messages are always printed when commands timeout.

Fixes: 1efa9bfe58c5 ("net: libwx: Implement interaction with firmware")
Signed-off-by: Jiawen Wu <jiawenwu@trustnetic.com>
---
 drivers/net/ethernet/wangxun/libwx/wx_hw.c | 22 ++++++++++------------
 1 file changed, 10 insertions(+), 12 deletions(-)

diff --git a/drivers/net/ethernet/wangxun/libwx/wx_hw.c b/drivers/net/ethernet/wangxun/libwx/wx_hw.c
index 1bf9c38e4125..7059e0100c7c 100644
--- a/drivers/net/ethernet/wangxun/libwx/wx_hw.c
+++ b/drivers/net/ethernet/wangxun/libwx/wx_hw.c
@@ -334,27 +334,25 @@ int wx_host_interface_command(struct wx *wx, u32 *buffer,
 	status = read_poll_timeout(rd32, hicr, hicr & WX_MNG_MBOX_CTL_FWRDY, 1000,
 				   timeout * 1000, false, wx, WX_MNG_MBOX_CTL);
 
+	buf[0] = rd32(wx, WX_MNG_MBOX);
+	if ((buf[0] & 0xff0000) >> 16 == 0x80) {
+		wx_dbg(wx, "It's unknown cmd.\n");
+		status = -EINVAL;
+		goto rel_out;
+	}
+
 	/* Check command completion */
 	if (status) {
 		wx_dbg(wx, "Command has failed with no status valid.\n");
-
-		buf[0] = rd32(wx, WX_MNG_MBOX);
-		if ((buffer[0] & 0xff) != (~buf[0] >> 24)) {
-			status = -EINVAL;
-			goto rel_out;
-		}
-		if ((buf[0] & 0xff0000) >> 16 == 0x80) {
-			wx_dbg(wx, "It's unknown cmd.\n");
-			status = -EINVAL;
-			goto rel_out;
-		}
-
 		wx_dbg(wx, "write value:\n");
 		for (i = 0; i < dword_len; i++)
 			wx_dbg(wx, "%x ", buffer[i]);
 		wx_dbg(wx, "read value:\n");
 		for (i = 0; i < dword_len; i++)
 			wx_dbg(wx, "%x ", buf[i]);
+		wx_dbg(wx, "check: %x %x\n", buffer[0] & 0xff, ~buf[0] >> 24);
+		if ((buffer[0] & 0xff) != (~buf[0] >> 24))
+			goto rel_out;
 	}
 
 	if (!return_data)
-- 
2.27.0

Re: [PATCH net] net: libwx: fix firmware mailbox abnormal return

[Jakub Kicinski]

On Thu, 26 Dec 2024 11:18:10 +0800 Jiawen Wu wrote:
> Firmware writes back 'firmware ready' and 'unknown command' in the mailbox
> message if there is an unknown command sent by driver. It tends to happen
> with the use of custom firmware. So move the check for 'unknown command'
> out of the poll timeout for 'firmware ready'. And adjust the debug log so
> that mailbox messages are always printed when commands timeout.

The commit message doesn't really explain what the problem is, 
just what the code does. What is the problem you're solving 
and how does it impact the user?

> Fixes: 1efa9bfe58c5 ("net: libwx: Implement interaction with firmware")
> Signed-off-by: Jiawen Wu <jiawenwu@trustnetic.com>
> ---
>  drivers/net/ethernet/wangxun/libwx/wx_hw.c | 22 ++++++++++------------
>  1 file changed, 10 insertions(+), 12 deletions(-)
> 
> diff --git a/drivers/net/ethernet/wangxun/libwx/wx_hw.c b/drivers/net/ethernet/wangxun/libwx/wx_hw.c
> index 1bf9c38e4125..7059e0100c7c 100644
> --- a/drivers/net/ethernet/wangxun/libwx/wx_hw.c
> +++ b/drivers/net/ethernet/wangxun/libwx/wx_hw.c
> @@ -334,27 +334,25 @@ int wx_host_interface_command(struct wx *wx, u32 *buffer,
>  	status = read_poll_timeout(rd32, hicr, hicr & WX_MNG_MBOX_CTL_FWRDY, 1000,
>  				   timeout * 1000, false, wx, WX_MNG_MBOX_CTL);
>  
> +	buf[0] = rd32(wx, WX_MNG_MBOX);
> +	if ((buf[0] & 0xff0000) >> 16 == 0x80) {
> +		wx_dbg(wx, "It's unknown cmd.\n");
> +		status = -EINVAL;
> +		goto rel_out;
> +	}

Why check this before the status check? If the poll timed out doesn't
it mean the FW did not respond?

>  	/* Check command completion */
>  	if (status) {
>  		wx_dbg(wx, "Command has failed with no status valid.\n");
> -
> -		buf[0] = rd32(wx, WX_MNG_MBOX);
> -		if ((buffer[0] & 0xff) != (~buf[0] >> 24)) {
> -			status = -EINVAL;
> -			goto rel_out;
> -		}
> -		if ((buf[0] & 0xff0000) >> 16 == 0x80) {
> -			wx_dbg(wx, "It's unknown cmd.\n");
> -			status = -EINVAL;
> -			goto rel_out;
> -		}
> 
> +		wx_dbg(wx, "check: %x %x\n", buffer[0] & 0xff, ~buf[0] >> 24);
> +		if ((buffer[0] & 0xff) != (~buf[0] >> 24))
> +			goto rel_out;

Inverse question here, I guess. Why is it only an error for FW not 
to be ready if cmd doesn't match?
-- 
pw-bot: cr

RE: [PATCH net] net: libwx: fix firmware mailbox abnormal return

[Jiawen Wu]

On Tue, Dec 31, 2024 10:12 AM, Jakub Kicinski wrote:
> On Thu, 26 Dec 2024 11:18:10 +0800 Jiawen Wu wrote:
> > Firmware writes back 'firmware ready' and 'unknown command' in the mailbox
> > message if there is an unknown command sent by driver. It tends to happen
> > with the use of custom firmware. So move the check for 'unknown command'
> > out of the poll timeout for 'firmware ready'. And adjust the debug log so
> > that mailbox messages are always printed when commands timeout.
> 
> The commit message doesn't really explain what the problem is,
> just what the code does. What is the problem you're solving
> and how does it impact the user?

The problem has not happened yet, but the code behavior is wrong. 
Follow this wrong flow, driver would never return error if there is a unknown
command. Since reading 'firmware ready' does not timeout. Thus driver would
mistakenly believe that the interaction has completed successfully.

> 
> > Fixes: 1efa9bfe58c5 ("net: libwx: Implement interaction with firmware")
> > Signed-off-by: Jiawen Wu <jiawenwu@trustnetic.com>
> > ---
> >  drivers/net/ethernet/wangxun/libwx/wx_hw.c | 22 ++++++++++------------
> >  1 file changed, 10 insertions(+), 12 deletions(-)
> >
> > diff --git a/drivers/net/ethernet/wangxun/libwx/wx_hw.c b/drivers/net/ethernet/wangxun/libwx/wx_hw.c
> > index 1bf9c38e4125..7059e0100c7c 100644
> > --- a/drivers/net/ethernet/wangxun/libwx/wx_hw.c
> > +++ b/drivers/net/ethernet/wangxun/libwx/wx_hw.c
> > @@ -334,27 +334,25 @@ int wx_host_interface_command(struct wx *wx, u32 *buffer,
> >  	status = read_poll_timeout(rd32, hicr, hicr & WX_MNG_MBOX_CTL_FWRDY, 1000,
> >  				   timeout * 1000, false, wx, WX_MNG_MBOX_CTL);
> >
> > +	buf[0] = rd32(wx, WX_MNG_MBOX);
> > +	if ((buf[0] & 0xff0000) >> 16 == 0x80) {
> > +		wx_dbg(wx, "It's unknown cmd.\n");
> > +		status = -EINVAL;
> > +		goto rel_out;
> > +	}
> 
> Why check this before the status check? If the poll timed out doesn't
> it mean the FW did not respond?

Firmware writes back 'firmware ready' and 'unknown command', so the poll
will not time out.

> 
> >  	/* Check command completion */
> >  	if (status) {
> >  		wx_dbg(wx, "Command has failed with no status valid.\n");
> > -
> > -		buf[0] = rd32(wx, WX_MNG_MBOX);
> > -		if ((buffer[0] & 0xff) != (~buf[0] >> 24)) {
> > -			status = -EINVAL;
> > -			goto rel_out;
> > -		}
> > -		if ((buf[0] & 0xff0000) >> 16 == 0x80) {
> > -			wx_dbg(wx, "It's unknown cmd.\n");
> > -			status = -EINVAL;
> > -			goto rel_out;
> > -		}
> >
> > +		wx_dbg(wx, "check: %x %x\n", buffer[0] & 0xff, ~buf[0] >> 24);
> > +		if ((buffer[0] & 0xff) != (~buf[0] >> 24))
> > +			goto rel_out;
> 
> Inverse question here, I guess. Why is it only an error for FW not
> to be ready if cmd doesn't match?

It is to check if the cmd has been processed by FW. If it matches, it means
that FW has already processed the cmd, but FWRDY timeout for some
unknown reason.
 

Re: [PATCH net] net: libwx: fix firmware mailbox abnormal return

[Jakub Kicinski]

On Thu, 2 Jan 2025 14:42:30 +0800 Jiawen Wu wrote:
> > >  	/* Check command completion */
> > >  	if (status) {
> > >  		wx_dbg(wx, "Command has failed with no status valid.\n");
> > > -
> > > -		buf[0] = rd32(wx, WX_MNG_MBOX);
> > > -		if ((buffer[0] & 0xff) != (~buf[0] >> 24)) {
> > > -			status = -EINVAL;
> > > -			goto rel_out;
> > > -		}
> > > -		if ((buf[0] & 0xff0000) >> 16 == 0x80) {
> > > -			wx_dbg(wx, "It's unknown cmd.\n");
> > > -			status = -EINVAL;
> > > -			goto rel_out;
> > > -		}
> > >
> > > +		wx_dbg(wx, "check: %x %x\n", buffer[0] & 0xff, ~buf[0] >> 24);
> > > +		if ((buffer[0] & 0xff) != (~buf[0] >> 24))
> > > +			goto rel_out;  
> > 
> > Inverse question here, I guess. Why is it only an error for FW not
> > to be ready if cmd doesn't match?  
> 
> It is to check if the cmd has been processed by FW. If it matches, it means
> that FW has already processed the cmd, but FWRDY timeout for some
> unknown reason.

Feels a little risky. AFAICT the value we're comparing is just 
a checksum (potentially always set to 0xff?) It doesn't seem to
identify the command very well. If the command timed out let's
always return an error.

The other explanations make sense, please include them in the commit
msg.