2011-12-06 01:30:23

by Larry Finger

Subject: Strange code in routine wlc_phy_radio_init_2057()

In looking at the support code for Type 2057 radios in brcmsmac, I noticed the
following routine:

static void wlc_phy_radio_init_2057(struct brcms_phy *pi)
{
struct radio_20xx_regs *regs_2057_ptr = NULL;

if (NREV_IS(pi->pubpi.phy_rev, 7)) {
regs_2057_ptr = regs_2057_rev4;
} else if (NREV_IS(pi->pubpi.phy_rev, 8)
|| NREV_IS(pi->pubpi.phy_rev, 9)) {
switch (pi->pubpi.radiorev) {
case 5:
if (pi->pubpi.radiover == 0x0)
regs_2057_ptr = regs_2057_rev5;
else if (pi->pubpi.radiover == 0x1)
regs_2057_ptr = regs_2057_rev5v1;
else
break;
case 7:
regs_2057_ptr = regs_2057_rev7;
break;
case 8:
regs_2057_ptr = regs_2057_rev8;
break;
default:
break;
}
}
wlc_phy_init_radio_regs_allbands(pi, regs_2057_ptr);
}

I do not think that the code for phy revisions 8 or 9, and radio revision 5 is
correct. When the radio version is 0, or 1, the case statement will fall through
to the radio revision 7 case, thereby negating the logic that sets regs_2057_ptr
for case 5.

I think the following patch will fix it:

Index: wireless-testing-new/drivers/net/wireless/brcm80211/brcmsmac/phy/phy_n.c
===================================================================
--- wireless-testing-new.orig/drivers/net/wireless/brcm80211/brcmsmac/phy/phy_n.c
+++ wireless-testing-new/drivers/net/wireless/brcm80211/brcmsmac/phy/phy_n.c
@@ -19996,8 +19996,8 @@ static void wlc_phy_radio_init_2057(stru
regs_2057_ptr = regs_2057_rev5;
else if (pi->pubpi.radiover == 0x1)
regs_2057_ptr = regs_2057_rev5v1;
- else
- break;
+
+ break;

case 7:


Larry

2011-12-14 20:07:39

by Arend van Spriel

Subject: Re: Strange code in routine wlc_phy_radio_init_2057()

On 12/06/2011 02:30 AM, Larry Finger wrote:
> In looking at the support code for Type 2057 radios in brcmsmac, I noticed the
> following routine:
>
> static void wlc_phy_radio_init_2057(struct brcms_phy *pi)
> {
> struct radio_20xx_regs *regs_2057_ptr = NULL;
>
> if (NREV_IS(pi->pubpi.phy_rev, 7)) {
> regs_2057_ptr = regs_2057_rev4;
> } else if (NREV_IS(pi->pubpi.phy_rev, 8)
> || NREV_IS(pi->pubpi.phy_rev, 9)) {
> switch (pi->pubpi.radiorev) {
> case 5:
> if (pi->pubpi.radiover == 0x0)
> regs_2057_ptr = regs_2057_rev5;
> else if (pi->pubpi.radiover == 0x1)
> regs_2057_ptr = regs_2057_rev5v1;
> else
> break;
> case 7:
> regs_2057_ptr = regs_2057_rev7;
> break;
> case 8:
> regs_2057_ptr = regs_2057_rev8;
> break;
> default:
> break;
> }
> }
> wlc_phy_init_radio_regs_allbands(pi, regs_2057_ptr);
> }
>
> I do not think that the code for phy revisions 8 or 9, and radio revision 5 is
> correct. When the radio version is 0, or 1, the case statement will fall through
> to the radio revision 7 case, thereby negating the logic that sets regs_2057_ptr
> for case 5.

You are right.

>
> I think the following patch will fix it:
>
> Index: wireless-testing-new/drivers/net/wireless/brcm80211/brcmsmac/phy/phy_n.c
> ===================================================================
> --- wireless-testing-new.orig/drivers/net/wireless/brcm80211/brcmsmac/phy/phy_n.c
> +++ wireless-testing-new/drivers/net/wireless/brcm80211/brcmsmac/phy/phy_n.c
> @@ -19996,8 +19996,8 @@ static void wlc_phy_radio_init_2057(stru
> regs_2057_ptr = regs_2057_rev5;
> else if (pi->pubpi.radiover == 0x1)
> regs_2057_ptr = regs_2057_rev5v1;
> - else
> - break;
> +
> + break;
>
> case 7:
>
>
> Larry
>

It is an improvement for sure, but it needs to be slightly different. I
will create a patch for this. Thanks for pointing us at this.

Gr. AvS