Re: [PATCH v1] wifi: ath9k: fix AR9003 mac hardware hang check register offset calculation

[Peter Seiderer]

Hello Simon,

On Sat, 22 Apr 2023 10:50:52 +0200, Simon Horman <simon.horman@xxxxxxxxxxxx> wrote:

> On Thu, Apr 20, 2023 at 10:43:16PM +0200, Peter Seiderer wrote:
> > Fix ath9k_hw_verify_hang()/ar9003_hw_detect_mac_hang() register offset
> > calculation (do not overflow the shift for the second register/queues
> > above five, use the register layout described in the comments above
> > ath9k_hw_verify_hang() instead).
> >
> > Fixes: 222e04830ff0 ("ath9k: Fix MAC HW hang check for AR9003")
> >
> > Reported-by: Gregg Wonderly <greggwonderly@xxxxxxxxxxxxxx>
> > Link: https://lore.kernel.org/linux-wireless/E3A9C354-0CB7-420C-ADEF-F0177FB722F4@xxxxxxxxxxxxxx/
> > Signed-off-by: Peter Seiderer <ps.report@xxxxxxx>
> > ---
> > Notes:
> >   - tested with MikroTik R11e-5HnD/Atheros AR9300 Rev:4 (lspci: 168c:0033
> >     Qualcomm Atheros AR958x 802.11abgn Wireless Network Adapter (rev 01))
> >     card
> > ---
> >  drivers/net/wireless/ath/ath9k/ar9003_hw.c | 27 ++++++++++++++--------
> >  1 file changed, 18 insertions(+), 9 deletions(-)
> >
> > diff --git a/drivers/net/wireless/ath/ath9k/ar9003_hw.c b/drivers/net/wireless/ath/ath9k/ar9003_hw.c
> > index 4f27a9fb1482..0ccf13a35fb4 100644
> > --- a/drivers/net/wireless/ath/ath9k/ar9003_hw.c
> > +++ b/drivers/net/wireless/ath/ath9k/ar9003_hw.c
> > @@ -1099,17 +1099,22 @@ static bool ath9k_hw_verify_hang(struct ath_hw *ah, unsigned int queue)
> >  {
> >  	u32 dma_dbg_chain, dma_dbg_complete;
> >  	u8 dcu_chain_state, dcu_complete_state;
> > +	unsigned int dbg_reg, reg_offset;
> >  	int i;
> >
> > -	for (i = 0; i < NUM_STATUS_READS; i++) {
> > -		if (queue < 6)
> > -			dma_dbg_chain = REG_READ(ah, AR_DMADBG_4);
> > -		else
> > -			dma_dbg_chain = REG_READ(ah, AR_DMADBG_5);
> > +	if (queue < 6) {
> > +		dbg_reg = AR_DMADBG_4;
> > +		reg_offset = i * 5;
>
> Hi Peter,
>
> unless my eyes are deceiving me, i is not initialised here.
>
> > +	} else {
> > +		dbg_reg = AR_DMADBG_5;
> > +		reg_offset = (i - 6) * 5;
>
> Or here.

You are absolutely right, it should be queue instead if i here...,
will provide (and test) an updated version of the patch...,
many thanks for review!

Regards,
Peter


>
> > +	}
> >
> > +	for (i = 0; i < NUM_STATUS_READS; i++) {
> > +		dma_dbg_chain = REG_READ(ah, dbg_reg);
> >  		dma_dbg_complete = REG_READ(ah, AR_DMADBG_6);
> >
> > -		dcu_chain_state = (dma_dbg_chain >> (5 * queue)) & 0x1f;
> > +		dcu_chain_state = (dma_dbg_chain >> reg_offset) & 0x1f;
> >  		dcu_complete_state = dma_dbg_complete & 0x3;
> >
> >  		if ((dcu_chain_state != 0x6) || (dcu_complete_state != 0x1))