Re: [PATCH v1] mt76: mt7915: wed: find rx token by physical address

[Felix Fietkau <nbd@nbd.name>]

On 12.03.25 02:41, Ping-Ke Shih wrote:
> Shengyu Qu <wiagn233@outlook.com> wrote:
>> 
>> Hi Shih,
>> 
>> Can you also give this patch a review? Seems I also need to convert to
>> reverse Xmas tree order for this patch, will do it in next version.
> 
> Yes. Please do it.
> 
>> 
>> Best regards,
>> Shengyu
>> 
>> 在 2025/2/22 19:38, Shengyu Qu 写道:
>> > The token id in RxDMAD may be incorrect when it is not the last frame
>> > due to WED HW bug. Lookup correct token id by physical address in sdp0.
>> >
>> > Downstream patch link:
>> https://git01.mediatek.com/plugins/gitiles/openwrt/feeds/mtk-openwrt-feeds/+/737340322ab22b138fd200e02
>> 0d61ffdbe3e36a9/autobuild/autobuild_5.4_mac80211_release/mt7988_wifi7_mac80211_mlo/package/kernel/mt76
>> /patches/0062-mtk-wifi-mt76-mt7915-wed-find-rx-token-by-physical-a.patch
>> >
>> > Signed-off-by: Peter Chiu <chui-hao.chiu@mediatek.com>
>> > Signed-off-by: Shengyu Qu <wiagn233@outlook.com>
>> > ---
>> >   drivers/net/wireless/mediatek/mt76/dma.c | 25 +++++++++++++++++++++++-
>> >   1 file changed, 24 insertions(+), 1 deletion(-)
>> >
>> > diff --git a/drivers/net/wireless/mediatek/mt76/dma.c b/drivers/net/wireless/mediatek/mt76/dma.c
>> > index 844af16ee551..5bf63014263c 100644
>> > --- a/drivers/net/wireless/mediatek/mt76/dma.c
>> > +++ b/drivers/net/wireless/mediatek/mt76/dma.c
>> > @@ -444,9 +444,32 @@ mt76_dma_get_buf(struct mt76_dev *dev, struct mt76_queue *q, int idx,
>> >   	mt76_dma_should_drop_buf(drop, ctrl, buf1, desc_info);
>> >
>> >   	if (mt76_queue_is_wed_rx(q)) {
>> > +		u32 id, find = 0;
>> >   		u32 token = FIELD_GET(MT_DMA_CTL_TOKEN, buf1);
>> > -		struct mt76_txwi_cache *t = mt76_rx_token_release(dev, token);
>> > +		struct mt76_txwi_cache *t;
>> > +
>> > +		if (*more) {
> 
> Because of scope, should define 'find' (or 'found'?) as bool here.
> 
>> > +			spin_lock_bh(&dev->rx_token_lock);
>> > +
>> > +			idr_for_each_entry(&dev->rx_token, t, id) {
> 
> idr_find() can only get a pointer associated with the ID. Not sure if it can
> have a reverse way -- get ID from pointer. With sequential search in data path,
> cost is higher.
> 
> Maybe you can consider to build association between argument 'idx' and
> pointer/token by another patch.
> 
> But logic is okay to me.

Adding to that, it matters how frequently the token ID is incorrect. If 
the token ID is correct most of the time, then it's worth optimizing for 
that as the default case. In that case you should avoid the extra idr 
scan if the pointer for the provided token id already matches.

- Felix