RE: rtw88: rtl8822cs AP mode not working

[Ping-Ke Shih <pkshih@realtek.com>]

+ Sascha for question 3)

(no other comments for now)

> -----Original Message-----
> From: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
> Sent: Wednesday, January 17, 2024 6:07 AM
> To: Ping-Ke Shih <pkshih@realtek.com>; Viacheslav <adeep@lexina.in>
> Cc: linux-wireless@vger.kernel.org; Jernej Škrabec <jernej.skrabec@gmail.com>
> Subject: Re: rtw88: rtl8822cs AP mode not working
> 
> Hi Ping-Ke and Viacheslav,
> 
> On Mon, Jan 15, 2024 at 3:17 AM Ping-Ke Shih <pkshih@realtek.com> wrote:
> [...]
> > > > I don't have this kind of wifi cards, could you help to check if AP mode
> > > > works in your side?
> > > I'll check that in the next few days.
> AP mode is also not working for me, I get the same problem as
> Viacheslav reported.
> 
> > > Also I'm wondering where code enables beacons (is it
> > > rtw_core_enable_beacon() or is there another relevant function?).
> > > Knowing that would be helpful to analyze this further.
> >
> > The main function to get and set beacon template to firmware is
> > rtw_fw_download_rsvd_page(). The basic concept is to put beacon frame via
> > qsel=BCN to a special TX FIFO area called "reserve page", and then
> > hardware/firmware will send beacon in interval of 100ms.
> Thanks for the explanation - that helped me find a better starting point!
> 
> I didn't have much time in the past few days, but I have some findings
> and questions:
> 1) I found the following comment/code in the downstream driver [0]:
> /*
>  * Disable Hw protection for a time which revserd for Hw sending beacon.
>  * Fix download reserved page packet fail that access collision with
> the protection time.
>  */
> val8 = rtw_read8(adapter, REG_BCN_CTRL_8822C);
> restore[1] = val8;
> val8 &= ~BIT_EN_BCN_FUNCTION_8822C;
> val8 |= BIT_DIS_TSF_UDT_8822C;
> rtw_write8(adapter, REG_BCN_CTRL_8822C, val8);
> 
> This is not part of the upstream rtw88 driver, so I made a patch and
> attached it.
> Unfortunately it doesn't fix the problem.
> 
> 2) PCI is the only HCI which does not need the checksum in the
> pkt_Info (USB and SDIO need the checksum).
> The checksum is added by rtw_tx_fill_txdesc_checksum() which is only
> called in usb.c and sdio.c.
> My understanding is that for reserved pages we can have more than one
> pkt_info in the buffer (my starting point for this thought is
> rtw_fill_rsvd_page_desc() from fw.c).
> In usb.c and sdio.c we're only calculating the checksum for the very
> first pkt_info, not for any subsequent ones (I didn't even know that
> it's possible to have more than one pkt_Info outside of RX and TX
> aggregation).
> However, it seems that the downstream code calculates the TX checksum
> for *all* pkt_info in the buffer, see [1]
> This code is missing from rtw88 at the moment. Since I didn't have
> time I did not try to implement this yet.
> 
> 3) Has anybody tried AP mode with rtw88 on a (supported) USB chipset?
> If my thought (from #2) is correct then AP mode would show the same
> problems there.
> 
> 4) Viacheslav, I think you previously mentioned that you did a bit of
> work with the downstream driver.
> It would be awesome if you could also take a look at the rtw88 and
> downstream driver code and start comparing them (logic that's
> different or completely missing from rtw88 is suspicious).
> 
> 
> Best regards,
> Martin
> 
> 
> [0]
> https://github.com/chewitt/RTL8822CS/blob/60cd82134d63aa9436b43c42933a86d6e5a191ba/hal/rtl8822c/rtl882
> 2c_ops.c#L1885-L1893
> [1] https://github.com/chewitt/RTL8822CS/blob/main/hal/rtl8822c/sdio/rtl8822cs_xmit.c#L311-L312