On Thu, 2022-06-23 at 11:50 +0800, Ping-Ke Shih wrote:
> On Mon, 2022-06-20 at 12:34 +0000, Ping-Ke Shih wrote:
> > On Mon, 2022-06-20 at 13:09 +0300, Kalle Valo wrote:
> > > Ping-Ke Shih <pkshih@xxxxxxxxxxx> writes:
> > >
> > > > From: Zong-Zhe Yang <kevin_yang@xxxxxxxxxxx>
> > > >
> > > > If a chip is configured to support mac80211 chanctx ops, we avoid
> > > > using older FW that does not support HW scan to make mac80211 stack
> > > > handle scanning as expected.
> > > >
> > > > Signed-off-by: Zong-Zhe Yang <kevin_yang@xxxxxxxxxxx>
> > > > Signed-off-by: Ping-Ke Shih <pkshih@xxxxxxxxxxx>
> > > > ---
> > > > drivers/net/wireless/realtek/rtw89/fw.c | 8 ++++++++
> > > > 1 file changed, 8 insertions(+)
> > > >
> > > > diff --git a/drivers/net/wireless/realtek/rtw89/fw.c
> > > > b/drivers/net/wireless/realtek/rtw89/fw.c
> > > > index 0e12629f789c3..a47451dc9d81d 100644
> > > > --- a/drivers/net/wireless/realtek/rtw89/fw.c
> > > > +++ b/drivers/net/wireless/realtek/rtw89/fw.c
> > > > @@ -250,6 +250,7 @@ static void rtw89_fw_recognize_features(struct rtw89_dev *rtwdev)
> > > >
> > > > int rtw89_fw_recognize(struct rtw89_dev *rtwdev)
> > > > {
> > > > + const struct rtw89_chip_info *chip = rtwdev->chip;
> > > > int ret;
> > > >
> > > > ret = __rtw89_fw_recognize(rtwdev, RTW89_FW_NORMAL);
> > > > @@ -261,6 +262,13 @@ int rtw89_fw_recognize(struct rtw89_dev *rtwdev)
> > > >
> > > > rtw89_fw_recognize_features(rtwdev);
> > > >
> > > > + if (chip->support_chanctx_num != 0 &&
> > > > + !RTW89_CHK_FW_FEATURE(SCAN_OFFLOAD, &rtwdev->fw)) {
> > > > + rtw89_err(rtwdev,
> > > > + "require newer FW to support HW scan for chanctx\n");
> > > > + return -ENOENT;
> > > > + }
> > >
> > > So if the user has not update the firmware a kernel upgrade will break
> > > their internet? That's not good, we should not break existing setups. So
> > > what firmware version is required?
> > >
> >
> > Firmware version 0.13.35.0 is required. The firmware has been in
> > linux-firmware repository on 2022-02-18. I think people being able
> > to update kernel can update firmware as well.
> >
> > The alternative ways could be
> > 1. add a module parameter, like no_channel_context. We can add a
> > prompt to note people can set it to 1 for old firmware.
> >
> > 2. wait version of request_firmware() as first step of pci probe.
> > The probe could cost longer time, because currently we use
> > request_firmware_nowait() and continue to initialize in parallel.
> > More, hw->priv isn't allocated at that moment, so it could be not
> > so straightforward.
> >
>
> To prevent break users' existing setups, modified approach 1 is adopted
> by patchset v2. We add a module parameter rtw89_use_chanctx, and disable
> it by default. So, users' setups with old firmware can still work.
>
> If a user wants channel context to support concurrency, he can set
> rtw89_use_chanctx=1 and upgrade firmware.
>
> I think this could be a better user experience of kernel.
>
We have a better idea that load firmware header and parse firmware
features when probing before ieee80211_register_hw(). If firmware
doesn't support hw_scan, it still can work as original. So, users
don't need to update firmware or set module parameters.
I have sent v3 with this idea that is friendly to users.
Ping-Ke
