Re: [PATCH wireless-next] wifi: brcm80211: change current_bss to a FAM

On Fri, Apr 17, 2026 at 12:15 PM Arend van Spriel
<[email protected]> wrote:
>
> On 01/04/2026 01:32, Rosen Penev wrote:
> > Change to a single allocation with kzalloc_flex and remove some
> > boilerplate.
> >
> > Done as a FAM instead of value to keep -> usage in codebase.
>
> I get it and it looks a bit like a trick, but....
Used in many places in the kernel:

git grep kzalloc_flex | grep ,\ 1 | wc -l
33
git grep struct_size | grep 1\) | wc -l
99
>
> > Signed-off-by: Rosen Penev <[email protected]>
> > ---
> >   .../broadcom/brcm80211/brcmsmac/main.c        | 32 ++-----------------
> >   .../broadcom/brcm80211/brcmsmac/main.h        |  2 +-
> >   2 files changed, 3 insertions(+), 31 deletions(-)
> >
> > diff --git a/drivers/net/wireless/broadcom/brcm80211/brcmsmac/main.c b/drivers/net/wireless/broadcom/brcm80211/brcmsmac/main.c
> > index c7eaf160e1fa..4f57d64b1dd3 100644
> > --- a/drivers/net/wireless/broadcom/brcm80211/brcmsmac/main.c
> > +++ b/drivers/net/wireless/broadcom/brcm80211/brcmsmac/main.c
> > @@ -418,21 +418,12 @@ static int brcms_chspec_bw(u16 chanspec)
> >       return BRCMS_10_MHZ;
> >   }
> >
> > -static void brcms_c_bsscfg_mfree(struct brcms_bss_cfg *cfg)
> > -{
> > -     if (cfg == NULL)
> > -             return;
> > -
> > -     kfree(cfg->current_bss);
> > -     kfree(cfg);
> > -}
> > -
>
> ... it does clearly give some code reduction.
>
> >   static void brcms_c_detach_mfree(struct brcms_c_info *wlc)
> >   {
> >       if (wlc == NULL)
> >               return;
> >
> > -     brcms_c_bsscfg_mfree(wlc->bsscfg);
> > +     kfree(wlc->bsscfg);
> >       kfree(wlc->pub);
> >       kfree(wlc->modulecb);
> >       kfree(wlc->default_bss);
>
> [...]
>
> > @@ -527,7 +499,7 @@ brcms_c_attach_malloc(uint unit, uint *err, uint devid)
> >               goto fail;
> >       }
> >
> > -     wlc->bsscfg = brcms_c_bsscfg_malloc(unit);
> > +     wlc->bsscfg = kzalloc_flex(*wlc->bsscfg, current_bss, 1, GFP_ATOMIC);
> >       if (wlc->bsscfg == NULL) {
> >               *err = 1011;
> >               goto fail;
> > diff --git a/drivers/net/wireless/broadcom/brcm80211/brcmsmac/main.h b/drivers/net/wireless/broadcom/brcm80211/brcmsmac/main.h
> > index b7ca0d9891c4..e9cbb95a5a17 100644
> > --- a/drivers/net/wireless/broadcom/brcm80211/brcmsmac/main.h
> > +++ b/drivers/net/wireless/broadcom/brcm80211/brcmsmac/main.h
> > @@ -614,7 +614,7 @@ struct brcms_bss_cfg {
> >       u8 SSID_len;
> >       u8 SSID[IEEE80211_MAX_SSID_LEN];
> >       u8 BSSID[ETH_ALEN];
> > -     struct brcms_bss_info *current_bss;
> > +     struct brcms_bss_info current_bss[];
>
> It does feel icky to use a flex array for 1:1 object relation. It seems
> the lifetime of both objects is exactly the same so I would consider
> using the value instead of FAM. It would give the same code reduction,
> but making the patch slightly larger.
>
> main.c: kfree(cfg->current_bss);
> main.c: cfg->current_bss = kzalloc_obj(*cfg->current_bss, GFP_ATOMIC);
> main.c: if (cfg->current_bss == NULL)
> main.c:                 wlc->bsscfg->current_bss->chanspec = chanspec;
> main.c:         rs = &wlc->bsscfg->current_bss->rateset;
> main.c:                 mcsset_bss = wlc->bsscfg->current_bss;
> main.c:         bi = wlc->bsscfg->current_bss->beacon_period << 10;
>
> So it is used in just 4 places not counting the first 3 hits for obvious
> reasons. Not too much effort to use value here so please do that.
Sure. Will send when the merge window opens.
>
> Regards,
> Arend