Arend van Spriel <arend.vanspriel@xxxxxxxxxxxx> writes:
> From: Arend van Spriel <aspriel@xxxxxxxxx>
>
> With the addition of TXQ stats in the per-tid statistics the struct
> station_info grew significantly. This resulted in stack limit warnings
> like below:
>
> CC [M] drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.o
> drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c:
> In function ‘brcmf_notify_connect_status_ap’:
> drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c:5530:1:
> error: the frame size of 1592 bytes is larger than 1024 bytes
>
> This patch adds an allocation function that those who want to provide
> per-tid stats should use to allocate the tid array, ie.
> struct station_info::pertid.
>
> Cc: Toke Høiland-Jørgensen <toke@xxxxxxx>
> Fixes: 52539ca89f36 ("cfg80211: Expose TXQ stats and parameters to
> userspace")
> Signed-off-by: Arend van Spriel <aspriel@xxxxxxxxx>
> ---
> + linux-wireless list
>
> Johannes, Toke,
>
> Here an alternative approach. Currently the only cfg80211-based driver
> providing per-tid stats is mac80211. This patch only changes mac80211
> and the other driver can keep using stack allocation. Even mac80211 could
> if wanted, but I left that part as is.
Hmm, yeah, that would work too. Though I worry that mixing dynamically
and statically allocated fields is going to result in errors further
down the road? I guess it also depends on whether struct station_info is
likely to grow more for other reasons, in which case on-stack allocation
needs to be changed anyway.
I'm fine with either approach; I'm happy to let it be up to our friendly
maintainers to decide which approach they prefer :)
-Toke
