iwlwifi: network not back after resuming from suspend since v6.15-rc2

[Coiby Xu <coiby.xu@gmail.com>]

Hi Emmanuel,

I notice since v6.15-rc2, most of the times the network won't be back
again after resuming from suspend. And sometimes I need to run
"systemctl suspend" for 8 or 9 consecutive times to make WIFI available
again. bisect and manual reverting show 15220a257319 ("wifi: iwlwifi:
don't warn if the NIC is gone in resume") is the 1st bad commit. 

Here are some other clues that may help resolve this issue,

     1. An iwlwifi setting [1] can make this issue disappear,
        options iwlwifi swcrypto=0
        options iwlwifi bt_coex_active=0
        options iwlwifi power_save=0
        options iwlwifi uapsd_disable=1
        options iwlmvm power_scheme=1

     2. Before v6.15-rc2, the issue rarely happens. And if it happens,
        simply suspending once can bring the network online again. Other
        users seem to have experienced the same issue e.g. [2][3][4].

     3. Suspending by "rtcwake -m mem -s 5" instead of by "systemcl suspend"
        can make it much easier to reproduce this issue for kernels older
        than v6.15-rc2.

If you need more details from me, feel free to let me know! And I'll
very happy to test any possible solutions. Thanks!

## Notes on bisect

The bisect is done with an experimental branch [5] of
kernel-auto-bisect tool [6].

1. bisect conf file
    TEST_STRATEGY="simple"
    RUNS_PER_COMMIT=6
    RUNS_PER_GOOD_COMMIT=3
    GOOD_COMMIT=6.14.4-300.fc42.x86_64
    BAD_COMMIT=6.19.14-200.fc43.x86_64
    REPRODUCER_SCRIPT=reproducer_wifi.sh

2. Reproducer 
    on_test() {
            # If sleep only 5 or 10 s, somehow the system will get suspended forever i.e.
            # "rtcwake -m on won't wake up the system .
            sleep 15
            ip --brief addr show wlp4s0 
            # "rtcwake -m mem -s 5" will quite often makes wifi unavailable. So use
            # systemctl suspend instead
            #rtcwake -m mem -s 5
            # Wake up the system after 4s
            rtcwake -m on -s 4 &
            systemctl suspend
            sleep 12
            ip --brief addr show wlp4s0 | grep -q " UP "
    }

kernel-auto-bisect went through a two-phase bisection,
     1. find 6.14.6-300.fc42.x86_64 as the 1st bad build
     2. find Fedora kernel commit 6453e892cf86c [7] as 1st bad commit

I can't bisect against upstream kernel directly because v6.15-rc2 won't
be able to mount my root partition. So I manually reverted upstream
15220a257319 ("wifi: iwlwifi: don't warn if the NIC is gone in resume")
in v6.15 to confirm it's indeed the bad commit.

[1] https://bbs.archlinux.org/viewtopic.php?pid=2227202#p2227202
[2] https://bbs.archlinux.org/viewtopic.php?id=301971
[3] https://discussion.fedoraproject.org/t/random-wi-fi-adapter-loss-on-fedora-42-intel-alder-lake-p-cnvi/152878
[4] https://forums.opensuse.org/t/wifi-suddenly-no-more-available/184399
[5] https://github.com/coiby/kernel-auto-bisect/tree/random_issue_bisect
[6] https://github.com/rhkdump/kernel-auto-bisect
[7] https://gitlab.com/cki-project/kernel-ark/-/commit/6453e892cf86c900d51f5884e9013e0dbeeea4ad


-- 
Best regards,
Coiby