| Message ID | 20201126171553.2097-1-youghand@codeaurora.org (mailing list archive) |
|---|---|
| State | Changes Requested |
| Delegated to: | Kalle Valo |
| Headers | show |
| Series | [v2] ath10k: skip the wait for completion to recovery in shutdown path | expand |
On Thu, Nov 26, 2020 at 9:16 AM Youghandhar Chintala <youghand@codeaurora.org> wrote: > --- a/drivers/net/wireless/ath/ath10k/snoc.c > +++ b/drivers/net/wireless/ath/ath10k/snoc.c > @@ -1790,9 +1790,6 @@ static int ath10k_snoc_remove(struct platform_device *pdev) > > reinit_completion(&ar->driver_recovery); > > - if (test_bit(ATH10K_SNOC_FLAG_RECOVERY, &ar_snoc->flags)) > - wait_for_completion_timeout(&ar->driver_recovery, 3 * HZ); Hmm, this is the only instance of waiting for this completion, which means that after this patch, 'ar->driver_recovery' is doing exactly nothing. Should you instead just remove it completely? Also, if your patch is correct, it seems like the completion was never needed in the first place. You should probably address such a claim in the commit message; is there truly no need to wait here? Or was there some purpose here, but that purpose was accomplished some other way? Or was there a purpose, and that purpose was misguided? It feels to me like it is indeed correct to remove this (shutdown should be performed promptly; we don't need to delay it just to try to "finish recovering"), but it's your job to convince the reader. Brian > - > set_bit(ATH10K_SNOC_FLAG_UNREGISTERING, &ar_snoc->flags); > > ath10k_core_unregister(ar);
Brian Norris <briannorris@chromium.org> writes: > On Thu, Nov 26, 2020 at 9:16 AM Youghandhar Chintala > <youghand@codeaurora.org> wrote: >> --- a/drivers/net/wireless/ath/ath10k/snoc.c >> +++ b/drivers/net/wireless/ath/ath10k/snoc.c >> @@ -1790,9 +1790,6 @@ static int ath10k_snoc_remove(struct platform_device *pdev) >> >> reinit_completion(&ar->driver_recovery); >> >> - if (test_bit(ATH10K_SNOC_FLAG_RECOVERY, &ar_snoc->flags)) >> - wait_for_completion_timeout(&ar->driver_recovery, 3 * HZ); > > Hmm, this is the only instance of waiting for this completion, which > means that after this patch, 'ar->driver_recovery' is doing exactly > nothing. Should you instead just remove it completely? > > Also, if your patch is correct, it seems like the completion was never > needed in the first place. You should probably address such a claim in > the commit message; is there truly no need to wait here? Or was there > some purpose here, but that purpose was accomplished some other way? > Or was there a purpose, and that purpose was misguided? It feels to me > like it is indeed correct to remove this (shutdown should be performed > promptly; we don't need to delay it just to try to "finish > recovering"), but it's your job to convince the reader. Exactly what I was thinking as well. To me this patch was just looks racy and all the commit log says that it's "unwanted delay".
On Thu, Nov 26, 2020 at 9:16 AM Youghandhar Chintala <youghand@codeaurora.org> wrote: > --- a/drivers/net/wireless/ath/ath10k/snoc.c > +++ b/drivers/net/wireless/ath/ath10k/snoc.c > @@ -1790,9 +1790,6 @@ static int ath10k_snoc_remove(struct platform_device *pdev) > > reinit_completion(&ar->driver_recovery); > > - if (test_bit(ATH10K_SNOC_FLAG_RECOVERY, &ar_snoc->flags)) > - wait_for_completion_timeout(&ar->driver_recovery, 3 * HZ); You are skipping recovery in ath10k_snoc_remove() which is a remove callback and also called in shutdown callback. So that means it is also called when you unload the ath10k module and not just when the system reboots/shutdown. While it makes sense to not skip recovery in shutdown/reboot sequence because the system is going down, it might very well be needed in case of unloading the module because we expect the system to be up and stable after unloading the ath10k module and we should be able to reload the ath10k module smoothly. If you remove that now and try to reload the ath10k module, won't that leave the system in possibly an inconsistent state because we skipped recovery in module remove and then we are trying to load the ath10k module when the recovery is not yet complete? In other words, you need to test ath10k module load/unload as well in addition to reboot tests to make sure this works as expected or else you will need a separate shutdown callback which skips the recovery part. Thanks, Sai
diff --git a/drivers/net/wireless/ath/ath10k/snoc.c b/drivers/net/wireless/ath/ath10k/snoc.c index 84666f72bdfa..15580a91fe98 100644 --- a/drivers/net/wireless/ath/ath10k/snoc.c +++ b/drivers/net/wireless/ath/ath10k/snoc.c @@ -1790,9 +1790,6 @@ static int ath10k_snoc_remove(struct platform_device *pdev) reinit_completion(&ar->driver_recovery); - if (test_bit(ATH10K_SNOC_FLAG_RECOVERY, &ar_snoc->flags)) - wait_for_completion_timeout(&ar->driver_recovery, 3 * HZ); - set_bit(ATH10K_SNOC_FLAG_UNREGISTERING, &ar_snoc->flags); ath10k_core_unregister(ar);
Currently in the shutdown callback we wait for recovery to complete before freeing up the resources. This can lead to unwanted delay during the shutdown and thereby increase the shutdown time. As an attempt to take less time during shutdown, remove the wait for recovery completion in the shutdown callback. Tested-on: WCN3990 hw1.0 SNOC WLAN.HL.3.1-01040-QCAHLSWMTPLZ-1 Signed-off-by: Youghandhar Chintala <youghand@codeaurora.org> --- Changes from v1: -Removed stray change-id text --- drivers/net/wireless/ath/ath10k/snoc.c | 3 --- 1 file changed, 3 deletions(-)