| Message ID | 1375430329-23352-1-git-send-email-bartosz.markowski@tieto.com (mailing list archive) |
|---|---|
| State | Not Applicable, archived |
| Headers | show |
On 2 August 2013 09:58, Bartosz Markowski <bartosz.markowski@tieto.com> wrote: > Unify the PCI options location. > > By default the SoC PS option is disabled to boost the > performance and due to poor stability on early HW revisions. > In future we can remove the module parameter and turn on/off > the PS for given hardware. > > This change also makes the pci module parameter for SoC PS static. > > Signed-off-by: Bartosz Markowski <bartosz.markowski@tieto.com> Hmm.. I'm worried about pci wake/sleep locking: (a) pci_wake() (awake count =1) (a) do something Now, if pci_sleep() and pci_wake() happen simultaneously/are preempted splitting primitive operations to: (a) pci_sleep(): decrease awake count (=0) (b) pci_wake(): increase awake count (=1), iowrite, wait for awake, return (a) pci_sleep(): iowrite(). return The flow in (b) now thinks the device is awake, but it's not guaranteed to be anymore because it has been put to sleep by (a). Pozdrawiam / Best regards, Micha? Kazior. -- To unsubscribe from this list: send the line "unsubscribe linux-wireless" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Michal Kazior <michal.kazior@tieto.com> writes: > On 2 August 2013 09:58, Bartosz Markowski <bartosz.markowski@tieto.com> wrote: >> Unify the PCI options location. >> >> By default the SoC PS option is disabled to boost the >> performance and due to poor stability on early HW revisions. >> In future we can remove the module parameter and turn on/off >> the PS for given hardware. >> >> This change also makes the pci module parameter for SoC PS static. >> >> Signed-off-by: Bartosz Markowski <bartosz.markowski@tieto.com> > > Hmm.. I'm worried about pci wake/sleep locking: > > (a) pci_wake() (awake count =1) > (a) do something > Now, if pci_sleep() and pci_wake() happen simultaneously/are > preempted splitting primitive operations to: > (a) pci_sleep(): decrease awake count (=0) > (b) pci_wake(): increase awake count (=1), iowrite, wait for awake, return > (a) pci_sleep(): iowrite(). return > The flow in (b) now thinks the device is awake, but it's not > guaranteed to be anymore because it has been put to sleep by (a). Yeah, there's a race alright. I wonder if I have ever seen a proper use of a atomic variable in a wireless driver :) But this doesn't prevent taking this patch as the race has existed since the beginning, right?
On 5 August 2013 18:40, Kalle Valo <kvalo@qca.qualcomm.com> wrote: > Michal Kazior <michal.kazior@tieto.com> writes: > >> On 2 August 2013 09:58, Bartosz Markowski <bartosz.markowski@tieto.com> wrote: >>> Unify the PCI options location. >>> >>> By default the SoC PS option is disabled to boost the >>> performance and due to poor stability on early HW revisions. >>> In future we can remove the module parameter and turn on/off >>> the PS for given hardware. >>> >>> This change also makes the pci module parameter for SoC PS static. >>> >>> Signed-off-by: Bartosz Markowski <bartosz.markowski@tieto.com> >> >> Hmm.. I'm worried about pci wake/sleep locking: >> >> (a) pci_wake() (awake count =1) >> (a) do something >> Now, if pci_sleep() and pci_wake() happen simultaneously/are >> preempted splitting primitive operations to: >> (a) pci_sleep(): decrease awake count (=0) >> (b) pci_wake(): increase awake count (=1), iowrite, wait for awake, return >> (a) pci_sleep(): iowrite(). return >> The flow in (b) now thinks the device is awake, but it's not >> guaranteed to be anymore because it has been put to sleep by (a). > > Yeah, there's a race alright. I wonder if I have ever seen a proper use > of a atomic variable in a wireless driver :) > > But this doesn't prevent taking this patch as the race has existed since > the beginning, right? I don't have anything against it. I just noticed the race and thought it's good to point out the issue. Pozdrawiam / Best regards, Micha? Kazior. -- To unsubscribe from this list: send the line "unsubscribe linux-wireless" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Michal Kazior <michal.kazior@tieto.com> writes: > On 5 August 2013 18:40, Kalle Valo <kvalo@qca.qualcomm.com> wrote: > >> But this doesn't prevent taking this patch as the race has existed since >> the beginning, right? > > I don't have anything against it. That's what I assumed as well, just wanted to be sure before I commit the patch. > I just noticed the race and thought it's good to point out the issue. Yes, that was a very good find. Please continue finding more races in the code :)
Bartosz Markowski <bartosz.markowski@tieto.com> writes: > Unify the PCI options location. > > By default the SoC PS option is disabled to boost the > performance and due to poor stability on early HW revisions. > In future we can remove the module parameter and turn on/off > the PS for given hardware. > > This change also makes the pci module parameter for SoC PS static. > > Signed-off-by: Bartosz Markowski <bartosz.markowski@tieto.com> Thanks, applied. But there were conflicts, please double check that I didn't break anything.
diff --git a/drivers/net/wireless/ath/ath10k/pci.c b/drivers/net/wireless/ath/ath10k/pci.c index d95439b..b351e23 100644 --- a/drivers/net/wireless/ath/ath10k/pci.c +++ b/drivers/net/wireless/ath/ath10k/pci.c @@ -32,7 +32,7 @@ #include "ce.h" #include "pci.h" -unsigned int ath10k_target_ps; +static unsigned int ath10k_target_ps; module_param(ath10k_target_ps, uint, 0644); MODULE_PARM_DESC(ath10k_target_ps, "Enable ath10k Target (SoC) PS option"); @@ -1740,6 +1740,7 @@ static void ath10k_pci_fw_interrupt_handler(struct ath10k *ar) static int ath10k_pci_hif_power_up(struct ath10k *ar) { + struct ath10k_pci *ar_pci = ath10k_pci_priv(ar); int ret; /* @@ -1758,13 +1759,9 @@ static int ath10k_pci_hif_power_up(struct ath10k *ar) if (ret) goto err; - if (ath10k_target_ps) { - ath10k_dbg(ATH10K_DBG_PCI, "on-chip power save enabled\n"); - } else { + if (!test_bit(ATH10K_PCI_FEATURE_SOC_POWER_SAVE, ar_pci->features)) /* Force AWAKE forever */ - ath10k_dbg(ATH10K_DBG_PCI, "on-chip power save disabled\n"); ath10k_do_pci_wake(ar); - } ret = ath10k_pci_ce_init(ar); if (ret) @@ -1785,7 +1782,7 @@ static int ath10k_pci_hif_power_up(struct ath10k *ar) err_ce: ath10k_pci_ce_deinit(ar); err_ps: - if (!ath10k_target_ps) + if (!test_bit(ATH10K_PCI_FEATURE_SOC_POWER_SAVE, ar_pci->features)) ath10k_do_pci_sleep(ar); err: return ret; @@ -1793,8 +1790,10 @@ err: static void ath10k_pci_hif_power_down(struct ath10k *ar) { + struct ath10k_pci *ar_pci = ath10k_pci_priv(ar); + ath10k_pci_ce_deinit(ar); - if (!ath10k_target_ps) + if (!test_bit(ATH10K_PCI_FEATURE_SOC_POWER_SAVE, ar_pci->features)) ath10k_do_pci_sleep(ar); } @@ -2244,6 +2243,9 @@ static void ath10k_pci_dump_features(struct ath10k_pci *ar_pci) case ATH10K_PCI_FEATURE_HW_1_0_WORKAROUND: ath10k_dbg(ATH10K_DBG_PCI, "QCA988X_1.0 workaround enabled\n"); break; + case ATH10K_PCI_FEATURE_SOC_POWER_SAVE: + ath10k_dbg(ATH10K_DBG_PCI, "QCA98XX SoC power save enabled\n"); + break; } } } @@ -2279,6 +2281,9 @@ static int ath10k_pci_probe(struct pci_dev *pdev, goto err_ar_pci; } + if (ath10k_target_ps) + set_bit(ATH10K_PCI_FEATURE_SOC_POWER_SAVE, ar_pci->features); + ath10k_pci_dump_features(ar_pci); ar = ath10k_core_create(ar_pci, ar_pci->dev, &ath10k_pci_hif_ops); diff --git a/drivers/net/wireless/ath/ath10k/pci.h b/drivers/net/wireless/ath/ath10k/pci.h index d3a2e6c..871bb33 100644 --- a/drivers/net/wireless/ath/ath10k/pci.h +++ b/drivers/net/wireless/ath/ath10k/pci.h @@ -153,6 +153,7 @@ struct service_to_pipe { enum ath10k_pci_features { ATH10K_PCI_FEATURE_MSI_X = 0, ATH10K_PCI_FEATURE_HW_1_0_WORKAROUND = 1, + ATH10K_PCI_FEATURE_SOC_POWER_SAVE = 2, /* keep last */ ATH10K_PCI_FEATURE_COUNT @@ -335,20 +336,22 @@ static inline u32 ath10k_pci_read32(struct ath10k *ar, u32 offset) return ioread32(ar_pci->mem + offset); } -extern unsigned int ath10k_target_ps; - void ath10k_do_pci_wake(struct ath10k *ar); void ath10k_do_pci_sleep(struct ath10k *ar); static inline void ath10k_pci_wake(struct ath10k *ar) { - if (ath10k_target_ps) + struct ath10k_pci *ar_pci = ath10k_pci_priv(ar); + + if (test_bit(ATH10K_PCI_FEATURE_SOC_POWER_SAVE, ar_pci->features)) ath10k_do_pci_wake(ar); } static inline void ath10k_pci_sleep(struct ath10k *ar) { - if (ath10k_target_ps) + struct ath10k_pci *ar_pci = ath10k_pci_priv(ar); + + if (test_bit(ATH10K_PCI_FEATURE_SOC_POWER_SAVE, ar_pci->features)) ath10k_do_pci_sleep(ar); }
Unify the PCI options location. By default the SoC PS option is disabled to boost the performance and due to poor stability on early HW revisions. In future we can remove the module parameter and turn on/off the PS for given hardware. This change also makes the pci module parameter for SoC PS static. Signed-off-by: Bartosz Markowski <bartosz.markowski@tieto.com> --- drivers/net/wireless/ath/ath10k/pci.c | 21 +++++++++++++-------- drivers/net/wireless/ath/ath10k/pci.h | 11 +++++++---- 2 files changed, 20 insertions(+), 12 deletions(-)