Runtime PM was added in 4.14. These patches fix a couple of issues reported by people trying out the new kernel on DB410c.
These should ideally go in one of the 4.14-rcs.
Archit Taneja (2): drm/msm/dsi: Use correct pm_runtime_put variant during host_init drm/msm/mdp5: Remove extra pm_runtime_put call in mdp5_crtc_cursor_set()
drivers/gpu/drm/msm/dsi/dsi_host.c | 2 +- drivers/gpu/drm/msm/mdp/mdp5/mdp5_crtc.c | 2 -- 2 files changed, 1 insertion(+), 3 deletions(-)
The DSI runtime PM suspend/resume callbacks check whether msm_host->cfg_hnd is non-NULL before trying to enable the bus clocks. This is done to accommodate early calls to these functions that may happen before the bus clocks are even initialized.
Calling pm_runtime_put_autosuspend() in dsi_host_init() can result in racy behaviour since msm_host->cfg_hnd is set very soon after. If the suspend callback happens too late, we end up trying to disable clocks that were never enabled, resulting in a bunch of WARN_ON splats.
Use pm_runtime_put_sync() so that the suspend callback is called immediately.
Reported-by: Nicolas Dechesne nicolas.dechesne@linaro.org Signed-off-by: Archit Taneja architt@codeaurora.org --- drivers/gpu/drm/msm/dsi/dsi_host.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/msm/dsi/dsi_host.c b/drivers/gpu/drm/msm/dsi/dsi_host.c index dbb31a014419..deaf869374ea 100644 --- a/drivers/gpu/drm/msm/dsi/dsi_host.c +++ b/drivers/gpu/drm/msm/dsi/dsi_host.c @@ -248,7 +248,7 @@ static const struct msm_dsi_cfg_handler *dsi_get_config( clk_disable_unprepare(ahb_clk); disable_gdsc: regulator_disable(gdsc_reg); - pm_runtime_put_autosuspend(dev); + pm_runtime_put_sync(dev); put_clk: clk_put(ahb_clk); put_gdsc:
On Fri, Oct 06, 2017 at 04:27:06PM +0530, Archit Taneja wrote:
The DSI runtime PM suspend/resume callbacks check whether msm_host->cfg_hnd is non-NULL before trying to enable the bus clocks. This is done to accommodate early calls to these functions that may happen before the bus clocks are even initialized.
Calling pm_runtime_put_autosuspend() in dsi_host_init() can result in racy behaviour since msm_host->cfg_hnd is set very soon after. If the suspend callback happens too late, we end up trying to disable clocks that were never enabled, resulting in a bunch of WARN_ON splats.
Sounds like the correct fix here is to block autosuspend until after everything is set up, including bus clocks. This patch just makes the race harder to hit in practice ... -Daniel
Use pm_runtime_put_sync() so that the suspend callback is called immediately.
Reported-by: Nicolas Dechesne nicolas.dechesne@linaro.org Signed-off-by: Archit Taneja architt@codeaurora.org
drivers/gpu/drm/msm/dsi/dsi_host.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/msm/dsi/dsi_host.c b/drivers/gpu/drm/msm/dsi/dsi_host.c index dbb31a014419..deaf869374ea 100644 --- a/drivers/gpu/drm/msm/dsi/dsi_host.c +++ b/drivers/gpu/drm/msm/dsi/dsi_host.c @@ -248,7 +248,7 @@ static const struct msm_dsi_cfg_handler *dsi_get_config( clk_disable_unprepare(ahb_clk); disable_gdsc: regulator_disable(gdsc_reg);
- pm_runtime_put_autosuspend(dev);
- pm_runtime_put_sync(dev);
put_clk: clk_put(ahb_clk); put_gdsc: -- The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum, hosted by The Linux Foundation
dri-devel mailing list dri-devel@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/dri-devel
On 10/06/2017 07:32 PM, Daniel Vetter wrote:
On Fri, Oct 06, 2017 at 04:27:06PM +0530, Archit Taneja wrote:
The DSI runtime PM suspend/resume callbacks check whether msm_host->cfg_hnd is non-NULL before trying to enable the bus clocks. This is done to accommodate early calls to these functions that may happen before the bus clocks are even initialized.
Calling pm_runtime_put_autosuspend() in dsi_host_init() can result in racy behaviour since msm_host->cfg_hnd is set very soon after. If the suspend callback happens too late, we end up trying to disable clocks that were never enabled, resulting in a bunch of WARN_ON splats.
Sounds like the correct fix here is to block autosuspend until after everything is set up, including bus clocks. This patch just makes the race harder to hit in practice ...
Thanks for the review. pm_runtime_put_sync() ensures that the suspend handler is called in the same context as that of the caller, right? msm_host->cfg_hnd is set to a non-NULL value only when we return from dsi_get_config(). The race would never happen in this case.
This call is a one time thing during DSI probe, we do a pm_runtime_get_sync() just so that we can read the block revision number. Once we have the revision number, we look at an internal table which maintains IP version specific resources, like what bus clocks to get, etc. Having pm_runtime_put_autosuspend() here didn't help much anyway.
Archit
-Daniel
Use pm_runtime_put_sync() so that the suspend callback is called immediately.
Reported-by: Nicolas Dechesne nicolas.dechesne@linaro.org Signed-off-by: Archit Taneja architt@codeaurora.org
drivers/gpu/drm/msm/dsi/dsi_host.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/msm/dsi/dsi_host.c b/drivers/gpu/drm/msm/dsi/dsi_host.c index dbb31a014419..deaf869374ea 100644 --- a/drivers/gpu/drm/msm/dsi/dsi_host.c +++ b/drivers/gpu/drm/msm/dsi/dsi_host.c @@ -248,7 +248,7 @@ static const struct msm_dsi_cfg_handler *dsi_get_config( clk_disable_unprepare(ahb_clk); disable_gdsc: regulator_disable(gdsc_reg);
- pm_runtime_put_autosuspend(dev);
- pm_runtime_put_sync(dev); put_clk: clk_put(ahb_clk); put_gdsc:
-- The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum, hosted by The Linux Foundation
dri-devel mailing list dri-devel@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/dri-devel
On Tue, Oct 10, 2017 at 08:58:03AM +0530, Archit Taneja wrote:
On 10/06/2017 07:32 PM, Daniel Vetter wrote:
On Fri, Oct 06, 2017 at 04:27:06PM +0530, Archit Taneja wrote:
The DSI runtime PM suspend/resume callbacks check whether msm_host->cfg_hnd is non-NULL before trying to enable the bus clocks. This is done to accommodate early calls to these functions that may happen before the bus clocks are even initialized.
Calling pm_runtime_put_autosuspend() in dsi_host_init() can result in racy behaviour since msm_host->cfg_hnd is set very soon after. If the suspend callback happens too late, we end up trying to disable clocks that were never enabled, resulting in a bunch of WARN_ON splats.
Sounds like the correct fix here is to block autosuspend until after everything is set up, including bus clocks. This patch just makes the race harder to hit in practice ...
Thanks for the review. pm_runtime_put_sync() ensures that the suspend handler is called in the same context as that of the caller, right? msm_host->cfg_hnd is set to a non-NULL value only when we return from dsi_get_config(). The race would never happen in this case.
This call is a one time thing during DSI probe, we do a pm_runtime_get_sync() just so that we can read the block revision number. Once we have the revision number, we look at an internal table which maintains IP version specific resources, like what bus clocks to get, etc. Having pm_runtime_put_autosuspend() here didn't help much anyway.
Maybe this stuff is different on arm than on pci, but on x86 you have to explicitly enable autosuspend. Before that, the device stays on.
What I mean with properly fixing this, is to delay enabling of autosuspend until you're fully set up. Which then allows you to drop the check for clocks and other stuff.
For i915, what we do is hold an artificial runtime pm reference that we grab first thing in ->probe and drop only once everything is fully set up (including asynchronous workers that load firmware). That way we make sure that our runtime pm code never sees a partially initiliazed driver.
Doing something similar sounds best too, i.e. instead of dropping the runtime pm reference here, only drop it once dsi_get_config has been called. Pronto, no race, no need for special functions. -Daniel
Archit
-Daniel
Use pm_runtime_put_sync() so that the suspend callback is called immediately.
Reported-by: Nicolas Dechesne nicolas.dechesne@linaro.org Signed-off-by: Archit Taneja architt@codeaurora.org
drivers/gpu/drm/msm/dsi/dsi_host.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/msm/dsi/dsi_host.c b/drivers/gpu/drm/msm/dsi/dsi_host.c index dbb31a014419..deaf869374ea 100644 --- a/drivers/gpu/drm/msm/dsi/dsi_host.c +++ b/drivers/gpu/drm/msm/dsi/dsi_host.c @@ -248,7 +248,7 @@ static const struct msm_dsi_cfg_handler *dsi_get_config( clk_disable_unprepare(ahb_clk); disable_gdsc: regulator_disable(gdsc_reg);
- pm_runtime_put_autosuspend(dev);
- pm_runtime_put_sync(dev); put_clk: clk_put(ahb_clk); put_gdsc:
-- The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum, hosted by The Linux Foundation
dri-devel mailing list dri-devel@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/dri-devel
-- Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux Foundation Collaborative Project
On Tue, Oct 10, 2017 at 4:59 AM, Daniel Vetter daniel@ffwll.ch wrote:
On Tue, Oct 10, 2017 at 08:58:03AM +0530, Archit Taneja wrote:
On 10/06/2017 07:32 PM, Daniel Vetter wrote:
On Fri, Oct 06, 2017 at 04:27:06PM +0530, Archit Taneja wrote:
The DSI runtime PM suspend/resume callbacks check whether msm_host->cfg_hnd is non-NULL before trying to enable the bus clocks. This is done to accommodate early calls to these functions that may happen before the bus clocks are even initialized.
Calling pm_runtime_put_autosuspend() in dsi_host_init() can result in racy behaviour since msm_host->cfg_hnd is set very soon after. If the suspend callback happens too late, we end up trying to disable clocks that were never enabled, resulting in a bunch of WARN_ON splats.
Sounds like the correct fix here is to block autosuspend until after everything is set up, including bus clocks. This patch just makes the race harder to hit in practice ...
Thanks for the review. pm_runtime_put_sync() ensures that the suspend handler is called in the same context as that of the caller, right? msm_host->cfg_hnd is set to a non-NULL value only when we return from dsi_get_config(). The race would never happen in this case.
This call is a one time thing during DSI probe, we do a pm_runtime_get_sync() just so that we can read the block revision number. Once we have the revision number, we look at an internal table which maintains IP version specific resources, like what bus clocks to get, etc. Having pm_runtime_put_autosuspend() here didn't help much anyway.
Maybe this stuff is different on arm than on pci, but on x86 you have to explicitly enable autosuspend. Before that, the device stays on.
On arm, we cannot assume that clks are enabled ever unless we've enabled them. And unclocked register access is insta-reboot. Although we probably shouldn't be using _autosuspend() on the display side of things. It makes sense for the gpu, where "booting up" the gpu is a heavier process and we might want to keep things alive for a bit longer incase another submit comes in.
BR, -R
What I mean with properly fixing this, is to delay enabling of autosuspend until you're fully set up. Which then allows you to drop the check for clocks and other stuff.
For i915, what we do is hold an artificial runtime pm reference that we grab first thing in ->probe and drop only once everything is fully set up (including asynchronous workers that load firmware). That way we make sure that our runtime pm code never sees a partially initiliazed driver.
Doing something similar sounds best too, i.e. instead of dropping the runtime pm reference here, only drop it once dsi_get_config has been called. Pronto, no race, no need for special functions. -Daniel
Archit
-Daniel
Use pm_runtime_put_sync() so that the suspend callback is called immediately.
Reported-by: Nicolas Dechesne nicolas.dechesne@linaro.org Signed-off-by: Archit Taneja architt@codeaurora.org
drivers/gpu/drm/msm/dsi/dsi_host.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/msm/dsi/dsi_host.c b/drivers/gpu/drm/msm/dsi/dsi_host.c index dbb31a014419..deaf869374ea 100644 --- a/drivers/gpu/drm/msm/dsi/dsi_host.c +++ b/drivers/gpu/drm/msm/dsi/dsi_host.c @@ -248,7 +248,7 @@ static const struct msm_dsi_cfg_handler *dsi_get_config( clk_disable_unprepare(ahb_clk); disable_gdsc: regulator_disable(gdsc_reg);
- pm_runtime_put_autosuspend(dev);
- pm_runtime_put_sync(dev); put_clk: clk_put(ahb_clk); put_gdsc:
-- The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum, hosted by The Linux Foundation
dri-devel mailing list dri-devel@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/dri-devel
-- Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux Foundation Collaborative Project
-- Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch
While converting mdp5_enable/disable() calls to pm_runtime_get/put() API, an extra call to pm_runtime_put_autosuspend() crept in mdp5_crtc_cursor_set(). This results in calling the suspend handler twice, and therefore clk_disables twice, which isn't a nice thing to do.
Fixes: d68fe15b1878 (drm/msm/mdp5: Use runtime PM get/put API instead ...)
Reported-by: Stanimir Varbanov stanimir.varbanov@linaro.org Signed-off-by: Archit Taneja architt@codeaurora.org --- drivers/gpu/drm/msm/mdp/mdp5/mdp5_crtc.c | 2 -- 1 file changed, 2 deletions(-)
diff --git a/drivers/gpu/drm/msm/mdp/mdp5/mdp5_crtc.c b/drivers/gpu/drm/msm/mdp/mdp5/mdp5_crtc.c index 6fcb58ab718c..440977677001 100644 --- a/drivers/gpu/drm/msm/mdp/mdp5/mdp5_crtc.c +++ b/drivers/gpu/drm/msm/mdp/mdp5/mdp5_crtc.c @@ -804,8 +804,6 @@ static int mdp5_crtc_cursor_set(struct drm_crtc *crtc,
spin_unlock_irqrestore(&mdp5_crtc->cursor.lock, flags);
- pm_runtime_put_autosuspend(&pdev->dev); - set_cursor: ret = mdp5_ctl_set_cursor(ctl, pipeline, 0, cursor_enable); if (ret) {
On Fri, Oct 6, 2017 at 12:57 PM, Archit Taneja architt@codeaurora.org wrote:
Runtime PM was added in 4.14. These patches fix a couple of issues reported by people trying out the new kernel on DB410c.
These should ideally go in one of the 4.14-rcs.
Archit Taneja (2): drm/msm/dsi: Use correct pm_runtime_put variant during host_init drm/msm/mdp5: Remove extra pm_runtime_put call in mdp5_crtc_cursor_set()
thanks. tested both patches on db410c and it fixes the issues we discussed yesterday. i can run glmark2 now with openbox. as such,
Tested-by: Nicolas Dechesne nicolas.dechesne@linaro.org
drivers/gpu/drm/msm/dsi/dsi_host.c | 2 +- drivers/gpu/drm/msm/mdp/mdp5/mdp5_crtc.c | 2 -- 2 files changed, 1 insertion(+), 3 deletions(-)
-- The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum, hosted by The Linux Foundation
-- To unsubscribe from this list: send the line "unsubscribe linux-arm-msm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
dri-devel@lists.freedesktop.org
-
Archit Taneja -
Daniel Vetter -
Nicolas Dechesne -
Rob Clark