| Message ID | 1427107800-21668-1-git-send-email-javier.martinez@collabora.co.uk (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
2015-03-23 11:49 GMT+01:00 Javier Martinez Canillas <javier.martinez@collabora.co.uk>: > This reverts commit 2d2c9a8d0a4f90e298315d2f4a282d8bd5d45e5c > ("ARM: dts: add display power domain for exynos5250"). > > The mentioned commit added a domain definition for the DISP1 > power domain and references to it in the appropriate devices > but this change breaks the display in at least the Exynos5250 > based Snow and Spring Chromebooks. > > On these machines, the boot-loader enables the DISP1 domain and > before the mentioned commit, the kernel didn't know about it so > the power domain remained always enabled. > > But after that commit when the exynos-dp probe is deferred, > the DISP1 domain is powered off and on again but the exynos-dp > driver fails to configure the video showing the following error: > > exynos-dp 145b0000.dp-controller: Timeout of video streamclk ok > exynos-dp 145b0000.dp-controller: unable to config video > > The same issue happens when the display is turned off and on > again using DPMS. > > So, it seems the DISP1 power domain definition is not complete > since the display works with the initialization made by the boot > loader but it does not work when the power domain is enabled by > the kernel. > > Having the definition in the DTS makes the power domain to be > powered on when needed and powered off when not needed which is > better in terms of power consumption but for now is safer to just > revert the commit to avoid adding a regression in some machines. > > Signed-off-by: Javier Martinez Canillas <javier.martinez@collabora.co.uk> I looked at the DP and FIMD drivers and with great help of Andrzej Hajda found the issue: the FIMD driver does not enable DP clock (DP_MIE_CLKCON register). The process looks like: 1. Bootloader sets the DP_MIE_CLKCON to proper value. 2. FIMD is probed, DP power domain is on. 3. FIMD is deffered, DP power domain is turned off. 4. The FIMD registers are reset so DP clock is turned off. 5. DP driver is probed, DP power domain is turned on but clock is not enabled. I'll do some more testing and sent a patch for it till end of Monday so I think we should not revert this commit. Best regards, Krzysztof
Hello Krzysztof, On 03/27/2015 03:29 PM, Krzysztof Kozlowski wrote: > 2015-03-23 11:49 GMT+01:00 Javier Martinez Canillas > > I looked at the DP and FIMD drivers and with great help of Andrzej > Hajda found the issue: the FIMD driver does not enable DP clock > (DP_MIE_CLKCON register). The process looks like: > 1. Bootloader sets the DP_MIE_CLKCON to proper value. > 2. FIMD is probed, DP power domain is on. > 3. FIMD is deffered, DP power domain is turned off. > 4. The FIMD registers are reset so DP clock is turned off. > 5. DP driver is probed, DP power domain is turned on but clock is not enabled. > Great! I knew it was some setup missing that was made by the boot-loader. > I'll do some more testing and sent a patch for it till end of Monday > so I think we should not revert this commit. > Agree, I of course preferred to fix the actual cause instead of relying on the boot-loader to do the initialization but I wasn't able to figure it out so I posted this patch to revert the commit in the meantime. > Best regards, > Krzysztof > Best regards, Javier
Javier Martinez Canillas <javier.martinez@collabora.co.uk> writes: > This reverts commit 2d2c9a8d0a4f90e298315d2f4a282d8bd5d45e5c > ("ARM: dts: add display power domain for exynos5250"). > > The mentioned commit added a domain definition for the DISP1 > power domain and references to it in the appropriate devices > but this change breaks the display in at least the Exynos5250 > based Snow and Spring Chromebooks. > > On these machines, the boot-loader enables the DISP1 domain and > before the mentioned commit, the kernel didn't know about it so > the power domain remained always enabled. > > But after that commit when the exynos-dp probe is deferred, > the DISP1 domain is powered off and on again but the exynos-dp > driver fails to configure the video showing the following error: > > exynos-dp 145b0000.dp-controller: Timeout of video streamclk ok > exynos-dp 145b0000.dp-controller: unable to config video > > The same issue happens when the display is turned off and on > again using DPMS. > > So, it seems the DISP1 power domain definition is not complete > since the display works with the initialization made by the boot > loader but it does not work when the power domain is enabled by > the kernel. > > Having the definition in the DTS makes the power domain to be > powered on when needed and powered off when not needed which is > better in terms of power consumption but for now is safer to just > revert the commit to avoid adding a regression in some machines. > > Signed-off-by: Javier Martinez Canillas <javier.martinez@collabora.co.uk> Tested-by: Kevin Hilman <khilman@linaro.org> FWIW, this patch fixes the boot panics when using MMC rootfs on exynos5800-peach-pi with current linux-next that have been happening for awhile. For several months now, DRM/display related stuff is very routinely breaking basic booting on exynos5, which gives the rather strong impression that the DRM stuff is not tested well enough to be merged. Kevin
On Wed, 2015-04-29 at 10:40 -0700, Kevin Hilman wrote: > Javier Martinez Canillas <javier.martinez@collabora.co.uk> writes: > Tested-by: Kevin Hilman <khilman@linaro.org> > > FWIW, this patch fixes the boot panics when using MMC rootfs on > exynos5800-peach-pi with current linux-next that have been happening > for awhile. That seems unlikely as this patch changes things in the exynos5250.dtsi, which isn't used by exynos5800-peach-pi.? > For several months now, DRM/display related stuff is very routinely > breaking basic booting on exynos5, which gives the rather strong > impression that the DRM stuff is not tested well enough to be merged. Unfortunately it seems to have been that way for ages, but only has started being visible since Javier turned on the various exynos DRM drivers on in exynos_defconfig.. So now we're actually seeing bugs rather then just having broken code sit around :/
Hello, On 04/29/2015 09:45 PM, Sjoerd Simons wrote: > On Wed, 2015-04-29 at 10:40 -0700, Kevin Hilman wrote: >> Javier Martinez Canillas <javier.martinez@collabora.co.uk> writes: > >> Tested-by: Kevin Hilman <khilman@linaro.org> >> >> FWIW, this patch fixes the boot panics when using MMC rootfs on >> exynos5800-peach-pi with current linux-next that have been happening >> for awhile. > > That seems unlikely as this patch changes things in the exynos5250.dtsi, > which isn't used by exynos5800-peach-pi.? > Indeed, it will be surprising if $subject changes any behavior on Exynos5420/5422/5800 machines. FWIW, this patch is not needed anymore since the right fix is Krzysztof's "drm/exynos: Enable DP clock to fix display on Exynos5250 and other" patch which is already in linux-next. The regression on Exynos5420/5422/5800 that has been for a while is fixed by "ARM: dts: Make DP a consumer of DISP1 power domain on Exynos5420" [0] which has been posted many weeks ago. >> For several months now, DRM/display related stuff is very routinely >> breaking basic booting on exynos5, which gives the rather strong >> impression that the DRM stuff is not tested well enough to be merged. > > Unfortunately it seems to have been that way for ages, but only has > started being visible since Javier turned on the various exynos DRM > drivers on in exynos_defconfig.. So now we're actually seeing bugs > rather then just having broken code sit around :/ > Agreed. Best regards, Javier [0]: https://lkml.org/lkml/2015/4/12/153
Javier Martinez Canillas <javier.martinez@collabora.co.uk> writes: > Hello, > > On 04/29/2015 09:45 PM, Sjoerd Simons wrote: >> On Wed, 2015-04-29 at 10:40 -0700, Kevin Hilman wrote: >>> Javier Martinez Canillas <javier.martinez@collabora.co.uk> writes: >> >>> Tested-by: Kevin Hilman <khilman@linaro.org> >>> >>> FWIW, this patch fixes the boot panics when using MMC rootfs on >>> exynos5800-peach-pi with current linux-next that have been happening >>> for awhile. >> >> That seems unlikely as this patch changes things in the exynos5250.dtsi, >> which isn't used by exynos5800-peach-pi.? >> > > Indeed, it will be surprising if $subject changes any behavior on > Exynos5420/5422/5800 machines. > > FWIW, this patch is not needed anymore since the right fix is Krzysztof's > "drm/exynos: Enable DP clock to fix display on Exynos5250 and other" patch > which is already in linux-next. > > The regression on Exynos5420/5422/5800 that has been for a while is fixed > by "ARM: dts: Make DP a consumer of DISP1 power domain on Exynos5420" [0] > which has been posted many weeks ago. Thanks for clarifying. I'm not sure how this worked for me. Seems like sometimes, even with the imprecise aborts it does actually finish booting to a shell, so I must've gotten successful boot and thought it was because of this revert. Kevin
Hello Kevin, On 04/30/2015 04:59 PM, Kevin Hilman wrote: > Javier Martinez Canillas <javier.martinez@collabora.co.uk> writes: >> >> The regression on Exynos5420/5422/5800 that has been for a while is fixed >> by "ARM: dts: Make DP a consumer of DISP1 power domain on Exynos5420" [0] >> which has been posted many weeks ago. > > Thanks for clarifying. > > I'm not sure how this worked for me. Seems like sometimes, even with > the imprecise aborts it does actually finish booting to a shell, so I > must've gotten successful boot and thought it was because of this > revert. That is correct, the imprecise external abort error most of the times lead to a kernel panic but sometimes it doesn't and the the system finish booting to user-space even after the error. I've pinged Kukjin again yesterday to pick the fix and he said that will do it this weekend [0]. So hopefully exynos5420 machines will boot again in linux-next soon after being broken for weeks. > > Kevin > Best regards, Javier [0]: https://lkml.org/lkml/2015/4/29/781
diff --git a/arch/arm/boot/dts/exynos5250.dtsi b/arch/arm/boot/dts/exynos5250.dtsi index 77f656eb8e6b..482fa97c6b3a 100644 --- a/arch/arm/boot/dts/exynos5250.dtsi +++ b/arch/arm/boot/dts/exynos5250.dtsi @@ -105,12 +105,6 @@ #power-domain-cells = <0>; }; - pd_disp1: disp1-power-domain@100440A0 { - compatible = "samsung,exynos4210-pd"; - reg = <0x100440A0 0x20>; - #power-domain-cells = <0>; - }; - clock: clock-controller@10010000 { compatible = "samsung,exynos5250-clock"; reg = <0x10010000 0x30000>; @@ -752,7 +746,6 @@ hdmi: hdmi { compatible = "samsung,exynos4212-hdmi"; reg = <0x14530000 0x70000>; - power-domains = <&pd_disp1>; interrupts = <0 95 0>; clocks = <&clock CLK_HDMI>, <&clock CLK_SCLK_HDMI>, <&clock CLK_SCLK_PIXEL>, <&clock CLK_SCLK_HDMIPHY>, @@ -765,7 +758,6 @@ mixer { compatible = "samsung,exynos5250-mixer"; reg = <0x14450000 0x10000>; - power-domains = <&pd_disp1>; interrupts = <0 94 0>; clocks = <&clock CLK_MIXER>, <&clock CLK_HDMI>, <&clock CLK_SCLK_HDMI>; @@ -779,7 +771,6 @@ }; dp: dp-controller@145B0000 { - power-domains = <&pd_disp1>; clocks = <&clock CLK_DP>; clock-names = "dp"; phys = <&dp_phy>; @@ -787,7 +778,6 @@ }; fimd: fimd@14400000 { - power-domains = <&pd_disp1>; clocks = <&clock CLK_SCLK_FIMD1>, <&clock CLK_FIMD1>; clock-names = "sclk_fimd", "fimd"; };
This reverts commit 2d2c9a8d0a4f90e298315d2f4a282d8bd5d45e5c ("ARM: dts: add display power domain for exynos5250"). The mentioned commit added a domain definition for the DISP1 power domain and references to it in the appropriate devices but this change breaks the display in at least the Exynos5250 based Snow and Spring Chromebooks. On these machines, the boot-loader enables the DISP1 domain and before the mentioned commit, the kernel didn't know about it so the power domain remained always enabled. But after that commit when the exynos-dp probe is deferred, the DISP1 domain is powered off and on again but the exynos-dp driver fails to configure the video showing the following error: exynos-dp 145b0000.dp-controller: Timeout of video streamclk ok exynos-dp 145b0000.dp-controller: unable to config video The same issue happens when the display is turned off and on again using DPMS. So, it seems the DISP1 power domain definition is not complete since the display works with the initialization made by the boot loader but it does not work when the power domain is enabled by the kernel. Having the definition in the DTS makes the power domain to be powered on when needed and powered off when not needed which is better in terms of power consumption but for now is safer to just revert the commit to avoid adding a regression in some machines. Signed-off-by: Javier Martinez Canillas <javier.martinez@collabora.co.uk> --- arch/arm/boot/dts/exynos5250.dtsi | 10 ---------- 1 file changed, 10 deletions(-)