remoteproc: qcom: fix QCOM_SMD dependencies
PATCH RESEND 1/1 remoteproc: qcom: fix QCOM_SMD dependencies
qcom_smd_register_edge() is provided by either QCOM_SMD or RPMSG_QCOM_SMD, and if both of them are disabled, it does nothing.
The check for the PIL drivers however only checks for QCOM_SMD, so it breaks with QCOM_SMD=n && RPMSG_QCOM_SMD=m:
drivers/remoteproc/built-in.o: In function `smd_subdev_remove':
qcom_wcnss_iris.c:(.text+0x231c): undefined reference to qcom_smd_unregister_edge' drivers/remoteproc/built-in.o: In function smd_subdev_probe':
qcom_wcnss_iris.c:(.text+0x2344): undefined reference to qcom_smd_register_edge' drivers/remoteproc/built-in.o: In function smd_subdev_probe':
qcom_q6v5_pil.c:(.text+0x3538): undefined reference to qcom_smd_register_edge' qcom_q6v5_pil.c:(.text+0x3538): relocation truncated to fit: R_AARCH64_CALL26 against undefined symbol qcom_smd_register_edge'
This clarifies the Kconfig dependency.
---
Submitted on Feb 28, still needed in mainline as of v4.11-rc2
--- drivers/remoteproc/Kconfig | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)
drivers/remoteproc/Kconfig +3 -3
@@ -76,7 +76,7 @@ config QCOM_ADSP_PIL depends on OF && ARCH_QCOM depends on REMOTEPROC depends on QCOM_SMEM - depends on QCOM_SMD || (COMPILE_TEST && QCOM_SMD=n) + depends on RPMSG_QCOM_SMD || QCOM_SMD || (COMPILE_TEST && QCOM_SMD=n && RPMSG_QCOM_SMD=n) select MFD_SYSCON select QCOM_MDT_LOADER select QCOM_RPROC_COMMON
@@ -93,7 +93,7 @@ config QCOM_Q6V5_PIL depends on OF && ARCH_QCOM depends on QCOM_SMEM depends on REMOTEPROC - depends on QCOM_SMD || (COMPILE_TEST && QCOM_SMD=n) + depends on RPMSG_QCOM_SMD || QCOM_SMD || (COMPILE_TEST && QCOM_SMD=n && RPMSG_QCOM_SMD=n) select MFD_SYSCON select QCOM_RPROC_COMMON select QCOM_SCM
@@ -104,7 +104,7 @@ config QCOM_Q6V5_PIL config QCOM_WCNSS_PIL tristate "Qualcomm WCNSS Peripheral Image Loader" depends on OF && ARCH_QCOM - depends on QCOM_SMD || (COMPILE_TEST && QCOM_SMD=n) + depends on RPMSG_QCOM_SMD || QCOM_SMD || (COMPILE_TEST && QCOM_SMD=n && RPMSG_QCOM_SMD=n) depends on QCOM_SMEM depends on REMOTEPROC select QCOM_MDT_LOADER
Signature · 1 line
2.9.0
Hi Arnd,
Quoted 2 lines · Arnd Bergmann
qcom_smd_register_edge()is provided by eitherQCOM_SMDorRPMSG_QCOM_SMD, and if both of them are disabled, it does nothing.
Actually the code itself looks wrong to me. There are two sets of stubs for qcom_smd_register_edge() and qcom_smd_unregister_edge() when the feature is disabled, one from include/linux/rpmsg/qcom_smd.h and one from include/linux/soc/qcom/smd.h. They have different definitions, and different conditions. The former is declared if neither backend is selected, while the latter is declared if QCOM_SMD isn't selected (regardless of the value of RPMSG_QCOM_SMD.)
So as it stands, QCOM_SMD=n && RPMSG_QCOM_SMD=m leads to 2 implementations of these functions, inline stubs from include/linux/soc/qcom/smd.h and actual implementations from drivers/rpmsg/qcom_smd.c. I know nothing about these drivers but this looks needlessly complex and error-prone to me. The stubs should be declared only once and only when no actual implementations are available. That is, assuming they are really supposed to be the same and it's not an unfortunate name collision.
Quoted 44 lines
The check for the PIL drivers however only checks for
QCOM_SMD, so it breaks with QCOM_SMD=n && RPMSG_QCOM_SMD=m:
drivers/remoteproc/built-in.o: In function `smd_subdev_remove':
qcom_wcnss_iris.c:(.text+0x231c): undefined reference toqcom_smd_unregister_edge' drivers/remoteproc/built-in.o: In functionsmd_subdev_probe':
qcom_wcnss_iris.c:(.text+0x2344): undefined reference toqcom_smd_register_edge' drivers/remoteproc/built-in.o: In functionsmd_subdev_probe':
qcom_q6v5_pil.c:(.text+0x3538): undefined reference toqcom_smd_register_edge' qcom_q6v5_pil.c:(.text+0x3538): relocation truncated to fit: R_AARCH64_CALL26 against undefined symbolqcom_smd_register_edge'This clarifies the
Kconfigdependency.Fixes 4b48921a8f74 ("remoteproc: qcom: Use common SMD edge handler")Signed-off-by Arnd Bergmann <[email protected]>---
Submitted on Feb 28, still needed in mainline as of v4.11-rc2--- drivers/remoteproc/Kconfig | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)drivers/remoteproc/Kconfig +2
-2@@ -76,7 +76,7 @@ config QCOM_ADSP_PIL depends on OF && ARCH_QCOM depends on REMOTEPROC depends on QCOM_SMEM - depends on QCOM_SMD || (COMPILE_TEST && QCOM_SMD=n) + depends on RPMSG_QCOM_SMD || QCOM_SMD || (COMPILE_TEST && QCOM_SMD=n && RPMSG_QCOM_SMD=n) select MFD_SYSCON select QCOM_MDT_LOADER select QCOM_RPROC_COMMON@@ -93,7 +93,7 @@ config QCOM_Q6V5_PIL depends on OF && ARCH_QCOM depends on QCOM_SMEM depends on REMOTEPROC - depends on QCOM_SMD || (COMPILE_TEST && QCOM_SMD=n) + depends on RPMSG_QCOM_SMD || QCOM_SMD || (COMPILE_TEST && QCOM_SMD=n && RPMSG_QCOM_SMD=n) select MFD_SYSCON select QCOM_RPROC_COMMON select QCOM_SCM
drivers/remoteproc/Kconfig config QCOM_Q6V5_PIL · line 110
Quoted 8 lines
drivers/remoteproc/Kconfig +1
-1@@ -104,7 +104,7 @@ config QCOM_Q6V5_PIL config QCOM_WCNSS_PIL tristate "Qualcomm WCNSS Peripheral Image Loader" depends on OF && ARCH_QCOM - depends on QCOM_SMD || (COMPILE_TEST && QCOM_SMD=n) + depends on RPMSG_QCOM_SMD || QCOM_SMD || (COMPILE_TEST && QCOM_SMD=n && RPMSG_QCOM_SMD=n) depends on QCOM_SMEM depends on REMOTEPROC select QCOM_MDT_LOADER
I don't think the COMPILE_TEST adds any value here. The whole set of drivers is architecture specific anyway so you won't gain much build test coverage. It may even prevent a legitimate combination of options on the intended target, if the feature provided by QCOM_SMD and RPMSG_QCOM_SMD is optional (if not, I would suggest to drop all the stubs and simply depend on RPMSG_QCOM_SMD || QCOM_SMD, for the sake of simplicity.)
Don't get me wrong, I am a big fan of COMPILE_TEST, but only when we can use it to get build testing coverage for free. If you have to change the code itself in order to be able to get the extra build testing coverage, I don't think it is a good idea. The kernel and the Kconfig dependencies can be complex enough as is.
Signature · 2 lines
Jean Delvare SUSE L3 Support
Quoted 6 lines · Jean Delvare
Hi Arnd,
Quoted 2 lines · Arnd Bergmann
qcom_smd_register_edge()is provided by eitherQCOM_SMDorRPMSG_QCOM_SMD, and if both of them are disabled, it does nothing.
Quoted 7 lines
Actually the code itself looks wrong to me. There are two sets of stubs for
qcom_smd_register_edge()andqcom_smd_unregister_edge()when the feature is disabled, one frominclude/linux/rpmsg/qcom_smd.hand one frominclude/linux/soc/qcom/smd.h. They have different definitions, and different conditions. The former is declared if neither backend is selected, while the latter is declared ifQCOM_SMDisn't selected (regardless of the value ofRPMSG_QCOM_SMD.)
This driver always includes the former header (linux/rpmsg/qcom_smd.h), which has both checks. The second header should be removed as soon as we have moved over the users to the new one.
Quoted 4 lines
So as it stands, QCOM_SMD=n && RPMSG_QCOM_SMD=m leads to 2 implementations of these functions, inline stubs from
include/linux/soc/qcom/smd.hand actual implementations fromdrivers/rpmsg/qcom_smd.c.
QCOM_SMD=n && RPMSG_QCOM_SMD=m should result in no code including include/linux/soc/qcom/smd.h, and I don't get any other failures here.
I know nothing about these drivers but this
Quoted 4 lines
looks needlessly complex and error-prone to me. The stubs should be declared only once and only when no actual implementations are available. That is, assuming they are really supposed to be the same and it's not an unfortunate name collision.
Agreed, but I think fixing that can be a separate effort. I don't know what Bjorn's time frame is for removing the soc/qcom/smd driver, but I'd guess we can already merge the headers into one as a first step.
Quoted 18 lines
drivers/remoteproc/Kconfig config QCOM_Q6V5_PIL · line 110
Quoted 9 lines
drivers/remoteproc/Kconfig +1
-1@@ -104,7 +104,7 @@ config QCOM_Q6V5_PIL config QCOM_WCNSS_PIL tristate "Qualcomm WCNSS Peripheral Image Loader" depends on OF && ARCH_QCOM - depends on QCOM_SMD || (COMPILE_TEST && QCOM_SMD=n) + depends on RPMSG_QCOM_SMD || QCOM_SMD || (COMPILE_TEST && QCOM_SMD=n && RPMSG_QCOM_SMD=n) depends on QCOM_SMEM depends on REMOTEPROC select QCOM_MDT_LOADERI don't think the
COMPILE_TESTadds any value here. The whole set of drivers is architecture specific anyway so you won't gain much build test coverage. It may even prevent a legitimate combination of options on the intended target, if the feature provided byQCOM_SMDandRPMSG_QCOM_SMDis optional (if not, I would suggest to drop all the stubs and simply depend onRPMSG_QCOM_SMD||QCOM_SMD, for the sake of simplicity.)
Quoted 5 lines
Don't get me wrong, I am a big fan of
COMPILE_TEST, but only when we can use it to get build testing coverage for free. If you have to change the code itself in order to be able to get the extra build testing coverage, I don't think it is a good idea. The kernel and theKconfigdependencies can be complex enough as is.
I think the problem is the ARCH_QCOM dependency here, which clearly gets in the way of COMPILE_TEST having any real effect.
I think generally speaking we either want to run the code and need both ARCH_QCOM and (RPMSG_QCOM_SMD || QCOM_SMD), or we do build-testing and need (COMPILE_TEST && RPMSG_QCOM_SMD=n
&& QCOM_SMD=n). We could add another Kconfig symbol that captures the dependency and then just add a simple dependency here.
Arnd
Quoted 1 line · Arnd Bergmann
On Tue, Mar 14, 2017 at 10:05 AM, Jean Delvare <[email protected]> wrote:
[..]
Quoted 17 lines
Quoted 8 lines
I don't think the
COMPILE_TESTadds any value here. The whole set of drivers is architecture specific anyway so you won't gain much build test coverage. It may even prevent a legitimate combination of options on the intended target, if the feature provided byQCOM_SMDandRPMSG_QCOM_SMDis optional (if not, I would suggest to drop all the stubs and simply depend onRPMSG_QCOM_SMD||QCOM_SMD, for the sake of simplicity.)Quoted 5 lines
Don't get me wrong, I am a big fan of
COMPILE_TEST, but only when we can use it to get build testing coverage for free. If you have to change the code itself in order to be able to get the extra build testing coverage, I don't think it is a good idea. The kernel and theKconfigdependencies can be complex enough as is.I think the problem is the
ARCH_QCOMdependency here, which clearly gets in the way ofCOMPILE_TESThaving any real effect.
Quoted 6 lines
I think generally speaking we either want to run the code and need both
ARCH_QCOMand (RPMSG_QCOM_SMD||QCOM_SMD), or we do build-testing and need (COMPILE_TEST&& RPMSG_QCOM_SMD=n
&& QCOM_SMD=n). We could add anotherKconfigsymbol that captures the dependency and then just add a simple dependency here.
The drivers doesn't depend on SMD for loading and booting the remoteprocs, the dependency is just to get the appropriate functions/stubs - although currently all firmware I've seen requires SMD to be in place.
I'll push forward with purging QCOM_SMD and I'll have a proper review of the dependencies as part of/following that process.
I believe it makes more sense to check for ARCH_QCOM || COMPILE_TEST and just have the others to check if we have each dependency build-in or module.
I forwarded your patch to Linus for v4.11, thanks.
Regards,
Bjorn