RT kernel build issue if RT_BASIC is enabled

[mkumar81]

Hi,
I'm trying to build the RT kernel source RPM from the link "http://vault.centos.org/7.3.1611/rt/Sou ... l7.src.rpm".
In default "CONFIG_PREEMPT_RT_FULL" is enabled. But when "CONFIG_PREEMPT_RTB" is enabled there is a COMPILATION ISSUE, I have to apply the below source patch to fix the build.
There are two RT source RPMs in the repository, both seems to have the same issue.
This seems a kernel source bug, please clarify. Please also let me know if there is a patch available that fix the issue or if I'm doing anything wrong.
The way I update the kernel config was using "make oldconfig" and "make menuconfig"

rt-basic-patch.txt

(2.49 KiB) Downloaded 95 times

Patch used to fix the build issue with PREEMT_RTB enabled
---------------------------------------------------------------------------
MCP Custom Changes - 5/8/2017
The following changes are to support compiling Linux with the basic realtime preemption scheme.
Without these changes, compilation errors occurred when building Linux with RT basic.

Code: Select all

diff -uNr linux-3.10.0.x86_64/include/drm/drm_backport.h linux-patch-dir/include/drm/drm_backport.h
--- linux-3.10.0.x86_64/include/drm/drm_backport.h	2016-12-14 15:12:03.000000000 -0500
+++ linux-patch-dir/include/drm/drm_backport.h	2017-05-08 10:25:08.453144931 -0400
@@ -164,9 +164,17 @@
 }
 
 #define cpu_relax_lowlatency() cpu_relax()
-#ifndef CONFIG_PREEMPT_RT_FULL
-#define pagefault_disabled()   in_atomic()
-#endif
+/*
+ * MCP Custom Change
+ * Comment out the definition of pagefault_disabled.
+ * Without this line being commented out, the compiler
+ * complains because pagefault_disabled() is defined in two places.
+ * One place is here as a macro. The other place is in include/sched.h
+ * as a function definition.
+ */
+//#ifndef CONFIG_PREEMPT_RT_FULL
+//#define pagefault_disabled()   in_atomic()
+//#endif
 
 static inline int arch_phys_wc_index(int handle)
 {
diff -uNr linux-3.10.0.x86_64/kernel/sched/core.c linux-patch-dir/kernel/sched/core.c
--- linux-3.10.0.x86_64/kernel/sched/core.c	2016-12-14 15:12:03.000000000 -0500
+++ linux-patch-dir/kernel/sched/core.c	2017-05-08 10:23:23.955151340 -0400
@@ -5519,7 +5519,15 @@
 #if defined(CONFIG_PREEMPT_RT_BASE)
 	/* As we have the actual cpumask trimmed down to nr_cpu_ids bits, we need
 	   to zero the remaining bits to avoid garbage being sent to userspace */
-	memset(mask, 0, sizeof(mask));
+
+    /* MCP Custom Change
+     * Comment out this memset.
+     * Without it being commented out, the compiler complains that the sizeof
+     * argument is referencing the size of a pointer rather than the underlying
+     * data type. This is correct since the data type of mask is typedef'd to
+     * a pointer.
+     */
+	//memset(mask, 0, sizeof(mask));
 #endif
 	if ((len * BITS_PER_BYTE) < nr_cpu_ids)
 		return -EINVAL;
diff -uNr linux-3.10.0.x86_64/net/core/dev.c linux-patch-dir/net/core/dev.c
--- linux-3.10.0.x86_64/net/core/dev.c	2016-12-14 15:12:03.000000000 -0500
+++ linux-patch-dir/net/core/dev.c	2017-05-08 10:21:58.075156608 -0400
@@ -2935,7 +2935,11 @@
 	__this_cpu_inc(xmit_recursion);
 }
 
-static inline int xmit_rec_dec(void)
+/* MCP Custom Change 
+ * Change return type from int to void
+ * so that the compiler doesn't complain. This looks like an typo.
+ */
+static inline void xmit_rec_dec(void)
 {
 	__this_cpu_dec(xmit_recursion);
 }

[mkumar81]

Hi,

Could anyone look into this build issue and confirm it's a genuine issue with official RT release?

Thanks,
Muthu

[duffsterlp]

I am running into the same issue.

The source kernel I used is here: http://vault.centos.org/7.3.1611/rt/Sou ... l7.src.rpm
I changed CONFIG_PREEMPT_RT_FULL to CONFIG_PREEMPT_RTB under:
Processor Types and Features -> Preemption Model
When I do an rpmbuild to generate the binary RPM, I get the following compilation error.

Code: Select all

kernel/sched/core.c: In function 'SYSC_sched_getaffinity':
kernel/sched/core.c:5522:24: error: argument to 'sizeof' in 'memset' call is the same expression as the destination; did you mean to dereference it? [-Werror=sizeof-pointer-memaccess]
  memset(mask, 0, sizeof(mask));

[hughesjr]

I did build this, but it is NOT officially supported.

The packages built as is .. here is the build logs and built binaries:

https://buildlogs.centos.org/c7-rt/kernel-rt/

I can build a newer version of the kernel-rt (and the add on packages) into that area if you like, BUT, they will not be officially supported.

[mkumar81]

@hughesjr thanks for trying. I downloaded your source rpm from the shared path and it lookslike you didn't modify the kernel config from CONFIG_PREEMPT_RT_FULL to CONFIG_PREEMPT_RTB under:
Processor Types and Features -> Preemption Model.
The issue is only if we enable "CONFIG_PREEMPT_RTB".

[hughesjr]

BTW, I built the kernel config as it comes from Red Hat sources, for the RT kernel. I assume that is the way they build it as well. If you modify the config file for a different (unsupported by them for RHEL) option, I'm not sure why you would expect that build.

[mkumar81]

@duffsterlp Thanks a lot for trying.
Yes that is one of the build error. Do we have any official patch for this? Is there any official fix?
[OR]
a discussion on the best fix for all the 3 build errors would be great too. In my original post I've shared the patch I used.

Thanks for your time.

[mkumar81]

@hughesjr doing a kernel config change should never break a build when the config changes are done following the rules. There is no point using a source build if it breaks when there is just a kernel config changes. I've done numerous changes to non-RT build, most of it just inserting drivers in kernel itself instead of an installable kernel modules. Everything has worked fine.
When RT is supported, there are 3 different options for the RT (FULL, RTB and LL), and I believe everything should build without issues when there is any of it is chosen. That is the only change I did.

# CONFIG_PREEMPT_NONE is not set
# CONFIG_PREEMPT_VOLUNTARY is not set
# CONFIG_PREEMPT__LL is not set
# CONFIG_PREEMPT_RTB is not set
CONFIG_PREEMPT_RT_FULL=y

[TrevorH]

If changing the config options on the kernel breaks the build then you need to complain to Redhat who provide the source and spec files for the kernel packages.

[mkumar81]

@TrevorH Is there an official Issue/Bug report link for RedHat that you can share with me that handles these kind of build issues?