From mahesh at linux.ibm.com  Sat Sep 20 03:19:43 2025
From: mahesh at linux.ibm.com (Mahesh Salgaonkar)
Date: Fri, 19 Sep 2025 22:49:43 +0530
Subject: [Skiboot] [PATCH 1/3] external/opal-prd: Fix compilation error
In-Reply-To: <20250919171945.3947759-1-mahesh@linux.ibm.com>
References: <20250919171945.3947759-1-mahesh@linux.ibm.com>
Message-ID: <20250919171945.3947759-2-mahesh@linux.ibm.com>

Upstream opal-prd fails with below compilation errors:

$ make -C external/opal-prd/
make: Entering directory '/home/mahesh/skiboot/external/opal-prd'
    [...]
    CC  thunk.o
/usr/include/bits/types.h: Assembler messages:
/usr/include/bits/types.h:30: Error: unrecognized opcode: `typedef'
/usr/include/bits/types.h:31: Error: unrecognized opcode: `typedef'
/usr/include/bits/types.h:32: Error: unrecognized opcode: `typedef'
    [...]

The file thunk.S includes <endian.h> which gets picked up from skiboot's
'include/endian.h' (instead of /usr/include/endian.h) which then
includes 'ccan/endian/endian.h' -> /usr/include/bits/types.h where
assember fails with compilation.

The 'include/endian.h' was introduced by commit 18c9b278e747 ("include:
Provide endian conversion functions") when support for PLDM was added.
This header file was introduced to be used by C files under pldm/,
core/pldm and libmctp/ only. In order to restrict this header file to
pldm/libmctp, move the endian.h header under 'pldm/include' folder. This
makes sure that except files under 'pldm' and 'libmctp', rest all other
C files would pick up default header file from '/usr/include/endian.h'
as it was prior to commit 18c9b278e747 ("include: Provide endian
conversion functions")

After moving endian.h under 'pldm/include', opal-prd gets successfully
compiled.

Fixes: 18c9b278e747 ("include: Provide endian conversion functions")
Signed-off-by: Mahesh Salgaonkar <mahesh at linux.ibm.com>
---
 {include => pldm/include}/endian.h | 0
 1 file changed, 0 insertions(+), 0 deletions(-)
 rename {include => pldm/include}/endian.h (100%)

diff --git a/include/endian.h b/pldm/include/endian.h
similarity index 100%
rename from include/endian.h
rename to pldm/include/endian.h
-- 
2.51.0


From mahesh at linux.ibm.com  Sat Sep 20 03:19:44 2025
From: mahesh at linux.ibm.com (Mahesh Salgaonkar)
Date: Fri, 19 Sep 2025 22:49:44 +0530
Subject: [Skiboot] [PATCH 2/3] external/opal-prd: Fix opal-prd service
 shutdown on memory errors
In-Reply-To: <20250919171945.3947759-1-mahesh@linux.ibm.com>
References: <20250919171945.3947759-1-mahesh@linux.ibm.com>
Message-ID: <20250919171945.3947759-3-mahesh@linux.ibm.com>

Whenever there is a memory error reported, opal-prd tries to spawn a
child process using fork to delegate the memory offline work to child
process. After handling memory error child process suppose to exit.
However, instead of delegating the task to child process the main thread
itself handles the memory error and exits. Thus causing opal-prd service
to go into stop/restart loop and eventually hits the systemd restart
limit leaving opal-prd service unavailable.

opal-prd[49096]: MEM: Memory error: range
0000000eeb445700-0000000eeb445700, type: correctable
opal-prd[49096]: MEM: Offlined 0000000eeb445700,0000000eeb455700, type
correctable: No such file or directory
systemd[1]: opal-prd.service: Service RestartSec=100ms expired,
scheduling restart.
systemd[1]: opal-prd.service: Scheduled restart job, restart counter is
at 7.
systemd[1]: opal-prd.service: Start request repeated too quickly.
systemd[1]: opal-prd.service: Failed with result 'start-limit-hit'.
systemd[1]: Failed to start OPAL PRD daemon

The fork() function, on success, returns pid of child process (pid > 0)
in the parent and 0 in the child. Instead of invoking memory worker
when return value pid == 0, it invokes worker when pid > 0 which is
parent process itself.

    pid = fork();
      if (pid > 0)
           exit(memory_error_worker(sysfsfile, typestr, i_start_addr,
							   i_endAddr));

The above logic causes the parent thread to exit after handling memory
error. Fix this by changing the if condition to (pid == 0).

Fixes: 8cbd0de88d16 ("opal-prd: Have a worker process handle page offlining")
Signed-off-by: Mahesh Salgaonkar <mahesh at linux.ibm.com>
---
 external/opal-prd/opal-prd.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/external/opal-prd/opal-prd.c b/external/opal-prd/opal-prd.c
index 1c610da4c..da947c827 100644
--- a/external/opal-prd/opal-prd.c
+++ b/external/opal-prd/opal-prd.c
@@ -755,9 +755,13 @@ int hservice_memory_error(uint64_t i_start_addr, uint64_t i_endAddr,
 	/*
 	 * HBRT expects the memory offlining process to happen in the background
 	 * after the notification is delivered.
+	 *
+	 * fork() return value:
+	 * On success, the PID of the child process is returned in the parent,
+	 * and 0 is returned in the child.
 	 */
 	pid = fork();
-	if (pid > 0)
+	if (pid == 0)
 		exit(memory_error_worker(sysfsfile, typestr, i_start_addr, i_endAddr));
 
 	if (pid < 0) {
-- 
2.51.0


From mahesh at linux.ibm.com  Sat Sep 20 03:19:42 2025
From: mahesh at linux.ibm.com (Mahesh Salgaonkar)
Date: Fri, 19 Sep 2025 22:49:42 +0530
Subject: [Skiboot] [PATCH 0/3] external/opal-prd fixes
Message-ID: <20250919171945.3947759-1-mahesh@linux.ibm.com>

Fix for opal-prd compilation failure and there are couple more fixes to
address issues in memory error handling code path.

Thanks,
-Mahesh.

Mahesh Salgaonkar (3):
  external/opal-prd: Fix compilation error
  external/opal-prd: Fix opal-prd service shutdown on memory errors
  external/opal-prd: remove misleading errno error message

 external/opal-prd/opal-prd.c       | 8 ++++++--
 {include => pldm/include}/endian.h | 0
 2 files changed, 6 insertions(+), 2 deletions(-)
 rename {include => pldm/include}/endian.h (100%)

-- 
2.51.0


From mahesh at linux.ibm.com  Sat Sep 20 03:19:45 2025
From: mahesh at linux.ibm.com (Mahesh Salgaonkar)
Date: Fri, 19 Sep 2025 22:49:45 +0530
Subject: [Skiboot] [PATCH 3/3] external/opal-prd: remove misleading errno
 error message
In-Reply-To: <20250919171945.3947759-1-mahesh@linux.ibm.com>
References: <20250919171945.3947759-1-mahesh@linux.ibm.com>
Message-ID: <20250919171945.3947759-4-mahesh@linux.ibm.com>

While handling memory error opal-prd displays misleading errno error
messages even when handler was successfully able to soft/hard offline the
requested memory page.

opal-prd[49096]: MEM: Memory error: range
0000000eeb445700-0000000eeb445700, type: correctable
opal-prd[49096]: MEM: Offlined 0000000eeb445700,0000000eeb455700, type
correctable: No such file or directory

In above example, an error message 'No such file or directory' was
displayed even after successfully offlining memory. This is because
printf in success case was using '%m' which prints errno based error
message. The value in errno is significant only when the return value of
the call indicated an error. The value of errno is never set to zero by
any system call or library function.

Hence, in success case do not use '%m' in printf to avoid misleading
error message

Signed-off-by: Mahesh Salgaonkar <mahesh at linux.ibm.com>
---
 external/opal-prd/opal-prd.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/external/opal-prd/opal-prd.c b/external/opal-prd/opal-prd.c
index da947c827..d85e3e9bd 100644
--- a/external/opal-prd/opal-prd.c
+++ b/external/opal-prd/opal-prd.c
@@ -721,7 +721,7 @@ static int memory_error_worker(const char *sysfsfile, const char *type,
 			ret = 1;
 		}
 	}
-	pr_log(LOG_CRIT, "MEM: Offlined %016lx,%016lx, type %s: %m\n",
+	pr_log(LOG_CRIT, "MEM: Offlined %016lx,%016lx, type %s\n",
 			i_start_addr, addr, type);
 
 	close(memfd);
-- 
2.51.0