Skip to contentSkip to conversation

log: allow for message continuation

PATCH Superseded Heinrich Schuchardt 4 participants 13 messages Sep 17, 2020 – Oct 11, 2020

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

cover log: allow for message continuation

With the first patch it becomes possible to continue a log message with the same log level and category as the previous messages.

We need this facility to convert pr_cont() to use our logging drivers.

The second patch provides a unit test.

This patch is a prerequisite:

[PATCH 1/1] log: mute messages generated by log drivers
https://lists.denx.de/pipermail/u-boot/2020-September/426418.html

Heinrich Schuchardt (2):
log: allow for message continuation
test: log: test message continuation

 common/log.c            | 23 ++++++++++++++----
 doc/develop/logging.rst |  6 +++++
 include/log.h           |  2 ++
 test/log/Makefile       |  4 +++-
 test/log/cont_test.c    | 52 +++++++++++++++++++++++++++++++++++++++++
 5 files changed, 81 insertions(+), 6 deletions(-)
 create mode 100644 test/log/cont_test.c

--
2.28.0

#1 lore raw

PATCH 1/2 log: allow for message continuation

Some drivers use macro pr_cont() for continuing a message sent via printk. Hence if we want to convert printk messaging to using the logging system, we must support continuation of log messages too.

As pr_cont() does not provide a message level we need a means of remembering the last log level.

With the patch a pseudo log level LOGL_CONT as well as a pseudo log category LOGC_CONT are introduced. Using these results in the application of the same log level and category as in the previous log message.

---
 common/log.c            | 23 ++++++++++++++++++-----
 doc/develop/logging.rst |  6 ++++++
 include/log.h           |  2 ++
 3 files changed, 26 insertions(+), 5 deletions(-)
common/log.c +18 -5
index 9a5f100da3..bafc09f263 100644
@@ -183,10 +183,12 @@ static bool log_passes_filters(struct log_device *ldev, struct log_rec *rec)
  * log_dispatch() - Send a log record to all log devices for processing
  *
  * The log record is sent to each log device in turn, skipping those which have
- * filters which block the record
+ * filters which block the record.
  *
- * @rec: Log record to dispatch
- * @return 0 (meaning success)
+ * All log messages created while processing log record @rec are ignored.
+ *
+ * @rec:	log record to dispatch
+ * Return:	0 msg sent, 1 msg not sent while already dispatching another msg
  */
 static int log_dispatch(struct log_rec *rec)
 {
@@ -199,7 +201,7 @@ static int log_dispatch(struct log_rec *rec)
 	 * as this might result in infinite recursion.
 	 */
 	if (processing_msg)
-		return 0;
+		return 1;
 
 	/* Emit message */
 	processing_msg = 1;
@@ -214,10 +216,18 @@ static int log_dispatch(struct log_rec *rec)
 int _log(enum log_category_t cat, enum log_level_t level, const char *file,
 	 int line, const char *func, const char *fmt, ...)
 {
+	static enum log_category_t logc_prev = LOGC_NONE;
+	static enum log_level_t logl_prev = LOGL_INFO;
 	char buf[CONFIG_SYS_CBSIZE];
 	struct log_rec rec;
 	va_list args;
 
+	/* Check for message continuation */
+	if (cat == LOGC_CONT)
+		cat = logc_prev;
+	if (level == LOGL_CONT)
+		level = logl_prev;
+
 	rec.cat = cat;
 	rec.level = level;
 	rec.file = file;
@@ -232,7 +242,10 @@ int _log(enum log_category_t cat, enum log_level_t level, const char *file,
 			gd->log_drop_count++;
 		return -ENOSYS;
 	}
-	log_dispatch(&rec);
+	if (!log_dispatch(&rec)) {
+		logc_prev = cat;
+		logl_prev = level;
+	}
 
 	return 0;
 }
doc/develop/logging.rst +6 -0
index 7ce8482ab6..c36f6bbbe4 100644
@@ -38,6 +38,9 @@ There are a number logging levels available, in increasing order of verbosity:
 * LOGL_DEBUG_CONTENT - Debug message showing full message content
 * LOGL_DEBUG_IO - Debug message showing hardware I/O access
 
+To continue a log message in a separate call of function log() use
+
+* LOGL_CONT - Use same log level as in previous call
 
 Logging category
 ----------------
@@ -56,6 +59,9 @@ The following main categories are defined:
 * LOGC_DT - Related to device tree control
 * LOGC_EFI - Related to EFI implementation
 
+To continue a log message in a separate call of function log() use
+
+* LOGC_CONT - Use same category as in previous call
 
 Enabling logging
 ----------------
include/log.h +2 -0
index 2859ce1f2e..567cd32d34 100644
@@ -35,6 +35,7 @@ enum log_level_t {
 
 	LOGL_FIRST = LOGL_EMERG,
 	LOGL_MAX = LOGL_DEBUG_IO,
+	LOGL_CONT = -1,		/* Use same log level as in previous call */
 };
 
 /**
@@ -60,6 +61,7 @@ enum log_category_t {
 
 	LOGC_COUNT,	/* Number of log categories */
 	LOGC_END,	/* Sentinel value for a list of log categories */
+	LOGC_CONT = -1,	/* Use same category as in previous call */
 };
 
 /* Helper to cast a uclass ID to a log category */

--
2.28.0

#2 lore raw

PATCH 1/2 log: allow for message continuation

Hi Heinrich,

Quoted 49 lines · Heinrich Schuchardt

Some drivers use macro pr_cont() for continuing a message sent via printk. Hence if we want to convert printk messaging to using the logging system, we must support continuation of log messages too.

As pr_cont() does not provide a message level we need a means of remembering the last log level.

With the patch a pseudo log level LOGL_CONT as well as a pseudo log category LOGC_CONT are introduced. Using these results in the application of the same log level and category as in the previous log message.

---
 common/log.c            | 23 ++++++++++++++++++-----
 doc/develop/logging.rst |  6 ++++++
 include/log.h           |  2 ++
 3 files changed, 26 insertions(+), 5 deletions(-)
common/log.c +6 -4
index 9a5f100da3..bafc09f263 100644
@@ -183,10 +183,12 @@ static bool log_passes_filters(struct log_device *ldev, struct log_rec *rec)
  * log_dispatch() - Send a log record to all log devices for processing
  *
  * The log record is sent to each log device in turn, skipping those which have
- * filters which block the record
+ * filters which block the record.
  *
- * @rec: Log record to dispatch
- * @return 0 (meaning success)
+ * All log messages created while processing log record @rec are ignored.
+ *
+ * @rec:       log record to dispatch
+ * Return:     0 msg sent, 1 msg not sent while already dispatching another msg
  */
 static int log_dispatch(struct log_rec *rec)
 {
@@ -199,7 +201,7 @@ static int log_dispatch(struct log_rec *rec)
         * as this might result in infinite recursion.
         */
        if (processing_msg)
-               return 0;
+               return 1;
 
        /* Emit message */
        processing_msg = 1;

common/log.c static int log_dispatch(struct log_rec *rec) · line 220

Quoted 5 lines
common/log.c +2 -0
@@ -214,10 +216,18 @@ static int log_dispatch(struct log_rec *rec)
 int _log(enum log_category_t cat, enum log_level_t level, const char *file,
         int line, const char *func, const char *fmt, ...)
 {
+       static enum log_category_t logc_prev = LOGC_NONE;
+       static enum log_level_t logl_prev = LOGL_INFO;

I don't think we can use static variables in logging. Perhaps we can use gobal_data?

common/log.c static int log_dispatch(struct log_rec *rec) · line 226

Quoted 6 lines
common/log.c +2 -0
        char buf[CONFIG_SYS_CBSIZE];
        struct log_rec rec;
        va_list args;
 
+       /* Check for message continuation */
+       if (cat == LOGC_CONT)

Regards,
Simon

#3 lore raw

PATCH 1/2 log: allow for message continuation

Quoted 49 lines · Simon Glass

Hi Heinrich,

Quoted 46 lines · Heinrich Schuchardt

Some drivers use macro pr_cont() for continuing a message sent via printk. Hence if we want to convert printk messaging to using the logging system, we must support continuation of log messages too.

As pr_cont() does not provide a message level we need a means of remembering the last log level.

With the patch a pseudo log level LOGL_CONT as well as a pseudo log category LOGC_CONT are introduced. Using these results in the application of the same log level and category as in the previous log message.

---
 common/log.c            | 23 ++++++++++++++++++-----
 doc/develop/logging.rst |  6 ++++++
 include/log.h           |  2 ++
 3 files changed, 26 insertions(+), 5 deletions(-)
common/log.c +6 -4
index 9a5f100da3..bafc09f263 100644
@@ -183,10 +183,12 @@ static bool log_passes_filters(struct log_device *ldev, struct log_rec *rec)
  * log_dispatch() - Send a log record to all log devices for processing
  *
  * The log record is sent to each log device in turn, skipping those which have
- * filters which block the record
+ * filters which block the record.
  *
- * @rec: Log record to dispatch
- * @return 0 (meaning success)
+ * All log messages created while processing log record @rec are ignored.
+ *
+ * @rec:       log record to dispatch
+ * Return:     0 msg sent, 1 msg not sent while already dispatching another msg
  */
 static int log_dispatch(struct log_rec *rec)
 {
@@ -199,7 +201,7 @@ static int log_dispatch(struct log_rec *rec)
         * as this might result in infinite recursion.
         */
        if (processing_msg)
-               return 0;
+               return 1;
 
Quoted 11 lines

common/log.c static int log_dispatch(struct log_rec *rec) · line 220

Quoted 8 lines
common/log.c +2 -0
        /* Emit message */
        processing_msg = 1;
@@ -214,10 +216,18 @@ static int log_dispatch(struct log_rec *rec)
 int _log(enum log_category_t cat, enum log_level_t level, const char *file,
         int line, const char *func, const char *fmt, ...)
 {
+       static enum log_category_t logc_prev = LOGC_NONE;
+       static enum log_level_t logl_prev = LOGL_INFO;

I don't think we can use static variables in logging. Perhaps we can use gobal_data?

Are you worried about relocation?

The initialization of the global data fields should be done in log_init() before gd->flags |= GD_FLG_LOG_READY; I assume.

Is the rest ok for you?

Best regards

Heinrich

Quoted 12 lines

common/log.c static int log_dispatch(struct log_rec *rec) · line 226

Quoted 8 lines
common/log.c +2 -0
        char buf[CONFIG_SYS_CBSIZE];
        struct log_rec rec;
        va_list args;
 
+       /* Check for message continuation */
+       if (cat == LOGC_CONT)

Regards,
Simon

#4 lore raw

PATCH 1/2 log: allow for message continuation

Hi Heinrich,

Quoted 63 lines · Heinrich Schuchardt
Quoted 49 lines · Simon Glass

Hi Heinrich,

Quoted 46 lines · Heinrich Schuchardt

Some drivers use macro pr_cont() for continuing a message sent via printk. Hence if we want to convert printk messaging to using the logging system, we must support continuation of log messages too.

As pr_cont() does not provide a message level we need a means of remembering the last log level.

With the patch a pseudo log level LOGL_CONT as well as a pseudo log category LOGC_CONT are introduced. Using these results in the application of the same log level and category as in the previous log message.

---
 common/log.c            | 23 ++++++++++++++++++-----
 doc/develop/logging.rst |  6 ++++++
 include/log.h           |  2 ++
 3 files changed, 26 insertions(+), 5 deletions(-)
common/log.c +6 -4
index 9a5f100da3..bafc09f263 100644
@@ -183,10 +183,12 @@ static bool log_passes_filters(struct log_device *ldev, struct log_rec *rec)
  * log_dispatch() - Send a log record to all log devices for processing
  *
  * The log record is sent to each log device in turn, skipping those which have
- * filters which block the record
+ * filters which block the record.
  *
- * @rec: Log record to dispatch
- * @return 0 (meaning success)
+ * All log messages created while processing log record @rec are ignored.
+ *
+ * @rec:       log record to dispatch
+ * Return:     0 msg sent, 1 msg not sent while already dispatching another msg
  */
 static int log_dispatch(struct log_rec *rec)
 {
@@ -199,7 +201,7 @@ static int log_dispatch(struct log_rec *rec)
         * as this might result in infinite recursion.
         */
        if (processing_msg)
-               return 0;
+               return 1;
 
Quoted 11 lines

common/log.c static int log_dispatch(struct log_rec *rec) · line 220

Quoted 8 lines
common/log.c +2 -0
        /* Emit message */
        processing_msg = 1;
@@ -214,10 +216,18 @@ static int log_dispatch(struct log_rec *rec)
 int _log(enum log_category_t cat, enum log_level_t level, const char *file,
         int line, const char *func, const char *fmt, ...)
 {
+       static enum log_category_t logc_prev = LOGC_NONE;
+       static enum log_level_t logl_prev = LOGL_INFO;

I don't think we can use static variables in logging. Perhaps we can use gobal_data?

Quoted 1 line

Are you worried about relocation?

Yes, and SPL.

Quoted 3 lines

The initialization of the global data fields should be done in log_init() before gd->flags |= GD_FLG_LOG_READY; I assume.

Yes.

Quoted 2 lines

Is the rest ok for you?

Yes. If you are adding new things to global_data you could convert default_log_level to a char to avoid using more space.

Quoted 18 lines

Best regards

Heinrich

Quoted 12 lines

common/log.c static int log_dispatch(struct log_rec *rec) · line 226

Quoted 8 lines
common/log.c +2 -0
        char buf[CONFIG_SYS_CBSIZE];
        struct log_rec rec;
        va_list args;
 
+       /* Check for message continuation */
+       if (cat == LOGC_CONT)

Regards,
Simon

#5 lore raw

PATCH 1/2 log: allow for message continuation

Hi Heinrich,

Quoted 75 lines · Simon Glass

Hi Heinrich,

Quoted 63 lines · Heinrich Schuchardt
Quoted 49 lines · Simon Glass

Hi Heinrich,

Quoted 46 lines · Heinrich Schuchardt

Some drivers use macro pr_cont() for continuing a message sent via printk. Hence if we want to convert printk messaging to using the logging system, we must support continuation of log messages too.

As pr_cont() does not provide a message level we need a means of remembering the last log level.

With the patch a pseudo log level LOGL_CONT as well as a pseudo log category LOGC_CONT are introduced. Using these results in the application of the same log level and category as in the previous log message.

---
 common/log.c            | 23 ++++++++++++++++++-----
 doc/develop/logging.rst |  6 ++++++
 include/log.h           |  2 ++
 3 files changed, 26 insertions(+), 5 deletions(-)
common/log.c +6 -4
index 9a5f100da3..bafc09f263 100644
@@ -183,10 +183,12 @@ static bool log_passes_filters(struct log_device *ldev, struct log_rec *rec)
  * log_dispatch() - Send a log record to all log devices for processing
  *
  * The log record is sent to each log device in turn, skipping those which have
- * filters which block the record
+ * filters which block the record.
  *
- * @rec: Log record to dispatch
- * @return 0 (meaning success)
+ * All log messages created while processing log record @rec are ignored.
+ *
+ * @rec:       log record to dispatch
+ * Return:     0 msg sent, 1 msg not sent while already dispatching another msg
  */
 static int log_dispatch(struct log_rec *rec)
 {
@@ -199,7 +201,7 @@ static int log_dispatch(struct log_rec *rec)
         * as this might result in infinite recursion.
         */
        if (processing_msg)
-               return 0;
+               return 1;
 
Quoted 11 lines

common/log.c static int log_dispatch(struct log_rec *rec) · line 220

Quoted 8 lines
common/log.c +2 -0
        /* Emit message */
        processing_msg = 1;
@@ -214,10 +216,18 @@ static int log_dispatch(struct log_rec *rec)
 int _log(enum log_category_t cat, enum log_level_t level, const char *file,
         int line, const char *func, const char *fmt, ...)
 {
+       static enum log_category_t logc_prev = LOGC_NONE;
+       static enum log_level_t logl_prev = LOGL_INFO;

I don't think we can use static variables in logging. Perhaps we can use gobal_data?

Quoted 1 line

Are you worried about relocation?

Yes, and SPL.

Quoted 3 lines

The initialization of the global data fields should be done in log_init() before gd->flags |= GD_FLG_LOG_READY; I assume.

Quoted 8 lines

Yes.

Quoted 2 lines

Is the rest ok for you?

Yes. If you are adding new things to global_data you could convert default_log_level to a char to avoid using more space.

Also I notice that the processing_msg variable in log.c crashes logging in SPL/TPL. Can you please move this to global_data too?

Regards,
Simon

#6 lore raw

PATCH 2/2 test: log: test message continuation

In reply to Heinrich Schuchardt

Provide a unit test checking that a continuation message will use the same log level and log category as the previous message.

---
 test/log/Makefile    |  4 +++-
 test/log/cont_test.c | 52 ++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 55 insertions(+), 1 deletion(-)
 create mode 100644 test/log/cont_test.c
test/log/Makefile +3 -1
index 4c92550f6e..1f91c90422 100644
@@ -12,7 +12,9 @@ ifdef CONFIG_SANDBOX
 obj-$(CONFIG_LOG_SYSLOG) += syslog_test.o
 endif
 
-ifndef CONFIG_LOG
+ifdef CONFIG_LOG
+obj-$(CONFIG_CONSOLE_RECORD) += cont_test.o
+else
 obj-$(CONFIG_CONSOLE_RECORD) += nolog_test.o
 endif
 
test/log/cont_test.c new file +52 -0

52 lines not on this page · raw message

--
2.28.0

#7 lore raw

PATCH 2/2 test: log: test message continuation

Quoted 10 lines · Heinrich Schuchardt

Provide a unit test checking that a continuation message will use the same log level and log category as the previous message.

---
 test/log/Makefile    |  4 +++-
 test/log/cont_test.c | 52 ++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 55 insertions(+), 1 deletion(-)
 create mode 100644 test/log/cont_test.c
#8 lore raw

PATCH 2/2 test: log: test message continuation

In reply to Heinrich Schuchardt
Quoted 5 lines · Heinrich Schuchardt

Provide a unit test checking that a continuation message will use the same log level and log category as the previous message.

[snip]

test/log/cont_test.c line 32

Quoted 1 line
test/log/cont_test.c new file +1 -0
+	log(LOGC_CONT, LOGL_CONT, "cc%d\n", 2);

These new values aren't defined, or I missed some other series. Thanks!

Signature · 8 lines
Tom
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 659 bytes
Desc: not available
URL: <https://lists.denx.de/pipermail/u-boot/attachments/20201010/4fa914cd/attachment.sig>
#9 lore raw

PATCH 2/2 test: log: test message continuation

Quoted 5 lines · Tom Rini
Quoted 3 lines · Heinrich Schuchardt

Provide a unit test checking that a continuation message will use the same log level and log category as the previous message.

Quoted 7 lines
Quoted 2 lines

[snip]

test/log/cont_test.c line 32

Quoted 1 line
test/log/cont_test.c new file +1 -0
+	log(LOGC_CONT, LOGL_CONT, "cc%d\n", 2);

These new values aren't defined, or I missed some other series. Thanks!

They should be defined in the first patch

Quoted 23 lines · Heinrich Schuchardt
include/log.h +2 -1
index 2859ce1f2e..567cd32d34 100644
@@ -35,6 +35,7 @@ enum log_level_t {
 
 	LOGL_FIRST = LOGL_EMERG,
 	LOGL_MAX = LOGL_DEBUG_IO,
+	LOGL_CONT = -1,		/* Use same log level as in previous call */
 };
 
 /**
@@ -60,6 +61,7 @@ enum log_category_t {
 
 	LOGC_COUNT,	/* Number of log categories */
 	LOGC_END,	/* Sentinel value for a list of log categories */
+	LOGC_CONT = -1,	/* Use same category as in previous call */
 };
 
 /* Helper to cast a uclass ID to a log category */
--

2.28.0

#10 lore raw

PATCH 2/2 test: log: test message continuation

Quoted 8 lines · Sean Anderson
Quoted 7 lines · Tom Rini
Quoted 5 lines · Heinrich Schuchardt

Provide a unit test checking that a continuation message will use the same log level and log category as the previous message.

Quoted 7 lines
Quoted 5 lines

[snip]

test/log/cont_test.c line 32

Quoted 1 line
test/log/cont_test.c new file +1 -0
+	log(LOGC_CONT, LOGL_CONT, "cc%d\n", 2);

These new values aren't defined, or I missed some other series. Thanks!

They should be defined in the first patch

and I see 1/2 has changes requested, so I shoulda deferred this one. Thanks!

Signature · 8 lines
Tom
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 659 bytes
Desc: not available
URL: <https://lists.denx.de/pipermail/u-boot/attachments/20201010/d41b36cb/attachment.sig>
#11 lore raw

PATCH 2/2 test: log: test message continuation

Quoted 15 lines · Tom Rini
Quoted 14 lines · Sean Anderson
Quoted 12 lines · Tom Rini
Quoted 5 lines · Heinrich Schuchardt

Provide a unit test checking that a continuation message will use the same log level and log category as the previous message.

[snip]

test/log/cont_test.c line 32

Quoted 1 line
test/log/cont_test.c new file +1 -0
+	log(LOGC_CONT, LOGL_CONT, "cc%d\n", 2);

These new values aren't defined, or I missed some other series. Thanks!

Quoted 5 lines
Quoted 1 line

They should be defined in the first patch

and I see 1/2 has changes requested, so I shoulda deferred this one. Thanks!

Hello Tom,

I am still waiting on the "doc: global data pointer" series to be merged as this will be the basis for adjusting the log continuation series.

https://patchwork.ozlabs.org/project/uboot/list/?series=205937

But I will be patient as it seems that you have been drowned in patches assigned to you.

Best regards

Heinrich

#12 lore raw

PATCH 2/2 test: log: test message continuation

In reply to Tom Rini
Quoted 5 lines · Tom Rini
Quoted 3 lines · Heinrich Schuchardt

Provide a unit test checking that a continuation message will use the same log level and log category as the previous message.

Quoted 7 lines
Quoted 2 lines

[snip]

test/log/cont_test.c line 32

Quoted 1 line
test/log/cont_test.c new file +1 -0
+	log(LOGC_CONT, LOGL_CONT, "cc%d\n", 2);

These new values aren't defined, or I missed some other series. Thanks!

See here:

https://patchwork.ozlabs.org/project/uboot/patch/20200917121902.57403-2-xypron.glpk at gmx.de/

But the series has to be reworked.

Best regards

Heinrich