CMN_ERR(9F) Kernel Functions for Drivers CMN_ERR(9F)

cmn_err, dev_err, vcmn_err, zcmn_err, vzcmn_err — display an error message or panic the system

#include <sys/cmn_err.h>
#include <sys/ddi.h>
#include <sys/sunddi.h>

void
cmn_err(int level, char *format, ...);

void
dev_err(dev_info_t *dip, int level, char *format, ...);

#include <sys/varargs.h>

void
vcmn_err(int level, char *format, va_list ap);

#include <sys/types.h>

void
zcmn_err(zoneid_t zoneid, int level, char *format, ...);

void
vzcmn_err(zoneid_t zoneid, int level, char *format, va_list ap);

Architecture independent level 1 (DDI/DKI).

()

level
A constant indicating the severity of the error condition.
format
Message to be displayed.

()

The dev_err() function works exactly like cmn_err(), but includes an additional argument:

dip
A pointer to the device's dev_info structure.

()

The vcmn_err() function takes level and format as described for cmn_err(), but its third argument is different:

ap
Variable argument list passed to the function.

()

The zcmn_err() function works exactly like cmn_err(), but includes an additional argument:

zoneid
Zone to which log messages should be directed. See zones(7).

()

The vzcmn_err() function works exactly like vcmn_err(), but includes an additional argument:

zoneid
Zone to which log messages should be directed. See zones(7).

()

The cmn_err() function displays a specified message on the console. cmn_err() can also panic the system. When the system panics, it attempts to save recent changes to data, display a "panic message" on the console, attempt to write a core file, and halt system processing. See the CE_PANIC level below.

level is a constant indicating the severity of the error condition. The four severity levels are:

Used to continue another message or to display an informative message not associated with an error. Note that multiple CE_CONT messages without a newline may or may not appear on the system console or in the system log as a single line message. A single line message may be produced by constructing the message with sprintf(9F) or vsprintf(9F) before calling ().
Used to display a message preceded with . This message is used to report system events that do not necessarily require user action, but may interest the system administrator. For example, a message saying that a sector on a disk needs to be accessed repeatedly before it can be accessed correctly might be noteworthy.
Used to display a message preceded with . This message is used to report system events that require immediate attention, such as those where if an action is not taken, the system may panic. For example, when a peripheral device does not initialize correctly, this level should be used.
Used to display a message preceded with "", and to panic the system. Drivers should specify this level only under the most severe conditions or when debugging a driver. A valid use of this level is when the system cannot continue to function. If the error is recoverable, or not essential to continued system operation, do not panic the system.

The format is identical to the one described in sprintf(9F) with additional meaning of the first character affecting where the message will be written:

!
The message goes only to the system log.
The message goes only to the console.
?
If level is also CE_CONT, the message is always sent to the system log, but is only written to the console when the system has been booted in verbose mode. See kernel(8). If neither condition is met, the "?" character has no effect and is simply ignored.

Refer to syslogd(8) to determine where the system log is written.

The () function sends log messages to the log of the global zone. cmn_err() appends a to each format, except when level is CE_CONT.

dev_err()

With the exception of its first argument (dip), dev_err() is identical to cmn_err(). dip is a pointer to a device's dev_info structure, which is used to prepend the driver name and instance number to the message. The driver name and instance number are retrieved using ddi_driver_name(9F) and ddi_get_instance(9F).

vcmn_err()

The vcmn_err() function is identical to cmn_err() except that its last argument, ap, is a pointer to a variable list of arguments. ap contains the list of arguments used by the conversion specifications in format. ap must be initialized by calling va_start(9F). va_end(9F) is used to clean up and must be called after each traversal of the list. Multiple traversals of the argument list, each bracketed by va_start(9F) and va_end(9F), are possible.

zcmn_err()

With the exception of its first argument (zoneid), zcmn_err() is identical to cmn_err(). zoneid is the numeric ID of the zone to which the message should be directed. Note that zoneid only has an effect if the message is sent to the system log. Using zoneid will cause messages to be sent to the log associated with the specified local zone rather than the log in the global zone. This is accomplished by the message being received and processed by the syslogd(8) process running in the specified zone instead of the one running in the global zone. You can retrieve a process zone ID from its credential structure using crgetzoneid(9F).

vzcmn_err()

With the exception of its first argument (zoneid), vzcmn_err() is identical to vcmn_err(). See the description of zcmn_err() above for an explanation on how the zoneid argument is handled.

The cmn_err() function can be called from user, kernel, interrupt, or high-level interrupt context.

None. However, if an unknown level is passed to cmn_err(), the following panic error message is displayed:

panic: unknown level in cmn_err (level=level, msg=format)

Using cmn_err()
This first example shows how cmn_err() can record tracing and debugging information only in the system log (lines 17); display problems with a device only on the system console (line 23); or display problems with the device on both the system console and in the system log (line 28). 
1  struct  reg {
2          uchar_t data;
3          uchar_t csr;
4  };
5
6  struct  xxstate {
7          ...
8          dev_info_t *dip;
9          struct reg *regp;
10         ...
11  };
12
13  dev_t dev;
14  struct xxstate *xsp;
15    ...
16  #ifdef DEBUG    /* in debugging mode, log function call */
17     cmn_err(CE_CONT, "!%s%d: xxopen function called.",
18          ddi_binding_name(xsp->dip), getminor(dev));
19  #endif  /* end DEBUG */
20    ...
21  /* display device power failure on system console */
22     if ((xsp->regp->csr & POWER) == OFF)
23          cmn_err(CE_NOTE, "^OFF.",
24               ddi_binding_name(xsp->dip), getminor(dev));
25    ...
26  /* display warning if device has bad VTOC */
27     if (xsp->regp->csr & BADVTOC)
28          cmn_err(CE_WARN, "%s%d: xxopen: Bad VTOC.",
29               ddi_binding_name(xsp->dip), getminor(dev));
Using the %b conversion specification
This example shows how to use the conversion specification. Because of the leading "?" character in the format string, this message will always be logged, but it will only be displayed when the kernel is booted in verbose mode. 
cmn_err(CE_CONT, "?reg=0x%b\n", regval, "\020\3Intr\2Err\1Enable");
Using regval
When regval is set to (decimal) 13, the following message would be displayed: 
reg=0xd<Intr,,Enable>
Error Routine
This example shows an error reporting routine which accepts a variable number of arguments and displays a single line error message both in the system log and on the system console. Note the use of vsprintf() to construct the error message before calling cmn_err(). 
#include <sys/varargs.h>
#include <sys/ddi.h>
#include <sys/sunddi.h>
#define MAX_MSG 256;

void
xxerror(dev_info_t *dip, int level, const char *fmt, ...)
{
    va_list     ap;
    int         instance;
    char        buf[MAX_MSG], *name;

instance = ddi_get_instance(dip);
name = ddi_binding_name(dip);

/* format buf using fmt and arguments contained in ap */

va_start(ap, fmt);
vsprintf(buf, fmt, ap);
va_end(ap);

/* pass formatted string to cmn_err(9F) */

cmn_err(level, "%s%d: %s", name, instance, buf);

}
Log to Current Zone
This example shows how messages can be sent to the log of the zone in which a thread is currently running, when applicable. Note that most hardware-related messages should instead be sent to the global zone using cmn_err(). 
zcmn_err(crgetzoneid(ddi_get_cred()), CE_NOTE, "out of processes");

zones(7), dmesg(8), kernel(8), ddi_binding_name(9F), ddi_cred(9F), ddi_driver_name(9F), ddi_get_instance(9F), sprintf(9F), va_arg(9F), va_end(9F), va_start(9F)

Writing Device Drivers

The () function with the CE_CONT argument can be used by driver developers as a driver code debugging tool. However, using cmn_err() in this capacity can change system timing characteristics.

Messages of arbitrary length can be generated using cmn_err(), but if the call to cmn_err() is made from high-level interrupt context and insufficient memory is available to create a buffer of the specified size, the message will be truncated to LOG_MSGSIZE bytes (see sys/log.h). For this reason, callers of cmn_err() that require complete and accurate message generation should post down from high-level interrupt context before calling cmn_err().

December 1, 2016 OmniOS