Re: [PATCH -v3 1/4] printk, Add pr_pfx for library functions

[Huang Ying <ying.huang@intel.com>]

Hi, Joe,

On Thu, 2010-12-02 at 16:37 +0800, Joe Perches wrote:
> On Thu, 2010-12-02 at 16:05 +0800, Huang Ying wrote:
> > For library functions doing printk, the log level and line prefix
> > usually need to be specified by the caller. So this patch adds
> > "pr_pfx" to make the life of these library functions easier.
> []
> > +#define pr_pfx(pfx, fmt, ...) \
> > +	printk("%s" fmt, pfx, ##__VA_ARGS__)
> 
> I think this would be a very error prone style.
> 
> pr_<foo> uses are log levels.
> 
> For an casual reader, is pfx a new log level or
> is it a prefix or is it both?  I do understand
> what you intend.

Maybe we can rename it to avoid confusing. Any suggestion? Or just
pfx_pr?

> Mutability of the KERN_<level> is pretty unusual.
> Most all of the uses are ?: style

If printk is used in library functions, the log level and some prefix
need to be specified by the caller. That is the intended user. Such as
that in 2 of the patch series.

> Here are the first uses in each file that has them:
> There aren't many.
> 
> $ grep -rP -m1 --include=*.[ch] "KERN_[A-Z]+\s*:\s*KERN" *
> arch/avr32/mm/fault.c:			       is_global_init(tsk) ? KERN_EMERG : KERN_INFO,
> arch/sparc/kernel/traps_64.c:	       (recoverable ? KERN_WARNING : KERN_CRIT), smp_processor_id(),
> arch/sparc/mm/fault_64.c:	       task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
> arch/sparc/mm/fault_32.c:	       task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
> arch/x86/kernel/signal.c:		       task_pid_nr(current) > 1 ? KERN_INFO : KERN_EMERG,
> arch/x86/mm/fault.c:		task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
> arch/parisc/kernel/traps.c:	level = user ? KERN_DEBUG : KERN_CRIT;
> drivers/pci/pcie/aer/aerdrv_errprint.c:		KERN_WARNING : KERN_ERR, dev_driver_string(&pdev->dev),	\
> include/linux/arcdevice.h:            		: x < D_DURING ? KERN_INFO : KERN_DEBUG, \
> mm/internal.h:		printk(level <= MMINIT_WARNING ? KERN_WARNING : KERN_DEBUG); \
> sound/core/misc.c:	const char *pfx = level ? KERN_DEBUG : KERN_DEFAULT;
> 
> I don't see why you can't use something like that
> with a simpler flag for level.

That is doable. I just think let caller specify log level and line
prefix is more flexible, and uses less code.

> You also miss a chance to use pr_pfx in ghes_print_estatus:

Sorry, forget that.

> +       if (__ratelimit(&ratelimit)) {
> +               printk(
> +       "%s""Hardware error from APEI Generic Hardware Error Source: %d\n",
> +       pfx, ghes->generic->header.source_id);
> +               apei_estatus_print(pfx, ghes->estatus);
> +
> 
> ghes_print_estatus is only called with a NULL first argument.
> so the code use of pr_pfx seems dubious in any case.

In some following-up patches (not sent out yet), ghes_print_estatus may
be called with non-NULL first argument.

Best Regards,
Huang Ying