Re: [PATCH v2 3/8] crash_core: change parse_crashkernel() to support crashkernel=,high|low parsing

[Baoquan He]

On 08/31/23 at 10:56am, Leizhen (ThunderTown) wrote:
> 
> 
> On 2023/8/29 20:16, Baoquan He wrote:
> > Now parse_crashkernel() is a real entry point for all kinds of
> > crahskernel parsing on any architecture.
> > 
> > And wrap the crahskernel=,high|low handling inside
> > CONFIG_ARCH_HAS_GENERIC_CRASHKERNEL_RESERVATION ifdeffery scope.
> > 
> > Signed-off-by: Baoquan He <bhe@xxxxxxxxxx>
> > ---
> >  include/linux/crash_core.h |  6 ++++++
> >  kernel/crash_core.c        | 28 +++++++++++++++++++++++++++-
> >  2 files changed, 33 insertions(+), 1 deletion(-)
> > 
> > diff --git a/include/linux/crash_core.h b/include/linux/crash_core.h
> > index 2e76289699ff..85260bf4a734 100644
> > --- a/include/linux/crash_core.h
> > +++ b/include/linux/crash_core.h
> > @@ -77,6 +77,12 @@ Elf_Word *append_elf_note(Elf_Word *buf, char *name, unsigned int type,
> >  			  void *data, size_t data_len);
> >  void final_note(Elf_Word *buf);
> >  
> > +#ifdef CONFIG_ARCH_HAS_GENERIC_CRASHKERNEL_RESERVATION
> > +#ifndef DEFAULT_CRASH_KERNEL_LOW_SIZE
> > +#define DEFAULT_CRASH_KERNEL_LOW_SIZE  (128UL << 20)
> > +#endif
> > +#endif
> > +
> >  int __init parse_crashkernel(char *cmdline, unsigned long long system_ram,
> >  		unsigned long long *crash_size, unsigned long long *crash_base,
> >  		unsigned long long *low_size, bool *high);
> > diff --git a/kernel/crash_core.c b/kernel/crash_core.c
> > index f6a5c219e2e1..355b0ab5189c 100644
> > --- a/kernel/crash_core.c
> > +++ b/kernel/crash_core.c
> > @@ -276,6 +276,9 @@ static int __init __parse_crashkernel(char *cmdline,
> >  /*
> >   * That function is the entry point for command line parsing and should be
> >   * called from the arch-specific code.
> > + *
> > + * If crashkernel=,high|low is supported on architecture, non-NULL values
> > + * should be passed to parameters 'low_size' and 'high'.
> >   */
> >  int __init parse_crashkernel(char *cmdline,
> >  			     unsigned long long system_ram,
> > @@ -291,7 +294,30 @@ int __init parse_crashkernel(char *cmdline,
> >  				crash_base, NULL);
> >  	if (!high)
> >  		return ret;
> > -
> > +#ifdef CONFIG_ARCH_HAS_GENERIC_CRASHKERNEL_RESERVATION
> > +	else if (ret == -ENOENT) {
> > +		ret = __parse_crashkernel(cmdline, 0, crash_size,
> > +				crash_base, suffix_tbl[SUFFIX_HIGH]);
> > +		if (ret || !*crash_size)
> > +			return -1;
> 
> Change -1 to -EINVAL?

Thanks a lot for careful reviewing, Zhen Lei.

Here, it's fine to me, parse_crashkernel() returns 0 on success, other value
on failure. '-1' or '-EINVAL' is not different to me in this case. I can
update if you think '-EINVAL' is better.

> 
> > +
> > +		/*
> > +		 * crashkernel=Y,low can be specified or not, but invalid value
> > +		 * is not allowed.
> > +		 */
> > +		ret = __parse_crashkernel(cmdline, 0, low_size,
> > +				crash_base, suffix_tbl[SUFFIX_LOW]);
> > +		if (ret == -ENOENT)
> > +			*low_size = DEFAULT_CRASH_KERNEL_LOW_SIZE;
> > +		else if (ret)
> > +			return -1;
> 
> return ret;

Ditto.

> 
> > +
> > +		*high = true;
> > +	} else if (ret || !*crash_size) {
> 
> This check can be moved outside of #ifdef. Because even '!high', it's completely
> applicable. The overall adjustment is as follows:

Hmm, the current logic is much easier to understand. However, I may not
100% get your suggestion. Can you paste the complete code in your
suggested way? Do not need 100% correct code, just the skeleton of code logic
so that I can better understand it and add inline comment.

> 
> -  	if (!high)
> -  		return ret;
> 
> #ifdef CONFIG_ARCH_HAS_GENERIC_CRASHKERNEL_RESERVATION
> 	if (high && ret == -ENOENT) {
> 		... ...
> 		if (ret || !*crash_size)	//parse HIGH
> 		... ...
> 	}
> 
> 	//At this point, *crash_size is not 0 and ret is 0.
> 	//We can also delete if (!*crash_size) above because it will be checked later.
> #endif
> 
> 	if (!*crash_size)
> 		ret = -EINVAL;
> 
> 	return ret;

When crashkernel=,high is specified while crashkernel=,low is omitted,
the ret==-ENOENT, but we can't return ret directly. That is still an
acceptable way.

> 
> -  	return 0;
> 
> > +		/* The specified value is invalid */
> > +		return -1;
> > +	}
> > +#endif
> >  	return 0;
> >  }
> >  
> > 
> 
> -- 
> Regards,
>   Zhen Lei
>