Re: Re: [PATCH 08/23] xen: statically initialize cpu_evtchn_mask_p

[Andrew Jones <drjones@redhat.com>]

On 01/25/2011 03:02 PM, Ian Campbell wrote:
> On Mon, 2011-01-24 at 17:44 +0000, Paolo Bonzini wrote:
>> On 10/12/2010 05:44 PM, Konrad Rzeszutek Wilk wrote:
>>> -static struct cpu_evtchn_s *cpu_evtchn_mask_p;
>>> +
>>> +static __initdata struct cpu_evtchn_s init_evtchn_mask = {
>>> +	.bits[0 ... (NR_EVENT_CHANNELS/BITS_PER_LONG)-1] = ~0ul,
>>> +};
>>> +static struct cpu_evtchn_s *cpu_evtchn_mask_p =&init_evtchn_mask;
>>> +
>>>   static inline unsigned long *cpu_evtchn_mask(int cpu)
>>>   {
>>>   	return cpu_evtchn_mask_p[cpu].bits;
>>
>> This causes a modpost warning:
>>
>>     WARNING: drivers/xen/built-in.o(.data+0x0): Section mismatch in
>>     reference from the variable cpu_evtchn_mask_p to the variable
>>     .init.data:init_evtchn_mask
>>
>>     The variable cpu_evtchn_mask_p references
>>     the variable __initdata init_evtchn_mask
>>
>>     If the reference is valid then annotate the
>>     variable with __init* or __refdata (see linux/init.h) or name the variable:
>>     *driver, *_template, *_timer, *_sht, *_ops, *_probe, *_probe_one, *_console, 
>>
>> This is harmless, the variable is initialized to non-init data
>> in an __init function.  The added noise is ugly, though.
> 
> Does this help? If I understand the comment which precedes  __initref
> correctly it is intended to address precisely this situation.
> 
> Ian.
> 8<---------
> 
> xen: events: mark cpu_evtchn_mask_p as __refdata
> 
> This variable starts out pointing at init_evtchn_mask which is marked
> __initdata but is set to point to a non-init data region in xen_init_IRQ
> which is itself an __init function so this is safe.
> 
> Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
> 
> diff --git a/drivers/xen/events.c b/drivers/xen/events.c
> index 31af0ac..5061af0 100644
> --- a/drivers/xen/events.c
> +++ b/drivers/xen/events.c
> @@ -114,7 +114,7 @@ struct cpu_evtchn_s {
>  static __initdata struct cpu_evtchn_s init_evtchn_mask = {
>  	.bits[0 ... (NR_EVENT_CHANNELS/BITS_PER_LONG)-1] = ~0ul,
>  };
> -static struct cpu_evtchn_s *cpu_evtchn_mask_p = &init_evtchn_mask;
> +static struct __refdata cpu_evtchn_s *cpu_evtchn_mask_p =
> &init_evtchn_mask;
>  
>  static inline unsigned long *cpu_evtchn_mask(int cpu)
>  {
> 

This does indeed fix it. Although you need __refdata to follow the
complete struct name 'struct cpu_evtchn_s' rather than just 'struct', i.e.

diff --git a/drivers/xen/events.c b/drivers/xen/events.c
index 7468147..a313890 100644
--- a/drivers/xen/events.c
+++ b/drivers/xen/events.c
@@ -114,7 +114,7 @@ struct cpu_evtchn_s {
 static __initdata struct cpu_evtchn_s init_evtchn_mask = {
        .bits[0 ... (NR_EVENT_CHANNELS/BITS_PER_LONG)-1] = ~0ul,
 };
-static struct cpu_evtchn_s *cpu_evtchn_mask_p = &init_evtchn_mask;
+static struct cpu_evtchn_s __refdata *cpu_evtchn_mask_p =
&init_evtchn_mask;

 static inline unsigned long *cpu_evtchn_mask(int cpu)
 {


Ian, are you going to push this to lkml in one of your batches?

Drew