[Gnewsense-dev] Re: [loongson-PATCH-v1 22/27] Hibernation Support in mips system

[Hongbing Hu]

yanh 写道:
> 在 2009-05-23六的 21:30 +0900，Atsushi Nemoto写道：
>   
> > On Sat, 23 May 2009 16:06:43 +0800, yanh <address@hidden> wrote:
> >     
> > > > > +unsigned long
> > > > > +       saved_ra,
> > > > > +       saved_sp,
> > > > >           
> > ...
> >     
> > > > > +       saved_v0,
> > > > > +       saved_v1;
> > > > >           
> > > > Instead of enumerating them, I would prefer something like "struct
> > > > pt_regs saved_regs" or "unsigned long saved_regs[32]".
> > > >         
> > > This implementation is referencing the x86 platform. 
> > > Not all the 32 reigsters are needed to save. 
> > > Maybe the whole registers needed to save can still be reduced.
> > >       
> > I did not mean save/restore all registers.  I just mean using only one
> > symbol (struct or array).  Though the struct or array contains some
> > unused members, it saves many instructions in
> > swsusp_arch_{suspend,resume}.
> >
> > For saving N registers, (N * 2) instructions are required to save to
> > individual variables, but (N + 2) instructions are required to save to
> > array or struct.
> >     
>
> Yes, the struct array method is more efficient, we will change to it.
>   
The length of registers  is different between  32bit kernel  and  64bit 
kernel.
That means the file hibernate.S wiil be divided into  hibernate_32.S  
and hibernate_64.S ?

> > > > > +void save_processor_state(void)
> > > > > +{
> > > > > +       saved_status = read_c0_status();
> > > > > +}
> > > > >           
> > > > No need to save/restore floating point registers?
> > > >         
> > > the floating point registers are not used by kernel, for user part, they
> > > are already saved while entering into kernel mode.
> > >       
> > No, floating point registers are not saved on entering into kernel.
> > They are saved on context switch.  
> >     
Yes, suspend to disk  will freeze processes at first,and i think the 
process wiil save the float point regs.
So there is no need to save them.
> It's my mistake. But I think, if there is need to save floating point
> registers, the kernel have already done for it. 
> What's need to save should be for the resuming parts. Those registers which are used while 
> in suspending path are needed to save.
>
> create_image()
>
>    in_suspend = 1;
>     save_processor_state();
>     error = swsusp_arch_suspend();
>     if (error)
>         printk(KERN_ERR "PM: Error %d creating hibernation image\n",
>             error);
>    
>   when resuming this functions is executed for righ after the  swsusp_arch_suspend().
>   This is not a normal function return.
>   So the registers need to save are those damaged by swsusp_arch_suspend and 
> what it calls.
>   
>   because this is not a normal function call, the callee saved registers can not be restored, so we need to save
> it explicitly. For caller saved temporary registers, they are already saved in 
> create_image.  Are this right?
>       
>   
---
> > Atsushi Nemoto
> >     
>
>
>
>   


--
---------------------------------------------------------
Hongbing,Hu (Software Department)
Tel:    0512-52308631
E-mail: address@hidden
MSN:    address@hidden
JiangSu Lemote Corp. Ltd.
MengLan, Yushan, Changshu, JiangSu Province, China
---------------------------------------------------------