ppc manual paging question

Wang, Baojun wangbj at lzu.edu.cn
Mon Oct 22 15:50:31 EST 2007


On Monday 22 October 2007 12:50:52, Benjamin Herrenschmidt wrote:
> On Mon, 2007-10-22 at 12:03 +0800, Wang, Baojun wrote:
> > hi,
> >
> >   I've got some qeustion about ppc(ppc44x) paging:
> >
> > how can I manually map a virtual address to a physical address through a
> > specific pgd? How does ppc translate virt address to physical one? I
> > think besides from tlb, the CPU will search the page table entries via
> > the pgd, can I alter the pgd value to change the memory translation?
> > under i386, it's very simple, we can just rewrite %%cr3, it even could
> > invalidate all tlb entries automatically, how can I do this under ppc?
> > I've tried rewrite
> > current->mm->pgd and current->thread.pgdir, but sounds like it still not
> > insufficiant, am I missing something vital?
>
> What the heck are you trying to do ? Please explain and I'll tell you
> how to do it properly :-)

I'm porting an adeos nano kernel named xtratum (http://www.xtratum.org) from 
x86 to ppc, I think I'm near the ending except the above problem. xtratum is 
doing things like xen but it's much simpler (it's aimed for realtime), it 
need provides memory space sperations for it's domains, so I need manually 
paging. Each domain is loaded by a userspace program (instead of the root 
domain as a kernel module), the loader will load the domain's (ELF staticly 
excutable) PT_LOAD section into memory, and then raise a properly system call 
(passing the structurized loaded data as arguments) to load the domain via 
load_domain_sys(), and at the last step of loading the domain, xtratum will 
jump to the entry code of the new domain(asm wrappered start() routine) and 
then everything should be fine. The problem now is as follow:

under my ppc (440GR/440EP) platform, start() is always at 0x100000a0, but I 
guess there is something wrong with my mm code so after the domain is loaded, 
the virt addres 0x100000a0 just point to garbage instead of the right start() 
routine. So how can I setup paging properly so that the virtual memory could 
be translated to proper data?

I can describe in more detail if neccessary.

Thanks very much for take care.

  Regards,
Wang

P.S:

The orignal xtratum (x86) code:

#define load_pd(pd) {\
  __asm__ __volatile__ ("movl %0,%%cr3": :"r" (__pa(pd))); \
}

#define save_pd(pd) {\
  __asm__ __volatile__ ("movl %%cr3, %0\n\t": "=r" (pd) :); \
  pd = (unsigned long) __va (pd); \
}

...

// Virtual address to page directory entry
#define va2pd(vaddress) ((vaddress) >> PGDIR_SHIFT)

// Virtual address to page table entry
#define va2pt(vaddress) (((vaddress) & 0x3FF000) >> PAGE_SHIFT)

// Page directory and page table to virtual address
#define pdpt2va(pd, pt) (((pd) << PGDIR_SHIFT) | ((pt) << PAGE_SHIFT))

// Next macro allows to obtain a pt address through the page directory
#define get_pd_addr(pd, pd_entry) \
  ((unsigned long) __va (((unsigned long *)(pd)) [(pd_entry)] & PAGE_MASK))

// And the following one allows to  obtain a page address via the page
// table
#define get_pt_addr(pt, pt_entry) \
  ((unsigned long)__va (((unsigned long *)*(pt)) [(pt_entry)] & PAGE_MASK))

...

static inline void fill_pd_entry (unsigned long pd,
                                     unsigned long pd_entry,
                                     unsigned long pt,
                                     unsigned long flags) {
  ((unsigned long *)pd) [pd_entry] =
    ((__pa (pt) & PAGE_MASK) | (flags & 0xFFF));
}

static inline void fill_pt_entry (unsigned long pt,
                                     unsigned long pt_entry,
                                     unsigned long page,
                                     unsigned long flags) {
  ((unsigned long *)pt) [pt_entry] =
    ((__pa (page) & PAGE_MASK) | (flags & 0xFFF));
}

...

static inline unsigned long create_page_directory (unsigned long
                                                   (*alloc_page) (void)) {
  unsigned long pd = (*alloc_page) (), c_pd;

  if (!pd) return pd;
  save_pd(c_pd);

  memset  ((unsigned char *) &((unsigned long *) pd)[0],
           0, 1024 * sizeof (unsigned long));

  memcpy ((unsigned char *) &((unsigned long *) pd)[va2pd(PAGE_OFFSET)],
          (unsigned char *) &((unsigned long *) c_pd)[va2pd(PAGE_OFFSET)],
          (1024 - va2pd(PAGE_OFFSET)) * sizeof (unsigned long));

  return pd;
}

static inline unsigned long allocate_user_page (unsigned long pd,
                                                unsigned long vaddress,
                                                unsigned long (*alloc_page)
                                                (void)) {
  unsigned long pt_entry = va2pt(vaddress),
    pd_entry = va2pd(vaddress), pt, page = 0;

  if (vaddress >= PAGE_OFFSET) return 0;

  // Check if there is already an allocated pt in the pd table
  if (!(((unsigned long *)pd) [pd_entry] & _PAGE_PRESENT)) {
    if (!(pt = (unsigned long) (*alloc_page) ())) return 0;
    fill_pd_entry (pd, pd_entry,
                   pt, _PAGE_PRESENT | _PAGE_RW | _PAGE_USER);
  } else
    pt = get_pd_addr(pd, pd_entry);

  // Check whether there is already an allocated page in the pt table
  if (!(((unsigned long *)pt)[pt_entry] & _PAGE_PRESENT)) {
    if (!(page = (unsigned long) (*alloc_page) ())) return 0;
    fill_pt_entry (pt, pt_entry, page,
                   _PAGE_PRESENT | _PAGE_RW | _PAGE_USER);
  }
  return page;
}


> Cheers,
> Ben.



-- 
Wang, Baojun                                        Lanzhou University
Distributed & Embedded System Lab              http://dslab.lzu.edu.cn
School of Information Science and Engeneering        wangbj at lzu.edu.cn
Tianshui South Road 222. Lanzhou 730000                     .P.R.China
Tel:+86-931-8912025                                Fax:+86-931-8912022



More information about the Linuxppc-dev mailing list