On Wed, 2003-06-25 at 18:59, Ravi wrote:
>
> --- Jim Bauer <jfbauer@nfr.com> wrote:
>
> > > You don't really need zap_page_range() to remove the old
> > > mapping. Since you know the virtual address at which the
> > > page is mapped, you can also find out where the corresponding
> > > PTE (page table entry) is. Then you can clear the entry
> > > with ptep_get_and_clear(). This lets you save the old PTE,
> > > from which you can get to the page using pte_page(). You
> > > can then free the page or do whatever else you want with it.
> >
> > Know of a good example of this?
>
> I couldn't find an example in any 2.5.x driver. But here is
> what is needed:
>
> {
> struct mm_struct *mm;
> pgd_t *pgd;
> pmd_t *pmd;
> pte_t *ptep, pte;
> struct page *old_page;
> struct page *new_page = ...; /* new page to be mapped in */
> pgprot_t prot = ...; /* PTE flags. _PAGE_RW, PAGE_SHARED, etc.
> * Set appropriately. */
>
> mm = current->mm;
> spin_lock(&mm->page_table_lock);
> pgd = pgd_offset(mm, address);
> pmd = pmd_offset(pgd, address);
> ptep = pte_offset_map(pmd, address);
> pte = ptep_get_and_clear(ptep);
> set_pte(ptep, mk_pte(new_page, prot));
> spin_unlock(&mm->page_table_lock);
> old_page = pte_page(pte);
> ...
> }
>
> Not many error checks are needed in this case since we
> know that the required address definitely has a mapping
> in the process page table.
>
I tried this, but the user-space program's mapping was still
refering to the old pages. However, if I ran it through gdb and
single stepped it then it saw the new buffer. It I used gdb, but
did not single step, then the prog being debugged saw the old
pages, but gdb saw the new ones?
Anyone care to explain that?
--
Kernelnewbies: Help each other learn about the Linux kernel.
Archive: http://mail.nl.linux.org/kernelnewbies/
FAQ: http://kernelnewbies.org/faq/
