On 8/27/05, Vincenzo Mallozzi <vinjunior@xxxxxxxx> wrote:
> I'm really sorry, but just when all seems to work well, the problem
> re-occurred.
>
> LIST_HEAD(vm_write_protected);
>
> static struct mtpmc_wrprotected_pages *mk_page(unsigned long
> addr)
> {
> struct mtpmc_wrprotected_pages *wr_page;
>
> wr_page = (...) kmalloc(.......);
>
ideally you should also initialize your list element in page structure
here only. Well this is not the place of OOPS rite, but its a probable
source of error. Do add the following in this function
INIT_LIST_HEAD(&wr_page->list); // If you dont do this, it will
still work, but its preferable to initilize all the elements.
> wr_page->address = addr;
> return wr_page;
> }
>
> struct mtpmc_vm_wrprotected *mk_vm(struct vm_area_struct
> *vm)
> {
> struct mtpmc_vm_wrprotected *wr_vma;
>
> wr_vma = (...) kmalloc(...........);
> INIT_LIST_HEAD(&wr_vma->pages_list);
>
along with this, also initialize the other list element of this
structure here only, the same thing which I mentioned for page
structure above. Also set the vm_start and vm_end elements here only,
rather than doing then later in function as your are doing.
So add the following lines to this function
INIT_LIST_HEAD(&wr_vma->list); // If you dont do this, it will
still work, but its preferable to initilize all the elements.
vma->vm_start = vm->vm_start;
vma->vm_end = vm->vm_end;
start and end should be set here, earlier you wee seting them after
adding the structure to the list and i think out side the control of
spinlocks, so an OOPS ..... ;-) this is the right place to do it
before adding the structure to list.
> return wr_vma;
> }
>
> void mtpmc_set_mm_not_writable(struct mm_struct *mm)
> {
> struct mtpmc_vm_wrprotected *vma;
> struct mtpmc_wrprotected_pages *page;
> struct vm_area_struct *vm;
> unsigned long addr;
> pte_t *pte;
>
> down_write(&mm->mmap_sem);
>
> for(vm = mm->mmap; vm != NULL; vm = vm->vm_next){
> vma = mk_vm(vm);
> for (addr = vm->vm_start; addr<vm->vm_end; addr+=PAGE_SIZE){
> page = mk_page(addr);
> list_add_tail(&page->list, &vma->pages_list);
> }
>
> list_add_tail(&vma->list, &vm_write_protected);
> vma->vm_start = vm->vm_start;
> vma->vm_end = vma->vm_end; /* THE PROLEM IS HERE - I'LL DESCRIBE IT
Dont set start and end here ..... you should set them in above
function as mentioned earlier.
Well I can not see the use of spinlocks here, but I hope you are using
them at right places.
As mentioned earlier also, do all the modification or reading of
lisits under the control of spinlocks. I think the error you are doing
is that you do add the vma and page structure in there corresponding
lists under the control of spinlock but then you start modifing the
elements of these structure directly without using the spinlocks, you
should not do that as the structures are on lists when you manupulate
them, so do hold the spinlock before doing any thing on list, not only
addition or deletion of structures from list, but rather manupulation
of elements of these structures as well which are on list.
If still the problem exist, I would advice you to send the whole code
without any modifications, just pate the whole code and someone on the
list might rectify that.
> BELOW */
> }
>
> up_write(&mm->mmap_sem);
>
> return;
> }
>
> See the line indicated above where I execute "vma->vm_end = vma->vm_end;".
> In this line there's an error. It should be "vma->vm_end = vm->vm_end;".
> So I've changed vma in vm. The result is that the system crashes with the NULL
> exception I've described in previous emails.
> I'm sorry if I'm posting many times this problem but when all seems go well
> some problem occurs.
> Now, I've posted all interesting code, by simplifying it.
> I initialize all values of pages and vma before inserting them into the
> appropriate lists.
> I'm using spin_lock_irqsave/spin_unlock_irqrestore while I'm creating the
> lists and spinlock/spin_unlock when I scan the lists from within the page
> fault handler.
> I hope you've the patience to reply to this mail.
Dont worry about this, this list is to help each other and that is
what we all are doing on this.
-Gaurav
> Thanks.
> Vincenzo Mallozzi.
>
>
>
>
>
> ___________________________________
> Yahoo! Mail: gratis 1GB per i messaggi e allegati da 10MB
> http://mail.yahoo.it
>
>
--
Kernelnewbies: Help each other learn about the Linux kernel.
Archive: http://mail.nl.linux.org/kernelnewbies/
FAQ: http://kernelnewbies.org/faq/
