Xen devel - Feb 2013 - Re: xen/pvh linux: Use ballooning to allocate grant table pages

Hello Mukesh Rathor,

The patch 40f633eb0def: "xen/pvh linux: Use ballooning to allocate
grant table pages" from Jan 31, 2013, leads to the following Sparse
warning: "drivers/xen/grant-table.c:1095:28: error: bad constant
	expression"

drivers/xen/grant-table.c
  1092  static int xlated_setup_gnttab_pages(int numpages, void **addr)
  1093  {
  1094          int i, rc;
  1095          unsigned long pfns[numpages];
  1096          struct page *pages[numpages];
                                   ^^^^^^^^
Because the kernel uses an 8k stack if this is over 500 it will cause
a kernel crash.  It will crash before we reach 500 actually.  It looks
like typical values for this are around 4 so we''re probably safe, but
it''s still a bit nasty.

  1097  
  1098          rc = alloc_xenballooned_pages(numpages, pages, 0);
  1099          if (rc != 0) {
  1100                  pr_warn("%s Could not balloon alloc %d pfns
rc:%d\n", __func__,

regards,
dan carpenter

On Thu, 7 Feb 2013 13:54:25 +0300
Dan Carpenter <dan.carpenter@oracle.com> wrote:
> Hello Mukesh Rathor,
> 
> The patch 40f633eb0def: "xen/pvh linux: Use ballooning to allocate
> grant table pages" from Jan 31, 2013, leads to the following Sparse
> warning: "drivers/xen/grant-table.c:1095:28: error: bad constant
> 	expression"
> 
> drivers/xen/grant-table.c
>   1092  static int xlated_setup_gnttab_pages(int numpages, void
> **addr) 1093  {
>   1094          int i, rc;
>   1095          unsigned long pfns[numpages];
>   1096          struct page *pages[numpages];
>                                    ^^^^^^^^
> Because the kernel uses an 8k stack if this is over 500 it will cause
> a kernel crash.  It will crash before we reach 500 actually.  It looks
> like typical values for this are around 4 so we''re probably safe,
but
> it''s still a bit nasty.
> 
>   1097  
>   1098          rc = alloc_xenballooned_pages(numpages, pages, 0);
>   1099          if (rc != 0) {
>   1100                  pr_warn("%s Could not balloon alloc %d pfns
> rc:%d\n", __func__,
> 
> regards,
> dan carpenter
Hi Dan,

Yeah, I know. Currently, there is a hard max on num of grant pages to
32, but on average are lot less. So I think it should be OK. Konrad, do
you still want me to change it to kmalloc?

Thanks
Mukesh

On Fri, Feb 08, 2013 at 03:50:43PM -0800, Mukesh Rathor
wrote:
> On Thu, 7 Feb 2013 13:54:25 +0300
> Dan Carpenter <dan.carpenter@oracle.com> wrote:
> 
> > Hello Mukesh Rathor,
> > 
> > The patch 40f633eb0def: "xen/pvh linux: Use ballooning to
allocate
> > grant table pages" from Jan 31, 2013, leads to the following
Sparse
> > warning: "drivers/xen/grant-table.c:1095:28: error: bad constant
> > 	expression"
> > 
> > drivers/xen/grant-table.c
> >   1092  static int xlated_setup_gnttab_pages(int numpages, void
> > **addr) 1093  {
> >   1094          int i, rc;
> >   1095          unsigned long pfns[numpages];
> >   1096          struct page *pages[numpages];
> >                                    ^^^^^^^^
> > Because the kernel uses an 8k stack if this is over 500 it will cause
> > a kernel crash.  It will crash before we reach 500 actually.  It looks
> > like typical values for this are around 4 so we''re probably
safe, but
> > it''s still a bit nasty.
> > 
> >   1097  
> >   1098          rc = alloc_xenballooned_pages(numpages, pages, 0);
> >   1099          if (rc != 0) {
> >   1100                  pr_warn("%s Could not balloon alloc %d
pfns
> > rc:%d\n", __func__,
> > 
> > regards,
> > dan carpenter
> 
> Hi Dan,
> 
> Yeah, I know. Currently, there is a hard max on num of grant pages to
> 32, but on average are lot less. So I think it should be OK. Konrad, do
> you still want me to change it to kmalloc?
If there is a hard limit then it should be ok.

The place where dynamically allocated arrays are forbiden is inside
loops.  On some arches the stack memory isn''t freed until the end of
the function.

regards,
dan carpenter
> 
> Thanks
> Mukesh