Re: [PATCH v6 3/5] vmcore: Introduce remap_oldmem_pfn_range()

[Michael Holzheu <holzheu@linux.vnet.ibm.com>]

On Mon, 15 Jul 2013 10:27:08 -0400
Vivek Goyal <vgoyal@redhat.com> wrote:

> On Mon, Jul 15, 2013 at 03:44:51PM +0200, Michael Holzheu wrote:
> > On Tue, 2 Jul 2013 11:42:14 -0400
> > Vivek Goyal <vgoyal@redhat.com> wrote:
> > 
> > > On Mon, Jul 01, 2013 at 09:32:37PM +0200, Michael Holzheu wrote:
> > > > For zfcpdump we can't map the HSA storage because it is only available
> > > > via a read interface. Therefore, for the new vmcore mmap feature we have
> > > > introduce a new mechanism to create mappings on demand.
> > > > 
> > > > This patch introduces a new architecture function remap_oldmem_pfn_range()
> > > > that should be used to create mappings with remap_pfn_range() for oldmem
> > > > areas that can be directly mapped. For zfcpdump this is everything besides
> > > > of the HSA memory. For the areas that are not mapped by remap_oldmem_pfn_range()
> > > > a generic vmcore a new generic vmcore fault handler mmap_vmcore_fault()
> > > > is called.
> > > > 
> > > > This handler works as follows:
> > > > 
> > > > * Get already available or new page from page cache (find_or_create_page)
> > > > * Check if /proc/vmcore page is filled with data (PageUptodate)
> > > > * If yes:
> > > >   Return that page
> > > > * If no:
> > > >   Fill page using __vmcore_read(), set PageUptodate, and return page
> > > > 
> > > > Signed-off-by: Michael Holzheu <holzheu@linux.vnet.ibm.com>
> > > 
> > > In general vmcore related changes look fine to me. I am not very familiar
> > > with the logic of finding pages in page cache and using page uptodate
> > > flag.
> > > 
> > > Hatayama, can you please review it.
> > > 
> > > Acked-by: Vivek Goyal <vgoyal@redhat.com>
> > 
> > Hello Vivek and Andrew,
> > 
> > We just realized that Hatayama's mmap patches went into v3.11-rc1. This currently
> > breaks s390 kdump because of the following two issues:
> > 
> > 1) The copy_oldmem_page() is now used for copying to vmalloc memory
> > 2) The mmap() implementation is not compatible with the current
> >    s390 crashkernel swap:
> >    See: http://marc.info/?l=kexec&m=136940802511603&w=2
> > 
> > The "kdump: Introduce ELF header in new memory feature" patch series will
> > fix both issues for s390.
> > 
> > There is the one small open discussion left:
> > 
> > http://www.mail-archive.com/linux-kernel@vger.kernel.org/msg464856.html
> > 
> > But once we have finished that, would it be possible to get the
> > patches in 3.11?
> 
> How about taking mmap() fault handler patches in 3.12. And in 3.11, deny
> mmap() on s390 forcing makedumpfile to fall back on read() interface. That
> way there will be no regression and mmap() related speedup will show up
> in next release on s390.

Hello Vivek and Hatayama,

But then we still would have to somehow fix the copy_oldmem_page() issue (1).

We would prefer to add the current patch series with "#ifndef CONFIG_S390" in
the fault handler.

@Vivek:

Since you are the kdump maintainer, could you tell us which of the both
variants you would like to have?

static int mmap_vmcore_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
{
...

#ifndef CONFIG_S390
	return VM_FAULT_SIGBUS;
#endif

or 

#ifndef CONFIG_S390
        WARN_ONCE(1, "vmcore: Unexpected call of mmap_vmcore_fault()");
#endif

For all architectures besides of s390 this would implement the requested behavior.

Regards,
Michael