On 01/09/18 at 09:09am, Dave Young wrote:
> On 01/09/18 at 03:13am, Kirill A. Shutemov wrote:
> > On Mon, Jan 08, 2018 at 08:46:53PM +0300, Kirill A. Shutemov wrote:
> > > On Mon, Jan 08, 2018 at 04:04:44PM +0000, Ingo Molnar wrote:
> > > >
> > > > hi Kirill,
> > > >
> > > > As Mike reported it below, your 5-level paging related upstream commit
> > > > 83e3c48729d9 and all its followup fixes:
> > > >
> > > > 83e3c48729d9: mm/sparsemem: Allocate mem_section at runtime for CONFIG_SPARSEMEM_EXTREME=y
> > > > 629a359bdb0e: mm/sparsemem: Fix ARM64 boot crash when CONFIG_SPARSEMEM_EXTREME=y
> > > > d09cfbbfa0f7: mm/sparse.c: wrong allocation for mem_section
> > > >
> > > > ... still breaks kexec - and that now regresses -stable as well.
> > > >
> > > > Given that 5-level paging now syntactically depends on having this commit, if we
> > > > fully revert this then we'll have to disable 5-level paging as well.
> >
> > This *should* help.
> >
> > Mike, could you test this? (On top of the rest of the fixes.)
> >
> > Sorry for the mess.
> >
> > From 100fd567754f1457be94732046aefca204c842d2 Mon Sep 17 00:00:00 2001
> > From: "Kirill A. Shutemov" <kirill.shutemov@xxxxxxxxxxxxxxx>
> > Date: Tue, 9 Jan 2018 02:55:47 +0300
> > Subject: [PATCH] kdump: Write a correct address of mem_section into vmcoreinfo
> >
> > Depending on configuration mem_section can now be an array or a pointer
> > to an array allocated dynamically. In most cases, we can continue to refer
> > to it as 'mem_section' regardless of what it is.
> >
> > But there's one exception: '&mem_section' means "address of the array" if
> > mem_section is an array, but if mem_section is a pointer, it would mean
> > "address of the pointer".
> >
> > We've stepped onto this in kdump code. VMCOREINFO_SYMBOL(mem_section)
> > writes down address of pointer into vmcoreinfo, not array as we wanted.
> >
> > Let's introduce VMCOREINFO_ARRAY() that would handle the situation
> > correctly for both cases.
> >
> > Signed-off-by: Kirill A. Shutemov <kirill.shutemov@xxxxxxxxxxxxxxx>
> > Fixes: 83e3c48729d9 ("mm/sparsemem: Allocate mem_section at runtime for CONFIG_SPARSEMEM_EXTREME=y")
> > ---
> > include/linux/crash_core.h | 2 ++
> > kernel/crash_core.c | 2 +-
> > 2 files changed, 3 insertions(+), 1 deletion(-)
> >
> > diff --git a/include/linux/crash_core.h b/include/linux/crash_core.h
> > index 06097ef30449..83ae04950269 100644
> > --- a/include/linux/crash_core.h
> > +++ b/include/linux/crash_core.h
> > @@ -42,6 +42,8 @@ phys_addr_t paddr_vmcoreinfo_note(void);
> > vmcoreinfo_append_str("PAGESIZE=%ld\n", value)
> > #define VMCOREINFO_SYMBOL(name) \
> > vmcoreinfo_append_str("SYMBOL(%s)=%lx\n", #name, (unsigned long)&name)
> > +#define VMCOREINFO_ARRAY(name) \
>
> Thanks for the patch, I have a similar patch but makedumpfile maintainer
> is looking at a userspace fix instead.
Seems we should add lkml to CC next time so that people can watch it.
> As for the macro name, VMCOREINFO_SYMBOL_ARRAY sounds better.
I still think using vmcoreinfo_append_str is better. Unless we replace
all array variables with the newly added macro.
vmcoreinfo_append_str("SYMBOL(mem_section)=%lx\n",
(unsigned long)mem_section);
>
> > + vmcoreinfo_append_str("SYMBOL(%s)=%lx\n", #name, (unsigned long)name)
> > #define VMCOREINFO_SIZE(name) \
> > vmcoreinfo_append_str("SIZE(%s)=%lu\n", #name, \
> > (unsigned long)sizeof(name))
> > diff --git a/kernel/crash_core.c b/kernel/crash_core.c
> > index b3663896278e..d4122a837477 100644
> > --- a/kernel/crash_core.c
> > +++ b/kernel/crash_core.c
> > @@ -410,7 +410,7 @@ static int __init crash_save_vmcoreinfo_init(void)
> > VMCOREINFO_SYMBOL(contig_page_data);
> > #endif
> > #ifdef CONFIG_SPARSEMEM
> > - VMCOREINFO_SYMBOL(mem_section);
> > + VMCOREINFO_ARRAY(mem_section);
> > VMCOREINFO_LENGTH(mem_section, NR_SECTION_ROOTS);
> > VMCOREINFO_STRUCT_SIZE(mem_section);
> > VMCOREINFO_OFFSET(mem_section, section_mem_map);
> > --
> > Kirill A. Shutemov
>
> Thanks
> Dave