Re: [PATCH] x86/mm: don't free p4d table when it is folded at runtime.

[Baoquan He <bhe@redhat.com>]

On 06/26/18 at 01:00pm, Kirill A. Shutemov wrote:
> On Tue, Jun 26, 2018 at 07:40:49AM +0000, Baoquan He wrote:
> > Hi Kirill,
> > 
> > On 06/25/18 at 02:50pm, Kirill A. Shutemov wrote:
> > > On Mon, Jun 25, 2018 at 01:24:27PM +0300, Andrey Ryabinin wrote:
> > > > When the p4d page table layer is folded at runtime, the p4d_free()
> > > > should do nothing, the same as in <asm-generic/pgtable-nop4d.h>.
> > > > 
> > > > It seems this bug should cause double-free in efi_call_phys_epilog(),
> > > > but I don't know how to trigger that code path, so I can't confirm that
> > > > by testing.
> > > 
> > > + Baoquan, Matt.
> > > 
> > > There's other bug in the efi_call_phys_epilog() that prevents the bug from
> > > being triggered.
> > > 
> > > With the patch below. You can trigger the bug with efi=old_map in kernel
> > > command line + KALSR and CONFIG_X86_5LEVEL=y:
> > > 
> > > 	page:fffff6bec0000000 count:0 mapcount:1 mapping:0000000000000000 index:0x0
> > > 	flags: 0x800(reserved)
> > > 	raw: 0000000000000800 fffff6bec0000008 fffff6bec0000008 0000000000000000
> > > 	raw: 0000000000000000 0000000000000000 0000000000000000 0000000000000000
> > > 	page dumped because: VM_BUG_ON_PAGE(page_ref_count(page) == 0)
> > > 	------------[ cut here ]------------
> > > 	kernel BUG at /home/kas/linux/la57/include/linux/mm.h:499!
> > > 	invalid opcode: 0000 [#1] PREEMPT SMP
> > > 	CPU: 0 PID: 0 Comm: swapper/0 Not tainted 4.18.0-rc2-00037-g6f0d349d922b-dirty #58
> > > 	Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 0.0.0 02/06/2015
> > > 	RIP: 0010:__free_pages+0x28/0x30
> > > 	Code: 00 00 8b 47 34 85 c0 74 15 f0 ff 4f 34 75 09 85 f6 74 06 e9 ca d8 ff ff c3 e9 64 ff ff ff 48
> > > 	RSP: 0000:ffffffff9a403e90 EFLAGS: 00000246
> > > 	RAX: 000000000000003e RBX: ffffffff9a41d000 RCX: 0000000000000002
> > > 	RDX: 0000000000000000 RSI: 0000000000000000 RDI: 00000000ffffffff
> > > 	RBP: 0000000000000000 R08: 0000000000000000 R09: 0000000000000000
> > > 	R10: 0000000000000000 R11: 0000000000000000 R12: ffff9092af089000
> > > 	R13: ffffffff9a598a80 R14: 0000000000000001 R15: 0000000000000001
> > > 	FS:  0000000000000000(0000) GS:ffff9092bfc00000(0000) knlGS:0000000000000000
> > > 	CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> > > 	CR2: ffff9092bffff000 CR3: 0000000198e1d000 CR4: 00000000000006b0
> > > 	Call Trace:
> > > 	 efi_call_phys_epilog+0x17d/0x1bb
> > > 	 efi_enter_virtual_mode+0x457/0x4ca
> > > 	 start_kernel+0x443/0x4dc
> > > 	 secondary_startup_64+0xb7/0xc0
> > > 	Modules linked in:
> > > 	---[ end trace 61e271260b11acdd ]---
> > > 
> > > I'll send patch for efi_call_phys_epilog().
> > > 
> > > > 
> > > > Fixes: 98219dda2ab5 ("x86/mm: Fold p4d page table layer at runtime")
> > > > Signed-off-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
> > > 
> > > Reviewed-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
> > > Cc: stable@vger.kernel.org # 4.17
> > > 
> > > > ---
> > > >  arch/x86/include/asm/pgalloc.h | 3 +++
> > > >  1 file changed, 3 insertions(+)
> > > > 
> > > > diff --git a/arch/x86/include/asm/pgalloc.h b/arch/x86/include/asm/pgalloc.h
> > > > index ada6410fd2ec..fbd578daa66e 100644
> > > > --- a/arch/x86/include/asm/pgalloc.h
> > > > +++ b/arch/x86/include/asm/pgalloc.h
> > > > @@ -184,6 +184,9 @@ static inline p4d_t *p4d_alloc_one(struct mm_struct *mm, unsigned long addr)
> > > >  
> > > >  static inline void p4d_free(struct mm_struct *mm, p4d_t *p4d)
> > > >  {
> > > > +	if (!pgtable_l5_enabled())
> > > > +		return;
> > > > +
> > > >  	BUG_ON((unsigned long)p4d & (PAGE_SIZE-1));
> > > >  	free_page((unsigned long)p4d);
> > > >  }
> > > 
> > > diff --git a/arch/x86/platform/efi/efi_64.c b/arch/x86/platform/efi/efi_64.c
> > > index e01f7ceb9e7a..77873ce700ae 100644
> > > --- a/arch/x86/platform/efi/efi_64.c
> > > +++ b/arch/x86/platform/efi/efi_64.c
> > > @@ -166,14 +166,14 @@ void __init efi_call_phys_epilog(pgd_t *save_pgd)
> > >                 pgd = pgd_offset_k(pgd_idx * PGDIR_SIZE);
> > >                 set_pgd(pgd_offset_k(pgd_idx * PGDIR_SIZE), save_pgd[pgd_idx]);
> > > 
> > > -               if (!(pgd_val(*pgd) & _PAGE_PRESENT))
> > > +               if (!pgd_present(*pgd))
> > >                         continue;
> > 
> > I may not understand boot-time p4d folding. Here p4d is folded, why
> > pgd_present() need always return 1 if p4d folded?
> 
> Yeah. This is confusing. Basically, we pretend that p4d level consist of 1
> entry, with the same value as pgd above. We say that pgd is always present
> and all checks happens on p4d level.
> 
> It's not specific to boot-time folding. Compilet-time folding doing the
> same. See include/asm-generic/pgtable-nop4d.h (and the rest -nop?d.h)

Thanks for telling, Kirill. Then the next p4d_present will do the real
_PAGE_PRESENT checking. If so, I think this is a good fix.