On Thu, Aug 30, 2018 at 03:25:42PM +0200, Thomas Gleixner wrote:
 
> And where is the problem? We know that we need the fixmap during early boot
> anyway, so allocating the PTE page statically and setting it up early
> enough is not really rocket science. No allocator required.

Further check shows the page table for fixmap is already setup in [-12M, -10M],
but fixmap's address space breaks the expection under certain kernel config.
So here is another fix, please review, thanks!

----------------------------------------------------------------

>From f63e4e5c6166a2f8d13ad3142b7a49b2fd1f8edf Mon Sep 17 00:00:00 2001
From: Feng Tang <feng.t...@intel.com>
Date: Fri, 31 Aug 2018 13:48:21 +0800
Subject: [PATCH] x86, mm: Make the real fixmap address consistent

We hit a kernel panic when enabling earlycon for a platform, the
call trace is:

          panic+0xd2/0x220
          __alloc_bootmem+0x31/0x34
          spp_getpage+0x60/0x8a
          fill_pte+0x71/0x130
          __set_pte_vaddr+0x1d/0x50
          set_pte_vaddr+0x3c/0x60
          __native_set_fixmap+0x23/0x30
          native_set_fixmap+0x30/0x40
          setup_earlycon+0x1e0/0x32f
          param_setup_earlycon+0x13/0x22
          do_early_param+0x5b/0x90
          parse_args+0x1f7/0x300
          parse_early_options+0x24/0x28
          parse_early_param+0x65/0x73
          setup_arch+0x31e/0x9f1
          start_kernel+0x58/0x44e

This panic happens as the earlycon's fixmap address has no
pmd/pte ready, and __set_fixmap will try to allocate memory to
setup the page table, and trigger panic due to no memory.

x86 kernel actually prepares the page table for fixmap in head_64.S:

        NEXT_PAGE(level2_fixmap_pgt)
                .fill   506,8,0
                .quad   level1_fixmap_pgt - __START_KERNEL_map + 
_PAGE_TABLE_NOENC
                /* 8MB reserved for vsyscalls + a 2MB hole = 4 + 1 entries */
                .fill   5,8,0

and it expects the fixmap address is in [-12M, -10M] range, but
current code in fixmap.h will break the expectation when
X86_VSYSCALL_EMULATION=n

        #ifdef CONFIG_X86_VSYSCALL_EMULATION
                VSYSCALL_PAGE = (FIXADDR_TOP - VSYSCALL_ADDR) >> PAGE_SHIFT,
        #endif

So removing the "#ifdef" will make the fixmap address space stable in
[-12M, -10M] and fix the issue.

Signed-off-by: Feng Tang <feng.t...@intel.com>
---
 arch/x86/include/asm/fixmap.h | 2 --
 1 file changed, 2 deletions(-)

diff --git a/arch/x86/include/asm/fixmap.h b/arch/x86/include/asm/fixmap.h
index e203169931c7..fcf27171f493 100644
--- a/arch/x86/include/asm/fixmap.h
+++ b/arch/x86/include/asm/fixmap.h
@@ -67,9 +67,7 @@ enum fixed_addresses {
 #ifdef CONFIG_X86_32
        FIX_HOLE,
 #else
-#ifdef CONFIG_X86_VSYSCALL_EMULATION
        VSYSCALL_PAGE = (FIXADDR_TOP - VSYSCALL_ADDR) >> PAGE_SHIFT,
-#endif
 #endif
        FIX_DBGP_BASE,
        FIX_EARLYCON_MEM_BASE,
-- 
2.14.1

Reply via email to