Re: [PATCH 05/13] efi/x86: don't map the entire kernel text RW for mixed mode

From: Ard Biesheuvel
Date: Wed Apr 08 2020 - 06:47:33 EST


On Wed, 8 Apr 2020 at 12:42, Jiri Slaby <jslaby@xxxxxxx> wrote:
>
> On 13. 01. 20, 18:22, Ard Biesheuvel wrote:
> > The mixed mode thunking routine requires a part of it to be
> > mapped 1:1, and for this reason, we currently map the entire
> > kernel .text read/write in the EFI page tables, which is bad.
> >
> > In fact, the kernel_map_pages_in_pgd() invocation that installs
> > this mapping is entirely redundant, since all of DRAM is already
> > 1:1 mapped read/write in the EFI page tables when we reach this
> > point, which means that .rodata is mapped read-write as well.
> >
> > So let's remap both .text and .rodata read-only in the EFI
> > page tables.
>
> This patch causes unhandled page faults in mixed mode:
>
> > BUG: unable to handle page fault for address: 000000001557ee88
> > #PF: supervisor write access in kernel mode
> > #PF: error_code(0x0003) - permissions violation
> > PGD fd52063 P4D fd52063 PUD fd53063 PMD 154000e1
> > Oops: 0003 [#1] SMP PTI
> > CPU: 1 PID: 191 Comm: systemd-escape Not tainted
> 5.6.2-20.gb22bc26-default #1 openSUSE Tumbleweed (unreleased)
> > Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 0.0.0
> 02/06/2015
> > RIP: 0008:0x3d2eed95
> > Code: 8b 45 d4 8b 4d 10 8b 40 04 89 01 89 3b 50 6a 00 8b 55 0c 6a 00
> 8b 45 08 0f b6 4d e4 6a 01 31 f6 e8 ee c5 fc ff 83 c4 10 eb 07 <89> 03
> be 05 00 00 80 a1 74 63 31 3d 83 c0 48 e8 44 d2 ff ff eb 05
> > RSP: 0018:000000000fd66fa0 EFLAGS: 00010002
> > RAX: 0000000000000001 RBX: 000000001557ee88 RCX: 000000003d1f1120
> > RDX: 0000000000000001 RSI: 0000000000000000 RDI: 0000000000000001
> > RBP: 000000000fd66fd8 R08: 000000001557ee88 R09: 0000000000000000
> > R10: 0000000000000055 R11: 0000000000000000 R12: 0000000015bcf000
> > R13: 0000000000000000 R14: 0000000000000000 R15: 0000000000000000
> > FS: 00007f36ee9dc940(0000) GS:ffff9b903d700000(0000)
> knlGS:0000000000000000
> > CS: 0008 DS: 0018 ES: 0018 CR0: 0000000080050033
> > CR2: 000000001557ee88 CR3: 000000000fd5e000 CR4: 00000000000006e0
> > DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
> > DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
> > Call Trace:
> > Modules linked in: efivarfs
> > CR2: 000000001557ee88
>
> EFI apparently tries to write to now read-only memory.
>
> See:
> https://bugzilla.suse.com/show_bug.cgi?id=1168645
>
> Reverting it on the top of 5.6 fixes the issue.
>
> I am using
> /usr/share/qemu/ovmf-ia32-code.bin
> /usr/share/qemu/ovmf-ia32-vars.bin
> from qemu-ovmf-ia32-202002-1.1.noarch rpm.
>

Do you have a git tree for Suse's OVMF fork? I did a lot of testing
with upstream OVMF, and never ran into this issue.


> > Signed-off-by: Ard Biesheuvel <ardb@xxxxxxxxxx>
> > ---
> > arch/x86/platform/efi/efi_64.c | 4 ++--
> > 1 file changed, 2 insertions(+), 2 deletions(-)
> >
> > diff --git a/arch/x86/platform/efi/efi_64.c b/arch/x86/platform/efi/efi_64.c
> > index c13fa2150976..6ec58ff60b56 100644
> > --- a/arch/x86/platform/efi/efi_64.c
> > +++ b/arch/x86/platform/efi/efi_64.c
> > @@ -391,11 +391,11 @@ int __init efi_setup_page_tables(unsigned long pa_memmap, unsigned num_pages)
> >
> > efi_scratch.phys_stack = page_to_phys(page + 1); /* stack grows down */
> >
> > - npages = (_etext - _text) >> PAGE_SHIFT;
> > + npages = (__end_rodata_aligned - _text) >> PAGE_SHIFT;
> > text = __pa(_text);
> > pfn = text >> PAGE_SHIFT;
> >
> > - pf = _PAGE_RW | _PAGE_ENC;
> > + pf = _PAGE_ENC;
> > if (kernel_map_pages_in_pgd(pgd, pfn, text, npages, pf)) {
> > pr_err("Failed to map kernel text 1:1\n");
> > return 1;
> >
>
> thanks,
> --
> js
> suse labs