[PATCH] riscv: mm: Fix W+X mappings at boot

[Jisheng Zhang]

On Tue, 18 May 2021 17:26:53 +0200
Alex Ghiti <alex at ghiti.fr> wrote:

> Hi Jisheng,

Hi Alex,

> 
> On 16/05/2021 11:00, Jisheng Zhang wrote:
> > From: Jisheng Zhang <jszhang at kernel.org>
> > 
> > When the kernel mapping was moved the last 2GB of the address space,
> > (__va(PFN_PHYS(max_low_pfn))) is much smaller than the .data section
> > start address, the last set_memory_nx() in protect_kernel_text_data()
> > will fail, thus the .data section is still mapped as W+X. This results
> > in below W+X mapping waring at boot. Fix it by passing the correct
> > .data section page num to the set_memory_nx().
> > 
> > [    0.396516] ------------[ cut here ]------------
> > [    0.396889] riscv/mm: Found insecure W+X mapping at address (____ptrval____)/0xffffffff80c00000
> > [    0.398347] WARNING: CPU: 0 PID: 1 at arch/riscv/mm/ptdump.c:258 note_page+0x244/0x24a
> > [    0.398964] Modules linked in:
> > [    0.399459] CPU: 0 PID: 1 Comm: swapper/0 Not tainted 5.13.0-rc1+ #14
> > [    0.400003] Hardware name: riscv-virtio,qemu (DT)
> > [    0.400591] epc : note_page+0x244/0x24a
> > [    0.401368]  ra : note_page+0x244/0x24a
> > [    0.401772] epc : ffffffff80007c86 ra : ffffffff80007c86 sp : ffffffe000e7bc30
> > [    0.402304]  gp : ffffffff80caae88 tp : ffffffe000e70000 t0 : ffffffff80cb80cf
> > [    0.402800]  t1 : ffffffff80cb80c0 t2 : 0000000000000000 s0 : ffffffe000e7bc80
> > [    0.403310]  s1 : ffffffe000e7bde8 a0 : 0000000000000053 a1 : ffffffff80c83ff0
> > [    0.403805]  a2 : 0000000000000010 a3 : 0000000000000000 a4 : 6c7e7a5137233100
> > [    0.404298]  a5 : 6c7e7a5137233100 a6 : 0000000000000030 a7 : ffffffffffffffff
> > [    0.404849]  s2 : ffffffff80e00000 s3 : 0000000040000000 s4 : 0000000000000000
> > [    0.405393]  s5 : 0000000000000000 s6 : 0000000000000003 s7 : ffffffe000e7bd48
> > [    0.405935]  s8 : ffffffff81000000 s9 : ffffffffc0000000 s10: ffffffe000e7bd48
> > [    0.406476]  s11: 0000000000001000 t3 : 0000000000000072 t4 : ffffffffffffffff
> > [    0.407016]  t5 : 0000000000000002 t6 : ffffffe000e7b978
> > [    0.407435] status: 0000000000000120 badaddr: 0000000000000000 cause: 0000000000000003
> > [    0.408052] Call Trace:
> > [    0.408343] [<ffffffff80007c86>] note_page+0x244/0x24a
> > [    0.408855] [<ffffffff8010c5a6>] ptdump_hole+0x14/0x1e
> > [    0.409263] [<ffffffff800f65c6>] walk_pgd_range+0x2a0/0x376
> > [    0.409690] [<ffffffff800f6828>] walk_page_range_novma+0x4e/0x6e
> > [    0.410146] [<ffffffff8010c5f8>] ptdump_walk_pgd+0x48/0x78
> > [    0.410570] [<ffffffff80007d66>] ptdump_check_wx+0xb4/0xf8
> > [    0.410990] [<ffffffff80006738>] mark_rodata_ro+0x26/0x2e
> > [    0.411407] [<ffffffff8031961e>] kernel_init+0x44/0x108
> > [    0.411814] [<ffffffff80002312>] ret_from_exception+0x0/0xc
> > [    0.412309] ---[ end trace 7ec3459f2547ea83 ]---
> > [    0.413141] Checked W+X mappings: failed, 512 W+X pages found
> > 
> > Fixes: 2bfc6cd81bd17e43 ("riscv: Move kernel mapping outside of linear mapping")
> > Signed-off-by: Jisheng Zhang <jszhang at kernel.org>
> > ---
> >   arch/riscv/mm/init.c | 8 ++++++--
> >   1 file changed, 6 insertions(+), 2 deletions(-)
> > 
> > diff --git a/arch/riscv/mm/init.c b/arch/riscv/mm/init.c
> > index 4faf8bd157ea..4c4c92ce0bb8 100644
> > --- a/arch/riscv/mm/init.c
> > +++ b/arch/riscv/mm/init.c
> > @@ -746,14 +746,18 @@ void __init protect_kernel_text_data(void)
> >   	unsigned long init_data_start = (unsigned long)__init_data_begin;
> >   	unsigned long rodata_start = (unsigned long)__start_rodata;
> >   	unsigned long data_start = (unsigned long)_data;
> > -	unsigned long max_low = (unsigned long)(__va(PFN_PHYS(max_low_pfn)));
> > +#if defined(CONFIG_64BIT) && defined(CONFIG_MMU)
> > +	unsigned long end_va = kernel_virt_addr + load_sz;
> > +#else
> > +	unsigned long end_va = (unsigned long)(__va(PFN_PHYS(max_low_pfn)));
> > +#endif
> >   
> >   	set_memory_ro(text_start, (init_text_start - text_start) >> PAGE_SHIFT);
> >   	set_memory_ro(init_text_start, (init_data_start - init_text_start) >> PAGE_SHIFT);
> >   	set_memory_nx(init_data_start, (rodata_start - init_data_start) >> PAGE_SHIFT);
> >   	/* rodata section is marked readonly in mark_rodata_ro */
> >   	set_memory_nx(rodata_start, (data_start - rodata_start) >> PAGE_SHIFT);
> > -	set_memory_nx(data_start, (max_low - data_start) >> PAGE_SHIFT);
> > +	set_memory_nx(data_start, (end_va - data_start) >> PAGE_SHIFT);
> >   }
> >   
> >   void mark_rodata_ro(void)
> >   
> 
> Thank you for taking the time to fix this, I had read a report here 
> https://github.com/starfive-tech/linux/issues/17 but had no time yet to 

I didn't know this github repo, in fact I don't have a beaglev board ;)
From the log, this is the same issue as I saw on Qemu.

> track this down.
> 
> Your fix seems good to me, but it intrigued me to see that for 32b 
> kernels, the whole linear mapping is mapped as executable and then here, 
> we remove this attribute. So I came up with a patch to map the kernel 
> correctly once at first time and avoid fixing this mapping afterwards. I 

Your solution looks better.

> added you in cc of this patch, any comment is welcome.
> 

I'm reading your patch. Will comment it soon.

Thanks