Re: [PATCH V2 1/3] riscv: Fixup static_obj() fail

From: Atish Patra
Date: Tue Oct 06 2020 - 16:39:05 EST

On Tue, Oct 6, 2020 at 9:46 AM Guo Ren <guoren@xxxxxxxxxx> wrote:
>
> On Tue, Oct 6, 2020 at 3:14 AM Atish Patra <atishp@xxxxxxxxxxxxxx> wrote:
> >
> > On Thu, Sep 24, 2020 at 9:19 AM Guo Ren <guoren@xxxxxxxxxx> wrote:
> > >
> > > How about this, revert the commit and don't free INIT_DATA_SECTION. I
> > > think the solution is safe enough, but wast a little memory.
> > >
> > > diff --git a/arch/riscv/kernel/vmlinux.lds.S b/arch/riscv/kernel/vmlinux.lds.S
> > > index f3586e3..34d00d9 100644
> > > --- a/arch/riscv/kernel/vmlinux.lds.S
> > > +++ b/arch/riscv/kernel/vmlinux.lds.S
> > > @@ -22,13 +22,11 @@ SECTIONS
> > > /* Beginning of code and text segment */
> > > . = LOAD_OFFSET;
> > > _start = .;
> > > - _stext = .;
> > > HEAD_TEXT_SECTION
> > > . = ALIGN(PAGE_SIZE);
> > >
> > > __init_begin = .;
> > > INIT_TEXT_SECTION(PAGE_SIZE)
> > > - INIT_DATA_SECTION(16)
> > > . = ALIGN(8);
> > > __soc_early_init_table : {
> > > __soc_early_init_table_start = .;
> > > @@ -55,6 +53,7 @@ SECTIONS
> > > . = ALIGN(SECTION_ALIGN);
> > > .text : {
> > > _text = .;
> > > + _stext = .;
> > > TEXT_TEXT
> > > SCHED_TEXT
> > > CPUIDLE_TEXT
> > > @@ -67,6 +66,8 @@ SECTIONS
> > > _etext = .;
> > > }
> > >
> > > + INIT_DATA_SECTION(16)
> > > +
> >
> > I think you need to move EXIT_DATA as well. Currently, we have init
> > data & text in one section.
> It's not related to this issue. There is two check code problem:

Yes. But we shouldn't move only INIT_DATA_SECTION out of __init section
while leaving percpu & exit data in the __init section. Here is what I
have in mind.

diff --git a/arch/riscv/kernel/vmlinux.lds.S
b/arch/riscv/kernel/vmlinux.lds.S
index 9795359cb9da..4432cef8184e 100644
--- a/arch/riscv/kernel/vmlinux.lds.S
+++ b/arch/riscv/kernel/vmlinux.lds.S
@@ -26,13 +26,11 @@ SECTIONS
/* Beginning of code and text segment */
. = LOAD_OFFSET;
_start = .;
_start = .;
- _stext = .;
HEAD_TEXT_SECTION
. = ALIGN(PAGE_SIZE);

__init_begin = .;
INIT_TEXT_SECTION(PAGE_SIZE)
- INIT_DATA_SECTION(16)
. = ALIGN(8);
__soc_early_init_table : {
__soc_early_init_table_start = .;
@@ -49,16 +47,13 @@ SECTIONS
{
EXIT_TEXT
}
- .exit.data :
- {
- EXIT_DATA
- }
- PERCPU_SECTION(L1_CACHE_BYTES)
+
__init_end = .;

. = ALIGN(SECTION_ALIGN);
.text : {
_text = .;
+ _stext = .;
TEXT_TEXT
SCHED_TEXT
CPUIDLE_TEXT
@@ -77,6 +72,17 @@ SECTIONS
#endif

/* Start of data section */
+ __init_data_begin = .;
+ INIT_DATA_SECTION(16)
+ .exit.data :
+ {
+ EXIT_DATA
+ }
+
+ PERCPU_SECTION(L1_CACHE_BYTES)
+
+ __init_data_end = .;
+

As you correctly pointed out, this wastes around ~200K init memory
that is wasted.
That is not an ideal solution.

The other alternative is to move __init_text section after _text as
well similar to other architectures. But that won't work
for RISC-V as we jump from _start to __start_kernel(in __init section)
in head.S. A JAL instruction can't be fit because
__start_kernel is now too far. We can't replace JAL with a JALR
because that would require an additional
instruction and violates image header format.

Any other ideas to solve this problem without wasting memory ?

> 1. static int static_obj(const void *obj)
> {
> unsigned long start = (unsigned long) &_stext,
> end = (unsigned long) &_end,
> addr = (unsigned long) obj;
>
> /*
> * static variable?
> */
> if ((addr >= start) && (addr < end))
> return 1;
>
> 2. /* Is this address range in the kernel text area? */
> static inline void check_kernel_text_object(const unsigned long ptr,
> unsigned long n, bool to_user)
> {
> unsigned long textlow = (unsigned long)_stext;
> unsigned long texthigh = (unsigned long)_etext;
> unsigned long textlow_linear, texthigh_linear;
>
> if (overlaps(ptr, n, textlow, texthigh))
> usercopy_abort("kernel text", NULL, to_user, ptr -
> textlow, n);
>
> The patch of commit: a0fa4027dc911 (riscv: Fixup static_obj() fail) broke 2th.
>
> > In general it is better idea to separate those similar to ARM64.
> > Additionally, ARM64 applies different mapping for init data & text
> > as the init data section is marked as non-executable[1]
> Yes, it's safer to protect init text & init data, but it's should be
> another patch.
>

Yes. I will send the patch based on this fix.

> >
> > However, we don't modify any permission for any init sections. Should
> > we do that as well ?
> Agree, we should do that.
>
> >
> > [1] https://patchwork.kernel.org/patch/9572869/
> >
> > > /* Start of data section */
> > > _sdata = .;
> > > RO_DATA(SECTION_ALIGN)
> > >
> > > On Thu, Sep 24, 2020 at 3:36 PM Andreas Schwab <schwab@xxxxxxxxxxxxxx> wrote:
> > > >
> > > > On Sep 14 2020, Aurelien Jarno wrote:
> > > >
> > > > > How should we proceed to get that fixed in time for 5.9? For the older
> > > > > branches where it has been backported (so far 5.7 and 5.8), should we
> > > > > just get that commit reverted instead?
> > > >
> > > > Can this please be resolved ASAP?
> > > >
> > > > Andreas.
> > > >
> > > > --
> > > > Andreas Schwab, schwab@xxxxxxxxxxxxxx
> > > > GPG Key fingerprint = 7578 EB47 D4E5 4D69 2510 2552 DF73 E780 A9DA AEC1
> > > > "And now for something completely different."
> > >
> > >
> > >
> > > --
> > > Best Regards
> > > Guo Ren
> > >
> > > ML: https://lore.kernel.org/linux-csky/
> > >
> > > _______________________________________________
> > > linux-riscv mailing list
> > > linux-riscv@xxxxxxxxxxxxxxxxxxx
> > > http://lists.infradead.org/mailman/listinfo/linux-riscv
> >
> >
> >
> > --
> > Regards,
> > Atish
>
>
>
> --
> Best Regards
> Guo Ren
>
> ML: https://lore.kernel.org/linux-csky/



--
Regards,
Atish