This change makes later calculations about where the kernel is located
easier to reason about. To better understand this change, we must first
clarify what 'VO' and 'ZO' are. These values were introduced in commits
by hpa:
77d1a49995 ("x86, boot: make symbols from the main vmlinux available")
37ba7ab5e3 ("x86, boot: make kernel_alignment adjustable; new bzImage fields")
Specifically:
All names prefixed with 'VO_':
- relate to the uncompressed kernel image
- the size of the VO image is: VO__end-VO__text ("VO_INIT_SIZE" define)
All names prefixed with 'ZO_':
- relate to the bootable compressed kernel image (boot/compressed/vmlinux),
which is composed of the following memory areas:
- head text
- compressed kernel (VO image and relocs table)
- decompressor code
- the size of the ZO image is: ZO__end - ZO_startup_32 ("ZO_INIT_SIZE" define, though see below)
The 'INIT_SIZE' value is used to find the larger of the two image sizes:
#define ZO_INIT_SIZE (ZO__end - ZO_startup_32 + ZO_z_extract_offset)
#define VO_INIT_SIZE (VO__end - VO__text)
#if ZO_INIT_SIZE > VO_INIT_SIZE
# define INIT_SIZE ZO_INIT_SIZE
#else
# define INIT_SIZE VO_INIT_SIZE
#endif
The current code uses extract_offset to decide where to position the
copied ZO (i.e. ZO starts at extract_offset). (This is why ZO_INIT_SIZE
currently includes the extract_offset.)
Why does z_extract_offset exist? It's needed because we are trying to minimize
the amount of RAM used for the whole act of creating an uncompressed, executable,
properly relocation-linked kernel image in system memory. We do this so that
kernels can be booted on even very small systems.
To achieve the goal of minimal memory consumption we have implemented an in-place
decompression strategy: instead of cleanly separating the VO and ZO images and
also allocating some memory for the decompression code's runtime needs, we instead
create this elaborate layout of memory buffers where the output (decompressed)
stream, as it progresses, overlaps with and destroys the input (compressed)
stream. This can only be done safely if the ZO image is placed to the end of the
VO range, plus a certain amount of safety distance to make sure that when the last
bytes of the VO range are decompressed, the compressed stream pointer is safely
beyond the end of the VO range.
z_extract_offset is calculated in arch/x86/boot/compressed/mkpiggy.c during
the build process, at a point when we know the exact compressed and
uncompressed size of the kernel images and can calculate this safe minimum
offset value. (Note that the mkpiggy.c calculation is not perfect, because
we don't know the decompressor used at that stage, so the z_extract_offset
calculation is necessarily imprecise and is mostly based on gzip internals -
we'll improve that in the next patch.)
When INIT_SIZE is bigger than VO_INIT_SIZE (uncommon but possible),
the copied ZO occupies the memory from extract_offset to the end of
decompression buffer. It overlaps with the soon-to-be-uncompressed kernel
like this:
|-----compressed kernel image------|
V V
0 extract_offset +INIT_SIZE
|-----------|---------------|-------------------------|--------|
| | | |
VO__text startup_32 of ZO VO__end ZO__end
^ ^
|-------uncompressed kernel image---------|
When INIT_SIZE is equal to VO_INIT_SIZE (likely) there's still space
left from end of ZO to the end of decompressing buffer, like below.
|-compressed kernel image-|
V V
0 extract_offset +INIT_SIZE
|-----------|---------------|-------------------------|--------|
| | | |
VO__text startup_32 of ZO ZO__end VO__end
^ ^
|------------uncompressed kernel image-------------|
To simplify calculations and avoid special cases, it is cleaner to
always place the compressed kernel image in memory so that ZO__end
is at the end of the decompression buffer, instead of placing t at
the start of extract_offset as is currently done.
This patch adds BP_init_size (which is the INIT_SIZE as passed in from
the boot_params) into asm-offsets.c to make it visible to the assembly
code.
Then when moving the ZO, it calculates the starting position of
the copied ZO (via BP_init_size and the ZO run size) so that the VO__end
will be at the end of the decompression buffer. To make the position
calculation safe, the end of ZO is page aligned (and a comment is added
to the existing VO alignment for good measure).
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
[ Rewrote changelog and comments. ]
Signed-off-by: Kees Cook <keescook@chromium.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Baoquan He <bhe@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Dave Young <dyoung@redhat.com>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vivek Goyal <vgoyal@redhat.com>
Cc: lasse.collin@tukaani.org
Link: http://lkml.kernel.org/r/1461888548-32439-3-git-send-email-keescook@chromium.org
[ Rewrote the changelog some more. ]
Signed-off-by: Ingo Molnar <mingo@kernel.org>
In order for global variables and functions to work in the
decompressor, we need to fix up the GOT in assembly code.
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
LKML-Reference: <4C57382E.8050501@zytor.com>
Rather than having X86_L1_CACHE_BYTES and X86_L1_CACHE_SHIFT
(with inconsistent defaults), just having the latter suffices as
the former can be easily calculated from it.
To be consistent, also change X86_INTERNODE_CACHE_BYTES to
X86_INTERNODE_CACHE_SHIFT, and set it to 7 (128 bytes) for NUMA
to account for last level cache line size (which here matters
more than L1 cache line size).
Finally, make sure the default value for X86_L1_CACHE_SHIFT,
when X86_GENERIC is selected, is being seen before that for the
individual CPU model options (other than on x86-64, where
GENERIC_CPU is part of the choice construct, X86_GENERIC is a
separate option on ix86).
Signed-off-by: Jan Beulich <jbeulich@novell.com>
Acked-by: Ravikiran Thirumalai <kiran@scalex86.org>
Acked-by: Nick Piggin <npiggin@suse.de>
LKML-Reference: <4AFD5710020000780001F8F0@vpn.id2.novell.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This has the consequence of changing the section name use for head
code from ".text.head" to ".head.text".
Linus suggested that we merge the ".text.head" section with ".text"
(presumably while preserving the fact that the head code starts at 0).
When I tried this it caused the kernel to not boot.
Signed-off-by: Tim Abbott <tabbott@ksplice.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Both on 32 and 64 bits, we copy all the way up to the end of bss,
except that on 64 bits there is a hack to avoid copying on top of the
page tables. There is no point in copying bss at all, especially
since we are just about to zero it all anyway.
To clean up and unify the handling, we now do:
- copy from startup_32 to _bss.
- zero from _bss to _ebss.
- the _ebss symbol is aligned to an 8-byte boundary.
- the page tables are moved to a separate section.
Use _bss as the copy endpoint since _edata may be misaligned.
[ Impact: cleanup, trivial performance improvement ]
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Aligning the .bss section makes it trivial to use large operation
sizes for moving the initialized sections and clearing the .bss.
The alignment chosen (L1 cache) is somewhat arbitrary, but should be
large enough to avoid all known performance traps and small enough to
not cause troubles.
[ Impact: trivial performance enhancement, future patch prep ]
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Look at the:
diff -u arch/x86/boot/compressed/vmlinux_*.lds
output and realize that they're basially exactly the same except for
trivial naming differences, and the fact that the 64-bit version has a
"pgtable" thing.
So unify them.
There's some trivial cleanup there (make the output format a Kconfig thing
rather than doing #ifdef's for it, and unify both 32-bit and 64-bit BSS
end to "_ebss", where 32-bit used to use the traditional "_end"), but
other than that it's really very mindless and straigt conversion.
For example, I think we should aim to remove "startup_32" vs "startup_64",
and just call it "startup", and get rid of one more difference. I didn't
do that.
Also, notice the comment in the unified vmlinux.lds.S talks about
"head_64" and "startup_32" which is an odd and incorrect mix, but that was
actually what the old 64-bit only lds file had, so the confusion isn't
new, and now that mixing is arguably more accurate thanks to the
vmlinux.lds.S file being shared between the two cases ;)
[ Impact: cleanup, unification ]
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Acked-by: Sam Ravnborg <sam@ravnborg.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Yinghai Lu
H. Peter Anvin
Denys Vlasenko
Jan Beulich
Tim Abbott
H. Peter Anvin
Sam Ravnborg
Linus Torvalds