Skip to content
This repository was archived by the owner on Apr 13, 2024. It is now read-only.

Commit aa7feb8

Browse files
tpimhtpimh
committed
patches: 4.4: x86_64: LLD specific patches
1 parent 82e90c2 commit aa7feb8

File tree

1 file changed

+390
-0
lines changed

1 file changed

+390
-0
lines changed

‎patches/4.4/x86_64/lld.patch

Lines changed: 390 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,390 @@
1+
From a011a363720775823e6a291ab62096ae94374bb8 Mon Sep 17 00:00:00 2001
2+
From: =?UTF-8?q?Rafael=20=C3=81vila=20de=20Esp=C3=ADndola?=
3+
<[email protected]>
4+
Date: Wed, 19 Dec 2018 11:01:43 -0800
5+
Subject: [PATCH 1/5] x86/build: Mark per-CPU symbols as absolute explicitly
6+
for LLD
7+
MIME-Version: 1.0
8+
Content-Type: text/plain; charset=UTF-8
9+
Content-Transfer-Encoding: 8bit
10+
11+
[ Upstream commit d071ae09a4a1414c1433d5ae9908959a7325b0ad ]
12+
13+
Accessing per-CPU variables is done by finding the offset of the
14+
variable in the per-CPU block and adding it to the address of the
15+
respective CPU's block.
16+
17+
Section 3.10.8 of ld.bfd's documentation states:
18+
19+
For expressions involving numbers, relative addresses and absolute
20+
addresses, ld follows these rules to evaluate terms:
21+
22+
Other binary operations, that is, between two relative addresses
23+
not in the same section, or between a relative address and an
24+
absolute address, first convert any non-absolute term to an
25+
absolute address before applying the operator."
26+
27+
Note that LLVM's linker does not adhere to the GNU ld's implementation
28+
and as such requires implicitly-absolute terms to be explicitly marked
29+
as absolute in the linker script. If not, it fails currently with:
30+
31+
ld.lld: error: ./arch/x86/kernel/vmlinux.lds:153: at least one side of the expression must be absolute
32+
ld.lld: error: ./arch/x86/kernel/vmlinux.lds:154: at least one side of the expression must be absolute
33+
Makefile:1040: recipe for target 'vmlinux' failed
34+
35+
This is not a functional change for ld.bfd which converts the term to an
36+
absolute symbol anyways as specified above.
37+
38+
Based on a previous submission by Tri Vo <[email protected]>.
39+
40+
Reported-by: Dmitry Golovin <[email protected]>
41+
Signed-off-by: Rafael Ávila de Espíndola <[email protected]>
42+
[ Update commit message per Boris' and Michael's suggestions. ]
43+
Signed-off-by: Nick Desaulniers <[email protected]>
44+
[ Massage commit message more, fix typos. ]
45+
Signed-off-by: Borislav Petkov <[email protected]>
46+
Tested-by: Dmitry Golovin <[email protected]>
47+
Cc: "H. Peter Anvin" <[email protected]>
48+
Cc: Andy Lutomirski <[email protected]>
49+
Cc: Brijesh Singh <[email protected]>
50+
Cc: Cao Jin <[email protected]>
51+
Cc: Ingo Molnar <[email protected]>
52+
Cc: Joerg Roedel <[email protected]>
53+
Cc: Masahiro Yamada <[email protected]>
54+
Cc: Masami Hiramatsu <[email protected]>
55+
Cc: Thomas Gleixner <[email protected]>
56+
Cc: Tri Vo <[email protected]>
57+
Cc: [email protected]
58+
Cc: [email protected]
59+
Cc: x86-ml <[email protected]>
60+
Link: https://lkml.kernel.org/r/[email protected]
61+
Signed-off-by: Sasha Levin <[email protected]>
62+
---
63+
arch/x86/kernel/vmlinux.lds.S | 2 +-
64+
1 file changed, 1 insertion(+), 1 deletion(-)
65+
66+
diff --git a/arch/x86/kernel/vmlinux.lds.S b/arch/x86/kernel/vmlinux.lds.S
67+
index a703842b54de..17e1e60b6b40 100644
68+
--- a/arch/x86/kernel/vmlinux.lds.S
69+
+++ b/arch/x86/kernel/vmlinux.lds.S
70+
@@ -365,7 +365,7 @@ SECTIONS
71+
* Per-cpu symbols which need to be offset from __per_cpu_load
72+
* for the boot processor.
73+
*/
74+
-#define INIT_PER_CPU(x) init_per_cpu__##x = x + __per_cpu_load
75+
+#define INIT_PER_CPU(x) init_per_cpu__##x = ABSOLUTE(x) + __per_cpu_load
76+
INIT_PER_CPU(gdt_page);
77+
INIT_PER_CPU(irq_stack_union);
78+
79+
--
80+
2.21.0
81+
82+
83+
From 9a607e600adedd84e58205ae812974322094efe5 Mon Sep 17 00:00:00 2001
84+
From: Nick Desaulniers <[email protected]>
85+
Date: Mon, 11 Feb 2019 11:30:04 -0800
86+
Subject: [PATCH 2/5] kbuild: clang: choose GCC_TOOLCHAIN_DIR not on LD
87+
88+
commit ad15006cc78459d059af56729c4d9bed7c7fd860 upstream.
89+
90+
This causes an issue when trying to build with `make LD=ld.lld` if
91+
ld.lld and the rest of your cross tools aren't in the same directory
92+
(ex. /usr/local/bin) (as is the case for Android's build system), as the
93+
GCC_TOOLCHAIN_DIR then gets set based on `which $(LD)` which will point
94+
where LLVM tools are, not GCC/binutils tools are located.
95+
96+
Instead, select the GCC_TOOLCHAIN_DIR based on another tool provided by
97+
binutils for which LLVM does not provide a substitute for, such as
98+
elfedit.
99+
100+
Fixes: 785f11aa595b ("kbuild: Add better clang cross build support")
101+
Link: https://github.com/ClangBuiltLinux/linux/issues/341
102+
Suggested-by: Nathan Chancellor <[email protected]>
103+
Reviewed-by: Nathan Chancellor <[email protected]>
104+
Tested-by: Nathan Chancellor <[email protected]>
105+
Signed-off-by: Nick Desaulniers <[email protected]>
106+
Signed-off-by: Masahiro Yamada <[email protected]>
107+
Signed-off-by: Nathan Chancellor <[email protected]>
108+
---
109+
Makefile | 2 +-
110+
1 file changed, 1 insertion(+), 1 deletion(-)
111+
112+
diff --git a/Makefile b/Makefile
113+
index 35be7983ef2d..7bf3fb717921 100644
114+
--- a/Makefile
115+
+++ b/Makefile
116+
@@ -610,7 +610,7 @@ all: vmlinux
117+
ifeq ($(cc-name),clang)
118+
ifneq ($(CROSS_COMPILE),)
119+
CLANG_TARGET := --target=$(notdir $(CROSS_COMPILE:%-=%))
120+
-GCC_TOOLCHAIN_DIR := $(dir $(shell which $(LD)))
121+
+GCC_TOOLCHAIN_DIR := $(dir $(shell which $(CROSS_COMPILE)elfedit))
122+
CLANG_PREFIX := --prefix=$(GCC_TOOLCHAIN_DIR)
123+
GCC_TOOLCHAIN := $(realpath $(GCC_TOOLCHAIN_DIR)/..)
124+
endif
125+
--
126+
2.21.0
127+
128+
129+
From c0cccc8cb3c2a4c74550f7cabbfe1a05426a0d45 Mon Sep 17 00:00:00 2001
130+
From: George Rimar <[email protected]>
131+
Date: Fri, 11 Jan 2019 12:10:12 -0800
132+
Subject: [PATCH 3/5] x86/build: Specify elf_i386 linker emulation explicitly
133+
for i386 objects
134+
135+
commit 927185c124d62a9a4d35878d7f6d432a166b74e3 upstream.
136+
137+
The kernel uses the OUTPUT_FORMAT linker script command in it's linker
138+
scripts. Most of the time, the -m option is passed to the linker with
139+
correct architecture, but sometimes (at least for x86_64) the -m option
140+
contradicts the OUTPUT_FORMAT directive.
141+
142+
Specifically, arch/x86/boot and arch/x86/realmode/rm produce i386 object
143+
files, but are linked with the -m elf_x86_64 linker flag when building
144+
for x86_64.
145+
146+
The GNU linker manpage doesn't explicitly state any tie-breakers between
147+
-m and OUTPUT_FORMAT. But with BFD and Gold linkers, OUTPUT_FORMAT
148+
overrides the emulation value specified with the -m option.
149+
150+
LLVM lld has a different behavior, however. When supplied with
151+
contradicting -m and OUTPUT_FORMAT values it fails with the following
152+
error message:
153+
154+
ld.lld: error: arch/x86/realmode/rm/header.o is incompatible with elf_x86_64
155+
156+
Therefore, just add the correct -m after the incorrect one (it overrides
157+
it), so the linker invocation looks like this:
158+
159+
ld -m elf_x86_64 -z max-page-size=0x200000 -m elf_i386 --emit-relocs -T \
160+
realmode.lds header.o trampoline_64.o stack.o reboot.o -o realmode.elf
161+
162+
This is not a functional change for GNU ld, because (although not
163+
explicitly documented) OUTPUT_FORMAT overrides -m EMULATION.
164+
165+
Tested by building x86_64 kernel with GNU gcc/ld toolchain and booting
166+
it in QEMU.
167+
168+
[ bp: massage and clarify text. ]
169+
170+
Suggested-by: Dmitry Golovin <[email protected]>
171+
Signed-off-by: George Rimar <[email protected]>
172+
Signed-off-by: Tri Vo <[email protected]>
173+
Signed-off-by: Borislav Petkov <[email protected]>
174+
Tested-by: Tri Vo <[email protected]>
175+
Tested-by: Nick Desaulniers <[email protected]>
176+
Cc: "H. Peter Anvin" <[email protected]>
177+
Cc: Ingo Molnar <[email protected]>
178+
Cc: Michael Matz <[email protected]>
179+
Cc: Thomas Gleixner <[email protected]>
180+
Cc: [email protected]
181+
Cc: [email protected]
182+
Cc: [email protected]
183+
Cc: x86-ml <[email protected]>
184+
Link: https://lkml.kernel.org/r/[email protected]
185+
[nc: Fix conflicts due to lack of commit 58ab5e0c2c40 ("Kbuild: arch:
186+
look for generated headers in obtree") in this tree]
187+
Signed-off-by: Nathan Chancellor <[email protected]>
188+
---
189+
arch/x86/boot/Makefile | 2 +-
190+
arch/x86/realmode/rm/Makefile | 2 +-
191+
2 files changed, 2 insertions(+), 2 deletions(-)
192+
193+
diff --git a/arch/x86/boot/Makefile b/arch/x86/boot/Makefile
194+
index 6da2cd0897f3..e94745321cac 100644
195+
--- a/arch/x86/boot/Makefile
196+
+++ b/arch/x86/boot/Makefile
197+
@@ -100,7 +100,7 @@ $(obj)/zoffset.h: $(obj)/compressed/vmlinux FORCE
198+
AFLAGS_header.o += -I$(obj)
199+
$(obj)/header.o: $(obj)/voffset.h $(obj)/zoffset.h
200+
201+
-LDFLAGS_setup.elf := -T
202+
+LDFLAGS_setup.elf := -m elf_i386 -T
203+
$(obj)/setup.elf: $(src)/setup.ld $(SETUP_OBJS) FORCE
204+
$(call if_changed,ld)
205+
206+
diff --git a/arch/x86/realmode/rm/Makefile b/arch/x86/realmode/rm/Makefile
207+
index 2730d775ef9a..228cb16962ba 100644
208+
--- a/arch/x86/realmode/rm/Makefile
209+
+++ b/arch/x86/realmode/rm/Makefile
210+
@@ -43,7 +43,7 @@ $(obj)/pasyms.h: $(REALMODE_OBJS) FORCE
211+
targets += realmode.lds
212+
$(obj)/realmode.lds: $(obj)/pasyms.h
213+
214+
-LDFLAGS_realmode.elf := --emit-relocs -T
215+
+LDFLAGS_realmode.elf := -m elf_i386 --emit-relocs -T
216+
CPPFLAGS_realmode.lds += -P -C -I$(obj)
217+
218+
targets += realmode.elf
219+
--
220+
2.21.0
221+
222+
223+
From ea44fbb53e24856dd3a2e90ed641acb0f61c695a Mon Sep 17 00:00:00 2001
224+
From: Alistair Strachan <[email protected]>
225+
Date: Fri, 3 Aug 2018 10:39:31 -0700
226+
Subject: [PATCH 4/5] x86: vdso: Use $LD instead of $CC to link
227+
228+
commit 379d98ddf41344273d9718556f761420f4dc80b3 upstream.
229+
230+
The vdso{32,64}.so can fail to link with CC=clang when clang tries to find
231+
a suitable GCC toolchain to link these libraries with.
232+
233+
/usr/bin/ld: arch/x86/entry/vdso/vclock_gettime.o:
234+
access beyond end of merged section (782)
235+
236+
This happens because the host environment leaked into the cross compiler
237+
environment due to the way clang searches for suitable GCC toolchains.
238+
239+
Clang is a retargetable compiler, and each invocation of it must provide
240+
--target=<something> --gcc-toolchain=<something> to allow it to find the
241+
correct binutils for cross compilation. These flags had been added to
242+
KBUILD_CFLAGS, but the vdso code uses CC and not KBUILD_CFLAGS (for various
243+
reasons) which breaks clang's ability to find the correct linker when cross
244+
compiling.
245+
246+
Most of the time this goes unnoticed because the host linker is new enough
247+
to work anyway, or is incompatible and skipped, but this cannot be reliably
248+
assumed.
249+
250+
This change alters the vdso makefile to just use LD directly, which
251+
bypasses clang and thus the searching problem. The makefile will just use
252+
${CROSS_COMPILE}ld instead, which is always what we want. This matches the
253+
method used to link vmlinux.
254+
255+
This drops references to DISABLE_LTO; this option doesn't seem to be set
256+
anywhere, and not knowing what its possible values are, it's not clear how
257+
to convert it from CC to LD flag.
258+
259+
Signed-off-by: Alistair Strachan <[email protected]>
260+
Signed-off-by: Thomas Gleixner <[email protected]>
261+
Acked-by: Andy Lutomirski <[email protected]>
262+
Cc: "H. Peter Anvin" <[email protected]>
263+
Cc: Greg Kroah-Hartman <[email protected]>
264+
Cc: [email protected]
265+
Cc: [email protected]
266+
Cc: Andi Kleen <[email protected]>
267+
Link: https://lkml.kernel.org/r/[email protected]
268+
Signed-off-by: Nathan Chancellor <[email protected]>
269+
---
270+
arch/x86/entry/vdso/Makefile | 22 +++++++++-------------
271+
1 file changed, 9 insertions(+), 13 deletions(-)
272+
273+
diff --git a/arch/x86/entry/vdso/Makefile b/arch/x86/entry/vdso/Makefile
274+
index 265c0ed68118..84c4a7105c2a 100644
275+
--- a/arch/x86/entry/vdso/Makefile
276+
+++ b/arch/x86/entry/vdso/Makefile
277+
@@ -41,10 +41,8 @@ targets += $(vdso_img_sodbg)
278+
279+
export CPPFLAGS_vdso.lds += -P -C
280+
281+
-VDSO_LDFLAGS_vdso.lds = -m64 -Wl,-soname=linux-vdso.so.1 \
282+
- -Wl,--no-undefined \
283+
- -Wl,-z,max-page-size=4096 -Wl,-z,common-page-size=4096 \
284+
- $(DISABLE_LTO)
285+
+VDSO_LDFLAGS_vdso.lds = -m elf_x86_64 -soname linux-vdso.so.1 --no-undefined \
286+
+ -z max-page-size=4096 -z common-page-size=4096
287+
288+
$(obj)/vdso64.so.dbg: $(src)/vdso.lds $(vobjs) FORCE
289+
$(call if_changed,vdso)
290+
@@ -90,10 +88,8 @@ CFLAGS_REMOVE_vvar.o = -pg
291+
#
292+
293+
CPPFLAGS_vdsox32.lds = $(CPPFLAGS_vdso.lds)
294+
-VDSO_LDFLAGS_vdsox32.lds = -Wl,-m,elf32_x86_64 \
295+
- -Wl,-soname=linux-vdso.so.1 \
296+
- -Wl,-z,max-page-size=4096 \
297+
- -Wl,-z,common-page-size=4096
298+
+VDSO_LDFLAGS_vdsox32.lds = -m elf32_x86_64 -soname linux-vdso.so.1 \
299+
+ -z max-page-size=4096 -z common-page-size=4096
300+
301+
# 64-bit objects to re-brand as x32
302+
vobjs64-for-x32 := $(filter-out $(vobjs-nox32),$(vobjs-y))
303+
@@ -121,7 +117,7 @@ $(obj)/vdsox32.so.dbg: $(src)/vdsox32.lds $(vobjx32s) FORCE
304+
$(call if_changed,vdso)
305+
306+
CPPFLAGS_vdso32.lds = $(CPPFLAGS_vdso.lds)
307+
-VDSO_LDFLAGS_vdso32.lds = -m32 -Wl,-m,elf_i386 -Wl,-soname=linux-gate.so.1
308+
+VDSO_LDFLAGS_vdso32.lds = -m elf_i386 -soname linux-gate.so.1
309+
310+
# This makes sure the $(obj) subdirectory exists even though vdso32/
311+
# is not a kbuild sub-make subdirectory.
312+
@@ -157,13 +153,13 @@ $(obj)/vdso32.so.dbg: FORCE \
313+
# The DSO images are built using a special linker script.
314+
#
315+
quiet_cmd_vdso = VDSO $@
316+
- cmd_vdso = $(CC) -nostdlib -o $@ \
317+
+ cmd_vdso = $(LD) -nostdlib -o $@ \
318+
$(VDSO_LDFLAGS) $(VDSO_LDFLAGS_$(filter %.lds,$(^F))) \
319+
- -Wl,-T,$(filter %.lds,$^) $(filter %.o,$^) && \
320+
+ -T $(filter %.lds,$^) $(filter %.o,$^) && \
321+
sh $(srctree)/$(src)/checkundef.sh '$(NM)' '$@'
322+
323+
-VDSO_LDFLAGS = -fPIC -shared $(call cc-ldoption, -Wl$(comma)--hash-style=both) \
324+
- $(call cc-ldoption, -Wl$(comma)--build-id) -Wl,-Bsymbolic $(LTO_CFLAGS)
325+
+VDSO_LDFLAGS = -shared $(call ld-option, --hash-style=both) \
326+
+ $(call ld-option, --build-id) -Bsymbolic
327+
GCOV_PROFILE := n
328+
329+
#
330+
--
331+
2.21.0
332+
333+
334+
From b22299ee0b94b2e2ed1492d38631a05492e42c63 Mon Sep 17 00:00:00 2001
335+
From: Nick Desaulniers <[email protected]>
336+
Date: Thu, 6 Dec 2018 11:12:31 -0800
337+
Subject: [PATCH 5/5] x86/vdso: Drop implicit common-page-size linker flag
338+
339+
commit ac3e233d29f7f77f28243af0132057d378d3ea58 upstream.
340+
341+
GNU linker's -z common-page-size's default value is based on the target
342+
architecture. arch/x86/entry/vdso/Makefile sets it to the architecture
343+
default, which is implicit and redundant. Drop it.
344+
345+
Fixes: 2aae950b21e4 ("x86_64: Add vDSO for x86-64 with gettimeofday/clock_gettime/getcpu")
346+
Reported-by: Dmitry Golovin <[email protected]>
347+
Reported-by: Bill Wendling <[email protected]>
348+
Suggested-by: Dmitry Golovin <[email protected]>
349+
Suggested-by: Rui Ueyama <[email protected]>
350+
Signed-off-by: Nick Desaulniers <[email protected]>
351+
Signed-off-by: Borislav Petkov <[email protected]>
352+
Acked-by: Andy Lutomirski <[email protected]>
353+
Cc: Andi Kleen <[email protected]>
354+
Cc: Fangrui Song <[email protected]>
355+
Cc: "H. Peter Anvin" <[email protected]>
356+
Cc: Ingo Molnar <[email protected]>
357+
Cc: Thomas Gleixner <[email protected]>
358+
Cc: x86-ml <[email protected]>
359+
Link: https://lkml.kernel.org/r/[email protected]
360+
Link: https://bugs.llvm.org/show_bug.cgi?id=38774
361+
Link: https://github.com/ClangBuiltLinux/linux/issues/31
362+
Signed-off-by: Nathan Chancellor <[email protected]>
363+
---
364+
arch/x86/entry/vdso/Makefile | 4 ++--
365+
1 file changed, 2 insertions(+), 2 deletions(-)
366+
367+
diff --git a/arch/x86/entry/vdso/Makefile b/arch/x86/entry/vdso/Makefile
368+
index 84c4a7105c2a..297dda4d5947 100644
369+
--- a/arch/x86/entry/vdso/Makefile
370+
+++ b/arch/x86/entry/vdso/Makefile
371+
@@ -42,7 +42,7 @@ targets += $(vdso_img_sodbg)
372+
export CPPFLAGS_vdso.lds += -P -C
373+
374+
VDSO_LDFLAGS_vdso.lds = -m elf_x86_64 -soname linux-vdso.so.1 --no-undefined \
375+
- -z max-page-size=4096 -z common-page-size=4096
376+
+ -z max-page-size=4096
377+
378+
$(obj)/vdso64.so.dbg: $(src)/vdso.lds $(vobjs) FORCE
379+
$(call if_changed,vdso)
380+
@@ -89,7 +89,7 @@ CFLAGS_REMOVE_vvar.o = -pg
381+
382+
CPPFLAGS_vdsox32.lds = $(CPPFLAGS_vdso.lds)
383+
VDSO_LDFLAGS_vdsox32.lds = -m elf32_x86_64 -soname linux-vdso.so.1 \
384+
- -z max-page-size=4096 -z common-page-size=4096
385+
+ -z max-page-size=4096
386+
387+
# 64-bit objects to re-brand as x32
388+
vobjs64-for-x32 := $(filter-out $(vobjs-nox32),$(vobjs-y))
389+
--
390+
2.21.0

0 commit comments

Comments
 (0)