Skip to contentSkip to conversation

x86/purgatory: do not use __builtin_memcpy and __builtin_memset

PATCH Superseded Nick Desaulniers 3 participants 7 messages Jul 23, 2019 – Jul 24, 2019

Subsystems (2)

v3 of 5 Superseded Latest is v5 →

lore
Share & tools
lore
#0 lore raw

PATCH v3 1/2 x86/purgatory: do not use __builtin_memcpy and __builtin_memset

Implementing memcpy and memset in terms of __builtin_memcpy and __builtin_memset is problematic.

GCC at -O2 will replace calls to the builtins with calls to memcpy and memset (but will generate an inline implementation at -Os). Clang will replace the builtins with these calls regardless of optimization level.

$ llvm-objdump -dr arch/x86/purgatory/string.o | tail

0000000000000339 memcpy:

     339: 48 b8 00 00 00 00 00 00 00 00 movabsq $0, %rax
                000000000000033b:  R_X86_64_64  memcpy
     343: ff e0                         jmpq    *%rax

0000000000000345 memset:

     345: 48 b8 00 00 00 00 00 00 00 00 movabsq $0, %rax
                0000000000000347:  R_X86_64_64  memset
     34f: ff e0

Such code results in infinite recursion at runtime. This is observed when doing kexec.

Instead, reuse an implementation from arch/x86/boot/compressed/string.c if we define warn as a symbol. Also, Clang may lower memcmp's that compare against 0 to bcmp's, so add a small definition, too. See also:
commit 5f074f3e192f ("lib/string.c: implement a basic bcmp")

---
Changes v2 -> v3:
* Add bcmp implementation.
* Drop tested-by tag (Vaibhav will help retest).
* Cc stable
Changes v1 -> v2:
* Add Fixes tag.
* Move this patch to first in the series.

 arch/x86/boot/string.c         |  7 +++++++
 arch/x86/purgatory/Makefile    |  3 +++
 arch/x86/purgatory/purgatory.c |  6 ++++++
 arch/x86/purgatory/string.c    | 23 -----------------------
 4 files changed, 16 insertions(+), 23 deletions(-)
 delete mode 100644 arch/x86/purgatory/string.c
arch/x86/boot/string.c +7 -0
index 401e30ca0a75..4c364cf63432 100644
@@ -37,6 +37,13 @@ int memcmp(const void *s1, const void *s2, size_t len)
 	return diff;
 }
 
+/*
+ * Clang may lower `memcmp == 0` to `bcmp == 0`.
+ */
+int bcmp(const void *s1, const void *s2, size_t len) {
+	return memcmp(s1, s2, len);
+}
+
 int strcmp(const char *str1, const char *str2)
 {
 	const unsigned char *s1 = (const unsigned char *)str1;
arch/x86/purgatory/Makefile +3 -0
index 3cf302b26332..91ef244026d2 100644
@@ -6,6 +6,9 @@ purgatory-y := purgatory.o stack.o setup-x86_$(BITS).o sha256.o entry64.o string
 targets += $(purgatory-y)
 PURGATORY_OBJS = $(addprefix $(obj)/,$(purgatory-y))
 
+$(obj)/string.o: $(srctree)/arch/x86/boot/compressed/string.c FORCE
+	$(call if_changed_rule,cc_o_c)
+
 $(obj)/sha256.o: $(srctree)/lib/sha256.c FORCE
 	$(call if_changed_rule,cc_o_c)
 
arch/x86/purgatory/purgatory.c +6 -0
index 6d8d5a34c377..b607bda786f6 100644
@@ -68,3 +68,9 @@ void purgatory(void)
 	}
 	copy_backup_region();
 }
+
+/*
+ * Defined in order to reuse memcpy() and memset() from
+ * arch/x86/boot/compressed/string.c
+ */
+void warn(const char *msg) {}
arch/x86/purgatory/string.c deleted +0 -23
deleted file mode 100644
index 01ad43873ad9..000000000000
@@ -1,23 +0,0 @@
-// SPDX-License-Identifier: GPL-2.0-only
-/*
- * Simple string functions.
- *
- * Copyright (C) 2014 Red Hat Inc.
- *
- * Author:
- *       Vivek Goyal <[email protected]>
- */
-
-#include <linux/types.h>
-
-#include "../boot/string.c"
-
-void *memcpy(void *dst, const void *src, size_t len)
-{
-	return __builtin_memcpy(dst, src, len);
-}
-
-void *memset(void *dst, int c, size_t len)
-{
-	return __builtin_memset(dst, c, len);
-}
Signature · 1 line
2.22.0.709.g102302147b-goog
#1 lore raw

PATCH v3 2/2 x86/purgatory: use CFLAGS_REMOVE rather than reset KBUILD_CFLAGS

KBUILD_CFLAGS is very carefully built up in the top level Makefile, particularly when cross compiling or using different build tools. Resetting KBUILD_CFLAGS via := assignment is an antipattern.

The comment above the reset mentions that -pg is problematic. Other Makefiles use CFLAGS_REMOVE_file.o = $(CC_FLAGS_FTRACE) when CONFIG_FUNCTION_TRACER is set. Prefer that pattern to wiping out all of the important KBUILD_CFLAGS then manually having to re-add them. Seems also that __stack_chk_fail references are generated when using CONFIG_STACKPROTECTOR or CONFIG_STACKPROTECTOR_STRONG.

Fixes 8fc5b4d4121c ("purgatory: core purgatory functionality")

---
Alternatively, we could put these in all in one variable and remove it without any conditional checks (I think that's ok to do so with CFLAGS_REMOVE).

Changes v2 -> v3:
* Prefer $(CC_FLAGS_FTRACE) which is exported to -pg.
* Also check CONFIG_STACKPROTECTOR and CONFIG_STACKPROTECTOR_STRONG.
* Cc stable.
Changes v1 -> v2:
Rather than manually add -mno-sse, -mno-mmx, -mno-sse2, prefer to filter -pg flags.

 arch/x86/purgatory/Makefile | 26 +++++++++++++++++++++-----
 1 file changed, 21 insertions(+), 5 deletions(-)
arch/x86/purgatory/Makefile +21 -5
index 91ef244026d2..6ef0ced59b9c 100644
@@ -20,11 +20,27 @@ KCOV_INSTRUMENT := n
 
 # Default KBUILD_CFLAGS can have -pg option set when FTRACE is enabled. That
 # in turn leaves some undefined symbols like __fentry__ in purgatory and not
-# sure how to relocate those. Like kexec-tools, use custom flags.
-
-KBUILD_CFLAGS := -fno-strict-aliasing -Wall -Wstrict-prototypes -fno-zero-initialized-in-bss -fno-builtin -ffreestanding -c -Os -mcmodel=large
-KBUILD_CFLAGS += -m$(BITS)
-KBUILD_CFLAGS += $(call cc-option,-fno-PIE)
+# sure how to relocate those.
+ifdef CONFIG_FUNCTION_TRACER
+CFLAGS_REMOVE_sha256.o += $(CC_FLAGS_FTRACE)
+CFLAGS_REMOVE_purgatory.o += $(CC_FLAGS_FTRACE)
+CFLAGS_REMOVE_string.o += $(CC_FLAGS_FTRACE)
+CFLAGS_REMOVE_kexec-purgatory.o += $(CC_FLAGS_FTRACE)
+endif
+
+ifdef CONFIG_STACKPROTECTOR
+CFLAGS_REMOVE_sha256.o += -fstack-protector
+CFLAGS_REMOVE_purgatory.o += -fstack-protector
+CFLAGS_REMOVE_string.o += -fstack-protector
+CFLAGS_REMOVE_kexec-purgatory.o += -fstack-protector
+endif
+
+ifdef CONFIG_STACKPROTECTOR_STRONG
+CFLAGS_REMOVE_sha256.o += -fstack-protector-strong
+CFLAGS_REMOVE_purgatory.o += -fstack-protector-strong
+CFLAGS_REMOVE_string.o += -fstack-protector-strong
+CFLAGS_REMOVE_kexec-purgatory.o += -fstack-protector-strong
+endif
 
 $(obj)/purgatory.ro: $(PURGATORY_OBJS) FORCE
 		$(call if_changed,ld)
Signature · 1 line
2.22.0.709.g102302147b-goog
#2 lore raw

Re: [PATCH v3 2/2] x86/purgatory: use CFLAGS_REMOVE rather than reset KBUILD_CFLAGS

arch/x86/purgatory/Makefile KCOV_INSTRUMENT := n · line 28

Quoted 5 lines · Nick Desaulniers
arch/x86/purgatory/Makefile +5 -0
+ifdef CONFIG_FUNCTION_TRACER
+CFLAGS_REMOVE_sha256.o += $(CC_FLAGS_FTRACE)
+CFLAGS_REMOVE_purgatory.o += $(CC_FLAGS_FTRACE)
+CFLAGS_REMOVE_string.o += $(CC_FLAGS_FTRACE)
+CFLAGS_REMOVE_kexec-purgatory.o += $(CC_FLAGS_FTRACE)

Nit. Can you please make that tabular?

CFLAGS_REMOVE_sha256.o += $(CC_FLAGS_FTRACE)
CFLAGS_REMOVE_purgatory.o += $(CC_FLAGS_FTRACE)
CFLAGS_REMOVE_string.o += $(CC_FLAGS_FTRACE)
CFLAGS_REMOVE_kexec-purgatory.o += $(CC_FLAGS_FTRACE)

The above confused my pattern recognition engine :)

Thanks,

tglx

Tested-by on v3 2/2
#3 lore raw

Re: [PATCH v3 2/2] x86/purgatory: use CFLAGS_REMOVE rather than reset KBUILD_CFLAGS

In reply to Nick Desaulniers
Quoted 71 lines · Nick Desaulniers

KBUILD_CFLAGS is very carefully built up in the top level Makefile, particularly when cross compiling or using different build tools. Resetting KBUILD_CFLAGS via := assignment is an antipattern.

The comment above the reset mentions that -pg is problematic. Other Makefiles use CFLAGS_REMOVE_file.o = $(CC_FLAGS_FTRACE) when CONFIG_FUNCTION_TRACER is set. Prefer that pattern to wiping out all of the important KBUILD_CFLAGS then manually having to re-add them. Seems also that __stack_chk_fail references are generated when using CONFIG_STACKPROTECTOR or CONFIG_STACKPROTECTOR_STRONG.

Fixes 8fc5b4d4121c ("purgatory: core purgatory functionality")

---
Alternatively, we could put these in all in one variable and remove it without any conditional checks (I think that's ok to do so with CFLAGS_REMOVE).

Changes v2 -> v3:
* Prefer $(CC_FLAGS_FTRACE) which is exported to -pg.
* Also check CONFIG_STACKPROTECTOR and CONFIG_STACKPROTECTOR_STRONG.
* Cc stable.
Changes v1 -> v2:
Rather than manually add -mno-sse, -mno-mmx, -mno-sse2, prefer to filter

(quoted diff) +0 -1
-pg flags.
 
 arch/x86/purgatory/Makefile | 26 +++++++++++++++++++++-----
 1 file changed, 21 insertions(+), 5 deletions(-)
arch/x86/purgatory/Makefile +21 -6
index 91ef244026d2..6ef0ced59b9c 100644
@@ -20,11 +20,27 @@ KCOV_INSTRUMENT := n
 
 # Default KBUILD_CFLAGS can have -pg option set when FTRACE is enabled. That
 # in turn leaves some undefined symbols like __fentry__ in purgatory and not
-# sure how to relocate those. Like kexec-tools, use custom flags.
-
-KBUILD_CFLAGS := -fno-strict-aliasing -Wall -Wstrict-prototypes -fno-zero-initialized-in-bss -fno-builtin -ffreestanding -c -Os -mcmodel=large
-KBUILD_CFLAGS += -m$(BITS)
-KBUILD_CFLAGS += $(call cc-option,-fno-PIE)
+# sure how to relocate those.
+ifdef CONFIG_FUNCTION_TRACER
+CFLAGS_REMOVE_sha256.o += $(CC_FLAGS_FTRACE)
+CFLAGS_REMOVE_purgatory.o += $(CC_FLAGS_FTRACE)
+CFLAGS_REMOVE_string.o += $(CC_FLAGS_FTRACE)
+CFLAGS_REMOVE_kexec-purgatory.o += $(CC_FLAGS_FTRACE)
+endif
+
+ifdef CONFIG_STACKPROTECTOR
+CFLAGS_REMOVE_sha256.o += -fstack-protector
+CFLAGS_REMOVE_purgatory.o += -fstack-protector
+CFLAGS_REMOVE_string.o += -fstack-protector
+CFLAGS_REMOVE_kexec-purgatory.o += -fstack-protector
+endif
+
+ifdef CONFIG_STACKPROTECTOR_STRONG
+CFLAGS_REMOVE_sha256.o += -fstack-protector-strong
+CFLAGS_REMOVE_purgatory.o += -fstack-protector-strong
+CFLAGS_REMOVE_string.o += -fstack-protector-strong
+CFLAGS_REMOVE_kexec-purgatory.o += -fstack-protector-strong
+endif
 
 $(obj)/purgatory.ro: $(PURGATORY_OBJS) FORCE
                $(call if_changed,ld)
--
Quoted 2 lines

2.22.0.709.g102302147b-goog

I tested the v3 patch series with clang compiled kernel for below scenarios:

1. kexec'ing into a new kernel.
2. Purposely crashing the running kernel to generate kdump logs.

Thanks,
Vaibhav

#4 lore raw

v3 cover Support kexec/kdump for clang built kernel

In reply to Nick Desaulniers

1. Reuse the implementation of memcpy and memset instead of relying on __builtin_memcpy and __builtin_memset as it causes infinite recursion in Clang (at any opt level) or GCC at -O2.
2. Don't reset KBUILD_CFLAGS, rather filter CONFIG_FUNCTION_TRACER, CONFIG_STACKPROTECTOR, and CONFIG_STACKPROTECTOR_STRONG flags via `CFLAGS_REMOVE_<file>.o'

A good test of this series (besides boot testing a kexec kernel):
* There should be no undefined symbols in arch/x86/purgatory/purgatory.ro:

$ nm arch/x86/purgatory/purgatory.ro
  particularly `warn`, `bcmp`, `__stack_chk_fail`, `memcpy` or `memset`.
* `-pg`, `-fstack-protector`, `-fstack-protector-strong` should not be
  added to the command line for the c source files under arch/x86/purgatory/
  when compiling with CONFIG_FUNCTION_TRACER=y, CONFIG_STACKPROTECTOR=y,
  and CONFIG_STACKPROTECTOR_STRONG=y.

V3 of: https://groups.google.com/forum/#!msg/clang-built-linux/GzThkAkZqqI/J727I6vpBwAJ

Nick Desaulniers (2):
x86/purgatory: do not use __builtin_memcpy and __builtin_memset x86/purgatory: use CFLAGS_REMOVE rather than reset KBUILD_CFLAGS

 arch/x86/boot/string.c         |  7 +++++++
 arch/x86/purgatory/Makefile    | 29 ++++++++++++++++++++++++-----
 arch/x86/purgatory/purgatory.c |  6 ++++++
 arch/x86/purgatory/string.c    | 23 -----------------------
 4 files changed, 37 insertions(+), 28 deletions(-)
 delete mode 100644 arch/x86/purgatory/string.c
Signature · 1 line
2.22.0.709.g102302147b-goog
#5 lore raw
In reply to Nick Desaulniers
Quoted 5 lines · Nick Desaulniers

Instead, reuse an implementation from arch/x86/boot/compressed/string.c if we define warn as a symbol. Also, Clang may lower memcmp's that compare against 0 to bcmp's, so add a small definition, too. See also:
commit 5f074f3e192f ("lib/string.c: implement a basic bcmp")

Quoted 9 lines

That SOB chain is weird. Is Vaibhav the author?

arch/x86/boot/string.c int memcmp(const void *s1, const void *s2, size_t len) · line 45

Quoted 6 lines
arch/x86/boot/string.c +6 -0
+/*
+ * Clang may lower `memcmp == 0` to `bcmp == 0`.
+ */
+int bcmp(const void *s1, const void *s2, size_t len) {
+	return memcmp(s1, s2, len);
+}

foo()
{
}

please.

Thanks,

tglx

#6 lore raw
Quoted 17 lines · Thomas Gleixner
Quoted 5 lines · Nick Desaulniers

Instead, reuse an implementation from arch/x86/boot/compressed/string.c if we define warn as a symbol. Also, Clang may lower memcmp's that compare against 0 to bcmp's, so add a small definition, too. See also:
commit 5f074f3e192f ("lib/string.c: implement a basic bcmp")

Quoted 9 lines
Quoted 2 lines

That SOB chain is weird. Is Vaibhav the author?

No, Nick Desaulniers is the author of v3.

Quoted 11 lines

arch/x86/boot/string.c int memcmp(const void *s1, const void *s2, size_t len) · line 45

Quoted 6 lines
arch/x86/boot/string.c +6 -0
+/*
+ * Clang may lower `memcmp == 0` to `bcmp == 0`.
+ */
+int bcmp(const void *s1, const void *s2, size_t len) {
+     return memcmp(s1, s2, len);
+}

foo()
{
}

Quoted 5 lines

please.

Thanks,

tglx

Thanks,
Vaibhav