Get --ftrace working on aarch64

This change implements a new approach to function call logging, that's based on the GCC flag: -fpatchable-function-entry. Read the commentary in build/config.mk to learn how it works.
2025-07-24 19:40:28 +00:00 · 2023-06-05 23:35:31 -07:00 · 2023-06-05 23:35:31 -07:00 · eb40cb371d
commit eb40cb371d
parent 5b908bc756
934 changed files with 2259 additions and 1268 deletions
--- a/tool/build/build.mk
+++ b/tool/build/build.mk
@ -105,7 +105,7 @@ o/$(MODE)/tool/build/blinkenlights.com:				\
 	@$(MAKE_SYMTAB_ZIP)

 o/$(MODE)/tool/build/emulator.o: private		\
-		OVERRIDE_COPTS +=			\
+		COPTS +=				\
 			-fno-sanitize=pointer-overflow

 o/$(MODE)/tool/build/dso/sandbox.so.zip.o		\
@ -127,7 +127,7 @@ o/$(MODE)/tool/build/dd.zip.o: private			\
 # we need pic because:
 #   so it can be an LD_PRELOAD payload
 o/$(MODE)/tool/build/dso/sandbox.o: private		\
-		OVERRIDE_CFLAGS +=			\
+		CFLAGS +=				\
 			-fPIC

 o/$(MODE)/tool/build/dso/sandbox.o:			\
--- a/tool/build/emubin/emubin.mk
+++ b/tool/build/emubin/emubin.mk
@ -61,7 +61,7 @@ o/tiny/tool/build/emubin/mdatest.bin.dbg:		\
 	@$(ELFLINK) -z max-page-size=0x10 -T tool/build/emucrt/real.lds

 $(TOOL_BUILD_EMUBIN_OBJS): private			\
-		OVERRIDE_CFLAGS +=			\
+		CFLAGS +=				\
 			$(NO_MAGIC)

 .PHONY: o/$(MODE)/tool/build/emubin
--- a/tool/build/fixupobj.c
+++ b/tool/build/fixupobj.c
@ -54,6 +54,17 @@ Usage: fixupobj.com [-h] ARGS...\n\
 #define MRS_TPIDR_EL0     0xd53bd040u
 #define MOV_REG(DST, SRC) (0xaa0003e0u | (SRC) << 16 | (DST))

+static const unsigned char kFatNops[8][8] = {
+    {},                                          //
+    {0x90},                                      // nop
+    {0x66, 0x90},                                // xchg %ax,%ax
+    {0x0f, 0x1f, 0x00},                          // nopl (%rax)
+    {0x0f, 0x1f, 0x40, 0x00},                    // nopl 0x00(%rax)
+    {0x0f, 0x1f, 0x44, 0x00, 0x00},              // nopl 0x00(%rax,%rax,1)
+    {0x66, 0x0f, 0x1f, 0x44, 0x00, 0x00},        // nopw 0x00(%rax,%rax,1)
+    {0x0f, 0x1f, 0x80, 0x00, 0x00, 0x00, 0x00},  // nopl 0x00000000(%rax)
+};
+
 void Write(const char *s, ...) {
  va_list va;
  va_start(va, s);
@ -110,6 +121,44 @@ void RewriteTlsCode(Elf64_Ehdr *elf, size_t elfsize) {
  }
 }

+/**
+ * Improve GCC11 `-fpatchable-function-entry` codegen.
+ *
+ * When using flags like `-fpatchable-function-entry=9,7` GCC v11 will
+ * insert two `nop` instructions, rather than merging them into faster
+ * "fat" nops.
+ *
+ * In order for this to work, the function symbol must be declared as
+ * `STT_FUNC` and `st_size` must have the function's byte length.
+ */
+void OptimizePatchableFunctionEntries(Elf64_Ehdr *elf, size_t elfsize) {
+#ifdef __x86_64__
+  long i, n;
+  int nopcount;
+  Elf64_Sym *syms;
+  Elf64_Shdr *shdr;
+  Elf64_Xword symcount;
+  unsigned char *p, *pe;
+  CHECK_NOTNULL((syms = GetElfSymbolTable(elf, elfsize, &symcount)));
+  for (i = 0; i < symcount; ++i) {
+    if (ELF64_ST_TYPE(syms[i].st_info) == STT_FUNC && syms[i].st_size) {
+      shdr = GetElfSectionHeaderAddress(elf, elfsize, syms[i].st_shndx);
+      p = GetElfSectionAddress(elf, elfsize, shdr);
+      p += syms[i].st_value;
+      pe = p + syms[i].st_size;
+      for (; p + 1 < pe; p += n) {
+        if (p[0] != 0x90) break;
+        if (p[1] != 0x90) break;
+        for (n = 2; p + n < pe && n < ARRAYLEN(kFatNops); ++n) {
+          if (p[n] != 0x90) break;
+        }
+        memcpy(p, kFatNops[n], n);
+      }
+    }
+  }
+#endif /* __x86_64__ */
+}
+
 void OptimizeRelocations(Elf64_Ehdr *elf, size_t elfsize) {
  char *strs;
  Elf64_Half i;
@ -191,6 +240,7 @@ void RewriteObject(const char *path) {
    }
    if (elf->e_machine == EM_NEXGEN32E) {
      OptimizeRelocations(elf, st.st_size);
+      OptimizePatchableFunctionEntries(elf, st.st_size);
    }
    if (elf->e_machine == EM_AARCH64) {
      RewriteTlsCode(elf, st.st_size);
--- a/tool/build/lib/asmdown.c
+++ b/tool/build/lib/asmdown.c
@ -16,10 +16,10 @@
 │ TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR             │
 │ PERFORMANCE OF THIS SOFTWARE.                                                │
 ╚─────────────────────────────────────────────────────────────────────────────*/
+#include "tool/build/lib/asmdown.h"
 #include "libc/mem/alg.h"
 #include "libc/mem/mem.h"
 #include "libc/str/str.h"
-#include "tool/build/lib/asmdown.h"
 #include "tool/build/lib/javadown.h"

 static bool IsSymbolChar1(char c) {
@ -76,6 +76,8 @@ struct Asmdown *ParseAsmdown(const char *code, size_t size) {
          state = SYM;
        } else if (code[i] == '\n') {
          ++line;
+        } else if (i + 10 < size && !memcmp(code + i, "\t.ftrace1\t", 10)) {
+          ++line;
        } else if (i + 8 < size && !memcmp(code + i, "\t.alias\t", 8)) {
          p1 = code + i + 8;
          if ((p2 = strchr(p1, ',')) && (p3 = strchr(p2, '\n'))) {
--- a/tool/lambda/lambda.mk
+++ b/tool/lambda/lambda.mk
@ -47,7 +47,7 @@ o/$(MODE)/tool/lambda/%.com.dbg:				\
 	@$(APELINK)

 o/$(MODE)/tool/lambda/tromp.o: private				\
-		OVERRIDE_CFLAGS +=				\
+		CFLAGS +=					\
 			-w

 $(TOOL_LAMBDA_OBJS):						\
--- a/tool/net/net.mk
+++ b/tool/net/net.mk
@ -131,7 +131,7 @@ o/$(MODE)/tool/net/redbean.com:							\
 	@$(TOOL_NET_REDBEAN_STANDARD_ASSETS_ZIP)

 o/$(MODE)/tool/net/lsqlite3.o: private						\
-		OVERRIDE_CFLAGS +=						\
+		CFLAGS +=							\
 			-DSQLITE_ENABLE_SESSION

 # REDBEAN-DEMO.COM
--- a/tool/plinko/lib/lib.mk
+++ b/tool/plinko/lib/lib.mk
@ -53,14 +53,14 @@ $(TOOL_PLINKO_LIB_A).pkg:				\
 		$(foreach x,$(TOOL_PLINKO_LIB_A_DIRECTDEPS),$($(x)_A).pkg)

 ifeq ($(MODE),)
-$(TOOL_PLINKO_LIB_A_OBJS): private OVERRIDE_CFLAGS += -fno-inline
+$(TOOL_PLINKO_LIB_A_OBJS): private CFLAGS += -fno-inline
 endif

 ifeq ($(MODE),dbg)
-$(TOOL_PLINKO_LIB_A_OBJS): private OVERRIDE_CFLAGS += -fno-inline
+$(TOOL_PLINKO_LIB_A_OBJS): private CFLAGS += -fno-inline
 endif

-$(TOOL_PLINKO_LIB_A_OBJS): private OVERRIDE_CFLAGS += -ffast-math -foptimize-sibling-calls -O2
+$(TOOL_PLINKO_LIB_A_OBJS): private CFLAGS += -ffast-math -foptimize-sibling-calls -O2

 TOOL_PLINKO_LIB_LIBS = $(foreach x,$(TOOL_PLINKO_LIB_ARTIFACTS),$($(x)))
 TOOL_PLINKO_LIB_SRCS = $(foreach x,$(TOOL_PLINKO_LIB_ARTIFACTS),$($(x)_SRCS))
--- a/tool/viz/lib/doublechrominance.S
+++ b/tool/viz/lib/doublechrominance.S
@ -27,9 +27,10 @@
 //	@param	%rsi is char[edi/16][16] output and %rsi==%rdx is OK
 //	@param	%rdx is char[edi/16][8] input
 //	@return	%rax is %rsi
+	.ftrace1
 doublechrominance:
+	.ftrace2
 	.leafprologue
-	.profilable
 	shr	$1,%edi			# backwards algorithm
 	jbe	1f			# do nothing if !n || n%2
 	mov	%edi,%ecx
--- a/tool/viz/lib/vizlib.mk
+++ b/tool/viz/lib/vizlib.mk
@ -50,7 +50,7 @@ TOOL_VIZ_LIB_A_DEPS :=					\
 	$(call uniq,$(foreach x,$(TOOL_VIZ_LIB_A_DIRECTDEPS),$($(x))))

 o/$(MODE)/tool/viz/lib/pmaddubsw.o: private		\
-		OVERRIDE_CFLAGS +=			\
+		CFLAGS +=				\
 			-fvect-cost-model=unlimited

 o/$(MODE)/tool/viz/lib/scale.o				\
@ -69,12 +69,12 @@ o/$(MODE)/tool/viz/lib/getxtermcodes.o			\
 o/$(MODE)/tool/viz/lib/lingamma.o			\
 o/$(MODE)/tool/viz/lib/perlin3.o			\
 o/$(MODE)/tool/viz/lib/resizegraphic.o: private		\
-		OVERRIDE_CFLAGS +=			\
+		CFLAGS +=				\
 			-DSTACK_FRAME_UNLIMITED		\
 			$(MATHEMATICAL)

 o/$(MODE)/tool/viz/lib/printmatrix.o: private		\
-		OVERRIDE_CFLAGS +=			\
+		CFLAGS +=				\
 			$(IEEE_MATH)

 $(TOOL_VIZ_LIB_A):					\
--- a/tool/viz/viz.mk
+++ b/tool/viz/viz.mk
@ -98,12 +98,12 @@ o/$(MODE)/tool/viz/printvideo.com:			\
 	@$(MAKE_SYMTAB_ZIP)

 o/$(MODE)/tool/viz/derasterize.o: private	\
-		OVERRIDE_CFLAGS +=		\
+		CFLAGS +=			\
 			-DSTACK_FRAME_UNLIMITED	\
 			$(MATHEMATICAL)

 o/$(MODE)/tool/viz/magikarp.o: private		\
-		OVERRIDE_CFLAGS +=		\
+		CFLAGS +=			\
 			$(MATHEMATICAL)

 $(TOOL_VIZ_OBJS):				\