Speed up coverage, not drastically

makefile

@@ -44,7 +44,7 @@ endif
 
 ifeq ($(HOST),linux)
 all: umka_shell umka_fuse umka_os umka_gen_devices_dat umka.sym umka.prp \
-     umka.lst tags covpreproc default.skn skin.skn
+     umka.lst tags default.skn skin.skn
 else ifeq ($(HOST),windows)
 all: umka_shell umka.sym umka.prp \
      umka.lst default.skn skin.skn
@@ -57,9 +57,6 @@ endif
 test: umka_shell
 	@cd test && make clean all && cd ../
 
-covpreproc: covpreproc.c
-	$(CC) $(CFLAGS) $(LDFLAGS) $^ -o $@
-
 umka_shell: umka_shell.o umka.o shell.o trace.o trace_lbr.o vdisk.o vnet.o \
             lodepng.o pci.o thread.o util.o optparse.o bestline.o
 	$(CC) $(LDFLAGS_32) $^ -o $@ -T umka.ld

test/makefile

@@ -67,7 +67,7 @@ else
 endif
 
 collect_coverage:
-	../covpreproc ../umka.lst 0x34 coverage.* > ../umka.cov
+	../tools/covpreproc ../umka.lst coverage.* > ../umka.cov
 
 clean:
 	rm -f *.out.log *.out.png coverage.*

tools/covpreproc.c

@@ -3,44 +3,33 @@
 
     UMKa - User-Mode KolibriOS developer tools
 
-    Copyright (C) 2020  Ivan Baravy <dunkaist@gmail.com>
+    Copyright (C) 2020,2022  Ivan Baravy <dunkaist@gmail.com>
 */
 
+#include <inttypes.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
-#include <inttypes.h>
 
-#define MAX_COVERED_CODE_SIZE (256*1024)
+#define COVERAGE_TABLE_SIZE (512*1024)
 
-typedef struct {
+struct coverage_branch {
-    uint64_t to_cnt, from_cnt;
+    uint64_t to_cnt;
-} branch;
+    uint64_t from_cnt;
+};
 
-branch branches[MAX_COVERED_CODE_SIZE];
+struct coverage_branch branches[COVERAGE_TABLE_SIZE];
 
-uint32_t coverage_offset, coverage_begin, coverage_end;
+uint32_t coverage_begin = 0x34; // TODD: detect in runtime
 
 void read_coverage_file(const char *fname) {
     FILE *f = fopen(fname, "r");
-    fseeko(f, 0, SEEK_END);
+    for (size_t i = 0; i < COVERAGE_TABLE_SIZE; i++) {
-    off_t fsize = ftello(f);
+        uint64_t from, to;
-    fseeko(f, 0, SEEK_SET);
+        fread(&to, 1, sizeof(uint64_t), f);
-    fread(&coverage_begin, sizeof(uint32_t), 1, f);
+        fread(&from, 1, sizeof(uint64_t), f);
-    fread(&coverage_end, sizeof(uint32_t), 1, f);
+        branches[i].to_cnt += to;
-    size_t branch_cnt = (fsize-4*2)/(2*4);
+        branches[i].from_cnt += from;
-    for (size_t i = 0; i < branch_cnt; i++) {
-        uint32_t from, to;
-        fread(&from, sizeof(uint32_t), 1, f);
-        fread(&to, sizeof(uint32_t), 1, f);
-        if (from >= coverage_begin && from < coverage_end) {
-            from = from - coverage_begin + coverage_offset;
-            branches[from].from_cnt++;
-        }
-        if (to >= coverage_begin && to < coverage_end) {
-            to = to - coverage_begin + coverage_offset;
-            branches[to].to_cnt++;
-        }
     }
     fclose(f);
 }
@@ -98,19 +87,19 @@ int is_cond_jump(const char *s) {
                 || !strncmp(s, "jpe", 3)
                 || !strncmp(s, "jnp", 3)
                 || !strncmp(s, "jpo", 3)
+                || !strncmp(s, "loop", 4)
                 || !strncmp(s, "jcxz", 4)
                 || !strncmp(s, "jecxz", 5);
     return found;
 }
 
 int main(int argc, char **argv) {
-    if (argc < 4) {
+    if (argc < 3) {
-        fprintf(stderr, "usage: covpreproc <listing file> <coverage_begin offset> <coverage files ...>\n");
+        fprintf(stderr, "usage: covpreproc <listing file> <coverage files ...>\n");
         exit(1);
     }
-    sscanf(argv[2], "%" SCNx32, &coverage_offset);
 
-    for (int i = 3; i < argc; i++) {
+    for (int i = 2; i < argc; i++) {
         read_coverage_file(argv[i]);
     }
 
@@ -131,9 +120,10 @@ int main(int argc, char **argv) {
             unsigned long pos = strtoul(tmp, NULL, 16);
             size_t total_to = 0, total_from = 0;
             for (size_t i = 0; i < inst_len; i++) {
-                if (pos + i < coverage_end - coverage_begin + coverage_offset) {
+                if (pos + i >= coverage_begin
-                    total_to += branches[pos + i].to_cnt;
+                    && pos + i < coverage_begin + COVERAGE_TABLE_SIZE) {
-                    total_from += branches[pos + i].from_cnt;
+                    total_to += branches[pos + i - coverage_begin].to_cnt;
+                    total_from += branches[pos + i - coverage_begin].from_cnt;
                 }
             }
             cur += total_to;
@@ -154,7 +144,8 @@ int main(int argc, char **argv) {
                 putchar('-');
             }
             if (is_cond) {
-                int spaces = 19 - printf("%10" PRIu64 "/%" PRIu64, taken, not_taken);
+                int spaces = 19 - printf("%10" PRIu64 "/%" PRIu64, taken,
+                                         not_taken);
                 while (spaces-- > 0)
                     putchar(' ');
             } else {

tools/makefile

@@ -7,7 +7,7 @@ CFLAGS=$(WARNINGS) $(NOWARNINGS) -std=c11 -O2 \
        -DNDEBUG -D_POSIX_C_SOURCE=200809L -fno-pie
 LDFLAGS=-no-pie
 
-all: mkdirrange mkfilepattern lfbviewx randdir
+all: mkdirrange mkfilepattern lfbviewx randdir covpreproc
 
 mkdirrange: mkdirrange.c
 	$(CC) $(CFLAGS) $(LDFLAGS) $< -o $@
@@ -21,7 +21,10 @@ lfbviewx: lfbviewx.c
 randdir: randdir.c
 	$(CC) $(CFLAGS) $(LDFLAGS) $< -o $@
 
+covpreproc: covpreproc.c
+	$(CC) $(CFLAGS) $(LDFLAGS) $< -o $@
+
 .PHONY: all clean
 
 clean:
-	rm -f *.o mkdirrange mkfilepattern lfbviewx
+	rm -f *.o mkdirrange mkfilepattern lfbviewx randdir covpreproc

trace_lbr.c

@@ -3,7 +3,7 @@
 
     UMKa - User-Mode KolibriOS developer tools
 
-    Copyright (C) 2019-2020  Ivan Baravy <dunkaist@gmail.com>
+    Copyright (C) 2019-2020,2022  Ivan Baravy <dunkaist@gmail.com>
     Copyright (C) 2021  Magomed Kostoev <mkostoevr@yandex.ru>
 */
 
@@ -18,11 +18,13 @@
 
 #include "umka.h"
 
-#define MSR_IA32_DEBUGCTLMSR        0x1d9
+#define MSR_IA32_DEBUGCTL           0x1d9
+#define MSR_IA32_DEBUGCTL_LBR       0x1    // enable profiling
+#define MSR_IA32_DEBUGCTL_BTF       0x2    // profile only branches if EFLAGS.TF
 #define MSR_IA32_LASTBRANCHFROMIP   0x1db
 #define MSR_IA32_LASTBRANCHTOIP     0x1dc
 
-int covfd, msrfd;
+int msrfd;
 
 uint64_t rdmsr(uint32_t reg)
 {
@@ -66,13 +68,15 @@ void handle_sigtrap() {
     uint64_t from = rdmsr(MSR_IA32_LASTBRANCHFROMIP);
     uint64_t to = rdmsr(MSR_IA32_LASTBRANCHTOIP);
 
-    if ((from >= (uintptr_t)coverage_begin && from < (uintptr_t)coverage_end) ||
+    if (from >= (uintptr_t)coverage_begin && from < (uintptr_t)coverage_end) {
-        (to >= (uintptr_t)coverage_begin && to < (uintptr_t)coverage_end)) {
+        coverage_table[from - (uintptr_t)coverage_begin].from_cnt++;
-        write(covfd, &from, 4);
-        write(covfd, &to, 4);
     }
 
-    wrmsr(MSR_IA32_DEBUGCTLMSR, 3);
+    if (to >= (uintptr_t)coverage_begin && to < (uintptr_t)coverage_end) {
+        coverage_table[to - (uintptr_t)coverage_begin].to_cnt++;
+    }
+
+    wrmsr(MSR_IA32_DEBUGCTL, MSR_IA32_DEBUGCTL_LBR + MSR_IA32_DEBUGCTL_BTF);
 #else
     printf("STUB: %s:%d", __FILE__, __LINE__);
 #endif
@@ -87,19 +91,12 @@ void trace_lbr_begin() {
     action.sa_flags = SA_SIGINFO;
     sigaction(SIGTRAP, &action, NULL);
 
-    wrmsr(MSR_IA32_DEBUGCTLMSR, 3);
+    wrmsr(MSR_IA32_DEBUGCTL, MSR_IA32_DEBUGCTL_LBR + MSR_IA32_DEBUGCTL_BTF);
     msrfd = open("/dev/cpu/0/msr", O_RDONLY);
     if (msrfd < 0) {
         perror("rdmsr: open");
         exit(1);
     }
-    char coverage_filename[32];
-    sprintf(coverage_filename, "coverage.%i", getpid());
-    covfd = open(coverage_filename, O_WRONLY | O_CREAT | O_TRUNC, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH | S_IWOTH);
-    void *coverage_begin_addr = coverage_begin;
-    void *coverage_end_addr = coverage_end;
-    write(covfd, &coverage_begin_addr, 4);
-    write(covfd, &coverage_end_addr, 4);
 #else
     printf("STUB: %s:%d", __FILE__, __LINE__);
 #endif
@@ -107,9 +104,8 @@ void trace_lbr_begin() {
 
 void trace_lbr_end() {
 #ifndef _WIN32
-    wrmsr(MSR_IA32_DEBUGCTLMSR, 0);
+    wrmsr(MSR_IA32_DEBUGCTL, 0);
     close(msrfd);
-    close(covfd);
 #else
     printf("STUB: %s:%d", __FILE__, __LINE__);
 #endif

umka.asm

@@ -88,12 +88,12 @@ pubsym irq_serv.irq_10, 'kos_irq_serv_irq10'
 pubsym attach_int_handler, 'kos_attach_int_handler', 12
 pubsym fs_execute, 'kos_fs_execute', 4
 pubsym set_keyboard_data, 'kos_set_keyboard_data'
-pubsym KEY_COUNT as 'kos_key_count'
+pubsym KEY_COUNT, 'kos_key_count'
-pubsym KEY_BUFF as 'kos_key_buff'
+pubsym KEY_BUFF, 'kos_key_buff'
-pubsym keyboard_mode as 'kos_keyboard_mode'
+pubsym keyboard_mode, 'kos_keyboard_mode'
-pubsym sys_getkey as 'kos_get_key'
+pubsym sys_getkey, 'kos_get_key'
-pubsym syslang as 'kos_syslang'
+pubsym syslang, 'kos_syslang'
-pubsym keyboard as 'kos_keyboard'
+pubsym keyboard, 'kos_keyboard'
 
 pubsym disk_add, 16
 pubsym disk_del, 4
@@ -991,6 +991,18 @@ restore org,sys_msg_board,delay_ms
 coverage_end:
 
 
+section '.bss.coverage' executable writable align 64
+struct coverage_branch
+    to_cnt DQ ?
+    from_cnt DQ ?
+ends
+
+COVERAGE_TABLE_SIZE = 512*1024  ; 512k should be enough for the kernel
+
+coverage_table rb COVERAGE_TABLE_SIZE * sizeof.coverage_branch
+pubsym coverage_table
+
+
 section '.data.boot' writeable align 0x1000
 BOOT boot_data
 virtual at BOOT

umka.h

@@ -833,6 +833,14 @@ umka_cli(void);
 void
 umka_sti(void);
 
+#define COVERAGE_TABLE_SIZE (512*1024)
+
+struct coverage_branch {
+    uint64_t to_cnt;
+    uint64_t from_cnt;
+};
+
+extern struct coverage_branch coverage_table[];
 extern uint8_t coverage_begin[];
 extern uint8_t coverage_end[];
 

umka_shell.c

@@ -13,6 +13,7 @@
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
+#include <unistd.h>
 #include "optparse.h"
 #include "shell.h"
 #include "umka.h"
@@ -104,8 +105,15 @@ main(int argc, char **argv) {
 
     run_test(&ctx);
 
-    if (coverage)
+    if (coverage) {
         trace_end();
+        char coverage_filename[32];
+        sprintf(coverage_filename, "coverage.%i", getpid());
+        FILE *f = fopen(coverage_filename, "w");
+        fwrite(coverage_table,
+               COVERAGE_TABLE_SIZE * sizeof(struct coverage_branch), 1, f);
+        fclose(f);
+    }
 
     return 0;
 }