4 files changed, 111 insertions, 28 deletions

diff --git a/libbb/Kbuild.src b/libbb/Kbuild.src
index 7785ef152..32fde90e6 100644
--- a/libbb/Kbuild.src
+++ b/libbb/Kbuild.src
@@ -58,6 +58,7 @@ lib-y += parse_mode.o
 lib-y += perror_msg.o
 lib-y += perror_nomsg_and_die.o
 lib-y += platform.o
+lib-y += popcnt.o
 lib-y += printable.o
 lib-y += printable_string.o
 lib-y += process_escape_sequence.o
@@ -99,6 +100,7 @@ lib-y += xgetcwd.o
 lib-y += xreadlink.o
 lib-y += xrealloc_vector.o
 
+lib-$(CONFIG_PLATFORM_POSIX) += alloc_affinity.o
 lib-$(CONFIG_PLATFORM_POSIX) += bb_askpass.o
 lib-$(CONFIG_PLATFORM_POSIX) += change_identity.o
 lib-$(CONFIG_PLATFORM_POSIX) += device_open.o
diff --git a/libbb/alloc_affinity.c b/libbb/alloc_affinity.c
new file mode 100644
index 000000000..b6d9f649a
--- /dev/null
+++ b/libbb/alloc_affinity.c
@@ -0,0 +1,29 @@
+/* vi: set sw=4 ts=4: */
+/*
+ * Utility routines.
+ *
+ * Copyright (C) 2024 Denys Vlasenko
+ *
+ * Licensed under GPLv2, see file LICENSE in this source tree.
+ */
+#include <sched.h>
+#include "libbb.h"
+
+unsigned long* FAST_FUNC get_malloc_cpu_affinity(int pid, unsigned *sz)
+{
+        unsigned long *mask = NULL;
+        unsigned sz_in_bytes = *sz;
+
+        for (;;) {
+                mask = xrealloc(mask, sz_in_bytes);
+                if (sched_getaffinity(pid, sz_in_bytes, (void*)mask) == 0)
+                        break; /* got it */
+                sz_in_bytes *= 2;
+                if (errno == EINVAL && (int)sz_in_bytes > 0)
+                        continue;
+                bb_perror_msg_and_die("can't %cet pid %d's affinity", 'g', pid);
+        }
+        //bb_error_msg("get mask[0]:%lx sz_in_bytes:%d", mask[0], sz_in_bytes);
+        *sz = sz_in_bytes;
+        return mask;
+}
diff --git a/libbb/lineedit.c b/libbb/lineedit.c
index 811e3e475..bd4a17348 100644
--- a/libbb/lineedit.c
+++ b/libbb/lineedit.c
@@ -351,7 +351,7 @@ static unsigned save_string(char *dst, unsigned maxsize)
                 return i;
         }
 }
-/* I thought just fputwc(c, stdout) would work. But no... */
+/* I thought just fputwc(c, stderr) would work. But no... */
 static void BB_PUTCHAR(wchar_t c)
 {
         if (unicode_status == UNICODE_ON) {
@@ -360,11 +360,11 @@ static void BB_PUTCHAR(wchar_t c)
                 ssize_t len = wcrtomb(buf, c, &mbst);
                 if (len > 0) {
                         buf[len] = '\0';
-                        fputs_stdout(buf);
+                        fputs(buf, stderr);
                 }
         } else {
                 /* In this case, c is always one byte */
-                putchar(c);
+                bb_putchar_stderr(c);
         }
 }
 # if ENABLE_UNICODE_COMBINING_WCHARS || ENABLE_UNICODE_WIDE_WCHARS
@@ -410,7 +410,7 @@ static void save_string(char *dst, unsigned maxsize)
         safe_strncpy(dst, command_ps, maxsize);
 }
 # endif
-# define BB_PUTCHAR(c) bb_putchar(c)
+# define BB_PUTCHAR(c) bb_putchar_stderr(c)
 /* Should never be called: */
 int adjust_width_and_validate_wc(unsigned *width_adj, int wc);
 #endif
@@ -469,7 +469,7 @@ static void put_cur_glyph_and_inc_cursor(void)
                 if (c == BB_NUL)
                         c = ' ';
                 BB_PUTCHAR(c);
-                bb_putchar('\b');
+                bb_putchar_stderr('\b');
 #endif
                 cmdedit_y++;
                 if (!ENABLE_UNICODE_WIDE_WCHARS || ofs_to_right == 0) {
@@ -531,12 +531,12 @@ static void goto_new_line(void)
         put_till_end_and_adv_cursor();
         /* "cursor == 0" is only if prompt is "" and user input is empty */
         if (cursor == 0 || cmdedit_x != 0)
-                bb_putchar('\n');
+                bb_putchar_stderr('\n');
 }
 
 static void beep(void)
 {
-        bb_putchar('\007');
+        bb_putchar_stderr('\007');
 }
 
 /* Full or last/sole prompt line, reset edit cursor, calculate terminal cursor.
@@ -544,7 +544,10 @@ static void beep(void)
  */
 static void put_prompt_custom(bool is_full)
 {
-        fputs_stdout((is_full ? cmdedit_prompt : prompt_last_line));
+        /* https://pubs.opengroup.org/onlinepubs/9699919799/utilities/V3_chap02.html
+         * says that shells must write $PSn to stderr, not stdout.
+         */
+        fputs((is_full ? cmdedit_prompt : prompt_last_line), stderr);
         cursor = 0;
         cmdedit_y = cmdedit_prmt_len / cmdedit_termw; /* new quasireal y */
         cmdedit_x = cmdedit_prmt_len % cmdedit_termw;
@@ -582,16 +585,16 @@ static void input_backward(unsigned num)
                         /* This is longer by 5 bytes on x86.
                          * Also gets miscompiled for ARM users
                          * (busybox.net/bugs/view.php?id=2274).
-                         * printf(("\b\b\b\b" + 4) - num);
+                         * fprintf(("\b\b\b\b" + 4) - num, stderr);
                          * return;
                          */
                         do {
-                                bb_putchar('\b');
+                                bb_putchar_stderr('\b');
                         } while (--num);
                         return;
                 }
 #endif
-                printf(ESC"[%uD", num);
+                fprintf(stderr, ESC"[%uD", num);
                 return;
         }
 
@@ -616,7 +619,7 @@ static void input_backward(unsigned num)
                  */
                 unsigned sv_cursor;
                 /* go to 1st column; go up to first line */
-                printf("\r" ESC"[%uA", cmdedit_y);
+                fprintf(stderr, "\r" ESC"[%uA", cmdedit_y);
                 cmdedit_y = 0;
                 sv_cursor = cursor;
                 put_prompt_last_line(); /* sets cursor to 0 */
@@ -631,12 +634,12 @@ static void input_backward(unsigned num)
                 cmdedit_x = (cmdedit_termw * cmdedit_y - num) % cmdedit_termw;
                 cmdedit_y -= lines_up;
                 /* go to 1st column; go up */
-                printf("\r" ESC"[%uA", lines_up);
+                fprintf(stderr, "\r" ESC"[%uA", lines_up);
                 /* go to correct column.
                  * xterm, konsole, Linux VT interpret 0 as 1 below! wow.
                  * need to *make sure* we skip it if cmdedit_x == 0 */
                 if (cmdedit_x)
-                        printf(ESC"[%uC", cmdedit_x);
+                        fprintf(stderr, ESC"[%uC", cmdedit_x);
         }
 }
 
@@ -644,11 +647,11 @@ static void input_backward(unsigned num)
 static void draw_custom(int y, int back_cursor, bool is_full)
 {
         if (y > 0) /* up y lines */
-                printf(ESC"[%uA", y);
-        bb_putchar('\r');
+                fprintf(stderr, ESC"[%uA", y);
+        bb_putchar_stderr('\r');
         put_prompt_custom(is_full);
         put_till_end_and_adv_cursor();
-        printf(SEQ_CLEAR_TILL_END_OF_SCREEN);
+        fputs(SEQ_CLEAR_TILL_END_OF_SCREEN, stderr);
         input_backward(back_cursor);
 }
 
@@ -693,7 +696,7 @@ static void input_delete(int save)
         command_len--;
         put_till_end_and_adv_cursor();
         /* Last char is still visible, erase it (and more) */
-        printf(SEQ_CLEAR_TILL_END_OF_SCREEN);
+        fputs(SEQ_CLEAR_TILL_END_OF_SCREEN, stderr);
         input_backward(cursor - j);     /* back to old pos cursor */
 }
 
@@ -1110,8 +1113,8 @@ static void remove_chunk(int16_t *int_buf, int beg, int end)
         if (dbg_bmp) {
                 int i;
                 for (i = 0; int_buf[i]; i++)
-                        bb_putchar((unsigned char)int_buf[i]);
-                bb_putchar('\n');
+                        bb_putchar_stderr((unsigned char)int_buf[i]);
+                bb_putchar_stderr('\n');
         }
 }
 /* Caller ensures that match_buf points to a malloced buffer
@@ -1294,7 +1297,7 @@ static void showfiles(void)
                 int nc;
 
                 for (nc = 1; nc < ncols && n+nrows < nfiles; n += nrows, nc++) {
-                        printf("%s%-*s", matches[n],
+                        fprintf(stderr, "%s%-*s", matches[n],
                                 (int)(column_width - unicode_strwidth(matches[n])), ""
                         );
                 }
@@ -1650,7 +1653,7 @@ void FAST_FUNC show_history(const line_input_t *st)
         if (!st)
                 return;
         for (i = 0; i < st->cnt_history; i++)
-                printf("%4d %s\n", i, st->history[i]);
+                fprintf(stderr, "%4d %s\n", i, st->history[i]);
 }
 
 # if ENABLE_FEATURE_EDITING_SAVEHISTORY
@@ -2090,7 +2093,7 @@ static void ask_terminal(void)
         pfd.events = POLLIN;
         if (safe_poll(&pfd, 1, 0) == 0) {
                 S.sent_ESC_br6n = 1;
-                fputs_stdout(ESC"[6n");
+                fputs(ESC"[6n", stderr);
                 fflush_all(); /* make terminal see it ASAP! */
         }
 }
@@ -2857,13 +2860,13 @@ int FAST_FUNC read_line_input(line_input_t *st, const char *prompt, char *comman
                         /* Control-k -- clear to end of line */
                         command_ps[cursor] = BB_NUL;
                         command_len = cursor;
-                        printf(SEQ_CLEAR_TILL_END_OF_SCREEN);
+                        fputs(SEQ_CLEAR_TILL_END_OF_SCREEN, stderr);
                         break;
                 case CTRL('L'):
                 vi_case(CTRL('L')|VI_CMDMODE_BIT:)
                         /* Control-l -- clear screen */
                         /* cursor to top,left; clear to the end of screen */
-                        printf(ESC"[H" ESC"[J");
+                        fputs(ESC"[H" ESC"[J", stderr);
                         draw_full(command_len - cursor);
                         break;
 #if MAX_HISTORY > 0
@@ -3050,8 +3053,8 @@ int FAST_FUNC read_line_input(line_input_t *st, const char *prompt, char *comman
                                 beep();
                         } else {
                                 command_ps[cursor] = ic;
-                                bb_putchar(ic);
-                                bb_putchar('\b');
+                                bb_putchar_stderr(ic);
+                                bb_putchar_stderr('\b');
                         }
                         break;
                 case '\x1b': /* ESC */
@@ -3249,7 +3252,10 @@ int FAST_FUNC read_line_input(line_input_t *st, const char *prompt, char *comman
 #undef read_line_input
 int FAST_FUNC read_line_input(const char* prompt, char* command, int maxsize)
 {
-        fputs_stdout(prompt);
+        /* https://pubs.opengroup.org/onlinepubs/9699919799/utilities/V3_chap02.html
+         * says that shells must write $PSn to stderr, not stdout.
+         */
+        fputs(prompt, stderr);
         fflush_all();
         if (!fgets(command, maxsize, stdin))
                 return -1;
diff --git a/libbb/popcnt.c b/libbb/popcnt.c
new file mode 100644
index 000000000..fe8cd240f
--- /dev/null
+++ b/libbb/popcnt.c
@@ -0,0 +1,46 @@
+/* vi: set sw=4 ts=4: */
+/*
+ * Utility routines.
+ *
+ * Copyright (C) 2024 Denys Vlasenko
+ *
+ * Licensed under GPLv2, see file LICENSE in this source tree.
+ */
+#include "libbb.h"
+
+unsigned FAST_FUNC bb_popcnt_32(uint32_t m)
+{
+        /* replace each 2 bit group with the count of set bits in it */
+        /* 00->00 01->01 10->01 11->10 */
+        m = m - ((m >> 1) & 0x55555555);
+        /* in each 4 bit group, add two 2-bit counts */
+        m = (m & 0x33333333) + ((m >> 2) & 0x33333333);
+        /* in each 8 bit group, add two 4-bit counts (in fact, 3-bit, 0nnn with n=0..4) */
+        m = (m + (m >> 4)) & 0x0f0f0f0f;
+#if 1 /* assume 32*32->32 multiply is fast */
+        m = m * 0x01010101; /* top byte = m + (m<<8) + (m<<16) + (m<<24) */
+        return m >> 24;
+#else
+        /*   0000aaaa0000bbbb0000cccc0000dddd */
+        /* +         0000aaaa0000bbbb0000cccc */
+        /* = 0000xxxx000_a+b_000xxxxx000_c+d_ (we don't care about x bits) */
+        m += m >> 8; /* in each 16-bit group, lowest 5 bits is the count */
+        /*   0000xxxx000_a+b_000xxxxx000_c+d_ */
+        /* +                 0000xxxx000_a+b_ */
+        /* = 0000xxxx000xxxxx00xxxxxx00a+b+cd */
+        m += m >> 16; /* in each 32-bit group, lowest 6 bits is the count */
+        return m & 0x3f; /* clear x bits */
+#endif
+}
+
+#if ULONG_MAX > 0xffffffff
+unsigned FAST_FUNC bb_popcnt_long(unsigned long m)
+{
+        BUILD_BUG_ON(sizeof(m) != 8);
+        /* 64-bit version of bb_popcnt_32 exists, but it uses 64-bit constants,
+         * which are awkward to generate on assembly level on most CPUs.
+         * For now, just add two 32-bit counts:
+         */
+        return bb_popcnt_32((uint32_t)m) + bb_popcnt_32((uint32_t)(m >> 32));
+}
+#endif