From eb251af8672e033f29322253f136e09c76f2b0a3 Mon Sep 17 00:00:00 2001
From: Minep <zelong56@gmail.com>
Date: Mon, 27 Jun 2022 01:24:40 +0100
Subject: [PATCH] fix: race condition and partial state issue on injecting
 signal context into user stack refactor: reduce the system timer frequency to
 1024. refactor: make the kprintf re-entrance and reduce the buffer size.
 improve: better debugging experience with qemu. improve: more information
 when debugging with iret.

---
 lunaix-os/includes/lunaix/timer.h    |  2 +-
 lunaix-os/kernel/asm/x86/interrupt.S |  1 +
 lunaix-os/kernel/kprintf.c           | 14 ++++++--------
 lunaix-os/kernel/sched.c             |  3 ++-
 lunaix-os/kernel/signal.c            | 25 ++++++++++++++++++++++++-
 lunaix-os/makefile                   |  4 ++--
 6 files changed, 36 insertions(+), 13 deletions(-)

diff --git a/lunaix-os/includes/lunaix/timer.h b/lunaix-os/includes/lunaix/timer.h
index c02d7f6..b499f3f 100644
--- a/lunaix-os/includes/lunaix/timer.h
+++ b/lunaix-os/includes/lunaix/timer.h
@@ -4,7 +4,7 @@
 #include <lunaix/ds/llist.h>
 #include <stdint.h>
 
-#define SYS_TIMER_FREQUENCY_HZ 2048
+#define SYS_TIMER_FREQUENCY_HZ 1024
 
 #define TIMER_MODE_PERIODIC 0x1
 
diff --git a/lunaix-os/kernel/asm/x86/interrupt.S b/lunaix-os/kernel/asm/x86/interrupt.S
index 8a43aa7..dc5c4c3 100644
--- a/lunaix-os/kernel/asm/x86/interrupt.S
+++ b/lunaix-os/kernel/asm/x86/interrupt.S
@@ -118,6 +118,7 @@
         movl %eax, %esp
 
 #ifdef __ASM_INTR_DIAGNOSIS
+        movl %eax, (debug_resv + 8)
         movl 56(%esp), %eax
         movl %eax, (debug_resv + 4)
 #endif
diff --git a/lunaix-os/kernel/kprintf.c b/lunaix-os/kernel/kprintf.c
index 374fff2..75b9fc4 100644
--- a/lunaix-os/kernel/kprintf.c
+++ b/lunaix-os/kernel/kprintf.c
@@ -3,14 +3,13 @@
 #include <lunaix/syslog.h>
 #include <lunaix/tty/tty.h>
 
-#define MAX_KPRINTF_BUF_SIZE 1024
-#define MAX_XFMT_SIZE 1024
-
-static char buf[MAX_KPRINTF_BUF_SIZE];
+#define MAX_KPRINTF_BUF_SIZE 512
+#define MAX_XFMT_SIZE 512
 
 void
 __kprintf(const char* component, const char* fmt, va_list args)
 {
+    char buf[MAX_KPRINTF_BUF_SIZE];
     if (!fmt)
         return;
     char log_level = '0';
@@ -74,16 +73,15 @@ __kprintf(const char* component, const char* fmt, va_list args)
 void
 kprint_panic(const char* fmt, ...)
 {
+    char buf[MAX_KPRINTF_BUF_SIZE];
     va_list args;
     va_start(args, fmt);
 
     tty_set_theme(VGA_COLOR_WHITE, VGA_COLOR_RED);
-    tty_clear_line(10);
-    tty_clear_line(11);
-    tty_clear_line(12);
+    tty_clear_line(24);
 
     __sprintf_internal(buf, fmt, MAX_KPRINTF_BUF_SIZE, args);
-    tty_put_str_at(buf, 0, 11);
+    tty_put_str_at(buf, 0, 24);
 
     va_end(args);
 }
\ No newline at end of file
diff --git a/lunaix-os/kernel/sched.c b/lunaix-os/kernel/sched.c
index 5081ccf..8e089d6 100644
--- a/lunaix-os/kernel/sched.c
+++ b/lunaix-os/kernel/sched.c
@@ -59,7 +59,8 @@ run(struct proc_info* proc)
     apic_done_servicing();
 
     asm volatile("pushl %0\n"
-                 "jmp switch_to\n" ::"r"(proc)); // kernel/asm/x86/interrupt.S
+                 "jmp switch_to\n" ::"r"(proc)
+                 : "memory"); // kernel/asm/x86/interrupt.S
 }
 
 int
diff --git a/lunaix-os/kernel/signal.c b/lunaix-os/kernel/signal.c
index de83438..cb47e8d 100644
--- a/lunaix-os/kernel/signal.c
+++ b/lunaix-os/kernel/signal.c
@@ -22,6 +22,7 @@ void* default_handlers[_SIG_NUM] = {
     [_SIGINT] = default_sighandler_term,
 };
 
+volatile isr_param __temp_save;
 // Referenced in kernel/asm/x86/interrupt.S
 void*
 signal_dispatch()
@@ -58,7 +59,29 @@ signal_dispatch()
     struct proc_sig* sig_ctx =
       (struct proc_sig*)(ustack - sizeof(struct proc_sig));
 
-    sig_ctx->prev_context = __current->intr_ctx;
+    /*
+        这是一个相当恶心的坑。
+        问题是出在原本的sig_ctx->prev_context = __current->intr_ctx的上面
+        这个语句会被gcc在编译时，用更加高效的 rep movsl 来代替。
+
+        由于我们采用按需分页，所以在很多情况下，用户栈实际被分配的空间不允许我们进行完整的
+        注入，而需要走page fault handler进行动态分页。
+
+        竞态条件就出现在这里！
+
+        假若我们的__current->intr_ctx注入了一半，然后产生page-fault中断，
+        那么这就会导致我们的__current->intr_ctx被这个page-fault中断导致的
+        上下文信息覆盖。那么当page-fault handler成功分配了一个页，返回，
+        拷贝也就得以进行。遗憾的是，只不过这次拷贝的内容和前面的拷贝是没有任何的关系
+        （因为此时的intr_ctx已经不是之前的intr_ctx了！）
+        而这就会导致我们保存在信号上下文中的进程上下文信息不完整，从而在soft_iret时
+        触发#GP。
+
+        解决办法就是先吧intr_ctx拷贝到一个静态分配的区域里，然后再注入到用户栈。
+    */
+    __temp_save = __current->intr_ctx;
+    sig_ctx->prev_context = __temp_save;
+
     sig_ctx->sig_num = sig_selected;
     sig_ctx->signal_handler = __current->sig_handler[sig_selected];
 
diff --git a/lunaix-os/makefile b/lunaix-os/makefile
index 0666dc4..f1a6dce 100644
--- a/lunaix-os/makefile
+++ b/lunaix-os/makefile
@@ -51,13 +51,13 @@ clean:
 	@sleep 2
 
 run: $(BUILD_DIR)/$(OS_ISO)
-	@qemu-system-i386 -cdrom $(BUILD_DIR)/$(OS_ISO) -monitor telnet::$(QEMU_MON_PORT),server,nowait &
+	@qemu-system-i386 -cdrom $(BUILD_DIR)/$(OS_ISO) -d cpu_reset -monitor telnet::$(QEMU_MON_PORT),server,nowait &
 	@sleep 1
 	@telnet 127.0.0.1 $(QEMU_MON_PORT)
 
 debug-qemu: all-debug
 	@${TOOLCHAIN}/i686-elf-objcopy --only-keep-debug $(BIN_DIR)/$(OS_BIN) $(BUILD_DIR)/kernel.dbg
-	@qemu-system-i386 -m 1G -rtc base=utc -s -S -cdrom $(BUILD_DIR)/$(OS_ISO) -monitor telnet::$(QEMU_MON_PORT),server,nowait &
+	@qemu-system-i386 -m 1G -rtc base=utc -d cpu_reset -s -S -cdrom $(BUILD_DIR)/$(OS_ISO) -monitor telnet::$(QEMU_MON_PORT),server,nowait &
 	@sleep 1
 	@$(QEMU_MON_TERM) -- telnet 127.0.0.1 $(QEMU_MON_PORT)
 	@gdb -s $(BUILD_DIR)/kernel.dbg -ex "target remote localhost:1234"
-- 
2.27.0