Date: Wed, 14 Aug 2024 11:25:24 +1200
From: Tao Liu <ltao@redhat.com>
Subject: [Crash-utility] [PATCH v2] Fix a segfault issue due to the
        incorrect irq_stack_size on ARM64
To: devel@lists.crash-utility.osci.io
Cc: "yeping . zheng" <yeping.zheng@nio.com>
Message-ID: <20240813232523.7237-1-ltao@redhat.com>
Content-Type: text/plain; charset="US-ASCII"; x-default=true

See the following stack trace:
(gdb) bt
 #0  0x00005635ac2b166b in arm64_unwind_frame (frame=0x7ffdaf35cb70,
     bt=0x7ffdaf35d430) at arm64.c:2821
 #1  arm64_back_trace_cmd (bt=0x7ffdaf35d430) at arm64.c:3306
 #2  0x00005635ac27b108 in back_trace (bt=bt@entry=0x7ffdaf35d430) at
     kernel.c:3239
 #3  0x00005635ac2880ae in cmd_bt () at kernel.c:2863
 #4  0x00005635ac1f16dc in exec_command () at main.c:893
 #5  0x00005635ac1f192a in main_loop () at main.c:840
 #6  0x00005635ac50df81 in captured_main (data=<optimized out>) at main.c:1284
 #7  gdb_main (args=<optimized out>) at main.c:1313
 #8  0x00005635ac50e000 in gdb_main_entry (argc=<optimized out>,
     argv=<optimized out>) at main.c:1338
 #9  0x00005635ac1ea2a5 in main (argc=5, argv=0x7ffdaf35dde8) at main.c:721

The issue may be encountered when thread_union symbol not found in vmlinux
due to compiling optimization.

This patch will try the following 2 methods to get the irq_stack_size
when thread_union symbol unavailable:

1. change the thread_shift when KASAN is enabled and with vmcoreinfo.
   In arm64/include/asm/memory.h:

   #if defined(CONFIG_KASAN_GENERIC) || defined(CONFIG_KASAN_SW_TAGS)
   ...
   #define IRQ_STACK_SIZE               THREAD_SIZE

   Since enabling the KASAN will affect the final value,
   this patch reset IRQ_STACK_SIZE according to the calculation process in
   kernel code.

2. Try getting the value from kernel code disassembly, to get
   THREAD_SHIFT directly from tbnz instruction.

   In arch/arm64/kernel/entry.S:
   .macro kernel_ventry, el:req, ht:req, regsize:req, label:req
   ...
         add     sp, sp, x0
         sub     x0, sp, x0
         tbnz    x0, #THREAD_SHIFT, 0f

   $ gdb vmlinux
   (gdb) disass vectors
   Dump of assembler code for function vectors:
      ...
      0xffff800080010804 <+4>:     add     sp, sp, x0
      0xffff800080010808 <+8>:     sub     x0, sp, x0
      0xffff80008001080c <+12>:    tbnz    w0, #16, 0xffff80008001081c <vectors+28>

Signed-off-by: yeping.zheng <yeping.zheng@nio.com>
Improved-by: Tao Liu <ltao@redhat.com>
---

v2 -> v1: Add stack_vm_area member check for task_struct, and kernel
          file reference.

Hi Yeping & lianbo,

Please have a test on this version of patch, I guess this will be the
last version if test is passed. Thanks in advance!

Thanks,
Tao Liu

---
 arm64.c | 63 +++++++++++++++++++++++++++++++++++++++++++++++++++++++--
 1 file changed, 61 insertions(+), 2 deletions(-)

diff --git a/arm64.c b/arm64.c
index 067c879..332e3f0 100644
--- a/arm64.c
+++ b/arm64.c
@@ -95,6 +95,7 @@ static int arm64_is_uvaddr(ulong, struct task_context *);
 static void arm64_calc_KERNELPACMASK(void);
 static void arm64_recalc_KERNELPACMASK(void);
 static int arm64_get_vmcoreinfo(unsigned long *vaddr, const char *label, int base);
+static ulong arm64_set_irq_stack_size(void);

 struct kernel_range {
        unsigned long modules_vaddr, modules_end;
@@ -2340,8 +2341,10 @@ arm64_irq_stack_init(void)
                if (MEMBER_EXISTS("thread_union", "stack")) {
                        if ((sz = MEMBER_SIZE("thread_union", "stack")) > 0)
                                ms->irq_stack_size = sz;
-               } else
-                       ms->irq_stack_size = ARM64_IRQ_STACK_SIZE;
+               } else {
+                       ulong res = arm64_set_irq_stack_size();
+                       ms->irq_stack_size = (res > 0) ? res : ARM64_IRQ_STACK_SIZE;
+               }

                machdep->flags |= IRQ_STACKS;

@@ -5056,6 +5059,62 @@ static void arm64_recalc_KERNELPACMASK(void){
        }
 }

+static ulong arm64_set_irq_stack_size(void)
+{
+       int min_thread_shift = 14;
+       ulong thread_shift = 0;
+       char buf1[BUFSIZE];
+       char *pos1, *pos2;
+       int errflag = 0;
+
+       if (kernel_symbol_exists("vmcoreinfo_data") &&
+           kernel_symbol_exists("vmcoreinfo_size")) {
+               /*
+                * Referring to arch/arm64/include/asm/memory.h
+                */
+               if (kernel_symbol_exists("kasan_enable_current")) {
+                       min_thread_shift += 1;
+               }
+               if (MEMBER_EXISTS("task_struct", "stack_vm_area") &&
+                   min_thread_shift < machdep->pageshift) {
+                       thread_shift = machdep->pageshift;
+               } else {
+                       thread_shift = min_thread_shift;
+               }
+       } else {
+               sprintf(buf1, "x/32i vectors");
+               open_tmpfile();
+               if (!gdb_pass_through(buf1, pc->tmpfile, GNU_RETURN_ON_ERROR)) {
+                       goto out;
+               }
+               rewind(pc->tmpfile);
+               while (fgets(buf1, BUFSIZE, pc->tmpfile)) {
+                       if ((pos1 = strstr(buf1, "tbnz"))) {
+                               if ((pos2 = strchr(pos1, '#'))) {
+                                       pos2 += 1;
+                                       for (pos1 = pos2;
+                                            *pos2 != '\0' && *pos2 != ',';
+                                            pos2++);
+                                       *pos2 = '\0';
+                                       thread_shift = stol(pos1,
+                                               RETURN_ON_ERROR|QUIET, &errflag);
+                                       if (errflag) {
+                                               thread_shift = 0;
+                                       }
+                                       break;
+                               }
+                       }
+               }
+out:
+               close_tmpfile();
+       }
+
+       if (thread_shift)
+               return ((1UL) << thread_shift);
+       else
+               return 0;
+}
+
 #endif  /* ARM64 */


--
2.40.1