scripts/symbolize.py: try to resolve abort address to symbol/section[+offset] #1767

jforissier · 2017-08-24T16:56:54Z

Use nm and objdump to find the symbol and ELF section that match the
address reported in the the abort line. This can help debug writes to
read-only data or unaligned accesses to global data, for example.

If the address can be resolved to a symbol plus some offset and/or a
section plus some offset, the abort line is printed again just before
the call stack with the symbolic information added. Not that the
translation cannot be done immediately when the abort line is seen
because at this point we don't know the architecture, and we don't have
the load address of the TA.

Here is an example (the line added by this patch is marked with >>):

   User TA data-abort at address 0x1314d0 (write permission fault)
    fsr 0x0000080f  ttbr0 0x0e07a06a  ttbr1 0x0e07406a  cidr 0x1
    cpu #0          cpsr 0x60000030
    r0 0x00000001      r4 0x00102780    r8 0x00000000   r12 0xb736e358
    r1 0x00102724      r5 0x00121e4f    r9 0x00000000    sp 0x001026e0
    r2 0x00000001      r6 0x001026dc   r10 0x00000000    lr 0x00105cf1
    r3 0x001314d0      r7 0x001026e0   r11 0x00000000    pc 0x00105790
   Status of TA 5b9e0e40-2636-11e1-ad9e-0002a5d5c51b (0xe073b70) (active)
    arch: arm  load address: 0x103000  ctx-idr: 1
    stack: 0x100000 10240
    region 0: va 0x100000 pa 0xe21e000 size 0x3000
    region 1: va 0x103000 pa 0xe100000 size 0x2e000
    region 2: va 0x131000 pa 0xe12e000 size 0xa000
    region 3: va 0x13b000 pa 0xe138000 size 0xe6000
    region 4: va 0 pa 0 size 0
    region 5: va 0 pa 0 size 0
    region 6: va 0 pa 0 size 0
    region 7: va 0 pa 0 size 0
>> User TA data-abort at address 0x1314d0 const_val+4 .rodata+4452 (write permission fault)
   Call stack:
    0x00105790 ta_entry_bad_mem_access at optee_test/ta/os_test/os_test.c:917
    0x00105cf1 TA_InvokeCommandEntryPoint at optee_test/ta/os_test/ta_entry.c:101
    0x00121e33 entry_invoke_command at optee_os/lib/libutee/arch/arm/user_ta_entry.c:207
    0x00121e8f __utee_entry at optee_os/lib/libutee/arch/arm/user_ta_entry.c:235

The test TA does the following:

  const int const_val[3] = { 1, };
  /* ... */
  ((int *)const_val)[1] = 2;

Suggested-by: Zeng Tao <[email protected]>
Signed-off-by: Jerome Forissier <[email protected]>

jforissier · 2017-08-25T12:39:04Z

Update: also print ELF section name [+offset] and added Suggested-by: tag.

jbech-linaro · 2017-08-29T12:09:27Z

scripts/symbolize.py

-            return '???'
-        reladdr = '0x{:x}'.format(int(addr, 0) - int(offs, 0))
+            return ''
+        return '0x{:x}'.format(int(addr, 0) - int(offs, 0))


Not super important and not really part of this PR, but I guess you always expect to get the numbers in hex? In that case the int(xyz, 0) can be changed to int(xyz, 16) which makes it a bit more clear what kind of string we are dealing with.

jbech-linaro · 2017-08-29T12:09:31Z

scripts/symbolize.py

+                idx, name, size, vma, lma, offs, algn = line.split()
+            except:
+                continue;
+	    ivma = int(vma, 16)


GitHub fooling me? Or is here an indent that isn't correct?

RIght, there's a tab. Will fix.

jbech-linaro · 2017-08-29T12:09:34Z

scripts/symbolize.py

+                continue;
+	    ivma = int(vma, 16)
+            isize = int(size, 16)
+            if ivma == iaddr:


Isn't this check superfluous? Wouldn't you get the same in the next if check but with offset 0?

Sure but I'd rather have symbol than symbol+0 in the tool output.

igoropaniuk · 2017-08-29T13:46:09Z

scripts/symbolize.py

+                break
+            prevsize = size
+            prevname = name
+        if ret is None:


duplication of the same checkif ret is None. I suggest to adda function like

def xstr(s): if s is None: return '' return str(s)

and then wrap , for example on lines 220, 221 with this function call:

if sym: ret += ' ' + xstr(sym)

I'll fix as above (initialize ret = '').

igoropaniuk · 2017-08-29T13:48:52Z

scripts/symbolize.py

@@ -128,13 +137,101 @@ def resolve(self, addr):
            ret = '!!!'
        return ret

+    def symbol_plus_offset(self, addr):
+        ret = None


why not to define ret = ''?

No particular reason. Will fix.

jforissier · 2017-08-30T11:40:37Z

Comments addressed. Sorry for the non-fast-forward push, I have inserted the int(str, 16) fix before the series since the issue applies to older code.

jbech-linaro · 2017-08-30T11:54:19Z

Reviewed-by: Joakim Bech <[email protected]>

igoropaniuk · 2017-08-31T08:03:34Z

Reviewed-by: Igor Opaniuk <[email protected]>

When converting a hex string formatted as '0x<hex>', a value of 0 may be given for base and Python will automatically assume a base-16 literal. However, since we're always dealing with hex strings (with or without a 0x prefix), it is more convenient to specify base-16 everywhere. Signed-off-by: Jerome Forissier <[email protected]> Reviewed-by: Joakim Bech <[email protected]> Reviewed-by: Igor Opaniuk <[email protected]>

…+offset] Use nm and objdump to find the symbol and ELF section that match the address reported in the the abort line. This can help debug writes to read-only data or unaligned accesses to global data, for example. If the address can be resolved to a symbol plus some offset and/or a section plus some offset, the abort line is printed again just before the call stack with the symbolic information added. Not that the translation cannot be done immediately when the abort line is seen because at this point we don't know the architecture, and we don't have the load address of the TA. Here is an example (the line added by this patch is marked with >>): User TA data-abort at address 0x1314d0 (write permission fault) fsr 0x0000080f ttbr0 0x0e07a06a ttbr1 0x0e07406a cidr 0x1 cpu #0 cpsr 0x60000030 r0 0x00000001 r4 0x00102780 r8 0x00000000 r12 0xb736e358 r1 0x00102724 r5 0x00121e4f r9 0x00000000 sp 0x001026e0 r2 0x00000001 r6 0x001026dc r10 0x00000000 lr 0x00105cf1 r3 0x001314d0 r7 0x001026e0 r11 0x00000000 pc 0x00105790 Status of TA 5b9e0e40-2636-11e1-ad9e-0002a5d5c51b (0xe073b70) (active) arch: arm load address: 0x103000 ctx-idr: 1 stack: 0x100000 10240 region 0: va 0x100000 pa 0xe21e000 size 0x3000 region 1: va 0x103000 pa 0xe100000 size 0x2e000 region 2: va 0x131000 pa 0xe12e000 size 0xa000 region 3: va 0x13b000 pa 0xe138000 size 0xe6000 region 4: va 0 pa 0 size 0 region 5: va 0 pa 0 size 0 region 6: va 0 pa 0 size 0 region 7: va 0 pa 0 size 0 >> User TA data-abort at address 0x1314d0 const_val+4 .rodata+4452 (write permission fault) Call stack: 0x00105790 ta_entry_bad_mem_access at optee_test/ta/os_test/os_test.c:917 0x00105cf1 TA_InvokeCommandEntryPoint at optee_test/ta/os_test/ta_entry.c:101 0x00121e33 entry_invoke_command at optee_os/lib/libutee/arch/arm/user_ta_entry.c:207 0x00121e8f __utee_entry at optee_os/lib/libutee/arch/arm/user_ta_entry.c:235 The test TA does the following: const int const_val[3] = { 1, }; /* ... */ ((int *)const_val)[1] = 2; Suggested-by: Zeng Tao <[email protected]> Signed-off-by: Jerome Forissier <[email protected]> Reviewed-by: Joakim Bech <[email protected]> Reviewed-by: Igor Opaniuk <[email protected]>

jbech-linaro · 2017-08-31T08:47:27Z

Great Jerome! Somewhere in the future it would be nice to have it written down somewhere telling people that this actually exists and then some example of how to use it. It could be a blog post and/or adding it to our debug.md file.

jforissier · 2017-08-31T08:53:10Z

@jbech-linaro good idea(s). I'll definitely propose something to add to debug.md.

jforissier force-pushed the symbolize-abort-addr branch 3 times, most recently from 5778064 to 38e47d7 Compare August 25, 2017 12:33

jforissier changed the title ~~scripts/symbolize.py: try to resolve abort address to symbol+offset~~ scripts/symbolize.py: try to resolve abort address to symbol/section[+offset] Aug 25, 2017

jbech-linaro reviewed Aug 29, 2017

View reviewed changes

igoropaniuk reviewed Aug 29, 2017

View reviewed changes

jforissier force-pushed the symbolize-abort-addr branch from 38e47d7 to cc3f47c Compare August 30, 2017 11:39

jforissier added 2 commits August 31, 2017 10:06

jforissier force-pushed the symbolize-abort-addr branch from cc3f47c to d9f57d7 Compare August 31, 2017 08:08

jforissier merged commit 142c5cc into OP-TEE:master Aug 31, 2017

jforissier deleted the symbolize-abort-addr branch August 31, 2017 08:10

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

scripts/symbolize.py: try to resolve abort address to symbol/section[+offset] #1767

scripts/symbolize.py: try to resolve abort address to symbol/section[+offset] #1767

jforissier commented Aug 24, 2017 •

edited

Loading

jforissier commented Aug 25, 2017

jbech-linaro Aug 29, 2017

jbech-linaro Aug 29, 2017

jforissier Aug 30, 2017

jbech-linaro Aug 29, 2017

jforissier Aug 30, 2017

igoropaniuk Aug 29, 2017

jforissier Aug 30, 2017

igoropaniuk Aug 29, 2017

jforissier Aug 30, 2017

jforissier commented Aug 30, 2017

jbech-linaro commented Aug 30, 2017

igoropaniuk commented Aug 31, 2017

jbech-linaro commented Aug 31, 2017

jforissier commented Aug 31, 2017

scripts/symbolize.py: try to resolve abort address to symbol/section[+offset] #1767

scripts/symbolize.py: try to resolve abort address to symbol/section[+offset] #1767

Conversation

jforissier commented Aug 24, 2017 • edited Loading

jforissier commented Aug 25, 2017

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jforissier commented Aug 30, 2017

jbech-linaro commented Aug 30, 2017

igoropaniuk commented Aug 31, 2017

jbech-linaro commented Aug 31, 2017

jforissier commented Aug 31, 2017

jforissier commented Aug 24, 2017 •

edited

Loading