RE: [PATCH v6 3/7] start/stop btrace with coresight etm and parse etm buffer. nat independant

["Metzger, Markus T" <markus.t.metzger@intel.com>]

Hello Zied,


+

+static void

+ftrace_remove_last_insn (struct btrace_thread_info *btinfo)

+{

+  /* If we didn't have a function, we return.  */

+  if (btinfo->functions.empty ())

+    return;



Should this be an error?



+

+  struct btrace_function *bfun = &btinfo->functions.back ();

+  /* If we had a gap before, we return.  */

+  if (bfun->errcode != 0)

+    return;



In which case can we have a gap?
[Zied] this is to make sure that following calls are safe to be executed.




+

+  if (!bfun->insn.empty ())

+    bfun->insn.pop_back ();

+  else

+    {

+      /* A valid function must have at least one instruction.  */

+      internal_error (__FILE__, __LINE__,

+                 _("Attempt to remove last instruction"

+                   "from an empty function"));



We just removed an instruction in the then statement, which

could well result in BFUN->INSN to become empty().



The statement about valid functions above may be a bit too

generic.  Also, ftrace_new_* () may create empty function

segments.



Isn't it rather that in our use-case, the caller knows that the

function segment cannot be empty?
[Zied] yes, we are sure that we have at least the last executed instruction that raised the breakpoint execption.

Would it make more sense to simply do ‘bfun->insn.pop_back ()’ inside the caller?

Apparently, the caller knows that there must be an instruction so it could safely remove it.  If we turn this into a separate function, we need to handle the general case.



+    case ocsd_isa_custom:

+      return BTRACE_INSN_FLAG_ISA_CUSTOM;

+

+    case ocsd_isa_unknown:

+      return BTRACE_INSN_FLAG_ISA_UNKNOWN;

+

+    default:

+      internal_error (__FILE__, __LINE__,

+                 _("Undefined elem->isa value returned by OpenCsd."));



This internal error kills GDB.  Should this be a normal error that just

stops trace processing for this unknown ISA?
[Zied] since I listed all possible values that may get returned by the decoder, the default case means that gdb and the decoder are doing something wrong and this should be handled as a fatal error.

The error is fatal for trace decode but not for GDB.  I think we should use error () instead of internal_error () for all those cases.



Even for normal trace off/on pairs, we'd want a gap if PC is moving.
[Zied] yes, but I do not have here a mean for checking if the PC has moved. The pc field is not valid in this kind of elements

In libipt, I compare the IPs of the stop and the resume trace packets.  If something like that doesn’t work in your case I won’t insist.

Regards,
Markus.
Intel Deutschland GmbH
Registered Address: Am Campeon 10, 85579 Neubiberg, Germany
Tel: +49 89 99 8853-0, www.intel.de <http://www.intel.de>
Managing Directors: Christin Eisenschmid, Sharon Heck, Tiffany Doon Silva  
Chairperson of the Supervisory Board: Nicole Lau
Registered Office: Munich
Commercial Register: Amtsgericht Muenchen HRB 186928