[Bug tdep/29738] New: Arm v8 M-profile secure extension dwarf2 unwinder performance

["tomas.vanek at fbl dot cz" <sourceware-bugzilla@sourceware.org>]

https://sourceware.org/bugzilla/show_bug.cgi?id=29738

            Bug ID: 29738
           Summary: Arm v8 M-profile secure extension dwarf2 unwinder
                    performance
           Product: gdb
           Version: HEAD
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: tdep
          Assignee: unassigned at sourceware dot org
          Reporter: tomas.vanek at fbl dot cz
  Target Milestone: ---

Commit a6e4a48c02acf29d6bec2ff63fc909b57cf4bc78
introduced synchronization sp with other xsp_x registers.
Unfortunately the code reads all four xsp_x registers by
get_frame_register_unsigned(). dwarf2 unwinder just passes this
request to the next inner frame. If the next inner frame is also
serviced by dwarf2, each of 4 xsp_x requests generates another 4 requests.

So e.g. 'backtrace' command on the chain of dwarf2 frames (normal debugging)
needs
  4 ^ number_of_frame
calls of arm_dwarf2_prev_register() just for stack unwinding!!!
The number of calls gets to very insane high values very soon,
8th and 9th frame unwinding is noticeable slow and 10 and more frames
looks as if gdb hangs.

Empirically tested with a counter added to the beginning of
arm_dwarf2_prev_register():

'bt' frames   total arm_dwarf2_prev_register() calls
        2       42
        3       181
        4       828
        5       3909
        6       18652
        7       89271
        8       427608
        9       2048659

-- 
You are receiving this mail because:
You are on the CC list for the bug.