drt: check_drc makes OR crashes

[titan73]

Describe the bug

Command drt::check_drc -output drc.txt

Same crash in 2 cases:

1. after opening an .odb
2. after the complete flow

Expected Behavior

Should not crash and generate the report.

Environment

OpenROAD v2.0-23770-g72a6f31f7d

To Reproduce

Test cases uploaded to PII sftp server:

• check_drc.zip for case 1)
• min_area.zip for case 2)

Relevant log output

[INFO DRT-0167] List of default vias:
  Layer v1
    default via: M2_M1_VH
  Layer v2
    default via: M3_M2_HV
  Layer v3B
    default via: M4_M3_VH
  Layer padopen
    default via: AP_M4
[INFO DRT-0162] Library cell analysis.
[INFO DRT-0163] Instance analysis.
[INFO DRT-0164] Number of unique instances = 188.
[INFO DRT-0168] Init region query.
[INFO DRT-0024]   Complete poly.
[INFO DRT-0024]   Complete contact.
[INFO DRT-0024]   Complete metal1.
[INFO DRT-0024]   Complete v1.
[INFO DRT-0024]   Complete metal2.
[INFO DRT-0024]   Complete v2.
[INFO DRT-0024]   Complete metal3.
[INFO DRT-0024]   Complete v3B.
[INFO DRT-0024]   Complete metal4B.
[INFO DRT-0024]   Complete padopen.
[INFO DRT-0024]   Complete alucap.
[INFO DRT-0033] poly shape region query size = 0.
[INFO DRT-0033] contact shape region query size = 0.
[INFO DRT-0033] metal1 shape region query size = 824531.
[INFO DRT-0033] v1 shape region query size = 162000.
[INFO DRT-0033] metal2 shape region query size = 98355.
[INFO DRT-0033] v2 shape region query size = 14150.
[INFO DRT-0033] metal3 shape region query size = 7165.
[INFO DRT-0033] v3B shape region query size = 9905.
[INFO DRT-0033] metal4B shape region query size = 4320.
[INFO DRT-0033] padopen shape region query size = 184.
[INFO DRT-0033] alucap shape region query size = 53.
[INFO DRT-0156] guideIn read 100000 guides.
[INFO DRT-0156] guideIn read 200000 guides.
[INFO DRT-0156] guideIn read 300000 guides.
[INFO DRT-0156] guideIn read 400000 guides.

[INFO DRT-0157] Number of guides:     441543

[INFO DRT-0176] GCELLGRID X 0 DO 208 STEP 6150 ;
[INFO DRT-0177] GCELLGRID Y 0 DO 132 STEP 6150 ;
Signal 11 received
Stack trace:
 0# 0x00000000011D1C48 in /prj/gpca_dig/users/costa/openroad/bin/openroad
 1# 0x000014D5BC0835B0 in /lib64/libc.so.6
 2# drt::FlexGCWorker::Impl::checkMetalShape_minArea(drt::gcPin*, bool) in /prj/gpca_dig/users/costa/openroad/bin/openroad
 3# drt::FlexGCWorker::Impl::checkMetalShape_main(drt::gcPin*, bool) in /prj/gpca_dig/users/costa/openroad/bin/openroad
 4# drt::FlexGCWorker::Impl::checkMetalShape(bool) in /prj/gpca_dig/users/costa/openroad/bin/openroad
 5# drt::FlexGCWorker::Impl::main() in /prj/gpca_dig/users/costa/openroad/bin/openroad
 6# 0x0000000002EB29F8 in /prj/gpca_dig/users/costa/openroad/bin/openroad
 7# GOMP_parallel in /sw/freetools/gcc/9.2.0/rh80_64/lib64/libgomp.so.1
 8# drt::TritonRoute::getDRCMarkers(std::__cxx11::list<std::unique_ptr<drt::frMarker, std::default_delete<drt::frMarker> >, std::allocator<std::unique_ptr<drt::frMarker, std::default_delete<drt::frMarker> > > >&, odb::Rect const&) in /prj/gpca_dig/users/costa/openroad/bin/openroad
 9# drt::TritonRoute::checkDRC(char const*, int, int, int, int, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, int) in /prj/gpca_dig/users/costa/openroad/bin/openroad
10# check_drc_cmd(char const*, int, int, int, int, char const*) in /prj/gpca_dig/users/costa/openroad/bin/openroad
11# 0x0000000001A3593C in /prj/gpca_dig/users/costa/openroad/bin/openroad
12# 0x000014D5C28C9008 in /lib64/libtcl8.6.so
13# TclNRRunCallbacks in /lib64/libtcl8.6.so
14# 0x000014D5C28C7B2F in /lib64/libtcl8.6.so
15# Tcl_EvalEx in /lib64/libtcl8.6.so
16# Tcl_Eval in /lib64/libtcl8.6.so
17# sta::sourceTclFile(char const*, bool, bool, Tcl_Interp*) in /prj/gpca_dig/users/costa/openroad/bin/openroad
18# 0x00000000011D147B in /prj/gpca_dig/users/costa/openroad/bin/openroad
19# Tcl_MainEx in /lib64/libtcl8.6.so
20# main in /prj/gpca_dig/users/costa/openroad/bin/openroad
21# __libc_start_main in /lib64/libc.so.6
22# _start in /prj/gpca_dig/users/costa/openroad/bin/openroad
/home/costa/tools/openroad/openroad: line 28: 3426978 Segmentation fault      $root/openroad/bin/openroad $*

Screenshots

No response

Additional Context

No response

[maliberty]

You should use check_drc not the swig wrapper drt::check_drc directly.

[titan73]

OR says check_drc is invalid:

OpenROAD v2.0-23776-gef472411ba 
Features included (+) or not (-): +GPU +GUI +Python
This program is licensed under the BSD-3 license. See the LICENSE file for details.
Components of this program may be licensed under more restrictive licenses which must be honored.
openroad> check_drc
invalid command name "check_drc"
while evaluating check_drc
openroad> 

That's why I tried using drt::check_drc after looking at the code.

[titan73]

The only use in test is here:
drt/test/drc_test.tcl:drt::check_drc -output_file $drc_file
check_drc is inside a name space as well as few other functions:

namespace eval drt {
proc step_dr { args } {
  # args match FlexDR::SearchRepairArgs
  if { [llength $args] != 9 } {
    utl::error DRT 308 "step_dr requires nine positional arguments."
  }

  drt::detailed_route_step_drt {*}$args
}

sta::define_cmd_args "check_drc" {
    [-box box]
    [-output_file filename]
    [-marker_name name]
} ;# checker off
proc check_drc { args } {
  sta::parse_key_args "check_drc" args \
    keys { -box -output_file -marker_name } \
    flags {} ;# checker off
  sta::check_argc_eq0 "check_drc" $args
  set box { 0 0 0 0 }
  if { [info exists keys(-box)] } {
    set box $keys(-box)
    if { [llength $box] != 4 } {
      utl::error DRT 612 "-box is a list of 4 coordinates."
    }
  }
  set marker_name "DRC"
  if { [info exists keys(-marker_name)] } {
    set marker_name $keys(-marker_name)
  }
  lassign $box x1 y1 x2 y2
  if { [info exists keys(-output_file)] } {
    set output_file $keys(-output_file)
  } else {
    utl::error DRT 613 "-output_file is required for check_drc command"
  }
  drt::check_drc_cmd $output_file $x1 $y1 $x2 $y2 $marker_name
}

proc fix_max_spacing { args } {
  sta::check_argc_eq0 "fix_max_spacing" $args
  drt::fix_max_spacing_cmd
}
}

[titan73]

It's been moved:

commit 58cdfd37152954aae30298b7b9a04613b0bdf5ff
Author: osamahammad21 <osama21@aucegypt.edu>
Date:   Sun Jul 24 20:01:24 2022 +0200

    drt: move check_drc under drt
    
    Signed-off-by: osamahammad21 <osama21@aucegypt.edu>

I'm wondering why as the reason of the move is not mentioned.

[maliberty]

I don't think we intend it to be for general users. It is mostly for testing the router's drc ability. It is not a general purpose drc engine nor intended to be one.

[titan73]

Ah ok. It's advertized (the wrong way) though:

openroad> help check_*
check_antennas [-verbose] [-report_violating_nets] [-report_file report_file]
   [-net net] 

check_axioms 
check_drc [-box box] [-output_file filename] [-marker_name name] 

check_placement [-verbose] [-disallow_one_site_gaps]
   [-report_file_name file_name] 
check_power_grid -net power_net [-error_file error_file] [-floorplanning]
   [-dont_require_terminals] 

check_setup [-verbose] [-no_input_delay] [-no_output_delay] [-multiple_clock]
   [-no_clock] [-unconstrained_endpoints] [-loops] [-generated_clocks]
   [> filename] [>> filename]  

and:

[maliberty]

That is unintentional, fixed in #8130

[ahmed532]

I have confirmed via bisect that commit b19cfc3 ("drt: correct use of drcBox_ inside FlexGCWorker") fixes the crash.
Parent Commit (b19cfc3^): Crashed with Signal 11 immediately after GCELLGRID logs.
Fix Commit (b19cfc3): Did NOT crash and ran successfully past the point of failure.

Why it crashed: The function checkMetalShape_minArea (src/drt/src/gc/FlexGC_main.cpp) was incorrectly using drWorker_->getDrcBox() to decide which polygons to process. This allowed some invalid or irrelevant polygons (which likely had null properties causing the segfault) to pass through the filter.

How it was fixed: The code was changed to use drcBox_ (a local member variable) instead.

-  if (!drWorker_->getDrcBox().contains(bbox2)) {
+  if (!drcBox_.contains(bbox2)) {

This stricter/correct boundary check ensures that only valid polygons within the specific worker's region are processed, effectively filtering out the ones that were causing the crash.

@osamahammad21 (the fix author) Please confirm to close this issue.