Segmentation fault

[wulffern]

I experience a segmentation fault when I try to netlist a layout.

I run

set VDD AVDD
set GND AVSS
set SUB BULKN
load ../design/SUN_PLL_SKY130NM/SUN_PLL.mag
#writeall force
extract all
ext2spice lvs
ext2spice -o lvs/SUN_PLL.spi
quit

On my design, and magic segfaults. I think it happens because somehow magic is trying to check the size of a empty cell, although I'm not sure, but the segfault happens in bpBins.c at

int ew = GEO_WIDTH(&e->e_rect);
int eh = GEO_HEIGHT(&e->e_rect);

I've fixed it with

https://github.com/wulffern/magic/commit/68b89ab774014c8e58459adebd0c4da8c1fb8bd7

[RTimothyEdwards]

I would want to be very careful with changes to the code around the bplane code. I will see if I can reproduce the problem from your repo.

[RTimothyEdwards]

@wulffern : I can't duplicate the issue because the design repository is incomplete; there are multiple cell layouts missing.

[wulffern]

@RTimothyEdwards yup, the PLL consists of multiple libraries. The full set of libraries for the PLL are

git clone https://github.com/wulffern/sun_pll_sky130nm
git clone https://github.com/wulffern/sun_tr_sky130nm
git clone https://github.com/wulffern/sun_trb_sky130nm
git clone https://github.com/wulffern/sun_sar9b_sky130nm
git clone https://github.com/wulffern/tech_sky130B
git clone https://github.com/wulffern/cpdk

cd sun_pll_sky130nm/work/
magic ../design/SUN_PLL_SKY130NM/SUN_PLL.mag &

When I then run the TCL, as shown below

wulff@work$  cat SUN_PLL_spi.tcl 
set VDD AVDD
set GND AVSS
set SUB BULKN
load ../design/SUN_PLL_SKY130NM/SUN_PLL.mag
extract all
ext2spice lvs
ext2spice -o SUN_PLL.spi
quit

wulff@work$ magic -noconsole -dnull SUN_PLL_spi.tcl
...
Extracting cut_M1M4_2x1 into ../design/SUN_PLL_SKY130NM/cut_M1M4_2x1.ext:
Extracting cut_M1M4_1x2 into ../design/SUN_PLL_SKY130NM/cut_M1M4_1x2.ext:
Extracting SUN_PLL_LPF into ../design/SUN_PLL_SKY130NM/SUN_PLL_LPF.ext:
Segmentation fault: 11

And via LLDB

Extracting SUN_PLL_LPF into ../design/SUN_PLL_SKY130NM/SUN_PLL_LPF.ext:
Process 17073 stopped
* thread #1, queue = 'com.apple.main-thread', stop reason = EXC_BAD_ACCESS (code=EXC_I386_GPFLT)
    frame #0: 0x00000001011bccd6 tclmagic.dylib`bpBinArraySizeIt(bbox=error: summary string parsing error, elements=0x0000600000349dc1, dxp=0x00007ff7bfefdca4, dyp=0x00007ff7bfefdca0, maxDXp=0x00007ff7bfefdc9c, maxDYp=0x00007ff7bfefdc98, numBinsp=0x00007ff7bfefdc94, countp=0x00007ff7bfefdc90) at bpBins.c:297:16
   294        //continue;
   295        //}
   296  
-> 297        int ew = GEO_WIDTH(&e->e_rect);
   298        int eh = GEO_HEIGHT(&e->e_rect);
   299  
   300        maxEX = MAX(maxEX,ew);
Target 0: (magicexec) stopped.
(lldb) 

It could of course be something with my magic files

[RTimothyEdwards]

Thank you for the detailed setup for duplicating the problem, which I can do. Interestingly, I can extract from interactive magic on the command line, but running in batch with the above instructions causes an immediate segfault. I will investigate.

[wulffern]

Thanks for looking into it. Found one more piece of info.

If I open the cell with magic ../design/SUN_PLL_SKY130NM/SUN_PLL.mag & and immediately do extract all, then it crashes

However, if I open, draw a box around the cells, expand all cells with hotkey x, wait for the DRC to finish, then extract all, it works fine.

[RTimothyEdwards]

And the even weirder thing that I found was that if I add the equivalent commands that make it work into the script, that is, select top cell and expand after the load command, then it still segfaults when running in batch.

[RTimothyEdwards]

And another weird thing is that I can run the script directly on SUN_PLL_LPF.mag and it works fine, even though that was the cell it was working on when it crashed. I may have to put this one on the back burner for a bit. I don't think I can solve it without running a debugger on the working and non-working methods and figuring out where they diverge. Your solution is a workaround but the real source of the problem is somewhere else.

[wulffern]

Found another workaround.

If I replace extract all with just extract, then first time it crashes, but second, and later times, it runs fine.

If I delete all *.ext files, and run, then it crashes the first time.

[RTimothyEdwards]

"extract" as opposed to "extract all" is an incremental extraction, so after the first run, it will not attempt to extract any cell for which a .ext exists and is more recent than the layout.

[RTimothyEdwards]

[Actually, I am finding that the SUN_PLL cell does not extract correctly even when magic doesn't crash (i.e., when subcells are expanded before extracting). The output is incomplete, and a warning is issued during "ext2spice".] Okay, never mind, that was the file produced when it crashed.

[RTimothyEdwards]

@wulffern : Finally had time to look into this, and I ended up patching the code to circumvent the issue in much the same way you did, which is pretty much what I suggested originally except that I wanted to make sure I understood the code surrounding the issue. Thank you for providing the example!