Re: [Gcl-devel] SGC on MacOS X

[Camm Maguire]

Hi Aurelien, and thanks for looking into this!

1) Does this happen on a subset of machines, or on all macosx?  If the
   latter, do you know when this first appeared?

2) The following code in sgc_start in sgbc.c:


  npp=page((&sgc_type_map[0]));
  if (npp<MAXPAGE && sgc_type_map[npp] != SGC_PERM_WRITABLE)
    perm_writable(&sgc_type_map[0],sizeof(sgc_type_map));


is supposed to mark this array permanently writable.  Can you discover
why this is not working?  (i.e. step through with a debugger in normal
gcl running (si::sgc-on t)?)

3) Would you like to put up a binary 2.6.2 build somewhere for
   inclusion on ftp.gnu.org?  Or does the likely proliferation of
   shared library versions in use make this of little value?

Take care,

Aurelien Chanudet <address@hidden> writes:

> Hi Camm & John,
> 
> I tried to investigate this infinite loop issue when building ACL2 with SGC 
> enabled. Here are my conclusions. Camm's
> comments appreciated, as always !
> 
> The infinite loop is occurring in the memprotect_handler routine which gets 
> called in order to handle stratified segfault
> (in my understanding). There's an infinite loop in this memory handler 
> because a segfault is occurring from within the
> handler itself :
> 
> Attaching to process 20950.
> Reading symbols for shared libraries . done
> Reading symbols for shared libraries ........ done
> 0x00013818 in memprotect_handler (sig=10, code=-1073753784, scp=0xbfffd188, 
> addr=0xbfffd148 "") at sgbc.c:1614
> 1614          sgc_type_map[p+j] = sgc_type_map[p+j] | SGC_TEMP_WRITABLE;
> (gdb) p/x (unsigned long) sgc_type_map
> $2 = 0x634890
> (gdb) call print_regions ()
> (gdb) p/x (unsigned long)(sgc_type_map+p+j)
> $4 = 0x63c0d8
> (gdb) p/x 0x639000+0x1b000
> $5 = 0x654000
> (gdb) p j
> $6 = 0
> 
> print_regions () is a function I wrote using Darwin's custom VM management 
> routines. It scans the whole address space,
> and displays memory attributes. The output of this routine is (the first 
> column of attributes denotes actual protection
> attributes, the second column denotes max protection attributes allowed).
> 
> <snip>
>    0x634000     0x1000 rwx  rwx  (no zone)
>    0x635000     0x1000 rwx  rwx  (no zone)
>    0x636000     0x1000 rwx  rwx  (no zone)
>    0x637000     0x1000 rwx  rwx  (no zone)
>    0x638000     0x1000 rwx  rwx  (no zone)
>    0x639000    0x1b000 r x  rwx  (no zone)
>    0x654000     0x1000 rwx  rwx  (no zone)
> <snip>
> 
> It looks as though sgc_type_map[p+j] falls in a non writable area 
> ([0x639000;0x654000[). As can be seen from the log
> above, sgc_type_map starts at 0x634890. My MAXPAGE has the default value 
> (128*1024).
> 
> Turing on DEBUG_MPROTECT in sgbc.c and trying to compile ACL2 yields the 
> following output :
> 
> <snip>
> mprotect(0x62c000,0x8000), sbrk(0)=0xeaae000
> prot[893,894,(5),writable]
> mprotect(0x634000,0x1000), sbrk(0)=0xeaae000
> prot[894,925,(0),not writable]
> mprotect(0x635000,0x1f000), sbrk(0)=0xeaae000
> prot[925,926,(5),writable]
> mprotect(0x654000,0x1000), sbrk(0)=0xeaae000
> prot[926,929,(0),not writable]
> <snip>
> [SGC on]
> fault:0x2b8404 [1] (0)  mprotect(0x2b8000,0x1000,0xeaae000)
> fault:0x12ac0b8 [4085] (0)  mprotect(0x12ac000,0x1000,0xeaae000)
> fault:0x12b2ec8 [4091] (0)  mprotect(0x12b2000,0x1000,0xeaae000)
> fault:0x713728 [1116] (0)  mprotect(0x713000,0x1000,0xeaae000)
> fault:0x2cec34 [23] (0)  mprotect(0x2ce000,0x1000,0xeaae000)
> fault:0x700fd4 [1097] (0)  mprotect(0x700000,0x1000,0xeaae000)
> fault:0x7aff000 [30792] (0)  mprotect(0x7aff000,0x1000,0xeaae000)
> 
> When running a separate saved_gcl and doing (si::sgc-on t), here is what I 
> get :
> 
>  >(si::sgc-on t)
> <snip>
> mprotect(0x62c000,0x8000), sbrk(0)=0xa302000
> prot[893,894,(5),writable]
> mprotect(0x634000,0x1000), sbrk(0)=0xa302000
> prot[894,925,(0),not writable]
> mprotect(0x635000,0x1f000), sbrk(0)=0xa302000
> prot[925,926,(5),writable]
> mprotect(0x654000,0x1000), sbrk(0)=0xa302000
> prot[926,929,(0),not writable]
> mprotect(0x655000,0x3000), sbrk(0)=0xa302000
> prot[929,931,(5),writable]
> mprotect(0x658000,0x2000), sbrk(0)=0xa302000
> prot[931,962,(0),not writable]
> <snip>
> fault:0x7182f8 [1121] (0)  mprotect(0x718000,0x1000,0xa302000)
> fault:0x70b3ac [1108] (0)  mprotect(0x70b000,0x1000,0xa302000)
> fault:0x7052d4 [1102] (0)  mprotect(0x705000,0x1000,0xa302000)
> fault:0x709c08 [1106] (0)  mprotect(0x709000,0x1000,0xa302000)
> fault:0x2c7a98 [16] (0)  mprotect(0x2c7000,0x1000,0xa302000)
> T
> fault:0x2ce2f8 [23] (0)  mprotect(0x2ce000,0x1000,0xa302000)
> fault:0x700cd0 [1097] (0)  mprotect(0x700000,0x1000,0xa302000)
> fault:0x2ccc88 [21] (0)  mprotect(0x2cc000,0x1000,0xa302000)
> fault:0x12dea5c [4135] (0)  mprotect(0x12de000,0x1000,0xa302000)
> fault:0x13806a0 [4297] (0)  mprotect(0x1380000,0x1000,0xa302000)
> fault:0x2b8c3c [1] (0)  mprotect(0x2b8000,0x1000,0xa302000)
> fault:0x131bf0c [4196] (0)  mprotect(0x131b000,0x1000,0xa302000)
> fault:0x13ec008 [4405] (0)  mprotect(0x13ec000,0x1000,0xa302000)
> 
> Aurelien
> 
> 
> 
> _______________________________________________
> Gcl-devel mailing list
> address@hidden
> http://lists.gnu.org/mailman/listinfo/gcl-devel
> 
> 
> 

-- 
Camm Maguire                                            address@hidden
==========================================================================
"The earth is but one country, and mankind its citizens."  --  Baha'u'llah