1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
|
This stuff worked with 2.1R2. It probably can be made to work with
newer versions, but I'm not able to do so or test it since I don't
have access to such a system.
Notes on compiling xlispstat on the IBM RT (AOS 4.3).
Bill Dunlap
Dept of Statistics, GN-22
University of Washington
Seattle, WA 98195
bill@stat.washington.edu
Fri Jan 19 14:27:18 PST 1990
AOS a.out format is almost close enough to standard BSD to use the same
foreign.h files as sun3 and vax. However, there are problems in
dynamic linking when xlisp scans the symbols table output by 'ld -A' to
find new pointers to functions.
Note that each function has two symbols: a text symbol, _.%s, which is
the address of the code, and a data symbol _%s, which points to the
address of the code. A C pointer to function must have the value of
this data symbol. (The names match the BSD convention, but the types
don't.)
To find all symbols representing pointers to functions, start by
scanning for all symbols of type "TEXT". If the symbol has a name of
the form "_.%s", then look for the symbol called "_%s". If it is not
of type "DATA", something is wrong. If it is of type "DATA", enter it
in the function pointer table along with the name %s.
The compilers bf77, hf77, and pcc place the text symbol before the data
symbol in the symbol table. hc places the data symbol first. There
are often several other symbols between the matched ones. 'ld -A' does
not change the order of the symbol tables. Thus a linear search
starting at the text symbol is not efficient; one should use a hashing
scheme. Perhaps, scan the symbol table once, hashing all data symbols
and values in a temporary hash table, then scan the symbol table again
looking for text symbols. Whenever a text symbol is found, look for
its mated data symbol in the temporary hash table. If there is one,
enter that data symbol in the (permanent) function pointer hash table.
I took the lazy way out and just put all data symbols into the function
pointer hash table, and let hell break loose if someone tried to call a
bogus function. xlisp alone has 6699 global data symbols and 6483
global text symbols, leaving about 216 "real" data symbols, so this should
be fairly safe. The change was to redefine SYMBOL_IS_GLOBAL_FUNCTION
from
#define SYM_IS_GLOBAL_FUNCTION(ldptr,symbol) \
(((symbol).n_type & N_TYPE) == N_TEXT && ((symbol).n_type & N_EXT))
to
#define SYM_IS_GLOBAL_FUNCTION(ldptr,symbol) \
(((symbol).n_type & N_TYPE) == N_DATA && ((symbol).n_type & N_EXT))
I made the redefinition in foreign.h and changed xsdyload.c so that
the definition of SYM_IS_GLOBAL_FUNCTION there will not override
the one in foreign.h.
I compiled all of xlispstat with the C compiler called hc (cc is
usually linked to hc), version 2.1n. It does not deal properly with
((double *)a.addr)[i] = makedouble(arg) ;
so I changed a line of that form to
{ double *dbl = &((double *)a.addr)[i] ; *dbl = makedouble(arg) ; }
(I #ifdef'ed this part of the code so it is used only on an IBM RT with
the HighC compiler.) This problem does not occur with the pcc
compiler, but that generally produces slower code than hc.
Both of these changes were in xsdynload.c. A patch file is in this
directory (machines/ibmrt_bsd/xsdynload.patch).
|
