Mailing-List: contact cygwin-help AT cygwin DOT com; run by ezmlm
List-Subscribe: <mailto:cygwin-subscribe AT cygwin DOT com>
List-Archive: <http://sources.redhat.com/ml/cygwin/>
List-Post: <mailto:cygwin AT cygwin DOT com>
List-Help: <mailto:cygwin-help AT cygwin DOT com>, <http://sources.redhat.com/ml/#faqs>
Sender: cygwin-owner AT cygwin DOT com
Mail-Followup-To: cygwin AT cygwin DOT com
Delivered-To: mailing list cygwin AT cygwin DOT com
From: "Ralf Habacker" <Ralf DOT Habacker AT freenet DOT de>
To: "Binutils" <binutils AT sources DOT redhat DOT com>, "cygwin" <cygwin AT cygwin DOT com>,
   "Kde-Cygwin" <kde-cygwin AT kde DOT org>
Cc: "Charles Wilson" <cwilson AT ece DOT gatech DOT edu>
Subject: [Patch] skipping import libraries for performance reasons - direct auto-import of dll's
Date: Mon, 25 Nov 2002 13:46:50 +0100
Message-ID: <00fe01c29480$b9453340$cd6407d5@BRAMSCHE>
MIME-Version: 1.0
Content-Type: multipart/mixed;
	boundary="----=_NextPart_000_00FF_01C29489.1B099B40"
X-Priority: 3 (Normal)
X-MSMail-Priority: Normal
Importance: Normal
X-MimeOLE: Produced By Microsoft MimeOLE V5.50.4133.2400
Note-from-DJ: This may be spam

------=_NextPart_000_00FF_01C29489.1B099B40
Content-Type: text/plain;
	charset="iso-8859-1"
Content-Transfer-Encoding: 7bit

Hi all,

the recent ld contains support importing of symbols located in other dll's
through import libraries.

Especially for big libraries using import libraries has some big disadvantages:

1. ld need much time and very much memory to process them  - This is because
each symbol is located in a seperate objectfile and ld opens a bfd for each.

2. The size of import libraries could be very huge (for example 10 MB for
qt-3) - This is because each symbol is located in a seperate objectfile.

current ld releases are able to skip import libraries by grabbing the needed
informations directly from a dll. The only issue which was open until now, is
that ld has no auto-import support for linking directly to a dll, but this is be
fixed with the appended patch.

Using dll's directly instead of import libraries has several advantages:

1. linking goes much faster (for qt-3 about 50%: from 2 minutes down to one
minute)

2. ld uses much less memory (for qt-3 about 70%: from 500 MB to 150 MB)

3. ld works more like the linux version. There are only static archives and
shared libraries which could be linked directly without the indirection of using
import libraries. This simplifies for example libtool handling.

How does it works:

See the example path layout. There are two dll's xxx.dll and yyy.dll in an
applications bin dir. In the lib dir there are the import library for xxx.dll
and a link to the dll cygyyy.dll.

../bin/
	cygxxx.dll
	cygyyy.dll

../lib/
	libxxx.dll.a
	[cyg|lib|]yyy.dll	(may be the dll  or a symbolic link to ../bin/cygyyy.dll)


Now the link line (you can see that using the dll directly is the same as using
the import libraries, there is no difference)

gcc -Wl,-verbose  -o a.exe -L../lib/ -lxxx -lyyy ...

The ld log below shows, that for xxx the import library is used and for yyy the
dll is directly used. This means, the only task of using the dll instead of the
import library is to replace the import library with a symbolic link to the dll.

<snip>
==================================================
attempt to open /usr/lib/crt0.o succeeded
/usr/lib/crt0.o
attempt to open client.o succeeded
client.o
attempt to open ../lib/libxxx.dll.a succeeded !!!
...
attempt to open ../lib/libyyy.dll.a failed
attempt to open ../lib/yyy.dll.a failed
attempt to open ../lib/libyyy.a failed
attempt to open ../lib/cygyyy.dll failed
attempt to open ../lib/libyyy.dll failed
attempt to open ../lib/yyy.dll succeeded !!!
<snip>
Info: resolving _var by linking to __imp__var (auto-import)
Info: resolving _foo by linking to __imp__foo (auto-import)
Info: resolving _func_ptr by linking to __imp__func_ptr (auto-import)

I have tested this patch with recent qt-2/3 and kde2 sources without any
problems. A sample test case is appended too.

Please give this patch a try and report problems to me o0r to this list.

Thank you

Ralf Habacker

KDE on cygwin http://cygwin.kde.org





------=_NextPart_000_00FF_01C29489.1B099B40
Content-Type: application/octet-stream;
	name="ld-auto-import-dll.patch"
Content-Transfer-Encoding: quoted-printable
Content-Disposition: attachment;
	filename="ld-auto-import-dll.patch"

Index: deffile.h=0A=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=0A=
RCS file: /cvs/src/src/ld/deffile.h,v=0A=
retrieving revision 1.5=0A=
diff -u -3 -p -B -u -b -B -r1.5 deffile.h=0A=
--- deffile.h	13 Mar 2001 06:14:27 -0000	1.5=0A=
+++ deffile.h	20 Nov 2002 21:23:45 -0000=0A=
@@ -52,6 +52,7 @@ typedef struct def_file_import {=0A=
   def_file_module *module;	/* always set */=0A=
   char *name;			/* may be NULL; either this or ordinal will be set */=0A=
   int ordinal;			/* may be -1 */=0A=
+  int data;					/* =3D 1 if data */ =0A=
 } def_file_import;=0A=
 =0A=
 typedef struct def_file {=0A=
Index: pe-dll.c=0A=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=0A=
RCS file: /cvs/src/src/ld/pe-dll.c,v=0A=
retrieving revision 1.48=0A=
diff -u -3 -p -B -u -b -B -r1.48 pe-dll.c=0A=
--- pe-dll.c	14 Nov 2002 18:03:16 -0000	1.48=0A=
+++ pe-dll.c	20 Nov 2002 21:24:20 -0000=0A=
@@ -2418,7 +2418,7 @@ pe_process_import_defs (output_bfd, link=0A=
 		exp.hint =3D exp.ordinal >=3D 0 ? exp.ordinal : 0;=0A=
 		exp.flag_private =3D 0;=0A=
 		exp.flag_constant =3D 0;=0A=
-		exp.flag_data =3D 0;=0A=
+    exp.flag_data =3D pe_def_file->imports[i].data;=0A=
 		exp.flag_noname =3D exp.name ? 0 : 1;=0A=
 		one =3D make_one (&exp, output_bfd);=0A=
 		add_bfd_to_link (one, one->filename, link_info);=0A=
@@ -2490,11 +2490,13 @@ pe_implied_import_dll (filename)=0A=
 {=0A=
   bfd *dll;=0A=
   unsigned long pe_header_offset, opthdr_ofs, num_entries, i;=0A=
-  unsigned long export_rva, export_size, nsections, secptr, expptr;=0A=
+  unsigned long export_rva, export_size, nsections, secptr, =
expptr,exp_funcbase;=0A=
   unsigned char *expdata, *erva;=0A=
   unsigned long name_rvas, ordinals, nexp, ordbase;=0A=
   const char *dll_name;=0A=
 =0A=
+  unsigned long data_start, data_end, bss_start, bss_end;=0A=
+=0A=
   /* No, I can't use bfd here.  kernel32.dll puts its export table in=0A=
      the middle of the .rdata section.  */=0A=
   dll =3D bfd_openr (filename, pe_details->target_name);=0A=
@@ -2511,11 +2513,8 @@ pe_implied_import_dll (filename)=0A=
       return false;=0A=
     }=0A=
 =0A=
-  dll_name =3D filename;=0A=
-  for (i =3D 0; filename[i]; i++)=0A=
-    if (filename[i] =3D=3D '/' || filename[i] =3D=3D '\\' || =
filename[i] =3D=3D ':')=0A=
-      dll_name =3D filename + i + 1;=0A=
-=0A=
+  /* get pe-header, optional header and numbers  =0A=
+     of export entries */=0A=
   pe_header_offset =3D pe_get32 (dll, 0x3c);=0A=
   opthdr_ofs =3D pe_header_offset + 4 + 20;=0A=
   num_entries =3D pe_get32 (dll, opthdr_ofs + 92);=0A=
@@ -2530,6 +2529,7 @@ pe_implied_import_dll (filename)=0A=
 	    pe_get16 (dll, pe_header_offset + 4 + 16));=0A=
   expptr =3D 0;=0A=
 =0A=
+  /* get the rva and size of the export section */ =0A=
   for (i =3D 0; i < nsections; i++)=0A=
     {=0A=
       char sname[8];=0A=
@@ -2550,6 +2550,36 @@ pe_implied_import_dll (filename)=0A=
 	}=0A=
     }=0A=
 =0A=
+  /* scan sections and store the base and =0A=
+     size of the data and bss segments in =0A=
+     data/base_start/end */ =0A=
+  for (i =3D 0; i < nsections; i++) {=0A=
+      unsigned long secptr1 =3D secptr + 40 * i;=0A=
+=0A=
+      unsigned long vsize =3D pe_get32 (dll, secptr1 + 8);=0A=
+      unsigned long vaddr =3D pe_get32 (dll, secptr1 + 12);=0A=
+      unsigned long flags =3D pe_get32 (dll, secptr1 + 36);=0A=
+      char sec_name[9];=0A=
+=0A=
+      sec_name[8] =3D '\0';=0A=
+      bfd_seek (dll, (file_ptr) secptr1 + 0, SEEK_SET);=0A=
+      bfd_bread (sec_name, (bfd_size_type) 8, dll);=0A=
+=0A=
+      if (strcmp(sec_name,".data") =3D=3D 0) {=0A=
+        data_start =3D vaddr;=0A=
+        data_end =3D vaddr + vsize;=0A=
+        if (pe_dll_extra_pe_debug)=0A=
+          printf("%s %s: 0x%08x-0x%08x =
(0x%08x)\n",__FUNCTION__,sec_name,vaddr,vaddr+vsize,flags);=0A=
+      }=0A=
+      else if (strcmp(sec_name,".bss") =3D=3D 0) {=0A=
+        bss_start =3D vaddr;=0A=
+        bss_end =3D vaddr + vsize;=0A=
+        if (pe_dll_extra_pe_debug)=0A=
+          printf("%s %s: 0x%08x-0x%08x =
(0x%08x)\n",__FUNCTION__,sec_name,vaddr,vaddr+vsize,flags);=0A=
+      }=0A=
+=0A=
+  }=0A=
+=0A=
   expdata =3D (unsigned char *) xmalloc (export_size);=0A=
   bfd_seek (dll, (file_ptr) expptr, SEEK_SET);=0A=
   bfd_bread (expdata, (bfd_size_type) export_size, dll);=0A=
@@ -2562,14 +2592,39 @@ pe_implied_import_dll (filename)=0A=
   name_rvas =3D pe_as32 (expdata + 32);=0A=
   ordinals =3D pe_as32 (expdata + 36);=0A=
   ordbase =3D pe_as32 (expdata + 16);=0A=
+  exp_funcbase =3D pe_as32 (expdata + 28);=0A=
 =0A=
+  /* use internal dll name instead of filename =0A=
+     to enable symbolic dll linking */ =0A=
+  dll_name =3D pe_as32 (expdata + 12) + erva ;=0A=
+=0A=
+	/* iterate through the list of symbols */ =0A=
   for (i =3D 0; i < nexp; i++)=0A=
     {=0A=
+    	/* pointer to the names vector */ =0A=
       unsigned long name_rva =3D pe_as32 (erva + name_rvas + i * 4);=0A=
       def_file_import *imp;=0A=
 =0A=
+    	/* pointer to the function address vector */ =0A=
+      unsigned long func_rva =3D pe_as32 (erva + exp_funcbase + i * 4);=0A=
+      int is_data =3D 0;=0A=
+=0A=
+        // skip unwanted symbols, which are exported in buggy =
auto-import releases=0A=
+        if (strstr(erva + name_rva,"_nm_") =3D=3D 0) {=0A=
+=0A=
+					/* is_data is true if the address is in the data or bss segment */ =0A=
+          is_data =3D (func_rva >=3D data_start && func_rva < data_end )=0A=
+                  || (func_rva >=3D bss_start && func_rva < bss_end);=0A=
+=0A=
       imp =3D def_file_add_import (pe_def_file, erva + name_rva, =
dll_name,=0A=
 				 i, 0);=0A=
+=0A=
+        	/* mark symbole type */=0A=
+          imp->data =3D is_data;=0A=
+=0A=
+          if (pe_dll_extra_pe_debug)=0A=
+            printf("%s dll-name: %s sym: %s addr: 0x%x =
%s\n",__FUNCTION__, dll_name, erva + name_rva, func_rva, is_data ? =
"(data)" : "");=0A=
+        }=0A=
     }=0A=
 =0A=
   return true;=0A=
Index: emultempl/pe.em=0A=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=0A=
RCS file: /cvs/src/src/ld/emultempl/pe.em,v=0A=
retrieving revision 1.69=0A=
diff -u -3 -p -B -u -b -B -r1.69 pe.em=0A=
--- emultempl/pe.em	14 Nov 2002 18:03:16 -0000	1.69=0A=
+++ emultempl/pe.em	20 Nov 2002 21:24:38 -0000=0A=
@@ -1009,9 +1009,10 @@ gld_${EMULATION_NAME}_after_open ()=0A=
   if (pe_enable_stdcall_fixup) /* -1=3Dwarn or 1=3Ddisable */=0A=
     pe_fixup_stdcalls ();=0A=
 =0A=
+  pe_process_import_defs(output_bfd, &link_info);=0A=
+=0A=
   pe_find_data_imports ();=0A=
 =0A=
-  pe_process_import_defs(output_bfd, &link_info);=0A=
   if (link_info.shared)=0A=
     pe_dll_build_sections (output_bfd, &link_info);=0A=
 =0A=

------=_NextPart_000_00FF_01C29489.1B099B40
Content-Type: application/octet-stream;
	name="ld-test.tar.bz2"
Content-Transfer-Encoding: base64
Content-Disposition: attachment;
	filename="ld-test.tar.bz2"
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------=_NextPart_000_00FF_01C29489.1B099B40
Content-Type: text/plain; charset=us-ascii

--
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple
Bug reporting:         http://cygwin.com/bugs.html
Documentation:         http://cygwin.com/docs.html
FAQ:                   http://cygwin.com/faq/
------=_NextPart_000_00FF_01C29489.1B099B40--