module.viomap support for ppc64

Current MODULE_DEVICE_TABLE(vio, $table); defines 2 char pointers. I'm not sure if depmod can really handle it. Where do they point to in the module binary? I could find an answer, so far. I just declared an array.

diff -p -purN module-init-tools-3.0-pre10.orig/moduleops_core.c module-init-tools-3.0-pre10/moduleops_core.c

moduleops_core.c +4 -0

@@ -176,6 +176,10 @@ static void PERBIT(fetch_tables)(struct
 	module->ccw_table = PERBIT(deref_sym)(module->data,
 					"__mod_ccw_device_table");
 
+	module->vio_size = PERBIT(VIO_DEVICE_SIZE);
+	module->vio_table = PERBIT(deref_sym)(module->data,
+					"__mod_vio_device_table");
+
 	module->ieee1394_size = PERBIT(IEEE1394_DEVICE_SIZE);
 	module->ieee1394_table = PERBIT(deref_sym)(module->data,
 					"__mod_ieee1394_device_table");

diff -p -purN module-init-tools-3.0-pre10.orig/tables.c module-init-tools-3.0-pre10/tables.c

tables.c +24 -0

@@ -127,7 +127,31 @@ void output_ieee1394_table(struct module
 			output_ieee1394_entry(fw, shortname, out);
 	}
 }
+static void output_vio_entry(struct vio_device_id *fw, char *name, FILE *out)
+{
+	fprintf(out, "%-20s %-32s %s\n",
+		name, fw->name, fw->compat);
+}
 
+void output_vio_table(struct module *modules, FILE *out)
+{
+	struct module *i;
+
+	fprintf(out, "%-20s %-32s compatible\n", "# vio module", "name");
+
+	for (i = modules; i; i = i->next) {
+		struct vio_device_id *fw;
+		char shortname[strlen(i->pathname) + 1];
+
+		if (!i->vio_table)
+			continue;
+
+		make_shortname(shortname, i->pathname);
+		for (fw = i->vio_table; *fw->name;
+		     fw = (void *) fw + i->vio_size)
+			output_vio_entry(fw, shortname, out);
+	}
+}
 
 /* We set driver_data to zero */
 static void output_ccw_entry(struct ccw_device_id *ccw, char *name, FILE *out)

Quoted 3 lines · Olaf Hering

Current MODULE_DEVICE_TABLE(vio, $table); defines 2 char pointers. I'm not sure if depmod can really handle it. Where do they point to in the module binary? I could find an answer, so far. I just declared an array.

Olaf explained on irc that output_vio_entry() below was finding NULL for the name and compat pointers. Perhaps some additional relocation needs to take place before those can be used?

2) Please modify scripts/mod/file2alias.c in the kernel source, not the module tools. The modules.XXXmap files are deprecated: device tables are supposed to be converted to aliases in the build process, and that is how userspace tools like hotplug are to find them.

3) I will still accept patches to module-init-tools if required for 2.4 compatibility, but they will be going away at some point!

Quoted 4 lines · Rusty Russell

2) Please modify scripts/mod/file2alias.c in the kernel source, not the module tools. The modules.XXXmap files are deprecated: device tables are supposed to be converted to aliases in the build process, and that is how userspace tools like hotplug are to find them.

I found no user of the modules.alias file. Hotplug still uses the map files. Parsing one big file will not improve performance, but thats a different story.

A hack for 2.6.8-rc4 is below. Can I read the alias file via while read a b c ; do : done < modules.alias ?
Is b supposed to contain not spaces? What special delimiter chars are allowed? The 'name' and 'compat' property can contain almost any char. I used '^' for the time being.

Quoted 2 lines

3) I will still accept patches to module-init-tools if required for 2.4 compatibility, but they will be going away at some point!

Noone cares about that old junk.

Quoted 10 lines · Olaf Hering

Quoted 4 lines · Rusty Russell

2) Please modify scripts/mod/file2alias.c in the kernel source, not the module tools. The modules.XXXmap files are deprecated: device tables are supposed to be converted to aliases in the build process, and that is how userspace tools like hotplug are to find them.

I found no user of the modules.alias file. Hotplug still uses the map files. Parsing one big file will not improve performance, but thats a different story.

You don't use the modules.alias file. You simply "modprobe vio:xyz^abc" and modprobe reads modules.alias if necessary (the user can also insert aliases in the modprobe.conf file, for example). Note that fnmatch is used, so you can actually use ? and * in your generated aliases.

Quoted 5 lines

A hack for 2.6.8-rc4 is below. Can I read the alias file via while read a b c ; do : done < modules.alias ?
Is b supposed to contain not spaces? What special delimiter chars are allowed? The 'name' and 'compat' property can contain almost any char. I used '^' for the time being.

Spaces are probably a bad idea, yes. ^ is a little odd, but probably not a bad choice. You could even use a full: "vio:name:%s:compat:%s" if you wanted to.

Quoted 4 lines

Quoted 2 lines

3) I will still accept patches to module-init-tools if required for 2.4 compatibility, but they will be going away at some point!

Noone cares about that old junk.

Shh... Marcelo might get offended 8)

Quoted 7 lines · Olaf Hering

Quoted 4 lines · Rusty Russell

2) Please modify scripts/mod/file2alias.c in the kernel source, not the module tools. The modules.XXXmap files are deprecated: device tables are supposed to be converted to aliases in the build process, and that is how userspace tools like hotplug are to find them.

Quoted 3 lines

I found no user of the modules.alias file. Hotplug still uses the map files. Parsing one big file will not improve performance, but thats a different story.

You don't use the modules.alias file. You simply "modprobe vio:xyz^abc" and modprobe reads modules.alias if necessary (the user can also insert aliases in the modprobe.conf file, for example). Note that fnmatch is used, so you can actually use ? and * in your generated aliases.

Quoted 8 lines · Rusty Russell

Quoted 7 lines · Olaf Hering

Quoted 4 lines · Rusty Russell

2) Please modify scripts/mod/file2alias.c in the kernel source, not the module tools. The modules.XXXmap files are deprecated: device tables are supposed to be converted to aliases in the build process, and that is how userspace tools like hotplug are to find them.

Quoted 8 lines

Quoted 3 lines

I found no user of the modules.alias file. Hotplug still uses the map files. Parsing one big file will not improve performance, but thats a different story.

You don't use the modules.alias file. You simply "modprobe vio:xyz^abc" and modprobe reads modules.alias if necessary (the user can also insert aliases in the modprobe.conf file, for example). Note that fnmatch is used, so you can actually use ? and * in your generated aliases.

But that complicates the parser. Current the hotplug agent script reads the simple modules.foomap and generates a list of possible drivers. Then it looks into the blacklist file to see if one of the possible modules should not be loaded, and skips this module.

Is there such functionality for the modules.alias file in module-init-tools? I played around with modprobe -n, but could not figure it out. Unfortunately, some hardware has more than one driver. bcm5700/tg3, eepro100/e100 and maybe more.

Current implementation of aliases is to load one at random: multiple alias resolution is undefined because noone knew what we should do (load them all? Load until one succeeds?). But note that that the base config file overrides anything extracted from the modules themselves, so users/distributions can always specify an exact match.

Quoted 4 lines

Is there such functionality for the modules.alias file in module-init-tools? I played around with modprobe -n, but could not figure it out. Unfortunately, some hardware has more than one driver. bcm5700/tg3, eepro100/e100 and maybe more.

OK, I think the difference here is that I feel modprobe should resolve it. What's the right answer? Do we need a new "unalias" config cmd which does the blacklist, or is the current positive method better? How do you currently decide?

Quoted 5 lines · Rusty Russell

Current implementation of aliases is to load one at random: multiple alias resolution is undefined because noone knew what we should do (load them all? Load until one succeeds?). But note that that the base config file overrides anything extracted from the modules themselves, so users/distributions can always specify an exact match.

How is the blacklist stuff supposed to work then? It must be possible to map an alias entry to a list of modules, and check if any of them is blacklisted. Then just 'for i in $DRIVERS ; do modprobe $i ; done'.