Skip to contentSkip to conversation

Suspend2 Merge: Driver model patches 2/2

PATCH In review Nigel Cunningham 3 participants 8 messages Sep 16, 2004 – Sep 17, 2004

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 Suspend2 Merge: Driver model patches 2/2

Hi.

This simple helper adds support for finding a class given its name. I use this to locate the frame buffer drivers and move them to the keep-alive tree while suspending other drivers.

Regards,

Nigel

drivers/base/class.c +16 -0
diff -ruN linux-2.6.9-rc1/drivers/base/class.c software-suspend-linux-2.6.9-rc1-rev3/drivers/base/class.c
@@ -460,6 +460,20 @@
 	kobject_put(&class_dev->kobj);
 }
 
+struct class * class_find(char * name)
+{
+	struct class * this_class;
+
+	if (!name)
+		return NULL;
+
+	list_for_each_entry(this_class, &class_subsys.kset.list, subsys.kset.kobj.entry) {
+		if (!(strcmp(this_class->name, name)))
+			return this_class;
+	}
+
+	return NULL;
+}
 
 int class_interface_register(struct class_interface *class_intf)
 {
@@ -542,3 +556,5 @@
 
 EXPORT_SYMBOL(class_interface_register);
 EXPORT_SYMBOL(class_interface_unregister);
+
+EXPORT_SYMBOL(class_find);
include/linux/device.h +1 -0
diff -ruN linux-2.6.9-rc1/include/linux/device.h software-suspend-linux-2.6.9-rc1-rev3/include/linux/device.h
@@ -163,6 +163,7 @@
 
 extern struct class * class_get(struct class *);
 extern void class_put(struct class *);
+extern struct class * class_find(char * name);
 
 
 struct class_attribute {
Signature · 7 lines
Nigel Cunningham
Pastoral Worker
Christian Reformed Church of Tuggeranong
PO Box 1004, Tuggeranong, ACT 2901

Many today claim to be tolerant. True tolerance, however, can cope with others
being intolerant.
#1 lore raw
Quoted 9 lines · Nigel Cunningham

This simple helper adds support for finding a class given its name. I use this to locate the frame buffer drivers and move them to the keep-alive tree while suspending other drivers.

drivers/base/class.c +4 -0
+struct class * class_find(char * name)
+{
+	struct class * this_class;
+

drivers/base/class.c line 476

Quoted 10 lines
drivers/base/class.c +10 -0
+	if (!name)
+		return NULL;
+
+	list_for_each_entry(this_class, &class_subsys.kset.list, subsys.kset.kobj.entry) {
+		if (!(strcmp(this_class->name, name)))
+			return this_class;
+	}
+
+	return NULL;
+}

Ick, no. I've been over this before with the fb people, and am not going to accept this patch (nevermind that it's broken...) See the lkml archives for more info on why I don't like this.

thanks,

greg k-h

#2 lore raw

Hi.

Quoted 16 lines · Greg KH
Quoted 15 lines · Nigel Cunningham

This simple helper adds support for finding a class given its name. I use this to locate the frame buffer drivers and move them to the keep-alive tree while suspending other drivers.

drivers/base/class.c +10 -0
+struct class * class_find(char * name)
+{
+	struct class * this_class;
+
+	if (!name)
+		return NULL;
+
+	list_for_each_entry(this_class, &class_subsys.kset.list, subsys.kset.kobj.entry) {
+		if (!(strcmp(this_class->name, name)))
+			return this_class;
Quoted 8 lines

drivers/base/class.c line 476

Quoted 4 lines
drivers/base/class.c +4 -0
+	}
+
+	return NULL;
+}

Ick, no. I've been over this before with the fb people, and am not going to accept this patch (nevermind that it's broken...) See the lkml archives for more info on why I don't like this.

Please excuse my ignorance but I don't see how it's broken (their patch just fills in a field that was left blank previously), and this patch just makes use of that change. What's the point to device_class if we don't use it?

That said, I do agree with using Pavel's new enum that includes _SNAPSHOT and can see that it's a cleaner way in that it requires less knowledge on suspend's part of what it wants to stay alive.

Regards,

Nigel

Signature · 7 lines
Nigel Cunningham
Pastoral Worker
Christian Reformed Church of Tuggeranong
PO Box 1004, Tuggeranong, ACT 2901

Many today claim to be tolerant. True tolerance, however, can cope with others
being intolerant.
#3 lore raw
Quoted 26 lines · Nigel Cunningham
Quoted 16 lines · Greg KH
Quoted 15 lines · Nigel Cunningham

This simple helper adds support for finding a class given its name. I use this to locate the frame buffer drivers and move them to the keep-alive tree while suspending other drivers.

drivers/base/class.c +10 -0
+struct class * class_find(char * name)
+{
+	struct class * this_class;
+
+	if (!name)
+		return NULL;
+
+	list_for_each_entry(this_class, &class_subsys.kset.list, subsys.kset.kobj.entry) {
+		if (!(strcmp(this_class->name, name)))
+			return this_class;
Quoted 8 lines

drivers/base/class.c line 476

Quoted 4 lines
drivers/base/class.c +4 -0
+	}
+
+	return NULL;
+}

Ick, no. I've been over this before with the fb people, and am not going to accept this patch (nevermind that it's broken...) See the lkml archives for more info on why I don't like this.

Quoted 1 line

Please excuse my ignorance but I don't see how it's broken

This function, as written is very broken. I will not accept it. Not to mention the fact that the functionality this function proposes to offer is not needed either.

Quoted 1 line

(their patch just fills in a field that was left blank previously),

What patch?

Quoted 2 lines

and this patch just makes use of that change. What's the point to device_class if we don't use it?

I don't see a use of device_class in this function. I'm confused.

thanks,

greg k-h

#4 lore raw

Hi.

Quoted 7 lines · Greg KH
Quoted 5 lines
Quoted 3 lines

Ick, no. I've been over this before with the fb people, and am not going to accept this patch (nevermind that it's broken...) See the lkml archives for more info on why I don't like this.

Please excuse my ignorance but I don't see how it's broken

This function, as written is very broken. I will not accept it. Not to

What's broken? (I want to learn what I've done wrong that I'm not seeing).

Quoted 6 lines

mention the fact that the functionality this function proposes to offer is not needed either.

Quoted 1 line

(their patch just fills in a field that was left blank previously),

What patch?

Attached. Sorry if I wrongly assumed this was the patch you're talking about.

Quoted 4 lines
Quoted 2 lines

and this patch just makes use of that change. What's the point to device_class if we don't use it?

I don't see a use of device_class in this function. I'm confused.

This patch finds the device_class that the frame buffer drivers register. It gets called by suspend code I haven't posted yet, which then moves the drivers in this class from one pm tree to another so that the frame buffer drivers don't get suspended until it's time for the atomic snapshot and can be resumed afterwards while we write the rest of the image, without resuming all drivers. Given Pavel's work with the new _SNAPSHOT flag, I guess this won't eventually be needed (provided, of course that drivers do the right thing when called with _SNAPSHOT).

Regards,

Nigel

Signature · 7 lines
Nigel Cunningham
Pastoral Worker
Christian Reformed Church of Tuggeranong
PO Box 1004, Tuggeranong, ACT 2901

Many today claim to be tolerant. True tolerance, however, can cope with others
being intolerant.
drivers/video/aty/atyfb_base.c +1 -1
diff -ruN linux-2.6.9-rc1/drivers/video/aty/atyfb_base.c software-suspend-linux-2.6.9-rc1-rev3/drivers/video/aty/atyfb_base.c
@@ -1972,7 +1972,7 @@
 
 			info->fix = atyfb_fix;
 			info->par = default_par;
-
+			info->device = &pdev->dev;
 #ifdef __sparc__
 			/*
 			 * Map memory-mapped registers.
drivers/video/chipsfb.c +1 -1
diff -ruN linux-2.6.9-rc1/drivers/video/chipsfb.c software-suspend-linux-2.6.9-rc1-rev3/drivers/video/chipsfb.c
@@ -416,7 +416,7 @@
 		release_mem_region(addr, size);
 		return -ENOMEM;
 	}
-
+	p->device = &dp->dev;
 	init_chips(p, addr);
 
 #ifdef CONFIG_PMAC_PBOOK
drivers/video/cyber2000fb.c +2 -0
diff -ruN linux-2.6.9-rc1/drivers/video/cyber2000fb.c software-suspend-linux-2.6.9-rc1-rev3/drivers/video/cyber2000fb.c
@@ -1399,6 +1399,8 @@
 		cfb->fb.var.xres, cfb->fb.var.yres,
 		h_sync / 1000, h_sync % 1000, v_sync);
 
+	if (cfb->dev)
+		cfb->fb.device = &cfb->dev->dev;
 	err = register_framebuffer(&cfb->fb);
 
 failed:
drivers/video/fbmem.c +2 -1
diff -ruN linux-2.6.9-rc1/drivers/video/fbmem.c software-suspend-linux-2.6.9-rc1-rev3/drivers/video/fbmem.c
@@ -1444,7 +1444,8 @@
 			break;
 	fb_info->node = i;
 
-	c = class_simple_device_add(fb_class, MKDEV(FB_MAJOR, i), NULL, "fb%d", i);
+	c = class_simple_device_add(fb_class, MKDEV(FB_MAJOR, i),
+				    fb_info->device, "fb%d", i);
 	if (IS_ERR(c)) {
 		/* Not fatal */
 		printk(KERN_WARNING "Unable to create class_device for framebuffer %d; errno = %ld\n", i, PTR_ERR(c));
drivers/video/fbsysfs.c +2 -0
diff -ruN linux-2.6.9-rc1/drivers/video/fbsysfs.c software-suspend-linux-2.6.9-rc1-rev3/drivers/video/fbsysfs.c
@@ -51,6 +51,8 @@
 	if (size)
 		info->par = p + fb_info_size;
 
+	info->device = dev;
+
 	return info;
 #undef PADDING
 #undef BYTES_PER_LONG
drivers/video/i810/i810_main.c +31 -40

89 lines not on this page · raw message

drivers/video/igafb.c +1 -0
diff -ruN linux-2.6.9-rc1/drivers/video/igafb.c software-suspend-linux-2.6.9-rc1-rev3/drivers/video/igafb.c
@@ -528,6 +528,7 @@
 	info->var = default_var;
 	info->fix = igafb_fix;
 	info->pseudo_palette = (void *)(par + 1);
+	info->device = &pdev->dev;
 
 	if (!iga_init(info, par)) {
 		iounmap((void *)par->io_base);
drivers/video/imsttfb.c +1 -0
diff -ruN linux-2.6.9-rc1/drivers/video/imsttfb.c software-suspend-linux-2.6.9-rc1-rev3/drivers/video/imsttfb.c
@@ -1524,6 +1524,7 @@
 	par->cmap_regs = (__u8 *)ioremap(addr + 0x840000, 0x1000);
 	info->par = par;
 	info->pseudo_palette = (void *) (par + 1);
+	info->device = &pdev->dev;
 	init_imstt(info);
 
 	pci_set_drvdata(pdev, info);
drivers/video/kyro/fbdev.c +1 -0
diff -ruN linux-2.6.9-rc1/drivers/video/kyro/fbdev.c software-suspend-linux-2.6.9-rc1-rev3/drivers/video/kyro/fbdev.c
@@ -735,6 +735,7 @@
 
 	fb_memset(info->screen_base, 0, size);
 
+	info->device = &pdev->dev;
 	if (register_framebuffer(info) < 0)
 		goto out_unmap;
 
drivers/video/matrox/matroxfb_base.c +1 -0
diff -ruN linux-2.6.9-rc1/drivers/video/matrox/matroxfb_base.c software-suspend-linux-2.6.9-rc1-rev3/drivers/video/matrox/matroxfb_base.c
@@ -1864,6 +1864,7 @@
 /* We do not have to set currcon to 0... register_framebuffer do it for us on first console
  * and we do not want currcon == 0 for subsequent framebuffers */
 
+	ACCESS_FBINFO(fbcon).device = &ACCESS_FBINFO(pcidev)->dev;
 	if (register_framebuffer(&ACCESS_FBINFO(fbcon)) < 0) {
 		goto failVideoIO;
 	}
drivers/video/pvr2fb.c +1 -0
diff -ruN linux-2.6.9-rc1/drivers/video/pvr2fb.c software-suspend-linux-2.6.9-rc1-rev3/drivers/video/pvr2fb.c
@@ -939,6 +939,7 @@
 
 	pvr2_fix.mmio_start	= pci_resource_start(pdev, 1);
 	pvr2_fix.mmio_len	= pci_resource_len(pdev, 1);
+	fbinfo->device = &pdev->dev;
 
 	return pvr2fb_common_init();
 }
drivers/video/radeonfb.c +1 -1
diff -ruN linux-2.6.9-rc1/drivers/video/radeonfb.c software-suspend-linux-2.6.9-rc1-rev3/drivers/video/radeonfb.c
@@ -3040,7 +3040,7 @@
 	pci_set_drvdata(pdev, rinfo);
 	rinfo->next = board_list;
 	board_list = rinfo;
-
+	((struct fb_info *) rinfo)->device = &pdev->dev;
 	if (register_framebuffer ((struct fb_info *) rinfo) < 0) {
 		printk ("radeonfb: could not register framebuffer\n");
 		iounmap ((void*)rinfo->fb_base);
drivers/video/riva/fbdev.c +7 -15
diff -ruN linux-2.6.9-rc1/drivers/video/riva/fbdev.c software-suspend-linux-2.6.9-rc1-rev3/drivers/video/riva/fbdev.c
@@ -1850,21 +1850,17 @@
 	NVTRACE_ENTER();
 	assert(pd != NULL);
 
-	info = kmalloc(sizeof(struct fb_info), GFP_KERNEL);
+	info = framebuffer_alloc(sizeof(struct riva_par), &pd->dev);
+
 	if (!info)
 		goto err_out;
 
-	default_par = kmalloc(sizeof(struct riva_par), GFP_KERNEL);
-	if (!default_par)
-		goto err_out_kfree;
-
-	memset(info, 0, sizeof(struct fb_info));
-	memset(default_par, 0, sizeof(struct riva_par));
+	default_par = (struct riva_par *) info->par;
 	default_par->pdev = pd;
 
 	info->pixmap.addr = kmalloc(64 * 1024, GFP_KERNEL);
 	if (info->pixmap.addr == NULL)
-		goto err_out_kfree1;
+		goto err_out_kfree;
 	memset(info->pixmap.addr, 0, 64 * 1024);
 
 	if (pci_enable_device(pd)) {
@@ -1888,7 +1884,7 @@
 
 	if(default_par->riva.Architecture == 0) {
 		printk(KERN_ERR PFX "unknown NV_ARCH\n");
-		goto err_out_kfree1;
+		goto err_out_free_base0;
 	}
 	if(default_par->riva.Architecture == NV_ARCH_10 ||
 	   default_par->riva.Architecture == NV_ARCH_20 ||
@@ -1994,7 +1990,6 @@
 	fb_destroy_modedb(info->monspecs.modedb);
 	info->monspecs.modedb_len = 0;
 	info->monspecs.modedb = NULL;
-
 	if (register_framebuffer(info) < 0) {
 		printk(KERN_ERR PFX
 			"error registering riva framebuffer\n");
@@ -2033,10 +2028,8 @@
 	pci_disable_device(pd);
 err_out_enable:
 	kfree(info->pixmap.addr);
-err_out_kfree1:
-	kfree(default_par);
 err_out_kfree:
-	kfree(info);
+	framebuffer_release(info);
 err_out:
 	return -ENODEV;
 }
@@ -2070,8 +2063,7 @@
 	pci_release_regions(pd);
 	pci_disable_device(pd);
 	kfree(info->pixmap.addr);
-	kfree(par);
-	kfree(info);
+	framebuffer_release(info);
 	pci_set_drvdata(pd, NULL);
 	NVTRACE_LEAVE();
 }
drivers/video/sstfb.c +1 -0
diff -ruN linux-2.6.9-rc1/drivers/video/sstfb.c software-suspend-linux-2.6.9-rc1-rev3/drivers/video/sstfb.c
@@ -1507,6 +1507,7 @@
 	fb_alloc_cmap(&info->cmap, 256, 0);
 
 	/* register fb */
+	info->device = &pdev->dev;
 	if (register_framebuffer(info) < 0) {
 		eprintk("can't register framebuffer.\n");
 		goto fail;
drivers/video/tgafb.c +1 -0
diff -ruN linux-2.6.9-rc1/drivers/video/tgafb.c software-suspend-linux-2.6.9-rc1-rev3/drivers/video/tgafb.c
@@ -1454,6 +1454,7 @@
 	tgafb_set_par(&all->info);
 	tgafb_init_fix(&all->info);
 
+	all->info.device = &pdev->dev;
 	if (register_framebuffer(&all->info) < 0) {
 		printk(KERN_ERR "tgafb: Could not register framebuffer\n");
 		ret = -EINVAL;
drivers/video/tridentfb.c +1 -0
diff -ruN linux-2.6.9-rc1/drivers/video/tridentfb.c software-suspend-linux-2.6.9-rc1-rev3/drivers/video/tridentfb.c
@@ -1164,6 +1164,7 @@
 		default_var.accel_flags &= ~FB_ACCELF_TEXT;
 	default_var.activate |= FB_ACTIVATE_NOW;
 	fb_info.var = default_var;
+	fb_info.device = &dev->dev;
 	if (register_framebuffer(&fb_info) < 0) {
 		output("Could not register Trident framebuffer\n");
 		return -EINVAL;
include/linux/fb.h +1 -0
diff -ruN linux-2.6.9-rc1/include/linux/fb.h software-suspend-linux-2.6.9-rc1-rev3/include/linux/fb.h
@@ -595,6 +595,7 @@
 	struct fb_pixmap sprite;	/* Cursor hardware mapper */
 	struct fb_cmap cmap;		/* Current cmap */
 	struct fb_ops *fbops;
+	struct device *device;
 	char *screen_base;		/* Virtual address */
 	int currcon;			/* Current VC. */
 	void *pseudo_palette;		/* Fake palette of 16 colors */ 
#5 lore raw
Quoted 7 lines · Nigel Cunningham

Hi.

Quoted 4 lines · Greg KH
Quoted 4 lines
Quoted 3 lines

Ick, no. I've been over this before with the fb people, and am not going to accept this patch (nevermind that it's broken...) See the lkml archives for more info on why I don't like this.

Quoted 6 lines
Quoted 3 lines
Quoted 1 line

Please excuse my ignorance but I don't see how it's broken

This function, as written is very broken. I will not accept it. Not to

What's broken? (I want to learn what I've done wrong that I'm not seeing).

- No locking when traversing the list.
- Reference count needs to be bumped before returning a pointer to the object you found.

Quoted 9 lines
Quoted 6 lines

mention the fact that the functionality this function proposes to offer is not needed either.

Quoted 1 line

(their patch just fills in a field that was left blank previously),

What patch?

Attached. Sorry if I wrongly assumed this was the patch you're talking about.

Ah, no, I've never seen this one, thanks. But it looks sane, I don't have a problem with it (sysfs will like it, it's not a suspend specific patch at all.)

thanks,

greg k-h

#6 lore raw

Hi.

Quoted 6 lines · Greg KH
Quoted 2 lines

What's broken? (I want to learn what I've done wrong that I'm not seeing).

- No locking when traversing the list.
- Reference count needs to be bumped before returning a pointer to the object you found.

Ah. Fair enough. I haven't seen any problems because the locking is more
abstract: processes are frozen when this runs for suspend. I'll fix it, but won't bother resubmitting it because Pavel's changes should obsolete this stuff.

Quoted 5 lines
Quoted 5 lines
Quoted 5 lines

mention the fact that the functionality this function proposes to offer is not needed either.

Quoted 1 line

(their patch just fills in a field that was left blank previously),

Quoted 8 lines
Quoted 4 lines
Quoted 1 line

What patch?

Attached. Sorry if I wrongly assumed this was the patch you're talking about.

Ah, no, I've never seen this one, thanks. But it looks sane, I don't have a problem with it (sysfs will like it, it's not a suspend specific patch at all.)

Antonio posted it to LKML last week IIRC, which is why I didn't include it in the device driver patches. Given Pavel's changes (again), I'm in two minds as to whether its needed. It's clearly the right thing to do, but not needed at the moment. Then again, as we noted already, the whole device_class thing doesn't get a lot of use at the moment.

Regards,

Nigel

Signature · 7 lines
Nigel Cunningham
Pastoral Worker
Christian Reformed Church of Tuggeranong
PO Box 1004, Tuggeranong, ACT 2901

Many today claim to be tolerant. True tolerance, however, can cope with others
being intolerant.
#7 lore raw

Hi!

Quoted 8 lines
Quoted 3 lines

Ah, no, I've never seen this one, thanks. But it looks sane, I don't have a problem with it (sysfs will like it, it's not a suspend specific patch at all.)

Antonio posted it to LKML last week IIRC, which is why I didn't include it in the device driver patches. Given Pavel's changes (again), I'm in two minds as to whether its needed. It's clearly the right thing to do, but not needed at the moment. Then again, as we noted already, the whole

If it is not needed right now, go for simple solution and drop that patch. Interested people can find it in list archives.

Signature · 1 line
64 bytes from 195.113.31.123: icmp_seq=28 ttl=51 time=448769.1 ms