Make prepare_namespace() wait for devices

Testing and/or comments welcome.

     -- Pierre Ossman

  Linux kernel, MMC maintainer        http://www.kernel.org
  PulseAudio, core developer          http://pulseaudio.org
  rdesktop, core developer          http://www.rdesktop.org

commit 7c542a5a027caa95bb00f8a8223c7e4aef88c931

Author: Pierre Ossman <[email protected]>
Date:   Mon May 14 14:15:46 2007 +0200

init: wait for asynchronously scanned block devices

Some buses (e.g. USB and MMC) do their scanning of devices in the background, causing a race between them and prepare_namespace(). In order to be able to use these buses without an initrd, we now wait for the device specified in root= to actually show up.

If the device never shows up than we will hang in an infinite loop. Previously we panic:ed instead, so this behaviour should be no worse.

index 46fe407..192b2d6 100644
@@ -438,11 +438,24 @@ void __init prepare_namespace(void)
 			root_device_name += 5;
 	}
 
-	is_floppy = MAJOR(ROOT_DEV) == FLOPPY_MAJOR;
-
 	if (initrd_load())
 		goto out;
 
+	/* wait for any asynchronous scanning to complete */
+	if (ROOT_DEV == 0) {
+		printk("Waiting for root device %s...\n", saved_root_name);
+		do {
+			while (driver_probe_done() != 0)
+				msleep(100);
+			ROOT_DEV = name_to_dev_t(saved_root_name);
+			if (ROOT_DEV == 0)
+				msleep(100);
+		} while (ROOT_DEV == 0);
+	}
+
+	is_floppy = MAJOR(ROOT_DEV) == FLOPPY_MAJOR;
+
+
 	if (is_floppy && rd_doload && rd_load_disk(0))
 		ROOT_DEV = Root_RAM0;
 

On Mon, May 14, 2007 at 02:19:37PM +0200, Pierre Ossman wrote:

First of all, please do not put patches in attachments.

As for the patch, an infinite loop is *much* worse than a panic. When one is installing a new kernel on remote box, having it booted with reboot on panic and loader falling back to old kernel on the next boot is a bloody good idea - gives you extra safety. With your patch it will get the box stuck.

My mailer tends to fsck them up if I just paste them. And it's attached without any base64 silliness, so most people can usually read it directly.

Ok. So a timeout or having the entire thing Kconfig optional?

Rgds

     -- Pierre Ossman

  Linux kernel, MMC maintainer        http://www.kernel.org
  PulseAudio, core developer          http://pulseaudio.org
  rdesktop, core developer          http://www.rdesktop.org

How about command-line optional?

Haavard

New suggestion.

     -- Pierre Ossman

  Linux kernel, MMC maintainer        http://www.kernel.org
  PulseAudio, core developer          http://pulseaudio.org
  rdesktop, core developer          http://www.rdesktop.org

commit bb8c44ee8b4d584295add58a4ea2f03b9938fc3c

Author: Pierre Ossman <[email protected]>
Date:   Mon May 14 14:15:46 2007 +0200

init: wait for asynchronously scanned block devices

Some buses (e.g. USB and MMC) do their scanning of devices in the background, causing a race between them and prepare_namespace(). In order to be able to use these buses without an initrd, we now wait for the device specified in root= to actually show up.

If the device never shows up than we will hang in an infinite loop. In order to not mess with setups that reboot on panic, the feature must be turned on via the command line option "rootwait".

index 46fe407..efc134c 100644
@@ -25,6 +25,7 @@ int __initdata rd_doload;	/* 1 = load RAM disk, 0 = don't load */
 int root_mountflags = MS_RDONLY | MS_SILENT;
 char * __initdata root_device_name;
 static char __initdata saved_root_name[64];
+int __initdata root_wait;
 
 dev_t ROOT_DEV;
 

@@ -216,6 +217,14 @@ static int __init root_dev_setup(char *line)
 
 __setup("root=", root_dev_setup);
 
+static int __init rootwait_setup(char *line)
+{
+	root_wait = simple_strtol(line,NULL,0);
+	return 1;
+}
+
+__setup("rootwait=", rootwait_setup);
+
 static char * __initdata root_mount_data;
 static int __init root_data_setup(char *str)
 {

@@ -438,11 +447,24 @@ void __init prepare_namespace(void)
 			root_device_name += 5;
 	}
 
-	is_floppy = MAJOR(ROOT_DEV) == FLOPPY_MAJOR;
-
 	if (initrd_load())
 		goto out;
 
+	/* wait for any asynchronous scanning to complete */
+	if ((ROOT_DEV == 0) && root_wait) {
+		printk(KERN_INFO "Waiting for root device %s...\n",
+			saved_root_name);
+		do {
+			while (driver_probe_done() != 0)
+				msleep(100);
+			ROOT_DEV = name_to_dev_t(saved_root_name);
+			if (ROOT_DEV == 0)
+				msleep(100);
+		} while (ROOT_DEV == 0);
+	}
+
+	is_floppy = MAJOR(ROOT_DEV) == FLOPPY_MAJOR;
+
 	if (is_floppy && rd_doload && rd_load_disk(0))
 		ROOT_DEV = Root_RAM0;
 

Works wonderfully here:

Waiting for root device /dev/mmcblk0p1...
<inserts card>
mmcblk0: mmc0:a95c SD128 123008KiB
mmcblk0: p1
VFS: Mounted root (ext2 filesystem).

Of course, it also fixes the case where the card _is_ inserted at bootup :-)

Haavard

Seconded. Love your work Pierre :)

Tested across 4 different brands of SD/MMC/RS-MMC card and 2 different configurations (both AVR32 arch) just for the hells of it. All plays nice.

Thanks,
--Ben.

Actually that's not correct. With panic=30 and lilo -R and a working backup kernel a system can recover from this. With your endless loop it can't.

Always add some kind of timeout.

-Andi

Hi there brave visitor from the future ;)

Did you check the second version? Is that sufficient or is a timeout a must in your book?

Rgds

     -- Pierre Ossman

  Linux kernel, MMC maintainer        http://www.kernel.org
  PulseAudio, core developer          http://pulseaudio.org
  rdesktop, core developer          http://www.rdesktop.org