[PATCH] mtd: prevent physmap from causing "request_module: runaway loop modprobe net-pf-1"

From: Gordon Farquharson
Date: Thu Mar 13 2008 - 03:14:44 EST


On Wed, Mar 12, 2008 at 1:19 AM, Andrew Morton
<akpm@xxxxxxxxxxxxxxxxxxxx> wrote:

> Just send the patch, please. Please retain you original analysis in its
> covering description.

On machines that register a physmap flash platform device (e.g. many
arm machines), parse_mtd_partitions() in drivers/mtd/mtdpart.c can
cause the following output in the boot log:

request_module: runaway loop modprobe net-pf-1
modprobe: FATAL: Could not load
/lib/modules/2.6.25-rc4-iop32x/modules.dep: No such file or directory

modprobe: FATAL: Could not load
/lib/modules/2.6.25-rc4-iop32x/modules.dep: No such file or directory

modprobe: FATAL: Could not load
/lib/modules/2.6.25-rc4-iop32x/modules.dep: No such file or directory
...

To illustrate how this can occur, I'll use the example where
CONFIG_KMOD=y, CONFIG_MTD_PARTITIONS=y, CONFIG_MTD_REDBOOT_PARTS=y,
and CONFIG_CMDLINE_PARTS is not set. I'll also refer to the following
code segment which is in drivers/mtd/mtdpart.c:parse_mtd_partitions():

for ( ; ret <= 0 && *types; types++) {
parser = get_partition_parser(*types);
#ifdef CONFIG_KMOD
if (!parser && !request_module("%s", *types))
parser = get_partition_parser(*types);
#endif
if (!parser) {
printk(KERN_NOTICE "%s partition parsing not available\n",
*types);
continue;
}
ret = (*parser->parse_fn)(master, pparts, origin);
if (ret > 0) {
printk(KERN_NOTICE "%d %s partitions found on MTD device %s\n",
ret, parser->name, master->name);
}
put_partition_parser(parser);

I should note that I am assuming (because I don't understand how) that
request_module() requires net-pf-1 to succeed. This assumption is
based on my observation of the request_module message
("request_module: runaway loop modprobe net-pf-1"), and that
af_unix_init() (net-pf-1 initialization) is called only after the call
to request_module() in this example.

Based on the configuration and assumptions described above, the
sequence of events that cause the boot log messages above are:

1. the first call to get_partition_parser() in parse_mtd_partitions()
does not find the parser in the the list of registered parsers, i.e.
get_partition_parser() returns a NULL pointer. In this example, the
only registered parser in the list is called Redboot;

2. parse_mtd_partitions() calls request_module() with (in this
example) the parameter "cmdlinepart";

3. request_module() complains (request_module: runaway loop modprobe
net-pf-1, etc.) because af_unix_init() has not been called. The boot
log shows that af_unix_init() is called only after
parse_mtd_partitions() finishes.

register_mtd_parser: RedBoot
physmap platform flash device: 00080000 at f0000000
Found: ST M29W400DB
physmap-flash.0: Found 1 x16 devices at 0x0 in 16-bit bank
number of JEDEC chips: 1
cfi_cmdset_0002: Disabling erase-suspend-program due to code brokenness.
parse_mtd_partitions:
get_partition_parser: name=cmdlinepart
get_partition_parser: p->name=RedBoot
get_partition_parser: p->owner=00000000
get_partition_parser: ret=00000000
parse_mtd_partitions: cmdlinepart
request_module: runaway loop modprobe net-pf-1
modprobe: FATAL: Could not load
/lib/modules/2.6.25-rc4-iop32x/modules.dep: No such file or directory

modprobe: FATAL: Could not load
/lib/modules/2.6.25-rc4-iop32x/modules.dep: No such file or directory

modprobe: FATAL: Could not load
/lib/modules/2.6.25-rc4-iop32x/modules.dep: No such file or directory

<--- 47 repeated modprobe messages deleted --->
parse_mtd_partitions: request_module called
cmdlinepart partition parsing not available
get_partition_parser: name=RedBoot
get_partition_parser: p->name=RedBoot
get_partition_parser: p->owner=00000000
get_partition_parser: ret->name=RedBoot
get_partition_parser: ret=c02b0f8c
parse_mtd_partitions: RedBoot
parse_mtd_partitions: request_module called
Searching for RedBoot partition table in physmap-flash.0 at offset 0x70000
No RedBoot partition table detected in physmap-flash.0
parse_mtd_partitions: end of function
mice: PS/2 mouse device common for all mice
i2c /dev entries driver
rtc-rs5c372 0-0032: rs5c372a found, 24hr, driver version 0.5
rtc-rs5c372 0-0032: rtc core: registered rtc-rs5c372 as rtc0
iop-adma iop-adma.0: Intel(R) IOP: ( cpy intr )
iop-adma iop-adma.1: Intel(R) IOP: ( cpy intr )
NET: Registered protocol family 26
TCP bic registered
af_unix_init:
NET: Registered protocol family 1
NET: Registered protocol family 17
XScale DSP coprocessor detected.
registered taskstats version 1
rtc-rs5c372 0-0032: setting system clock to 2008-03-11 03:16:46 UTC (1205205406)

The reason parse_mtd_partitions() tries to use cmdlinepart, even
though it is not a registered parsing scheme when CONFIG_CMDLINE_PARTS
is not set, is because cmdlinepart is hard coded in
drivers/mtd/maps/physmap.c:

#ifdef CONFIG_MTD_PARTITIONS
static const char *part_probe_types[] = { "cmdlinepart", "RedBoot", NULL };
#endif

The patch below changes CONFIG_MTD_REDBOOT_PARTS,
CONFIG_MTD_AFS_PARTS, and CONFIG_MTD_OF_PARTS from tristate to bool,
and removes the call to request_module() in
drivers/mtd/mtdpart.c:parse_mtd_partitions(). The changes to the
definitions of the configuration variables ensure that MTD parsers
compiled into the kernel are registered before parse_mtd_partitions()
is called, which negates the need to have parse_mtd_partitions() call
request_module(), and thereby eliminating the call to request_module()
before af_unix_init(). The patch also changes the setting of
CONFIG_MTD_REDBOOT_PARTS in arch/mips/configs/mtx1_defconfig from 'm'
to 'y'. This defconfig is the only one that set any of the
partitioning schemes to be compiled as a module.

Signed-off-by: Gordon Farquharson <gordonfarquharson@xxxxxxxxx>

---

diff --git a/arch/mips/configs/mtx1_defconfig b/arch/mips/configs/mtx1_defconfig
index fa3aa39..ace51d3 100644
--- a/arch/mips/configs/mtx1_defconfig
+++ b/arch/mips/configs/mtx1_defconfig
@@ -743,7 +743,7 @@ CONFIG_MTD=m
# CONFIG_MTD_DEBUG is not set
CONFIG_MTD_CONCAT=m
CONFIG_MTD_PARTITIONS=y
-CONFIG_MTD_REDBOOT_PARTS=m
+CONFIG_MTD_REDBOOT_PARTS=y
CONFIG_MTD_REDBOOT_DIRECTORY_BLOCK=-1
# CONFIG_MTD_REDBOOT_PARTS_UNALLOCATED is not set
# CONFIG_MTD_REDBOOT_PARTS_READONLY is not set
diff --git a/drivers/mtd/Kconfig b/drivers/mtd/Kconfig
index e850334..20b4574 100644
--- a/drivers/mtd/Kconfig
+++ b/drivers/mtd/Kconfig
@@ -48,7 +48,7 @@ config MTD_PARTITIONS
'normal' form of partitioning used on a block device.

config MTD_REDBOOT_PARTS
- tristate "RedBoot partition table parsing"
+ bool "RedBoot partition table parsing"
depends on MTD_PARTITIONS
---help---
RedBoot is a ROM monitor and bootloader which deals with multiple
@@ -135,7 +135,7 @@ config MTD_CMDLINE_PARTS
If unsure, say 'N'.

config MTD_AFS_PARTS
- tristate "ARM Firmware Suite partition parsing"
+ bool "ARM Firmware Suite partition parsing"
depends on ARM && MTD_PARTITIONS
---help---
The ARM Firmware Suite allows the user to divide flash devices into
@@ -151,7 +151,7 @@ config MTD_AFS_PARTS
'armflash' map driver (CONFIG_MTD_ARMFLASH) does this, for example.

config MTD_OF_PARTS
- tristate "Flash partition map based on OF description"
+ bool "Flash partition map based on OF description"
depends on PPC_OF && MTD_PARTITIONS
help
This provides a partition parsing function which derives
diff --git a/drivers/mtd/mtdpart.c b/drivers/mtd/mtdpart.c
index c66902d..a3b3331 100644
--- a/drivers/mtd/mtdpart.c
+++ b/drivers/mtd/mtdpart.c
@@ -555,10 +555,6 @@ int parse_mtd_partitions(struct mtd_info *master,
const char **types,

for ( ; ret <= 0 && *types; types++) {
parser = get_partition_parser(*types);
-#ifdef CONFIG_KMOD
- if (!parser && !request_module("%s", *types))
- parser = get_partition_parser(*types);
-#endif
if (!parser) {
printk(KERN_NOTICE "%s partition parsing not available\n",
*types);

--
Gordon Farquharson
GnuPG Key ID: 32D6D676
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/