Problems with 'mtd: warn when registering the same master many times'

From: Guenter Roeck
Date: Mon Nov 02 2015 - 10:41:58 EST


Brian,

I see the following warnings in recent mips qemu tests on linux-next.

WARNING: CPU: 0 PID: 1 at drivers/mtd/mtdcore.c:619 mtd_device_parse_register+0x160/0x16c()
MTD already registered

Looking into the code, this is the result of your patch 'mtd: warn when registering
the same master many times'.

This patch is supposed to warn if mtd_device_parse_register is called multiple times
with the same mtd_info structure pointer as parameter.

At the surface, the check appears to make sense. However, it has a problem.
The output of "git grep reboot_notifier.notifier_call" in drivers/mtd results in

chips/cfi_cmdset_0001.c: mtd->reboot_notifier.notifier_call = cfi_intelext_reboot;
chips/cfi_cmdset_0002.c: mtd->reboot_notifier.notifier_call = cfi_amdstd_reboot;
mtdcore.c: WARN_ONCE(mtd->reboot_notifier.notifier_call, "MTD already registered\n");
mtdcore.c: if (mtd->_reboot && !mtd->reboot_notifier.notifier_call) {
mtdcore.c: mtd->reboot_notifier.notifier_call = mtd_reboot_notifier;

As it turns out, the observed warning is not seen because mtd_device_parse_register()
is called multiple times. It is seen because mtd->reboot_notifier.notifier_call
is already set to cfi_intelext_reboot by the time it is checked.

I don't know if this is a bug in the mtd code, or if this is a valid use case.
Whatever it is, the warning does not warn about the driver, it warns about a potential
inconsistency in the mtd code itself (and if it is is a valid use case, the warning
is inappropriate). Maybe it would make sense to change the warning as follows ?

WARN_ONCE(mtd->_reboot && mtd->reboot_notifier.notifier_call, "MTD already registered\n");

I am not sure, though, if that is correct, since even that might be a valid use case
(a caller registering a cfi based mtd device with a _reboot callback).

I also see the warning in crisv32 runtime tests. This is because the code in
arch/cris/arch-v32/drivers/axisflashmap.c calls mtd_device_register() multiple times
with the same mtd_info argument, each time to register a different partition.
I am not sure if the check is appropriate for this case either, since the code calls
mtd_device_register(), both 'type' and 'parser_data' are NULL, parse_mtd_partitions()
does not do anything, and the problem you are concerned about does not apply.

How about changing the warning to something like the following ?

WARN_ONCE(types && mtd->_reboot && mtd->reboot_notifier.notifier_call, "MTD already registered\n");

This would eliminate (what I think are) false positives and only warn if there
is a real problem.

Thanks,
Guenter
--
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/