Skip to contentSkip to conversation

scsi: megaraid_mbox: return -ENOMEM on megaraid_init_mbox() allocation failure

PATCH In review Jiapeng Chong 3 participants 5 messages Oct 19, 2021 – Oct 20, 2021

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

PATCH 1/1 scsi: megaraid_mbox: return -ENOMEM on megaraid_init_mbox() allocation failure

From: chongjiapeng <[email protected]>

Fixes the following smatch warning:

drivers/scsi/megaraid/megaraid_mbox.c:715 megaraid_init_mbox() warn:
returning -1 instead of -ENOMEM is sloppy.

Fixes dd00cc486ab1 ("some kmalloc/memset ->kzalloc (tree wide)")
---
 drivers/scsi/megaraid/megaraid_mbox.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/scsi/megaraid/megaraid_mbox.c +2 -1
index 14f930d27ca1..d98b223eab9a 100644
@@ -712,7 +712,8 @@ megaraid_init_mbox(adapter_t *adapter)
 	 * controllers
 	 */
 	raid_dev = kzalloc(sizeof(mraid_device_t), GFP_KERNEL);
-	if (raid_dev == NULL) return -1;
+	if (raid_dev == NULL)
+		return -ENOMEM;
 
 
 	/*
Signature · 1 line
2.19.1.6.gb485710b
#1 lore raw
Quoted 6 lines · Jiapeng Chong

From: chongjiapeng <[email protected]>

Fixes the following smatch warning:

drivers/scsi/megaraid/megaraid_mbox.c:715 megaraid_init_mbox() warn:
returning -1 instead of -ENOMEM is sloppy.

Why is this a problem? megaraid_init_mbox() is called using this pattern:

	// Start the mailbox based controller
	if (megaraid_init_mbox(adapter) != 0) {
		con_log(CL_ANN, (KERN_WARNING
			"megaraid: mailbox adapter did not initialize\n"));

		goto out_free_adapter;
	}

So the only meaningful returns are 0 on success and anything else
(although megaraid uses -1 for this) on failure. Since -1 is the
conventional failure return, why alter that to something different that still won't be printed or acted on? And worse still, if we make this change, it will likely excite other static checkers to complain we're losing error information ...

James

#2 lore raw
Quoted 19 lines · James Bottomley
Quoted 6 lines · Jiapeng Chong

From: chongjiapeng <[email protected]>

Fixes the following smatch warning:

drivers/scsi/megaraid/megaraid_mbox.c:715 megaraid_init_mbox() warn:
returning -1 instead of -ENOMEM is sloppy.

Why is this a problem? megaraid_init_mbox() is called using this pattern:

	// Start the mailbox based controller
	if (megaraid_init_mbox(adapter) != 0) {
		con_log(CL_ANN, (KERN_WARNING
			"megaraid: mailbox adapter did not initialize\n"));

		goto out_free_adapter;
	}
Quoted 2 lines

So the only meaningful returns are 0 on success and anything else (although megaraid uses -1 for this) on failure.

I think you're arguing for a bool (?)

Smatch apparently did not think of that -- probably needs a holiday.

Quoted 5 lines

Since -1 is the conventional failure return, why alter that to something different that still won't be printed or acted on? And worse still, if we make this change, it will likely excite other static checkers to complain we're losing error information ...

... and arguably they would be correct.

#3 lore raw
Quoted 20 lines · Finn Thain
Quoted 18 lines · James Bottomley
Quoted 7 lines · Jiapeng Chong

From: chongjiapeng <[email protected]>

Fixes the following smatch warning:

drivers/scsi/megaraid/megaraid_mbox.c:715 megaraid_init_mbox() warn:
returning -1 instead of -ENOMEM is sloppy.

Why is this a problem? megaraid_init_mbox() is called using this pattern:

	// Start the mailbox based controller
	if (megaraid_init_mbox(adapter) != 0) {
		con_log(CL_ANN, (KERN_WARNING
			"megaraid: mailbox adapter did not

initialize\n"));

Quoted 7 lines
Quoted 5 lines
		goto out_free_adapter;
	}

So the only meaningful returns are 0 on success and anything else (although megaraid uses -1 for this) on failure.

I think you're arguing for a bool (?)

I'm not arguing for anything ... I'm just explaining how the current code works that makes this change pointless. Megaraid is an older driver, so even if the current return is two state, changing it to bool would be unnecessary churn.

Quoted 9 lines

Smatch apparently did not think of that -- probably needs a holiday.

Quoted 5 lines

Since -1 is the conventional failure return, why alter that to something different that still won't be printed or acted on? And worse still, if we make this change, it will likely excite other static checkers to complain we're losing error information ...

... and arguably they would be correct.

Well, yes ... that's why I don't want one "fix" that generates a cascading sequence of further "fixes".

James

#4 lore raw
Quoted 6 lines · James Bottomley
Quoted 2 lines

... and arguably they would be correct.

Well, yes ... that's why I don't want one "fix" that generates a cascading sequence of further "fixes".

OTOH, if you don't "fix" it, it generates a cascading sequence of copy-and-paste antipatterns in new code, and poor training data for those of us reading old code.

Anyway, I agree that the churn would be too risky. But it sure would be nice if automatic tools were able to perform a program transformation of this kind at the source level, being that the compiler will surely do it anyway at a lower level.

There's a lot to be said for source code that reflects the compiler's understanding of the logic, rather than the human's.