Skip to contentSkip to conversation

libsrp: fix compile failure

PATCH In review James Bottomley 6 participants 12 messages Dec 30, 2009 – Jan 4, 2010 SCSI

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 libsrp: fix compile failure

commit 45465487897a1c6d508b14b904dc5777f7ec7e04

Author: Stefani Seibold <[email protected]>
Date:   Mon Dec 21 14:37:26 2009 -0800

kfifo: move struct kfifo in place

Caused a compile failure in ibmvscsitgt.c because it changed a pointer to kfifo in the libsrp.h structure to a direct inclusion without including <linux/kfifo.h>. The fix is simple, just add the include, but how did this happen? This change, introduced at -rc2, hardly looks like a bug fix, and it clearly didn't go through linux-next, which would have picked up this compile failure (it only occurs on ppc because of the ibm virtual scsi target).

James

---

include/scsi/libsrp.h +1 -0
index 07e3add..f4105c9 100644
@@ -2,6 +2,7 @@
 #define __LIBSRP_H__
 
 #include <linux/list.h>
+#include <linux/kfifo.h>
 #include <scsi/scsi_cmnd.h>
 #include <scsi/scsi_host.h>
 #include <scsi/srp.h>
#1 lore raw
Quoted 5 lines · James Bottomley

The fix is simple, just add the include, but how did this happen? This change, introduced at -rc2, hardly looks like a bug fix, and it clearly didn't go through linux-next, which would have picked up this compile failure (it only occurs on ppc because of the ibm virtual scsi target).

It came through Andrew - and apparently parts of Andrews chain weren't in next. Don't know why.

Quoted 1 line

James

Sign-off?

Linus

#2 lore raw
Quoted 8 lines · Linus Torvalds
Quoted 5 lines · James Bottomley

The fix is simple, just add the include, but how did this happen? This change, introduced at -rc2, hardly looks like a bug fix, and it clearly didn't go through linux-next, which would have picked up this compile failure (it only occurs on ppc because of the ibm virtual scsi target).

Quoted 6 lines

It came through Andrew - and apparently parts of Andrews chain weren't in next. Don't know why.

Quoted 1 line

James

Sign-off?

Sorry, forgot ...

James

#3 lore raw
In reply to Linus Torvalds
Quoted 7 lines · Linus Torvalds
Quoted 4 lines

The fix is simple, just add the include, but how did this happen? This change, introduced at -rc2, hardly looks like a bug fix, and it clearly didn't go through linux-next, which would have picked up this compile failure (it only occurs on ppc because of the ibm virtual scsi target).

It came through Andrew - and apparently parts of Andrews chain weren't in next. Don't know why.

Uhm ... are they supposed to be? -mm is being built on top of -next, not vice versa, right?

Signature · 2 lines
Jiri Kosina
SUSE Labs, Novell Inc.
#4 lore raw
Quoted 7 lines · Jiri Kosina
Quoted 5 lines · Linus Torvalds
Quoted 4 lines

The fix is simple, just add the include, but how did this happen? This change, introduced at -rc2, hardly looks like a bug fix, and it clearly didn't go through linux-next, which would have picked up this compile failure (it only occurs on ppc because of the ibm virtual scsi target).

Quoted 5 lines
Quoted 2 lines

It came through Andrew - and apparently parts of Andrews chain weren't in next. Don't know why.

Uhm ... are they supposed to be? -mm is being built on top of -next, not vice versa, right?

Well, the fact that the compile failure wasn't detected before it went upstream should answer that ...

But to be more specific: linux-next is our integration tree (and also the obscure architecture compile tree). To ensure the best possible integration, every tree should be built and tested in linux-next at least once before it goes to Linus. There were originally technical reasons why -mm wasn't in ... I just thought they'd been fixed by now.

James

#5 lore raw
Quoted 2 lines

least once before it goes to Linus. There were originally technical reasons why -mm wasn't in ... I just thought they'd been fixed by now.

No - mm also caused problems with the kfifo merge of a core API change that didn't go via -next.

Alan

#6 lore raw

Am Montag, den 04.01.2010, 17:35 +0000 schrieb Alan Cox:

Quoted 7 lines
Quoted 2 lines

least once before it goes to Linus. There were originally technical reasons why -mm wasn't in ... I just thought they'd been fixed by now.

No - mm also caused problems with the kfifo merge of a core API change that didn't go via -next.

Alan

I tried my best to port everything to the new kfifo API. But it was not possible to check it against every architecture. x86 and x86_64 was compile clean. I think the missing #include was not a big thing.

Stefani

#7 lore raw
Quoted 11 lines · Stefani Seibold

Am Montag, den 04.01.2010, 17:35 +0000 schrieb Alan Cox:

Quoted 7 lines
Quoted 2 lines

least once before it goes to Linus. There were originally technical reasons why -mm wasn't in ... I just thought they'd been fixed by now.

No - mm also caused problems with the kfifo merge of a core API change that didn't go via -next.

Alan

I tried my best to port everything to the new kfifo API. But it was not possible to check it against every architecture.

That's what linux-next is for: to make at least this compile checking against all of our architectures more of a reality. It's still missing bits, and some of the ports (*cough* parisc *cough*) compile against linux-next rather infrequently, but it's far better than nothing.

Quoted 2 lines

x86 and x86_64 was
compile clean. I think the missing #include was not a big thing.

It's easily fixable, yes ... breaking the build *is* a pretty big thing because of the fallout it causes (for ppc: about ~100 instant WTF moments followed by grubbing around in git to find the cause).

It's also an indicator of future problems. Realistically, all API changes should go into linux-next both to prevent this sort of thing from happening and to see if we have any other pending conflicts that may cause problems.

James

#8 lore raw
In reply to Stefani Seibold
Quoted 4 lines · On Mon, 04 Jan 2010 20:24:14 +0100 Stefani Seibold <[email protected]> wrote:

Am Montag, den 04.01.2010, 17:35 +0000 schrieb Alan Cox:

Quoted 3 lines
Quoted 2 lines

least once before it goes to Linus. There were originally technical reasons why -mm wasn't in ... I just thought they'd been fixed by now.

Quoted 8 lines
Quoted 4 lines

No - mm also caused problems with the kfifo merge of a core API change that didn't go via -next.

Alan

I tried my best to port everything to the new kfifo API. But it was not possible to check it against every architecture. x86 and x86_64 was compile clean. I think the missing #include was not a big thing.

Not your fault - -next is there so stuff is visible for merging and to avoid collisions. Andrew not going via -next broke various pending patches.

Conversion is easy enough, it just wasn't visible at the right time.

#9 lore raw
In reply to James Bottomley

[ added Stephen to CC ]

Quoted 8 lines · James Bottomley

Well, the fact that the compile failure wasn't detected before it went upstream should answer that ...

But to be more specific: linux-next is our integration tree (and also the obscure architecture compile tree). To ensure the best possible integration, every tree should be built and tested in linux-next at least once before it goes to Linus. There were originally technical reasons why -mm wasn't in ... I just thought they'd been fixed by now.

/me checks ...

Yes, it indeed is that way -- Andew pulls whole linux-next as one of the patches into -mm series.

To make linux-next really working the way it is intended to work we need to have -mm part of it, as it is used as a last point for a non-trivial amount of patches before they enter Linus' tree.

Andrew, why do we have the current setup, and not the other way around?

Thanks,

Signature · 2 lines
Jiri Kosina
SUSE Labs, Novell Inc.
#10 lore raw
Quoted 16 lines · On Mon, 4 Jan 2010 22:12:26 +0100 (CET) Jiri Kosina <[email protected]> wrote:

[ added Stephen to CC ]

Quoted 8 lines · James Bottomley

Well, the fact that the compile failure wasn't detected before it went upstream should answer that ...

But to be more specific: linux-next is our integration tree (and also the obscure architecture compile tree). To ensure the best possible integration, every tree should be built and tested in linux-next at least once before it goes to Linus. There were originally technical reasons why -mm wasn't in ... I just thought they'd been fixed by now.

/me checks ...

Quoted 9 lines

Yes, it indeed is that way -- Andew pulls whole linux-next as one of the patches into -mm series.

To make linux-next really working the way it is intended to work we need to have -mm part of it, as it is used as a last point for a non-trivial amount of patches before they enter Linus' tree.

Andrew, why do we have the current setup, and not the other way around?

Because I suck. I haven't yet got around to feeding -mm into linux-next. It's a bit tricky, because -mm is based on linux-next. Probably we'll address this by adding a "linux-next before the mm bits" marker to linux-next.

#11 lore raw
Quoted 7 lines · Andrew Morton
Quoted 7 lines

Yes, it indeed is that way -- Andew pulls whole linux-next as one of the patches into -mm series.

To make linux-next really working the way it is intended to work we need to have -mm part of it, as it is used as a last point for a non-trivial amount of patches before they enter Linus' tree.

Quoted 6 lines
Quoted 1 line

Andrew, why do we have the current setup, and not the other way around?

Because I suck. I haven't yet got around to feeding -mm into linux-next. It's a bit tricky, because -mm is based on linux-next. Probably we'll address this by adding a "linux-next before the mm bits" marker to linux-next.

Yup. That should work and would be really helpful.

Signature · 2 lines
Jiri Kosina
SUSE Labs, Novell Inc.