Skip to contentSkip to conversation

Kconfig bug (was: Re: fault-injection framework on ia64)

DISCUSSION Blaisorblade 2 participants 5 messages Feb 16, 2007 – Feb 17, 2007

lore
Share & tools
lore
#0 lore raw
Quoted 5 lines · Akinobu Mita
Quoted 5 lines

But if STACKTRACE depends on STACKTRACE_SUPPORT, which isn't defined on ia64, kconfig should disable FAULT_INJECTION altogether with my patch (as it did before it). I don't think that FAULT_INJECTION could be enabled before my patch, right?

Quoted 7 lines
Quoted 4 lines

_BUT_ probably STACKTRACE_SUPPORT should be defined, with default n and no prompt, on ia64, or Kconfig gets confused (it should output a warning, like 'undefined symbol STACKTRACE_SUPPORT'). We on UML used to get warning when INPUT was used but not defined. Hmmm....

Really? I intentionaly removed the STACKTRACE_SUPPORT in x86_64 Kconfig, then I did make menuconfig. But I didn't get such warning.

Well, I got that warning, and I don't get it on STACKTRACE_SUPPORT. Strange. But it was time ago...

However, more important, if I remove STACKTRACE_SUPPORT, or if I make it 'default n', FAULT_INJECTION can still be enabled, even if it selects STACKTRACE which has a failed dependency (tested on UML). Which is a Kconfig bug - if A selects B and B depends on C, no dependency of A on C is deduced. Right Roman?

This Kconfig bug explains how your bug could surface.

This also means that my patch cannot to go -stable because it would cause problems on IA64 and any arch without STACKTRACE_SUPPORT.
See if yours can, Akinobu.

Signature · 5 lines
Inform me of my mistakes, so I can add them to my list!
Paolo Giarrusso, aka Blaisorblade
http://www.user-mode-linux.org/~blaisorblade
Chiacchiera con i tuoi amici in tempo reale! 
 http://it.yahoo.com/mail_it/foot/*http://it.messenger.yahoo.com 
#1 lore raw

Hi,

Quoted 5 lines · Blaisorblade

However, more important, if I remove STACKTRACE_SUPPORT, or if I make it 'default n', FAULT_INJECTION can still be enabled, even if it selects STACKTRACE which has a failed dependency (tested on UML). Which is a Kconfig bug - if A selects B and B depends on C, no dependency of A on C is deduced. Right Roman?

Correct, but could someone please give me some more context, what the bug is supposed to be here?

bye, Roman

#2 lore raw
In reply to Blaisorblade
Quoted 11 lines · Roman Zippel

Hi,

Quoted 5 lines · Blaisorblade

However, more important, if I remove STACKTRACE_SUPPORT, or if I make it 'default n', FAULT_INJECTION can still be enabled, even if it selects STACKTRACE which has a failed dependency (tested on UML). Which is a Kconfig bug - if A selects B and B depends on C, no dependency of A on C is deduced. Right Roman?

Correct, but could someone please give me some more context, what the bug is supposed to be here?

Hmm.
You agree that "if A selects B and B depends on C, no dependency of A on C is deduced", right?

The implication is that the user can enable A, which selects B, while C is still disabled. The B -> C dependency is violated, and the build will fail or bad things will happen.

I hope things are now clearer, even without the description of the testcase where this happened.

Signature · 5 lines
Inform me of my mistakes, so I can add them to my list!
Paolo Giarrusso, aka Blaisorblade
http://www.user-mode-linux.org/~blaisorblade
Chiacchiera con i tuoi amici in tempo reale! 
 http://it.yahoo.com/mail_it/foot/*http://it.messenger.yahoo.com 
#3 lore raw
In reply to Blaisorblade

Hi,

Quoted 12 lines · Blaisorblade
Quoted 11 lines · Roman Zippel

Hi,

Quoted 5 lines · Blaisorblade

However, more important, if I remove STACKTRACE_SUPPORT, or if I make it 'default n', FAULT_INJECTION can still be enabled, even if it selects STACKTRACE which has a failed dependency (tested on UML). Which is a Kconfig bug - if A selects B and B depends on C, no dependency of A on C is deduced. Right Roman?

Correct, but could someone please give me some more context, what the bug is supposed to be here?

Quoted 7 lines

Hmm.
You agree that "if A selects B and B depends on C, no dependency of A on C is deduced", right?

The implication is that the user can enable A, which selects B, while C is still disabled. The B -> C dependency is violated, and the build will fail or bad things will happen.

Indeed, that's why in this case everything that selects STACKTRACE depends on STACKTRACE_SUPPORT.

bye, Roman

#4 lore raw
In reply to Blaisorblade
Quoted 20 lines · Roman Zippel

Hi,

Quoted 17 lines · Blaisorblade
Quoted 11 lines · Roman Zippel

Hi,

Quoted 5 lines · Blaisorblade

However, more important, if I remove STACKTRACE_SUPPORT, or if I make it 'default n', FAULT_INJECTION can still be enabled, even if it selects STACKTRACE which has a failed dependency (tested on UML). Which is a Kconfig bug - if A selects B and B depends on C, no dependency of A on C is deduced. Right Roman?

Correct, but could someone please give me some more context, what the bug is supposed to be here?

Hmm.
You agree that "if A selects B and B depends on C, no dependency of A on C is deduced", right?

Quoted 6 lines
Quoted 3 lines

The implication is that the user can enable A, which selects B, while C is still disabled. The B -> C dependency is violated, and the build will fail or bad things will happen.

Indeed, that's why in this case everything that selects STACKTRACE depends on STACKTRACE_SUPPORT.

Ah. Ok... so this 'bug' can be workarounded. But couldn't it be fixed or at least documented as pitfall? A depends on B is similar, from this point of view, to "A selects B", and they could be treated together for recursive dependency analisys.

Signature · 5 lines
Inform me of my mistakes, so I can add them to my list!
Paolo Giarrusso, aka Blaisorblade
http://www.user-mode-linux.org/~blaisorblade
Chiacchiera con i tuoi amici in tempo reale! 
 http://it.yahoo.com/mail_it/foot/*http://it.messenger.yahoo.com