671711 - cc is not a register

Status: RESOLVED FIXED

(NSS :: Build, defect, P1)

Description

[Rafael Ávila de Espíndola (:espindola) (not reading bugmail)]

gcc accepts

  __asm__ ( "" : : : "%cc");

or
  __asm__ ( "" : : : "%memory");

but I couldn't find any documentation support for using % in here. It is only used for registers. Clang currently rejects those constructs.

Comment 1

[Rafael Ávila de Espíndola (:espindola) (not reading bugmail)]

Attachment: patch I am testing

Comment 2

[Rafael Ávila de Espíndola (:espindola) (not reading bugmail)]

With this patch a build of m-c with clang just finished successfully.

Comment 3

[(no longer active)]

This is a patch that we want to have on m-c as soon as possible to prevent build bustages.  If we can't do an NSS merge soon, we need to take this as a spot fix on m-c once it gets reviewed...

Comment 4

[Rafael Ávila de Espíndola (:espindola) (not reading bugmail)]

Should anyone else be cced on this? Is wtc the right person to review the patch? This bug is once again the only blocker to building with clang.

Comment 5

[Wan-Teh Chang]

Comment on attachment 546050 [details] [diff] [review]
patch I am testing

Rafael: thank you for the patch.  I am not familiar with GCC inline assembly syntax,
so you need to explain why the change is correct.

For example, if cc is not a register (the Summary of this bug report), then what is cc?
Searching for "cc" in the file
http://mxr.mozilla.org/security/source/security/nss/lib/freebl/mpi/mp_comba.c,
I found no occurrences of "cc" except in these macros.

A Bing search showed that "cc" refers to the condition code register (or condition
code flags) and can be used in the clobber list.  Is that correct?

Comment 6

[Rafael Ávila de Espíndola (:espindola) (not reading bugmail)]

It is used to represent the condition code. The only "documentation" I could find is a comment in varasm.c:

/* Decode an `asm' spec for a declaration as a register name.
   Return the register number, or -1 if nothing specified,
   or -2 if the ASMSPEC is not `cc' or `memory' and is not recognized,
   or -3 if ASMSPEC is `cc' and is not recognized,
   or -4 if ASMSPEC is `memory' and is not recognized.
   Accept an exact spelling or a decimal number.
   Prefixes such as % are optional.  */

I could not find exactly where in gcc -3 is used.

Comment 7

[Eric Christopher]

The % syntax nominally goes before a register, in this case cc, like "memory" isn't a true register it's just a way of specifying that the condition code registers for the target are clobbered. If you had a target with multiple condition code registers like ARM you wouldn't want to have to explicitly list each register. Therefore it isn't really a register that you'd prefix with % like you do the rest of the x86 registers. Also, for example, again on ARM you wouldn't really want to preface them with % at all there either.

Comment 8

[Christian :Biesinger (don't email me, ping me on IRC)]

cc'ing wtc so he sees the above comments

Comment 9

[Wan-Teh Chang]

Comment on attachment 546050 [details] [diff] [review]
patch I am testing

r=wtc.  Thank you for the explanations.

Patch checked in on the NSS trunk (NSS 3.13) and
NSS_3_12_BRANCH (NSS 3.12.11).

Checking in mp_comba.c;
/cvsroot/mozilla/security/nss/lib/freebl/mpi/mp_comba.c,v  <--  mp_comba.c
new revision: 1.4; previous revision: 1.3
done

Checking in mp_comba.c;
/cvsroot/mozilla/security/nss/lib/freebl/mpi/mp_comba.c,v  <--  mp_comba.c
new revision: 1.3.56.1; previous revision: 1.3
done

Comment 10

[Rafael Ávila de Espíndola (:espindola) (not reading bugmail)]

Can/should we backport this to m-c? When is the next nss merge planned to happen?

Comment 11

[Rafael Ávila de Espíndola (:espindola) (not reading bugmail)]

Ping. Can we backport this?

Comment 12

[Wan-Teh Chang]

Rafael: please ask Brian Smith about the schedule of the NSS 3.12.11 release.

Comment 13

[Rafael Ávila de Espíndola (:espindola) (not reading bugmail)]

Brian, what is the schedule for the 3.12.11 release?

This bug currently blocks building with clang. I have hacked the version of clang we have on the bots, but having to hack the compiler instead of fixing our own code is not the ideal situation.

Comment 14

[Rafael Ávila de Espíndola (:espindola) (not reading bugmail)]

This will be fixed when it lands in m-c