Skip to contentSkip to conversation

powerpc: fix register usage in some inline assembly code

PATCH New Timur Tabi 4 participants 6 messages Dec 14, 2010 – Dec 17, 2010 PowerPC

lore
Share & tools
lore

Timur Tabi

author
#0 lore raw

PATCH 1/1 powerpc: fix register usage in some inline assembly code

In some usages of inline assembly, hard-coded registers were specified when a scratch register should have been used instead.

---

I can't test the kgdb changes, but the rest appears to be okay.

 arch/powerpc/lib/kgdb.c |   25 +++++++++++++++----------
 arch/powerpc/lib/time.c |    5 ++++-
 2 files changed, 19 insertions(+), 11 deletions(-)
arch/powerpc/lib/kgdb.c +15 -10
index 1ec6818..19a56db 100644
@@ -12,11 +12,13 @@ void breakinst(void);
 int
 kgdb_setjmp(long *buf)
 {
-	asm ("mflr 0; stw 0,0(%0);"
-	     "stw 1,4(%0); stw 2,8(%0);"
-	     "mfcr 0; stw 0,12(%0);"
-	     "stmw 13,16(%0)"
-	     : : "r" (buf));
+	unsigned long temp;
+
+	asm volatile("mflr %0; stw %0,0(%1);"
+	     "stw %%r1,4(%1); stw %%r2,8(%1);"
+	     "mfcr %0; stw %0,12(%1);"
+	     "stmw %%r13,16(%1)"
+	     : "=&r"(temp) : "r" (buf));
 	/* XXX should save fp regs as well */
 	return 0;
 }
@@ -24,13 +26,16 @@ kgdb_setjmp(long *buf)
 void
 kgdb_longjmp(long *buf, int val)
 {
+	unsigned long temp;
+
 	if (val == 0)
 		val = 1;
-	asm ("lmw 13,16(%0);"
-	     "lwz 0,12(%0); mtcrf 0x38,0;"
-	     "lwz 0,0(%0); lwz 1,4(%0); lwz 2,8(%0);"
-	     "mtlr 0; mr 3,%1"
-	     : : "r" (buf), "r" (val));
+
+	asm volatile("lmw %%r13,16(%1);"
+	     "lwz %0,12(%1); mtcrf 0x38,%0;"
+	     "lwz %0,0(%1); lwz %%r1,4(%1); lwz %%r2,8(%1);"
+	     "mtlr %0; mr %%r3,%2"
+	     : "=&r"(temp) : "r" (buf), "r" (val));
 }
 
 static inline unsigned long
arch/powerpc/lib/time.c +4 -1
index 2909961..34633c3 100644
@@ -78,6 +78,8 @@ unsigned long ticks2usec(unsigned long ticks)
 
 int init_timebase (void)
 {
+	unsigned long temp;
+
 #if defined(CONFIG_5xx) || defined(CONFIG_8xx)
 	volatile immap_t *immap = (immap_t *) CONFIG_SYS_IMMR;
 
@@ -86,7 +88,8 @@ int init_timebase (void)
 #endif
 
 	/* reset */
-	asm ("li 3,0 ; mttbu 3 ; mttbl 3 ;");
+	asm volatile("li %0,0 ; mttbu %0 ; mttbl %0;"
+	     : "=&r"(temp) );
 
 #if defined(CONFIG_5xx) || defined(CONFIG_8xx)
 	/* enable */
Signature · 1 line
1.7.2.3
#1 lore raw

PATCH 1/1 powerpc: fix register usage in some inline assembly code

Dear Timur Tabi,

Quoted 2 lines · In message <[email protected]> you wrote:

In some usages of inline assembly, hard-coded registers were specified when a scratch register should have been used instead.

Is this just a cosmetic improvement, or are you fixing any real bug? If so, what exactly is the problem?

Best regards,

Wolfgang Denk

Signature · 6 lines
DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd at denx.de
Conceptual integrity in turn dictates that the  design  must  proceed
from  one  mind,  or  from  a  very small number of agreeing resonant
minds.               - Frederick Brooks Jr., "The Mythical Man Month"
#2 lore raw

PATCH 1/1 powerpc: fix register usage in some inline assembly code

Quoted 6 lines · Wolfgang Denk
Quoted 2 lines · In message<[email protected]> you wrote:
Quoted 2 lines

In some usages of inline assembly, hard-coded registers were specified when a scratch register should have been used instead.

Is this just a cosmetic improvement, or are you fixing any real bug? If so, what exactly is the problem?

Well, I'm not a expert on inline assembly, so I could be missing something, but IMHO it is fixing a real bug, although the current code does technically work.

The problem is that the code was modifying registers and gcc was unaware of that. So there is a possibility that gcc could have put something important in those registers. It appears that the authors of that code knew which registers were unused, and just made sure to use those.

Signature · 2 lines
Timur Tabi
Linux kernel developer
#3 lore raw

PATCH 1/1 powerpc: fix register usage in some inline assembly code

Quoted 11 lines · On Dec 15, 2010, at 6:32 AM, Tabi Timur-B04825 wrote:
Quoted 6 lines · Wolfgang Denk
Quoted 2 lines · In message<[email protected]> you wrote:
Quoted 2 lines

In some usages of inline assembly, hard-coded registers were specified when a scratch register should have been used instead.

Is this just a cosmetic improvement, or are you fixing any real bug? If so, what exactly is the problem?

Well, I'm not a expert on inline assembly, so I could be missing something, but IMHO it is fixing a real bug, although the current code does technically work.

The problem is that the code was modifying registers and gcc was unaware of that. So there is a possibility that gcc could have put something important in those registers. It appears that the authors of that code knew which registers were unused, and just made sure to use those.

I doubt we disagree that the change is a good one, just trying to understand if it was because of an actual issue you where seeing or just because of some code review that you made it.

ie, is this needed in v2010.12 or goes into 'next'

- k

author
#4 lore raw

PATCH 1/1 powerpc: fix register usage in some inline assembly code

Quoted 3 lines · Kumar Gala

I doubt we disagree that the change is a good one, just trying to understand if it was because of an actual issue you where seeing or just because of some code review that you made it.

Just a code review.

Quoted 2 lines

ie, is this needed in v2010.12 or goes into 'next'

Definitely 'next'.

Signature · 2 lines
Timur Tabi
Linux kernel developer at Freescale
#5 lore raw

PATCH 1/1 powerpc: fix register usage in some inline assembly code

In reply to Timur Tabi

Dear Timur Tabi,

Quoted 11 lines · In message <[email protected]> you wrote:

In some usages of inline assembly, hard-coded registers were specified when a scratch register should have been used instead.

---

I can't test the kgdb changes, but the rest appears to be okay.

 arch/powerpc/lib/kgdb.c |   25 +++++++++++++++----------
 arch/powerpc/lib/time.c |    5 ++++-
 2 files changed, 19 insertions(+), 11 deletions(-)

Applied to "next", thanks.

Best regards,

Wolfgang Denk

Signature · 6 lines
DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd at denx.de
Every program has at least one bug and can be shortened by  at  least
one  instruction  --  from  which,  by induction, one can deduce that
every program can be reduced to one instruction which doesn't work.