Skip to contentSkip to conversation

printf: fix Woverride-init warning for EDEADLK errno

PATCH In review Arnd Bergmann 5 participants 9 messages Oct 26, 2020 – Oct 27, 2020 Library code

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 printf: fix Woverride-init warning for EDEADLK errno

From: Arnd Bergmann <[email protected]>

On most architectures, gcc -Wextra warns about the list of error numbers containing both EDEADLK and EDEADLOCK:

lib/errname.c:15:67: warning: initialized field overwritten [-Woverride-init]
15 | #define E(err) [err + BUILD_BUG_ON_ZERO(err <= 0 || err > 300)] = "-" #err
| ^~~
lib/errname.c:172:2: note: in expansion of macro 'E'

  172 |  E(EDEADLK), /* EDEADLOCK */
      |  ^

lib/errname.c:15:67: note: (near initialization for 'names_0[35]')
15 | #define E(err) [err + BUILD_BUG_ON_ZERO(err <= 0 || err > 300)] = "-" #err
| ^~~
lib/errname.c:172:2: note: in expansion of macro 'E'

  172 |  E(EDEADLK), /* EDEADLOCK */
      |  ^

Make that line conditional on the two values being distinct.

Fixes 57f5677e535b ("printf: add support for printing symbolic error names")
---
 lib/errname.c | 2 ++
 1 file changed, 2 insertions(+)
lib/errname.c +2 -0
index 0c4d3e66170e..6adff0bf2445 100644
@@ -169,7 +169,9 @@ static const char *names_0[] = {
 	E(ECANCELED), /* ECANCELLED */
 	E(EAGAIN), /* EWOULDBLOCK */
 	E(ECONNREFUSED), /* EREFUSED */
+#if EDEADLK != EDEADLOCK
 	E(EDEADLK), /* EDEADLOCK */
+#endif
 };
 #undef E
 
Signature · 1 line
2.27.0
#1 lore raw

Good morning Arnd,

Quoted 9 lines · Arnd Bergmann

From: Arnd Bergmann <[email protected]>

On most architectures, gcc -Wextra warns about the list of error numbers containing both EDEADLK and EDEADLOCK:

lib/errname.c:15:67: warning: initialized field overwritten [-Woverride-init]
15 | #define E(err) [err + BUILD_BUG_ON_ZERO(err <= 0 || err > 300)] = "-" #err
| ^~~
lib/errname.c:172:2: note: in expansion of macro 'E'

Quoted 8 lines
  172 |  E(EDEADLK), /* EDEADLOCK */
      |  ^

lib/errname.c:15:67: note: (near initialization for 'names_0[35]')
15 | #define E(err) [err + BUILD_BUG_ON_ZERO(err <= 0 || err > 300)] = "-" #err
| ^~~
lib/errname.c:172:2: note: in expansion of macro 'E'

  172 |  E(EDEADLK), /* EDEADLOCK */
      |  ^

bad performance of gcc to warn twice about the same line and not mentioning the line that has E(EDEADLOCK).

Quoted 13 lines

Make that line conditional on the two values being distinct.

Fixes 57f5677e535b ("printf: add support for printing symbolic error names")
---
 lib/errname.c | 2 ++
 1 file changed, 2 insertions(+)
lib/errname.c +0 -0
index 0c4d3e66170e..6adff0bf2445 100644

lib/errname.c static const char *names_0[] = { · line 174

Quoted 6 lines
lib/errname.c +2 -0
@@ -169,7 +169,9 @@ static const char *names_0[] = {
 	E(ECANCELED), /* ECANCELLED */
 	E(EAGAIN), /* EWOULDBLOCK */
 	E(ECONNREFUSED), /* EREFUSED */
+#if EDEADLK != EDEADLOCK
 	E(EDEADLK), /* EDEADLOCK */
+#endif

The comments suggest that duplicates are expected. Would it make sense to add similar conditions to the other three entries?

Best regards
Uwe

Quoted 3 lines
lib/errname.c +0 -0
 };
 #undef E
 
#3 lore raw
In reply to Arnd Bergmann
Quoted 11 lines · Arnd Bergmann

From: Arnd Bergmann <[email protected]>

On most architectures, gcc -Wextra warns about the list of error numbers containing both EDEADLK and EDEADLOCK:

lib/errname.c:15:67: warning: initialized field overwritten [-Woverride-init]
15 | #define E(err) [err + BUILD_BUG_ON_ZERO(err <= 0 || err > 300)] = "-" #err
| ^~~
lib/errname.c:172:2: note: in expansion of macro 'E'

  172 |  E(EDEADLK), /* EDEADLOCK */
      |  ^
Quoted 9 lines

lib/errname.c:15:67: note: (near initialization for 'names_0[35]')
15 | #define E(err) [err + BUILD_BUG_ON_ZERO(err <= 0 || err > 300)] = "-" #err
| ^~~
lib/errname.c:172:2: note: in expansion of macro 'E'

  172 |  E(EDEADLK), /* EDEADLOCK */
      |  ^

Make that line conditional on the two values being distinct.

NAK. That would end up using the "EDEADLOCK" string for the value 35 on those architectures where they are the same, despite EDEADLK being the by far the most used symbol. See the comments and original commit log, the placement of these is deliberate.

How about we do this instead?

From: Rasmus Villemoes <[email protected]>

The table of errno value->name contains a few duplicate entries since e.g. EDEADLK == EDEADLOCK on most architectures. For the known cases, the most used symbolic constant is listed last so that takes precedence - the idea being that if someone sees "can't do that:
-EDEADLK" in dmesg, grepping for EDEADLK is more likely to find the place where that error was generated (grepping for "can't do that" will find the printk() that emitted it, but the source would often be a few calls down).

However, that means one gets

warning: initialized field overwritten [-Woverride-init]

when building with W=1. As the use of multiple initializers for the same entry here is quite deliberate, explicitly disable that warning for errname.o.

Fixes 57f5677e535b ("printf: add support for printing symbolic error

names")

---
 lib/Makefile | 1 +
 1 file changed, 1 insertion(+)
lib/Makefile +1 -0
index ce45af50983a2a5e3582..a98119519e100103818d 100644
@@ -224,6 +224,7 @@ obj-$(CONFIG_HAVE_ARCH_TRACEHOOK) += syscall.o
 
 obj-$(CONFIG_DYNAMIC_DEBUG_CORE) += dynamic_debug.o
 obj-$(CONFIG_SYMBOLIC_ERRNAME) += errname.o
+CFLAGS_errname.o += $(call cc-disable-warning, override-init)
 
 obj-$(CONFIG_NLATTR) += nlattr.o
 
Signature · 1 line
2.23.0
#4 lore raw
Quoted 3 lines · Rasmus Villemoes

How about we do this instead?

From: Rasmus Villemoes <[email protected]>

Missed:

Subject: lib/errname.o: disable -Woverride-init

Quoted 38 lines

The table of errno value->name contains a few duplicate entries since e.g. EDEADLK == EDEADLOCK on most architectures. For the known cases, the most used symbolic constant is listed last so that takes precedence - the idea being that if someone sees "can't do that:
-EDEADLK" in dmesg, grepping for EDEADLK is more likely to find the place where that error was generated (grepping for "can't do that" will find the printk() that emitted it, but the source would often be a few calls down).

However, that means one gets

warning: initialized field overwritten [-Woverride-init]

when building with W=1. As the use of multiple initializers for the same entry here is quite deliberate, explicitly disable that warning for errname.o.

Fixes 57f5677e535b ("printf: add support for printing symbolic error

names")

---
 lib/Makefile | 1 +
 1 file changed, 1 insertion(+)
lib/Makefile +1 -0
index ce45af50983a2a5e3582..a98119519e100103818d 100644
@@ -224,6 +224,7 @@ obj-$(CONFIG_HAVE_ARCH_TRACEHOOK) += syscall.o
 
 obj-$(CONFIG_DYNAMIC_DEBUG_CORE) += dynamic_debug.o
 obj-$(CONFIG_SYMBOLIC_ERRNAME) += errname.o
+CFLAGS_errname.o += $(call cc-disable-warning, override-init)
 
 obj-$(CONFIG_NLATTR) += nlattr.o
 
Signature · 8 lines
Rasmus Villemoes
Software Developer
Prevas A/S
Hedeager 3
DK-8200 Aarhus N
+45 51210274
[email protected]
www.prevas.dk
#2 lore raw
In reply to Uwe Kleine-König
Quoted 9 lines · Uwe Kleine-König
Quoted 8 lines · Arnd Bergmann

Make that line conditional on the two values being distinct.

Fixes 57f5677e535b ("printf: add support for printing symbolic error names")
---
 lib/errname.c | 2 ++
 1 file changed, 2 insertions(+)
Quoted 14 lines

lib/errname.c static const char *names_0[] = { · line 174

Quoted 11 lines
lib/errname.c +2 -0
index 0c4d3e66170e..6adff0bf2445 100644
@@ -169,7 +169,9 @@ static const char *names_0[] = {
      E(ECANCELED), /* ECANCELLED */
      E(EAGAIN), /* EWOULDBLOCK */
      E(ECONNREFUSED), /* EREFUSED */
+#if EDEADLK != EDEADLOCK
      E(EDEADLK), /* EDEADLOCK */
+#endif

The comments suggest that duplicates are expected. Would it make sense to add similar conditions to the other three entries?

The other ones are always aliases, so there is no point in having an #ifdef. The reason we need to handle EDEADLK separately is that it's an alias on most architectures but not on on others, specifically mips, powerpc and sparc.

Arnd

#5 lore raw
In reply to Rasmus Villemoes
Quoted 7 lines · Rasmus Villemoes
Quoted 1 line · Arnd Bergmann

From: Arnd Bergmann <[email protected]>

NAK. That would end up using the "EDEADLOCK" string for the value 35 on those architectures where they are the same, despite EDEADLK being the by far the most used symbol. See the comments and original commit log, the placement of these is deliberate.

Ok, I see.

Quoted 34 lines

How about we do this instead?

From: Rasmus Villemoes <[email protected]>

The table of errno value->name contains a few duplicate entries since e.g. EDEADLK == EDEADLOCK on most architectures. For the known cases, the most used symbolic constant is listed last so that takes precedence - the idea being that if someone sees "can't do that:
-EDEADLK" in dmesg, grepping for EDEADLK is more likely to find the place where that error was generated (grepping for "can't do that" will find the printk() that emitted it, but the source would often be a few calls down).

However, that means one gets

warning: initialized field overwritten [-Woverride-init]

when building with W=1. As the use of multiple initializers for the same entry here is quite deliberate, explicitly disable that warning for errname.o.

Fixes 57f5677e535b ("printf: add support for printing symbolic error

names")

---
 lib/Makefile | 1 +
 1 file changed, 1 insertion(+)
lib/Makefile +0 -0
index ce45af50983a2a5e3582..a98119519e100103818d 100644

lib/Makefile obj-$(CONFIG_HAVE_ARCH_TRACEHOOK) += syscall.o · line 228

Quoted 5 lines
lib/Makefile +1 -0
@@ -224,6 +224,7 @@ obj-$(CONFIG_HAVE_ARCH_TRACEHOOK) += syscall.o
 
 obj-$(CONFIG_DYNAMIC_DEBUG_CORE) += dynamic_debug.o
 obj-$(CONFIG_SYMBOLIC_ERRNAME) += errname.o
+CFLAGS_errname.o += $(call cc-disable-warning, override-init)
 

This works, but it conflicts with a different patch series I have, and it disables a potentially useful warning in case we get another conflict in this file, so I'd prefer to find a way to avoid the warning rather than force-disabling it.

How about adding the #ifdef around the EDEADLOCK line instead of the EDEADLK one? Something like

lib/errname.c +3 -1
index 0c4d3e66170e..93043fb960cc 100644
@@ -38,7 +38,10 @@ static const char *names_0[] = {
        E(ECOMM),
        E(ECONNABORTED),
        E(ECONNRESET),
+       E(EDEADLK), /* EDEADLOCK */
+#if EDEADLK != EDEADLOCK /* mips, sparc, powerpc */
        E(EDEADLOCK),
+#endif
        E(EDESTADDRREQ),
        E(EDOM),
        E(EDOTDOT),
@@ -169,7 +172,6 @@ static const char *names_0[] = {
        E(ECANCELED), /* ECANCELLED */
        E(EAGAIN), /* EWOULDBLOCK */
        E(ECONNREFUSED), /* EREFUSED */
-       E(EDEADLK), /* EDEADLOCK */
 };
 #undef E
 

Arnd

#6 lore raw
Quoted 9 lines · Arnd Bergmann
Quoted 7 lines · Rasmus Villemoes
Quoted 1 line · Arnd Bergmann

From: Arnd Bergmann <[email protected]>

NAK. That would end up using the "EDEADLOCK" string for the value 35 on those architectures where they are the same, despite EDEADLK being the by far the most used symbol. See the comments and original commit log, the placement of these is deliberate.

Good point.

Quoted 14 lines

Ok, I see.

Quoted 12 lines

How about we do this instead?

when building with W=1. As the use of multiple initializers for the same entry here is quite deliberate, explicitly disable that warning for errname.o.

lib/Makefile +0 -0
index ce45af50983a2a5e3582..a98119519e100103818d 100644
@@ -224,6 +224,7 @@ obj-$(CONFIG_HAVE_ARCH_TRACEHOOK) += syscall.o
 
Quoted 8 lines

lib/Makefile obj-$(CONFIG_HAVE_ARCH_TRACEHOOK) += syscall.o · line 228

Quoted 4 lines
lib/Makefile +1 -0
 obj-$(CONFIG_DYNAMIC_DEBUG_CORE) += dynamic_debug.o
 obj-$(CONFIG_SYMBOLIC_ERRNAME) += errname.o
+CFLAGS_errname.o += $(call cc-disable-warning, override-init)
 

This works, but it disables a potentially useful warning in case we get another conflict in this file, so I'd prefer to find a way to avoid the warning rather than force-disabling it.

Yeah, I think that it is better to explicitely disable the less used variant in the code than hiding the double initialization. It will be clear what is going on.

Quoted 19 lines

How about adding the #ifdef around the EDEADLOCK line instead of the EDEADLK one? Something like

lib/errname.c +3 -0
index 0c4d3e66170e..93043fb960cc 100644
@@ -38,7 +38,10 @@ static const char *names_0[] = {
        E(ECOMM),
        E(ECONNABORTED),
        E(ECONNRESET),
+       E(EDEADLK), /* EDEADLOCK */
+#if EDEADLK != EDEADLOCK /* mips, sparc, powerpc */
        E(EDEADLOCK),
+#endif
        E(EDESTADDRREQ),
        E(EDOM),
        E(EDOTDOT),

lib/errname.c static const char *names_0[] = { · line 172

Quoted 4 lines
lib/errname.c +0 -1
@@ -169,7 +172,6 @@ static const char *names_0[] = {
        E(ECANCELED), /* ECANCELLED */
        E(EAGAIN), /* EWOULDBLOCK */
        E(ECONNREFUSED), /* EREFUSED */
-       E(EDEADLK), /* EDEADLOCK */

This should stay :-)

And we should remove the ECANCELLED definition. It is always the same as ECANCELED and replaced. We do not define EWOULDBLOCK and EREFUSED either.

Best Regards,
Petr

#7 lore raw
Quoted 58 lines · Petr Mladek
Quoted 9 lines · Arnd Bergmann
Quoted 7 lines · Rasmus Villemoes
Quoted 1 line · Arnd Bergmann

From: Arnd Bergmann <[email protected]>

NAK. That would end up using the "EDEADLOCK" string for the value 35 on those architectures where they are the same, despite EDEADLK being the by far the most used symbol. See the comments and original commit log, the placement of these is deliberate.

Good point.

Quoted 14 lines

Ok, I see.

Quoted 12 lines

How about we do this instead?

when building with W=1. As the use of multiple initializers for the same entry here is quite deliberate, explicitly disable that warning for errname.o.

lib/Makefile +0 -0
index ce45af50983a2a5e3582..a98119519e100103818d 100644
@@ -224,6 +224,7 @@ obj-$(CONFIG_HAVE_ARCH_TRACEHOOK) += syscall.o
 
Quoted 8 lines

lib/Makefile obj-$(CONFIG_HAVE_ARCH_TRACEHOOK) += syscall.o · line 228

Quoted 4 lines
lib/Makefile +1 -0
 obj-$(CONFIG_DYNAMIC_DEBUG_CORE) += dynamic_debug.o
 obj-$(CONFIG_SYMBOLIC_ERRNAME) += errname.o
+CFLAGS_errname.o += $(call cc-disable-warning, override-init)
 

This works, but it disables a potentially useful warning in case we get another conflict in this file, so I'd prefer to find a way to avoid the warning rather than force-disabling it.

Yeah, I think that it is better to explicitely disable the less used variant in the code than hiding the double initialization. It will be clear what is going on.

Quoted 17 lines

How about adding the #ifdef around the EDEADLOCK line instead of the EDEADLK one? Something like

lib/errname.c +3 -0
index 0c4d3e66170e..93043fb960cc 100644
@@ -38,7 +38,10 @@ static const char *names_0[] = {
        E(ECOMM),
        E(ECONNABORTED),
        E(ECONNRESET),
+       E(EDEADLK), /* EDEADLOCK */
+#if EDEADLK != EDEADLOCK /* mips, sparc, powerpc */
        E(EDEADLOCK),
+#endif
        E(EDESTADDRREQ),
        E(EDOM),
Quoted 9 lines

lib/errname.c static const char *names_0[] = { · line 172

Quoted 6 lines
lib/errname.c +0 -1
        E(EDOTDOT),
@@ -169,7 +172,6 @@ static const char *names_0[] = {
        E(ECANCELED), /* ECANCELLED */
        E(EAGAIN), /* EWOULDBLOCK */
        E(ECONNREFUSED), /* EREFUSED */
-       E(EDEADLK), /* EDEADLOCK */

This should stay :-)

No, Arnd moved it next to EDEADLOCK, which is fine (it can lose the comment /* EDEADLOCK */, though; the comment on the ifdef is sufficient). Especially when:

Quoted 3 lines

And we should remove the ECANCELLED definition. It is always the same as ECANCELED and replaced. We do not define EWOULDBLOCK and EREFUSED either.

Yes, I'm not sure why I elided EWOULDBLOCK and EREFUSED but not ECANCELLED. So let's move EAGAIN, ECONNREFUSED and ECANCELED to their proper alphabetic place. But I also want to add a check that the things we've elided match some value that we do handle. So add something like

#ifdef EREFUSED /* parisc */
static_assert(EREFUSED == ECONNREFUSED);
#endif

#ifdef ECANCELLED /* parisc */
static_assert(ECANCELLED == ECANCELED);
#endif

static_assert(EAGAIN == EWOULDBLOCK); /* everywhere */

so that if we ever import some arch that defines EREFUSED to something other than ECONNREFUSED, it would be caught. Essentially, errname.c should mention every #define E* that appears in any errno*.h.

Rasmus

#8 lore raw
Quoted 90 lines · Rasmus Villemoes
Quoted 58 lines · Petr Mladek
Quoted 9 lines · Arnd Bergmann
Quoted 7 lines · Rasmus Villemoes
Quoted 1 line · Arnd Bergmann

From: Arnd Bergmann <[email protected]>

NAK. That would end up using the "EDEADLOCK" string for the value 35 on those architectures where they are the same, despite EDEADLK being the by far the most used symbol. See the comments and original commit log, the placement of these is deliberate.

Good point.

Quoted 14 lines

Ok, I see.

Quoted 12 lines

How about we do this instead?

when building with W=1. As the use of multiple initializers for the same entry here is quite deliberate, explicitly disable that warning for errname.o.

lib/Makefile +0 -0
index ce45af50983a2a5e3582..a98119519e100103818d 100644
@@ -224,6 +224,7 @@ obj-$(CONFIG_HAVE_ARCH_TRACEHOOK) += syscall.o
 
Quoted 8 lines

lib/Makefile obj-$(CONFIG_HAVE_ARCH_TRACEHOOK) += syscall.o · line 228

Quoted 4 lines
lib/Makefile +1 -0
 obj-$(CONFIG_DYNAMIC_DEBUG_CORE) += dynamic_debug.o
 obj-$(CONFIG_SYMBOLIC_ERRNAME) += errname.o
+CFLAGS_errname.o += $(call cc-disable-warning, override-init)
 

This works, but it disables a potentially useful warning in case we get another conflict in this file, so I'd prefer to find a way to avoid the warning rather than force-disabling it.

Yeah, I think that it is better to explicitely disable the less used variant in the code than hiding the double initialization. It will be clear what is going on.

Quoted 17 lines

How about adding the #ifdef around the EDEADLOCK line instead of the EDEADLK one? Something like

lib/errname.c +3 -0
index 0c4d3e66170e..93043fb960cc 100644
@@ -38,7 +38,10 @@ static const char *names_0[] = {
        E(ECOMM),
        E(ECONNABORTED),
        E(ECONNRESET),
+       E(EDEADLK), /* EDEADLOCK */
+#if EDEADLK != EDEADLOCK /* mips, sparc, powerpc */
        E(EDEADLOCK),
+#endif
        E(EDESTADDRREQ),
        E(EDOM),
Quoted 9 lines

lib/errname.c static const char *names_0[] = { · line 172

Quoted 6 lines
lib/errname.c +0 -1
        E(EDOTDOT),
@@ -169,7 +172,6 @@ static const char *names_0[] = {
        E(ECANCELED), /* ECANCELLED */
        E(EAGAIN), /* EWOULDBLOCK */
        E(ECONNREFUSED), /* EREFUSED */
-       E(EDEADLK), /* EDEADLOCK */

This should stay :-)

No, Arnd moved it next to EDEADLOCK, which is fine (it can lose the comment /* EDEADLOCK */, though; the comment on the ifdef is sufficient). Especially when:

Quoted 3 lines

And we should remove the ECANCELLED definition. It is always the same as ECANCELED and replaced. We do not define EWOULDBLOCK and EREFUSED either.

Yes, I'm not sure why I elided EWOULDBLOCK and EREFUSED but not ECANCELLED. So let's move EAGAIN, ECONNREFUSED and ECANCELED to their proper alphabetic place. But I also want to add a check that the things we've elided match some value that we do handle. So add something like

#ifdef EREFUSED /* parisc */
static_assert(EREFUSED == ECONNREFUSED);
#endif

#ifdef ECANCELLED /* parisc */
static_assert(ECANCELLED == ECANCELED);
#endif

Quoted 5 lines

static_assert(EAGAIN == EWOULDBLOCK); /* everywhere */

so that if we ever import some arch that defines EREFUSED to something other than ECONNREFUSED, it would be caught. Essentially, errname.c should mention every #define E* that appears in any errno*.h.

Sounds like a good plan.

Arnd, are you going to take care of this or should we clean up it ourself?

Best Regards,
Petr