Skip to contentSkip to conversation

OMAP3+: boards: Fix the __initconst build error for devicetree/test branch

RFC In review Cousson, Benoit 2 participants 2 messages Aug 19, 2011

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

RFC 1/1 OMAP3+: boards: Fix the __initconst build error for devicetree/test branch

Hi Grant,

The current devicetree/test branch is broken due to the introduction of the __initconst for dt_compat table.

My original basic fix is the following one, but Felipe figured out the root cause. So it can be fixed by adding an extra const to make the whole table const:
- const char **dt_compat; /* array of device tree
+ const char * const *dt_compat; /* array of device tree

That fix will require fixing all the current users of dt_compat with something like that:

-static const char *omap3_beagle_dt_match[] __initconst = { +static const char * const omap3_beagle_dt_match[] __initconst = {

A couple of of_XXX APIs will have to updated as well, that's why I was wondering if you wanted to go that way.

Any thought?

Regards,
Benoit

---

Quoted 1 line

From fcee9dbfeb44dc5c5b79f002e63be6e00d8f82b6 Mon Sep 17 00:00:00 2001

From: Benoit Cousson <[email protected]>
Date: Mon, 8 Aug 2011 11:30:45 +0200
Subject: [PATCH] OMAP3+: boards: Fix the __initconst build error

The __initconst introduced some build errors with the following compiler version:
arm-none-linux-gnueabi-gcc (Sourcery G++ Lite 2010.09-50) 4.5.1

arch/arm/mach-omap2/board-omap3beagle.c:566:20: error:
omap3_beagle_dt_match causes a section type conflict

Replace them by __initdata.

---
 arch/arm/mach-omap2/board-igep0020.c    |    2 +-
 arch/arm/mach-omap2/board-omap3beagle.c |    2 +-
 arch/arm/mach-omap2/board-omap4panda.c  |    2 +-
 arch/arm/mach-omap2/board-overo.c       |    2 +-
 4 files changed, 4 insertions(+), 4 deletions(-)
arch/arm/mach-omap2/board-igep0020.c +1 -1
index 249165a..2521823 100644
@@ -671,7 +671,7 @@ static void __init igep_init(void)
 	}
 }
 
-static const char *igep2_dt_compat[] __initconst = {
+static const char *igep2_dt_compat[] __initdata = {
 	"ISEE,igep-v2",
 	NULL
 };
arch/arm/mach-omap2/board-omap3beagle.c +1 -1
index b3b9cbc..a7923ca 100644
@@ -563,7 +563,7 @@ static void __init omap3_beagle_init(void)
 	beagle_opp_init();
 }
 
-static const char *omap3_beagle_dt_match[] __initconst = {
+static const char *omap3_beagle_dt_match[] __initdata = {
 	"ti,omap3-beagle",
 	NULL
 };
arch/arm/mach-omap2/board-omap4panda.c +1 -1
index b526aa9..1c583c7 100644
@@ -581,7 +581,7 @@ static void __init omap4_panda_map_io(void)
 	omap44xx_map_common_io();
 }
 
-static const char *omap4_panda_match[] __initconst = {
+static const char *omap4_panda_match[] __initdata = {
 	"ti,omap4-panda",
 	NULL,
 };
arch/arm/mach-omap2/board-overo.c +1 -1
index 18ae2c2..f0b8489 100644
@@ -560,7 +560,7 @@ static void __init overo_init(void)
 					"OVERO_GPIO_USBH_CPEN\n");
 }
 
-static const char *omap3_overo_dt_match[] __initconst = {
+static const char *omap3_overo_dt_match[] __initdata = {
 	"gumstix,omap3-overo",
 	NULL
 };
Signature · 1 line
1.7.0.4
#1 lore raw

HI,

Quoted 17 lines · Benoit

Hi Grant,

The current devicetree/test branch is broken due to the introduction of the __initconst for dt_compat table.

My original basic fix is the following one, but Felipe figured out the root cause. So it can be fixed by adding an extra const to make the whole table const:

(quoted diff) +1 -1
-	const char		**dt_compat;	/* array of device tree
+	const char		* const *dt_compat; /* array of device tree

That fix will require fixing all the current users of dt_compat with something like that:

(quoted diff) +1 -1
-static const char *omap3_beagle_dt_match[] __initconst = {
+static const char * const omap3_beagle_dt_match[] __initconst = {
Quoted 4 lines

A couple of of_XXX APIs will have to updated as well, that's why I was wondering if you wanted to go that way.

Any thought?

the funny thing is that this was not happening with CodeSourcery 2011.03 when I sent the patch.

Maybe it's either a compiler bug on one of the toolchains or newer GCCs are smart enough to create another section ??

But then again, 2011.03 isn't compiling sleep34xx.S:

arch/arm/mach-omap2/sleep34xx.S: Assembler messages:
arch/arm/mach-omap2/sleep34xx.S:150: Error: selected processor does not support ARM mode smc #1' arch/arm/mach-omap2/sleep34xx.S:418: Error: selected processor does not support ARM mode smc #1' arch/arm/mach-omap2/sleep34xx.S:429: Error: selected processor does not support ARM mode smc #1' arch/arm/mach-omap2/sleep34xx.S:454: Error: selected processor does not support ARM mode smc #0' arch/arm/mach-omap2/sleep34xx.S:460: Error: selected processor does not support ARM mode smc #0' arch/arm/mach-omap2/sleep34xx.S:465: Error: selected processor does not support ARM mode smc #0' make[1]: *** [arch/arm/mach-omap2/sleep34xx.o] Error 1

so it might be better to keep 2010.09 working.

Signature · 1 line
balbi