Skip to contentSkip to conversation

Support for (au1100) 64-bit physical address space broken on 2.6.12?

PATCH Rodolfo Giometti 3 participants 8 messages Jul 16, 2005 – Jul 18, 2005

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

Hello,

switching from linux-mips 2.6.12-rc3 to 2.6.12 I notice that the following patch has been applied:

http://www.linux-mips.org/archives/linux-mips/2005-06/msg00207.html

But, on my system, recompiling the source I noticed that compilation stops with errors. Even downloading a clean version of source code from linux-mips's CVS and choosing, for instance, the board DB1100, I got the same result.

The problem is that the above patch works well if the 64-bit physical address space support is disabled, but, if enabled, it breaks compilation stage.

Here what I get after getting source form CVS and doing the commands:

   # make pb1100_defconfig   (this board turn on CONFIG_64BIT_PHYS_ADDR option)
   # make
   ...
   include/asm-mips/mach-au1x00/ioremap.h:25: warning: static declaration of 'fixup_bigphys_addr' follows non-static declaration
   include/asm/pgtable.h:363: warning: 'fixup_bigphys_addr' declared inline after being called
   include/asm/pgtable.h:363: warning: previous declaration of 'fixup_bigphys_addr' was here
   include/asm-mips/mach-au1x00/ioremap.h: In function `fixup_bigphys_addr':
   include/asm-mips/mach-au1x00/ioremap.h:26: warning: implicit declaration of function `__fixup_bigphys_addr'
   arch/mips/au1000/common/setup.c: At top level:
   arch/mips/au1000/common/setup.c:159: error: conflicting types for '__fixup_bigphys_addr'
   include/asm-mips/mach-au1x00/ioremap.h:26: error: previous implicit declaration of '__fixup_bigphys_addr' was here
   arch/mips/au1000/common/setup.c: In function `__fixup_bigphys_addr':
   ...

After a little job I implemented the attached patch (patch-64BIT_PHYS_ADDR) that works on my system on both settings (CONFIG_64BIT_PHYS_ADDR on or off).

I don't know if it can resolve the above problem for others CPUs (I tested it on au1100) but, at least, on this processor the PCMCIA support now is functional. :)

I also suggest to apply the second patch (patch-PCMCIA_Kconfig) who simply auto enable 64 bit support when choosing PCMCIA support.

Ciao,

Rodolfo

Signature · 4 lines
GNU/Linux Solutions                  e-mail:    [email protected]
Linux Device Driver                             [email protected]
Embedded Systems                     home page: giometti.enneenne.com
UNIX programming                     phone:     +39 349 2432127

Index: arch/mips/au1000/common/setup.c
===================================================================
RCS file: /home/develop/cvs_private/linux-mips-exadron/arch/mips/au1000/common/setup.c,v retrieving revision 1.1.1.1

arch/mips/au1000/common/setup.c +0 -35
diff -u -r1.1.1.1 setup.c
@@ -152,38 +152,3 @@
 	while (au_readl(SYS_COUNTER_CNTRL) & SYS_CNTRL_T0S);
 	au_writel(0, SYS_TOYTRIM);
 }
-
-#if defined(CONFIG_64BIT_PHYS_ADDR)
-/* This routine should be valid for all Au1x based boards */
-phys_t __fixup_bigphys_addr(phys_t phys_addr, phys_t size)
-{
-	u32 start, end;
-
-	/* Don't fixup 36 bit addresses */
-	if ((phys_addr >> 32) != 0) return phys_addr;
-
-#ifdef CONFIG_PCI
-	start = (u32)Au1500_PCI_MEM_START;
-	end = (u32)Au1500_PCI_MEM_END;
-	/* check for pci memory window */
-	if ((phys_addr >= start) && ((phys_addr + size) < end)) {
-		return (phys_t)((phys_addr - start) + Au1500_PCI_MEM_START);
-	}
-#endif
-
-	/* All Au1x SOCs have a pcmcia controller */
-	/* We setup our 32 bit pseudo addresses to be equal to the
-	 * 36 bit addr >> 4, to make it easier to check the address
-	 * and fix it.
-	 * The Au1x socket 0 phys attribute address is 0xF 4000 0000.
-	 * The pseudo address we use is 0xF400 0000. Any address over
-	 * 0xF400 0000 is a pcmcia pseudo address.
-	 */
-	if ((phys_addr >= 0xF4000000) && (phys_addr < 0xFFFFFFFF)) {
-		return (phys_t)(phys_addr << 4);
-	}
-
-	/* default nop */
-	return phys_addr;
-}
-#endif

Index: include/asm-mips/pgtable.h
===================================================================
RCS file: /home/develop/cvs_private/linux-mips-exadron/include/asm-mips/pgtable.h,v retrieving revision 1.1.1.1

include/asm-mips/pgtable.h +1 -1
diff -u -r1.1.1.1 pgtable.h
@@ -17,6 +17,7 @@
 #endif
 
 #include <asm/pgtable-bits.h>
+#include <ioremap.h>
 
 #define PAGE_NONE	__pgprot(_PAGE_PRESENT | _CACHE_CACHABLE_NONCOHERENT)
 #define PAGE_SHARED	__pgprot(_PAGE_PRESENT | _PAGE_READ | _PAGE_WRITE | \
@@ -360,7 +361,6 @@
 #endif
 
 #ifdef CONFIG_64BIT_PHYS_ADDR
-extern phys_t fixup_bigphys_addr(phys_t phys_addr, phys_t size);
 extern int remap_pfn_range(struct vm_area_struct *vma, unsigned long from, unsigned long pfn, unsigned long size, pgprot_t prot);
 
 static inline int io_remap_page_range(struct vm_area_struct *vma,

Index: include/asm-mips/mach-au1x00/ioremap.h
===================================================================
RCS file: /home/develop/cvs_private/linux-mips-exadron/include/asm-mips/mach-au1x00/ioremap.h,v retrieving revision 1.1.1.1

include/asm-mips/mach-au1x00/ioremap.h +36 -1
diff -u -r1.1.1.1 ioremap.h
@@ -11,7 +11,42 @@
 
 #include <linux/types.h>
 
-#ifndef CONFIG_64BIT_PHYS_ADDR
+#ifdef CONFIG_64BIT_PHYS_ADDR
+/* This routine should be valid for all Au1x based boards */
+static inline phys_t __fixup_bigphys_addr(phys_t phys_addr, phys_t size)
+{
+#ifdef CONFIG_PCI
+	u32 start, end;
+#endif
+
+	/* Don't fixup 36 bit addresses */
+	if ((phys_addr >> 32) != 0) return phys_addr;
+
+#ifdef CONFIG_PCI
+	start = (u32)Au1500_PCI_MEM_START;
+	end = (u32)Au1500_PCI_MEM_END;
+	/* check for pci memory window */
+	if ((phys_addr >= start) && ((phys_addr + size) < end)) {
+		return (phys_t)((phys_addr - start) + Au1500_PCI_MEM_START);
+	}
+#endif
+
+	/* All Au1x SOCs have a pcmcia controller */
+	/* We setup our 32 bit pseudo addresses to be equal to the
+	 * 36 bit addr >> 4, to make it easier to check the address
+	 * and fix it.
+	 * The Au1x socket 0 phys attribute address is 0xF 4000 0000.
+	 * The pseudo address we use is 0xF400 0000. Any address over
+	 * 0xF400 0000 is a pcmcia pseudo address.
+	 */
+	if ((phys_addr >= 0xF4000000) && (phys_addr < 0xFFFFFFFF)) {
+		return (phys_t)(phys_addr << 4);
+	}
+
+	/* default nop */
+	return phys_addr;
+}
+#else
 static inline phys_t __fixup_bigphys_addr(phys_t phys_addr, phys_t size)
 {
 	return phys_addr;

Index: drivers/pcmcia/Kconfig
===================================================================
RCS file: /home/develop/cvs_private/linux-mips-exadron/drivers/pcmcia/Kconfig,v retrieving revision 1.1.1.1

drivers/pcmcia/Kconfig +1 -0
diff -u -r1.1.1.1 Kconfig
@@ -137,6 +137,7 @@
 config PCMCIA_AU1X00
 	tristate "Au1x00 pcmcia support"
 	depends on SOC_AU1X00 && PCMCIA
+	select 64BIT_PHYS_ADDR
 
 config PCMCIA_SA1100
 	tristate "SA1100 support"
#1 lore raw
Quoted 7 lines · Rodolfo Giometti

Hello,

switching from linux-mips 2.6.12-rc3 to 2.6.12 I notice that the following patch has been applied:

   http://www.linux-mips.org/archives/linux-mips/2005-06/msg00207.html
Quoted 8 lines

But, on my system, recompiling the source I noticed that compilation stops with errors. Even downloading a clean version of source code from linux-mips's CVS and choosing, for instance, the board DB1100, I got the same result.

The problem is that the above patch works well if the 64-bit physical address space support is disabled, but, if enabled, it breaks compilation stage.

I fixed this is the latest tree a couple of days ago.

Pete

Quoted 31 lines

Here what I get after getting source form CVS and doing the commands:

   # make pb1100_defconfig   (this board turn on CONFIG_64BIT_PHYS_ADDR option)
   # make
   ...
   include/asm-mips/mach-au1x00/ioremap.h:25: warning: static declaration of 'fixup_bigphys_addr' follows non-static declaration
   include/asm/pgtable.h:363: warning: 'fixup_bigphys_addr' declared inline after being called
   include/asm/pgtable.h:363: warning: previous declaration of 'fixup_bigphys_addr' was here
   include/asm-mips/mach-au1x00/ioremap.h: In function `fixup_bigphys_addr':
   include/asm-mips/mach-au1x00/ioremap.h:26: warning: implicit declaration of function `__fixup_bigphys_addr'
   arch/mips/au1000/common/setup.c: At top level:
   arch/mips/au1000/common/setup.c:159: error: conflicting types for '__fixup_bigphys_addr'
   include/asm-mips/mach-au1x00/ioremap.h:26: error: previous implicit declaration of '__fixup_bigphys_addr' was here
   arch/mips/au1000/common/setup.c: In function `__fixup_bigphys_addr':
   ...

After a little job I implemented the attached patch (patch-64BIT_PHYS_ADDR) that works on my system on both settings (CONFIG_64BIT_PHYS_ADDR on or off).

I don't know if it can resolve the above problem for others CPUs (I tested it on au1100) but, at least, on this processor the PCMCIA support now is functional. :)

I also suggest to apply the second patch (patch-PCMCIA_Kconfig) who simply auto enable 64 bit support when choosing PCMCIA support.

Ciao,

Rodolfo

#2 lore raw
Quoted 1 line · Pete Popov

I fixed this is the latest tree a couple of days ago.

Great! :)

Did you already publish it? I checked the linux-mips CVS before sending my patches but I saw nothing about it. :-o

Where can I get your patch in order to compare the two solutions?

Thanks,

Rodolfo

Signature · 4 lines
GNU/Linux Solutions                  e-mail:    [email protected]
Linux Device Driver                             [email protected]
Embedded Systems                     home page: giometti.enneenne.com
UNIX programming                     phone:     +39 349 2432127
#3 lore raw
Quoted 9 lines · Rodolfo Giometti
Quoted 1 line · Pete Popov

I fixed this is the latest tree a couple of days ago.

Great! :)

Did you already publish it? I checked the linux-mips CVS before sending my patches but I saw nothing about it. :-o

Where can I get your patch in order to compare the two solutions?

Just do a cvs update in your directory and you'll get the patch.

Pete

#4 lore raw
Quoted 1 line · Pete Popov

Just do a cvs update in your directory and you'll get the patch.

Mmm... I just did as you suggest but I got nothing... ok, maybe it's a my problem.

I'll take a better look to the on line CVS. :)

Ciao,

Rodolfo

Signature · 4 lines
GNU/Linux Solutions                  e-mail:    [email protected]
Linux Device Driver                             [email protected]
Embedded Systems                     home page: giometti.enneenne.com
UNIX programming                     phone:     +39 349 2432127
#5 lore raw
In reply to Pete Popov
Quoted 1 line · Pete Popov

I fixed this is the latest tree a couple of days ago.

Something is still wrong... I just downloaded the whole linux-mips tree from the CVS (to avoid conflics with my local reporitory) and after the commands:

   # make pb1100_defconfig
   # make

I get:

arch/mips/mm/ioremap.c: In function `__ioremap':
include/asm-mips/mach-au1x00/ioremap.h:15: sorry, unimplemented: inlining failed in call to '__fixup_bigphys_addr': function body not available arch/mips/mm/ioremap.c:28: sorry, unimplemented: called from here
make[1]: *** [arch/mips/mm/ioremap.o] Error 1
make: *** [arch/mips/mm] Error 2

Ciao,

Rodolfo

Signature · 4 lines
GNU/Linux Solutions                  e-mail:    [email protected]
Linux Device Driver                             [email protected]
Embedded Systems                     home page: giometti.enneenne.com
UNIX programming                     phone:     +39 349 2432127
#6 lore raw
Quoted 10 lines · Rodolfo Giometti
Quoted 1 line · Pete Popov

I fixed this is the latest tree a couple of days ago.

Something is still wrong... I just downloaded the whole linux-mips tree from the CVS (to avoid conflics with my local reporitory) and after the commands:

   # make pb1100_defconfig
   # make
Quoted 7 lines

I get:

arch/mips/mm/ioremap.c: In function `__ioremap':
include/asm-mips/mach-au1x00/ioremap.h:15: sorry, unimplemented: inlining failed in call to '__fixup_bigphys_addr': function body not available arch/mips/mm/ioremap.c:28: sorry, unimplemented: called from here
make[1]: *** [arch/mips/mm/ioremap.o] Error 1
make: *** [arch/mips/mm] Error 2

Oh, pb boards ... I haven't done any maintenance on the Pb boards in 2.6. AMD wasn't interested in moving them forward and I don't have time to build and test kernels for so many boards, so I'm not sure what to do with them. The db1500/1550 should build just fine. Take a look at the kernel support for those boards, and it shouldn't be too hard to update the pb1100.

Thanks,

Pete

#7 lore raw
In reply to Rodolfo Giometti
Quoted 7 lines · Rodolfo Giometti

I get:

arch/mips/mm/ioremap.c: In function `__ioremap':
include/asm-mips/mach-au1x00/ioremap.h:15: sorry, unimplemented: inlining failed in call to '__fixup_bigphys_addr': function body not available arch/mips/mm/ioremap.c:28: sorry, unimplemented: called from here
make[1]: *** [arch/mips/mm/ioremap.o] Error 1
make: *** [arch/mips/mm] Error 2

Well, "extern inline __attribute__((always_inline))" certainly makes no
sense. Pete?

Maciej