Skip to contentSkip to conversation

make headers_install headers problem on sparc64

DISCUSSION [email protected] 3 participants 9 messages Oct 18, 2006 – Dec 30, 2006

lore
Share & tools
lore
#0 lore raw

Hi david, lkml.

I found a message where you said

"If you have problems with the exported headers (other than that some userspace abuses headers which it shouldn't), then _shout_. Please don't just let them go unreported."

So,

Using headers exported from vanilla linux 2.6.18.1 and compiling the elftoaout part of the debian sparc-utils package, I get some breakage here:

In file included from /usr/include/asm/elf.h:5,
from /usr/include/linux/elf.h:7,
from elftoaout.c:24:

due to tlb_type, cheetah et al being undefined in this part of asm-sparc64/elf.h:

bash-3.1# tail -n30 /pkg/linux/include/asm-sparc64/elf.h

/* This yields a mask that user programs can use to figure out what
instruction set this cpu supports. */

/* On Ultra, we support all of the v8 capabilities. */ static __inline__ unsigned int sparc64_elf_hwcap(void)

{
        unsigned int cap = (HWCAP_SPARC_FLUSH | HWCAP_SPARC_STBAR |
                            HWCAP_SPARC_SWAP | HWCAP_SPARC_MULDIV |
                            HWCAP_SPARC_V9);

        if (tlb_type == cheetah || tlb_type == cheetah_plus)
                cap |= HWCAP_SPARC_ULTRA3;
        else if (tlb_type == hypervisor)
                cap |= HWCAP_SPARC_BLKINIT;

        return cap;
}

#define ELF_HWCAP sparc64_elf_hwcap();

/* This yields a string that ld.so will use to load implementation
specific libraries for optimization. This is more specific in intent than poking at uname or /proc/cpuinfo. */

#define ELF_PLATFORM (NULL)

#endif /* !(__ASM_SPARC64_ELF_H) */

Hope this is useful

Andrew Walrond

#1 lore raw
Quoted 7 lines · On Wed, 2006-10-18 at 22:37 +0000, [email protected] wrote:

Hi david, lkml.

I found a message where you said

"If you have problems with the exported headers (other than that some userspace abuses headers which it shouldn't), then _shout_. Please don't just let them go unreported."

Thank you.

Quoted 6 lines

So,

Using headers exported from vanilla linux 2.6.18.1 and compiling the elftoaout part of the debian sparc-utils package, I get some breakage here:

Quoted 8 lines

In file included from /usr/include/asm/elf.h:5,
from /usr/include/linux/elf.h:7,
from elftoaout.c:24:

due to tlb_type, cheetah et al being undefined in this part of asm-sparc64/elf.h:

Hm, yes. There's still a lot of crap being exposed there that we really don't need to be showing. Try this entirely untested patch...

The sparc64 version could do with something similar, too.

----
[SPARC] Clean up asm-sparc/elf.h pollution in userspace.

We don't need to export sparc_elf_hwcap() to userspace, and it doesn't build there. Remove it by moving it inside #ifdef __KERNEL__, along with some other things which don't need to be exported.

include/asm-sparc/elf.h +20 -21
index 83a3dd1..aaf6ef4 100644
@@ -8,11 +8,6 @@ #define __ASMSPARC_ELF_H
 
 #include <asm/ptrace.h>
 
-#ifdef __KERNEL__
-#include <asm/mbus.h>
-#include <asm/uaccess.h>
-#endif
-
 /*
  * Sparc section types
  */
@@ -77,6 +72,23 @@ typedef unsigned long elf_greg_t;
 #define ELF_NGREG 38
 typedef elf_greg_t elf_gregset_t[ELF_NGREG];
 
+typedef struct {
+	union {
+		unsigned long	pr_regs[32];
+		double		pr_dregs[16];
+	} pr_fr;
+	unsigned long __unused;
+	unsigned long	pr_fsr;
+	unsigned char	pr_qcnt;
+	unsigned char	pr_q_entrysize;
+	unsigned char	pr_en;
+	unsigned int	pr_q[64];
+} elf_fpregset_t;
+
+#ifdef __KERNEL__
+#include <asm/mbus.h>
+#include <asm/uaccess.h>
+
 /* Format is:
  * 	G0 --> G7
  *	O0 --> O7
@@ -99,20 +111,7 @@ do {	unsigned long *dest = &(__elf_regs[
 	dest[34] = src->npc;				\
 	dest[35] = src->y;				\
 	dest[36] = dest[37] = 0; /* XXX */		\
-} while(0); /* Janitors: Don't touch this colon. */
-
-typedef struct {
-	union {
-		unsigned long	pr_regs[32];
-		double		pr_dregs[16];
-	} pr_fr;
-	unsigned long __unused;
-	unsigned long	pr_fsr;
-	unsigned char	pr_qcnt;
-	unsigned char	pr_q_entrysize;
-	unsigned char	pr_en;
-	unsigned int	pr_q[64];
-} elf_fpregset_t;
+} while(0); /* Janitors: Don't touch this semicolon. */
 
 #define ELF_CORE_COPY_TASK_REGS(__tsk, __elf_regs)	\
 	({ ELF_CORE_COPY_REGS((*(__elf_regs)), (__tsk)->thread.kregs); 1; })
@@ -165,8 +164,8 @@ #define ELF_HWCAP	((ARCH_SUN4C_SUN4) ? 0
 
 #define ELF_PLATFORM	(NULL)
 
-#ifdef __KERNEL__
 #define SET_PERSONALITY(ex, ibcs2) set_personality((ibcs2)?PER_SVR4:PER_LINUX)
-#endif
+
+#endif /* __KERNEL__ */
 
 #endif /* !(__ASMSPARC_ELF_H) */
Signature · 1 line
dwmw2
#2 lore raw
Quoted 6 lines · David Woodhouse

Hm, yes. There's still a lot of crap being exposed there that we really don't need to be showing. Try this entirely untested patch...

The sparc64 version could do with something similar, too.

I'm on sparc64 (pure 64bit) so I can't test your patch. For reference, I used this (not as smart as yours) patch:

include/asm-sparc64/elf.h 2006-10-19 08:57:11.000000000 +0000 renamed from include/asm-sparc64/elf.h 2006-09-20 03:42:06.000000000 +0000 +3 -0
diff -Naur linux-2.6.18.1/include/asm-sparc64/elf.h linux-2.6.18.1-fix/include/asm-sparc64/elf.h
@@ -142,6 +142,7 @@
 #define ELF_ET_DYN_BASE         0x0000010000000000UL
 #endif
 
+#ifdef __KERNEL__
 
 /* This yields a mask that user programs can use to figure out what
    instruction set this cpu supports.  */
@@ -163,6 +164,8 @@
 
 #define ELF_HWCAP      sparc64_elf_hwcap();
 
+#endif
+
 /* This yields a string that ld.so will use to load implementation
    specific libraries for optimization.  This is more specific in
    intent than poking at uname or /proc/cpuinfo.  */

Andrew Walrond

#3 lore raw
In reply to [email protected]

Another problem; when compiling reiserfsprogs 3.6.19 there is a header missing:

../include/reiserfs_fs.h:41:27: error: asm/unaligned.h: No such file or directory

I see this file exists in the kernel sources, but not in the exported headers.

Andrew

#4 lore raw
Quoted 11 lines · On Thu, Oct 19, 2006 at 10:54:41AM +0000, [email protected] wrote:

Another problem; when compiling reiserfsprogs 3.6.19 there is a header missing:

../include/reiserfs_fs.h:41:27: error: asm/unaligned.h: No such file or directory

I see this file exists in the kernel sources, but not in the exported headers.

Simply copying asm-generic/unaligned.h into the sanitised header tree under asm/ solves the problem, so should headers_install just include unaligned.h ?

Quoted 1 line

Andrew

PS Great job BTW; I have encountered amazingly few problems using the sanitised headers created with headers_install.

#5 lore raw
Quoted 3 lines · On Thu, 2006-10-19 at 11:10 +0000, [email protected] wrote:

Simply copying asm-generic/unaligned.h into the sanitised header tree under asm/ solves the problem, so should headers_install just include unaligned.h ?

No. That header should not be exposed to userspace. Just fix reiserfsprogs instead. It's not as if unaligned access is _hard_ -- you just have to ask the compiler to do it for you:

http://cvs.fedora.redhat.com/viewcvs/rpms/reiserfs-utils/devel/header-fix.patch?view=markup

Signature · 1 line
dwmw2
#7 lore raw
Quoted 9 lines · On Thu, 2006-10-19 at 11:37 +0000, [email protected] wrote:
Quoted 7 lines · David Woodhouse

No. That header should not be exposed to userspace. Just fix reiserfsprogs instead. It's not as if unaligned access is _hard_ -- you just have to ask the compiler to do it for you:

http://cvs.fedora.redhat.com/viewcvs/rpms/reiserfs-utils/devel/header-fix.patch?view=markup

Thanks for the link

Fedora Core 6 is already shipping with the output of 'make headers_install'. This means that in general, if userspace is misbehaving and it's something we ship in Fedora Core, Extras or Livna, we're going to have fixed it up already. So Fedora CVS is a good place to look, although such fixes all should have been pushed upstream rather than just languishing in our packages.

Signature · 1 line
dwmw2
#8 lore raw
In reply to David Woodhouse
Quoted 3 lines · David Woodhouse

No. That header should not be exposed to userspace. Just fix reiserfsprogs instead. It's not as if unaligned access is _hard_ -- you just have to ask the compiler to do it for you:

reiserfsprogs 3.6.20 already handles the case where asm/unaligned.h isnt installed ... i just wouldnt suggest using that version as it wont even compile on big endian machines and doesnt work with non-standard journals ;)
-mike