Bug#824363: glibc: Update debian xfail list for hppa

[John David Anglin]

Source: glibc
Version: 2.22
Severity: normal
Tags: patch

Dear Maintainer,

The attach patch updated the hppa xfail list based on the failures seen
in various builds for 2.22.  It adds two tests that sometimes fail.

Both tests fail during the unwind phase of cancellation.  It seems the
stack is being clobbered in a timing dependent manner and this breaks
the unwind.  The cause of the unwind failure is still undertermined.

This appears to be the reason many cancellation and cleanup tests fail
on hppa.

With this change and the other two patches provided today, glibc should
now build successfully on hppa.

Regards,
Dave Anglin

-- System Information:
Debian Release: stretch/sid
  APT prefers unstable
  APT policy: (500, 'unstable')
Architecture: hppa (parisc64)

Kernel: Linux 3.18.29+ (SMP w/4 CPU cores)
Locale: LANG=en_US.UTF-8, LC_CTYPE=en_US.UTF-8 (charmap=UTF-8) (ignored: LC_ALL 
set to en_CA.utf8)
Shell: /bin/sh linked to /bin/dash
Init: sysvinit (via /sbin/init)
--- testsuite-xfail-debian.mk.save	2016-05-11 17:13:39.0 -0400
+++ testsuite-xfail-debian.mk	2016-05-11 20:18:33.848340159 -0400
@@ -160,8 +160,10 @@
 test-xfail-tst-cancel5 = yes
 test-xfail-tst-cancelx20 = yes
 test-xfail-tst-cancelx21 = yes
+test-xfail-tst-cancelx3 = yes
 test-xfail-tst-cancelx4 = yes
 test-xfail-tst-cancelx5 = yes
+test-xfail-tst-cleanup4 = yes
 test-xfail-tst-cleanupx4 = yes
 test-xfail-tst-cputimer2 = yes
 test-xfail-tst-cputimer3 = yes

Bug#824362: glibc: Fix failure of debug/backtrace-tst on hppa

[John David Anglin]

Source: glibc
Version: 2.22
Severity: normal
Tags: patch

Dear Maintainer,

The failure of the debug/backtrace-tst was introduced by the
hppa/submitted-dladdr.diff patch.  The code was not prepared to handle
being passed an arbitrary address.  As a result, _dl_lookup_address
generates a segmentation fault.

I believe this fixes one or two other tests as well.

This is BZ 20098:
https://sourceware.org/bugzilla/show_bug.cgi?id=20098

A patch to fix this problem in the upstream source is here:
https://sourceware.org/ml/libc-alpha/2016-05/txti8b7kcvurW.txt

Attached is a replacement for the current hppa/submitted-dladdr.diff patch.

Please update current patch.

Thanks,
Dave

-- System Information:
Debian Release: stretch/sid
  APT prefers unstable
  APT policy: (500, 'unstable')
Architecture: hppa (parisc64)

Kernel: Linux 3.18.29+ (SMP w/4 CPU cores)
Locale: LANG=en_US.UTF-8, LC_CTYPE=en_US.UTF-8 (charmap=UTF-8) (ignored: LC_ALL 
set to en_CA.utf8)
Shell: /bin/sh linked to /bin/dash
Init: sysvinit (via /sbin/init)
2016-01-02  John David Anglin  

	[BZ #19415]
	* sysdeps/hppa/dl-fptr.c (_dl_fixup): Declare.
	(elf_machine_resolve): New.  Return address of _dl_runtime_resolve.
	(_dl_lookup_address): Rewrite using function resolver trampoline.
	* sysdeps/hppa/dl-lookupcfg.h (DL_LOOKUP_ADDRESS): Don't clear bottom
	two bits in address.

Index: glibc-2.22/sysdeps/hppa/dl-fptr.c
===
--- glibc-2.22.orig/sysdeps/hppa/dl-fptr.c
+++ glibc-2.22/sysdeps/hppa/dl-fptr.c
@@ -321,23 +321,76 @@ _dl_unmap (struct link_map *map)
   map->l_mach.fptr_table = NULL;
 }
 
+extern ElfW(Addr) _dl_fixup (struct link_map *, ElfW(Word)) attribute_hidden;
+
+static inline Elf32_Addr
+elf_machine_resolve (void)
+{
+  Elf32_Addr addr;
+
+  asm ("b,l 1f,%0\n"
+"	depi	0,31,2,%0\n"
+"1:	addil	L'_dl_runtime_resolve - ($PIC_pcrel$0 - 8),%0\n"
+"	ldo	R'_dl_runtime_resolve - ($PIC_pcrel$0 - 12)(%%r1),%0\n"
+   : "=r" (addr) : : "r1");
+
+  return addr;
+}
+
+static inline int
+_dl_read_access_allowed (unsigned int *addr)
+{
+  int result;
+
+  asm ("proberi(%1),3,%0" : "=r" (result) : "r" (addr) : );
+
+  return result;
+}
 
 ElfW(Addr)
 _dl_lookup_address (const void *address)
 {
   ElfW(Addr) addr = (ElfW(Addr)) address;
-  struct fdesc_table *t;
-  unsigned long int i;
+  unsigned int *desc, *gptr;
 
-  for (t = local.root; t != NULL; t = t->next)
-{
-  i = (struct fdesc *) addr - >fdesc[0];
-  if (i < t->first_unused && addr == (ElfW(Addr)) >fdesc[i])
-	{
-	  addr = t->fdesc[i].ip;
-	  break;
-	}
-}
+  /* Return ADDR if the least-significant two bits of ADDR are not consistent
+ with ADDR being a linker defined function pointer.  The normal value for
+ a code address in a backtrace is 3.  */
+  if (((unsigned int) addr & 3) != 2)
+return addr;
 
-  return addr;
+  /* Handle special case where ADDR points to page 0.  */
+  if ((unsigned int) addr < 4096)
+return addr;
+
+  /* Clear least-significant two bits from descriptor address.  */
+  desc = (unsigned int *) ((unsigned int) addr & ~3);
+  if (!_dl_read_access_allowed (desc))
+return addr;
+
+  /* Load first word of candidate descriptor.  It should be a pointer
+ with word alignment and point to memory that can be read.  */
+  gptr = (unsigned int *) desc[0];
+  if (((unsigned int) gptr & 3) != 0
+  || !_dl_read_access_allowed (gptr))
+return addr;
+
+  /* See if descriptor requires resolution.  The following trampoline is
+ used in each global offset table for function resolution:
+
+		ldw 0(r20),r22
+		bv r0(r22)
+		ldw 4(r20),r21
+ tramp:	b,l .-12,r20
+		depwi 0,31,2,r20
+		.word _dl_runtime_resolve
+		.word "_dl_runtime_resolve ltp"
+ got:	.word _DYNAMIC
+		.word "struct link map address" */
+  if (gptr[0] == 0xea9f1fdd			/* b,l .-12,r20 */
+  && gptr[1] == 0xd6801c1e			/* depwi 0,31,2,r20 */
+  && (ElfW(Addr)) gptr[2] == elf_machine_resolve ())
+_dl_fixup ((struct link_map *) gptr[5], (ElfW(Word)) desc[1]);
+
+  return (ElfW(Addr)) desc[0];
 }
Index: glibc-2.22/sysdeps/hppa/dl-lookupcfg.h
===
--- glibc-2.22.orig/sysdeps/hppa/dl-lookupcfg.h
+++ glibc-2.22/sysdeps/hppa/dl-lookupcfg.h
@@ -31,9 +31,7 @@ rtld_hidden_proto (_dl_symbol_address)
 
 Elf32_Addr _dl_lookup_address (const void *address);
 
-/* Clear the bottom two bits so generic code can find the fdesc entry */
-#define DL_LOOKUP_ADDRESS(addr) \
-  (_dl_lookup_address ((void *)((unsigned long)addr & ~3)))
+#define DL_LOOKUP_ADDRESS(addr) _dl_lookup_address ((const void *) addr)
 
 void attribute_hidden _dl_unmap (struct link_map *map);
 

Bug#824344: glibc: FTBFS on hppa due to stdlib/tst-setcontext3 test failure

[John David Anglin]

Source: glibc
Version: 2.22
Severity: normal
Tags: patch

Dear Maintainer,

Glibc fails to build on hppa because the stdlib/tst-setcontext3 test fails
and doesn't produce a .out file.  See, for example,
.

This problem is BZ 18124:
.

There is a submitted upstream patch here:
.

Please add attached patch to the hppa patch set.

Thanks,
Dave

-- System Information:
Debian Release: stretch/sid
  APT prefers unstable
  APT policy: (500, 'unstable')
Architecture: hppa (parisc64)

Kernel: Linux 3.18.29+ (SMP w/4 CPU cores)
Locale: LANG=en_US.UTF-8, LC_CTYPE=en_US.UTF-8 (charmap=UTF-8) (ignored: LC_ALL 
set to en_CA.utf8)
Shell: /bin/sh linked to /bin/dash
Init: sysvinit (via /sbin/init)
Index: glibc-2.22/sysdeps/unix/sysv/linux/hppa/setcontext.S
===
--- glibc-2.22.orig/sysdeps/unix/sysv/linux/hppa/setcontext.S
+++ glibc-2.22/sysdeps/unix/sysv/linux/hppa/setcontext.S
@@ -141,7 +141,7 @@ ENTRY(__setcontext)
 
 	/* No further context available. Exit now.  */
 	bl	HIDDEN_JUMPTARGET(exit), %r2
-	ldi	-1, %r26
+	ldi	0, %r26
 
 
 .Lerror: