svn commit: r339876 - head/libexec/rtld-elf

[Mark Millard]

On 2018-Nov-3, at 12:04 PM, Mark Millard <marklmi26-fbsd at yahoo.com> wrote:

> On 2018-Nov-3, at 8:49 AM, Konstantin Belousov <kostikbel at gmail.com> wrote:
> 
>> On Fri, Nov 02, 2018 at 05:51:25PM -0700, Mark Millard wrote:
>>> On 2018-Nov-2, at 11:50 AM, Konstantin Belousov <kostikbel at gmail.com> wrote:
>>> 
>>>> On Fri, Nov 02, 2018 at 10:38:08AM -0700, Mark Millard wrote:
>>>>> . . .
>>>> 
>>>> There seems to be an issue with the direct execution mode on ppc.
>>>> Even otherwise working ld-elf.so.1 segfaults if I try to use it as
>>>> standalone binary.
>>>> 
>>>> But if I specify patched ld-elf.so.1 as the interpreter for some program,
>>>> using 'cc -Wl,-I,<path>/ld-elf.so.1' it works.  So I see there two bugs,
>>>> one is regression due to textsize calculation, which should be fixed by
>>>> my patch.  Another is the direct exec problem.
>>> 
>>> My head -r339076 based powerpc64 and armv7 contexts also
>>> fail for:
>>> 
>>> # /libexec/ld-elf.so.1 /bin/ls
>>> 
>>> The armv7 (a Cortext-A7 context) is interestingly different
>>> in how it fails:
>>> 
>>> # /libexec/ld-elf.so.1 /bin/ls
>>> ld-elf.so.1: /bin/ls: mmap of entire address space failed: Cannot allocate memory
>> Can you show the ktrace/kdump for this ?
> 
> Sure, in the Cortex-A7 context . . .
> 
> . . .
> 
> 
>>> My aarch64 context (a Cortext-A53 context) had no problem.
>>> 
>>> (All 3 examples are without any of the the recent updates
>>> or patches to ld-elf.so.1 source code.)
>> 
>> And still, does the patch for isync range works ?  You can test the new
>> ld-elf.so.1 standalone by hard-coding its path into the binary.  Build e.g.
>> only ls(1) by using make in its directory, then re-issue the linking
>> command with the additional flag '-Wl,-I,<path to patched ld-elf.so.1>
>> and try to run ls.full.
> 
> Looks like the old PowerMac is available again for
> such activity. So I'll see about testing.

I tried installing a buildworld that has the patches
via using the make arguments:

installworld distrib-dirs distribution DB_FROM_SRC=1 DESTDIR=/usr/obj/DESTDIRs/gcc421-powerpc-installworld-poud

(I normally have poudriere-devel use this gcc421-powerpc-installworld-poud
in this environment.)

Then:

# chroot /usr/obj/DESTDIRs/gcc421-powerpc-installworld-poud
# pwd
/
# ls
.cshrc		COPYRIGHT	boot		etc		libexec		mnt		proc		root		tmp		var
.profile	bin		dev		lib		media		net		rescue		sbin		usr
# df -m
Filesystem             1M-blocks  Used Avail Capacity  Mounted on
/dev/ufs/FBSDG4Srootfs     95203 38217 49369    44%    /
devfs                          0     0     0   100%    /dev
# ls -lTd /bin/ls
-r-xr-xr-x  1 root  wheel  36816 Nov  3 19:19:26 2018 /bin/ls
# ls -lTd /libexec/ld-elf.so.1*
-r-xr-xr-x  1 root  wheel  136100 Nov  3 19:17:53 2018 /libexec/ld-elf.so.1
-r-xr-xr-x  1 root  wheel  135932 Nov  2 02:10:34 2018 /libexec/ld-elf.so.1.old
# /usr/libexec/gdb /libexec/ld-elf.so.1
. . .
(gdb) disass reloc_non_plt
Dump of assembler code for function reloc_non_plt:
. . .
0x00002b48 <reloc_non_plt+204>:	bl      0xef78 <free>
0x00002b4c <reloc_non_plt+208>:	lwz     r31,64(r29)
0x00002b50 <reloc_non_plt+212>:	lwz     r9,68(r29)
0x00002b54 <reloc_non_plt+216>:	mr      r11,r31
0x00002b58 <reloc_non_plt+220>:	add     r0,r31,r9
0x00002b5c <reloc_non_plt+224>:	cmplw   cr7,r31,r0
0x00002b60 <reloc_non_plt+228>:	blt+    cr7,0x2b78 <reloc_non_plt+252>
0x00002b64 <reloc_non_plt+232>:	b       0x2bbc <reloc_non_plt+320>
0x00002b68 <reloc_non_plt+236>:	addi    r31,r31,32
0x00002b6c <reloc_non_plt+240>:	add     r0,r11,r9
0x00002b70 <reloc_non_plt+244>:	cmplw   cr7,r0,r31
0x00002b74 <reloc_non_plt+248>:	ble-    cr7,0x2bbc <reloc_non_plt+320>
0x00002b78 <reloc_non_plt+252>:	lwz     r0,0(r31)
0x00002b7c <reloc_non_plt+256>:	cmpwi   cr7,r0,1
0x00002b80 <reloc_non_plt+260>:	bne+    cr7,0x2b68 <reloc_non_plt+236>
0x00002b84 <reloc_non_plt+264>:	lwz     r0,24(r31)
0x00002b88 <reloc_non_plt+268>:	andi.   r10,r0,1
---Type <return> to continue, or q <return> to quit---
0x00002b8c <reloc_non_plt+272>:	beq+    0x2b68 <reloc_non_plt+236>
0x00002b90 <reloc_non_plt+276>:	lwz     r0,52(r29)
0x00002b94 <reloc_non_plt+280>:	lwz     r3,8(r31)
0x00002b98 <reloc_non_plt+284>:	lwz     r4,20(r31)
0x00002b9c <reloc_non_plt+288>:	addi    r31,r31,32
0x00002ba0 <reloc_non_plt+292>:	add     r3,r0,r3
0x00002ba4 <reloc_non_plt+296>:	bl      0x12858 <__syncicache>
0x00002ba8 <reloc_non_plt+300>:	lwz     r11,64(r29)
0x00002bac <reloc_non_plt+304>:	lwz     r9,68(r29)
0x00002bb0 <reloc_non_plt+308>:	add     r0,r11,r9
0x00002bb4 <reloc_non_plt+312>:	cmplw   cr7,r0,r31
0x00002bb8 <reloc_non_plt+316>:	bgt+    cr7,0x2b78 <reloc_non_plt+252>
. . .


So, unless the chroot somehow bypasses use of that local
/libexec/ld-elf.so.1 , the __synicache in a loop seems to
be working under the chroot use.

Let me know if this test is insufficient for some reason.

===
Mark Millard
marklmi at yahoo.com
( dsl-only.net went
away in early 2018-Mar)