Bug 211867

LTC Owner is: skannery.com
LTC Originator is: qiyao.com

Problem description:
According to the ptrace man page, the last arugments should be address and data.

if we use ptrace like this,
  long reg[32];
  ptrace (PPC_PTRACE_GETREGS, pid, 0, �[0]);
but the return value is always -1.

If we use ptrace like this,
  ptrace (PPC_PTRACE_GETREGS, pid, �[0], 0);
It works!

It seems that the order last two arguments is reversed!

Provide output from "uname -a", if possible:
[qiyao@plinuxt6 ~/ptrace]$ uname -a
Linux plinuxt6 2.6.17-1.2630.fc6 #1 SMP Wed Sep 6 17:17:26 EDT 2006 ppc64 ppc64
ppc64 GNU/Linux

------------------------------------------------------------------------------------------------
Compile test case in 64-bit mode.

[qiyao@plinuxt6 ~/ptrace]$ gcc -m64 test-ppc_ptrace_getregs.c -o
test-ppc_ptrace_getregs.64
[qiyao@plinuxt6 ~/ptrace]$ ./test-ppc_ptrace_getregs.64 
Child pid: 9933
Start of child.
Child stopped by signal 5
Test PPC_PTRACE_GETREGS:
ERROR: Parent's GETREGS failed with the 'correct' arg order: Bad address
ERROR: Parent's GETREGS succeeded, reversed arg order
Test PPC_PTRACE_GETFPREGS:
ERROR: Parent's GETFPREGS failed with the 'correct' arg order: Bad address
ERROR: Parent's GETFPREGS succeeded, reversed arg order
Test PPC_PTRACE_SETREGS:
ERROR: Parent's SETREGS failed with the 'correct' arg order: Bad address
ERROR: Parent's SETREGS succeeded, reversed arg order
Test PPC_PTRACE_SETFPREGS:
ERROR: Parent's SETFPREGS failed with the 'correct' arg order: Bad address
ERROR: Parent's SETFPREGS succeeded, reversed arg order


We find that ptrace operations(GETREGS, SETREGS, GETFPREGS, SETFPREGS) are
succeeded, but with reversed argument order.

Compile this case in 32-bit mode.
[qiyao@plinuxt6 ~/ptrace]$ gcc -m32 test-ppc_ptrace_getregs.c -o
test-ppc_ptrace_getregs.32
[qiyao@plinuxt6 ~/ptrace]$ ./test-ppc_ptrace_getregs.32 
Start of child.
Child pid: 9940
Child stopped by signal 5
Test PPC_PTRACE_GETREGS:
ERROR: Parent's GETREGS failed with the 'correct' arg order: Bad address
ERROR: Parent's GETREGS succeeded, reversed arg order
Test PPC_PTRACE_GETFPREGS:
ERROR: Parent's GETFPREGS failed with the 'correct' arg order: Bad address
ERROR: Parent's GETFPREGS succeeded, reversed arg order
Test PPC_PTRACE_SETREGS:
ERROR: Parent's SETREGS failed with the 'correct' arg order: Bad address
ERROR: Parent's SETREGS failed with reversed arg order: Bad address
Test PPC_PTRACE_SETFPREGS:
ERROR: Parent's SETFPREGS failed with the 'correct' arg order: Bad address
ERROR: Parent's SETFPREGS succeeded, reversed arg order

We find that ptrace operations are succeeded with reversed argument order,
except PPC_PTRACE_SETREGS.

------------------------------------------------------------------------------------------------
Yao,
 PPC_PTRACE_GETREGS option for powerpc is implemented such that general purpose
registers of the child process get copied to the address variable instead of
data variable.
==============================================================================
#ifdef CONFIG_PPC64
        case PPC_PTRACE_GETREGS: { /* Get GPRs 0 - 31. */
...
                unsigned long __user *tmp = (unsigned long __user *)addr;
                                                                                
                for (i = 0; i < 32; i++) {
                        ret = put_user(*reg, tmp);
                        if (ret)
                                break;
==============================================================================
whereas for i386 architecture these registers get copied to data variable:
==============================================================================
unsigned long __user *datap = (unsigned long __user *)data;
...
case PTRACE_GETREGS: { /* Get all gp regs from the child. */
...
                for ( i = 0; i < FRAME_SIZE*sizeof(long); i += sizeof(long) ) {
                        __put_user(getreg(child, i), datap);
                        datap++;
                }
                ret = 0;
                break;
===============================================================================

The prepared test patch is for aligning
PPC_PTRACE_*REGS options with the general format of ptrace.
Rgds, Supriya

---------------------------------------------------------------------------------------------

I code a patch with your patch as a reference, and rebuild the kernel.	I find
that this patch works.	Here are some steps,

1) rpmbuild -bp SPECS/kernel-2.6.spec
2) Apply this patch, and cd to BUILD/kernel-2.6.17/linux-2.6.17.ppc64
3) cp ./boot/config-2.6.17-1.2630.fc6 .config 
4) make oldconfig & make
5) make modules_install
6) cp vmlinux.strip /boot/vmlinuz-qiyao
7) cp System.map /boot/System.map-qiyao
8) mkinitrd initrd-qiyao 2.6.17-prep
9) edit /etc/yaboot.conf to add kernel image and initrd

Reboot machine,

[qiyao@plinuxt6 ~/ptrace]$ uname -a
Linux plinuxt6 2.6.17-prep #1 SMP Sun Oct 22 22:32:54 EDT 2006 ppc64 ppc64
ppc64 GNU/Linux

And run the test case,
[qiyao@plinuxt6 ~/ptrace]$ gcc test-ppc_ptrace_getregs.c -o
test-ppc_ptrace_getregs.32 
[qiyao@plinuxt6 ~/ptrace]$ gcc -m64 test-ppc_ptrace_getregs.c -o
test-ppc_ptrace_getregs.64
[qiyao@plinuxt6 ~/ptrace]$ ./test-ppc_ptrace_getregs.32 
Start of child.
Child pid: 2877
Child stopped by signal 5
Test PPC_PTRACE_GETREGS:
OK: Parent's GETREGS succeeded, 'correct' arg order
ERROR: Parent's GETREGS failed with reversed arg order: Bad address
Test PPC_PTRACE_GETFPREGS:
OK: Parent's GETFPREGS succeeded, 'correct' arg order
ERROR: Parent's GETFPREGS failed with reversed arg order: Bad address
Test PPC_PTRACE_SETREGS:
ERROR: Parent's SETREGS failed with the 'correct' arg order: Bad address
ERROR: Parent's SETREGS failed with reversed arg order: Bad address
Test PPC_PTRACE_SETFPREGS:
OK: Parent's SETFPREGS succeeded, 'correct' arg order
ERROR: Parent's SETFPREGS failed with reversed arg order: Bad address
[qiyao@plinuxt6 ~/ptrace]$ ./test-ppc_ptrace_getregs.64 
Start of child.
Child pid: 2879
Child stopped by signal 5
Test PPC_PTRACE_GETREGS:
OK: Parent's GETREGS succeeded, 'correct' arg order
ERROR: Parent's GETREGS failed with reversed arg order: Bad address
Test PPC_PTRACE_GETFPREGS:
OK: Parent's GETFPREGS succeeded, 'correct' arg order
ERROR: Parent's GETFPREGS failed with reversed arg order: Bad address
Test PPC_PTRACE_SETREGS:
OK: Parent's SETREGS succeeded, 'correct' arg order
ERROR: Parent's SETREGS failed with reversed arg order: Bad address
Test PPC_PTRACE_SETFPREGS:
OK: Parent's SETFPREGS succeeded, 'correct' arg order
ERROR: Parent's SETFPREGS failed with reversed arg order: Bad address

---------------------------------------------------------------------------------------------
Red Hat,

Mirroring this bug for your review of the patch

Patch information:

Problem:
PPC_PTRACE_GETREGS option for powerpc is implemented such that general purpose
registers of the child process get copied to the address variable instead of
data variable.

Arch: PPC-64 
Kernel: 2.6.17-1.2630.fc6

Test Case: <Attached>
Test Results : Succesful, see above
Patch against FC6: <Attached>

Submission to the community: In progress.

-thanks.