(2012/01/27 3:28), Denys Vlasenko wrote:

On Thu, Jan 26, 2012 at 3:14 PM, Masami Hiramatsu
[off-list ref] wrote:

quoted

(2012/01/26 0:32), Denys Vlasenko wrote:

quoted

On Wed, Jan 25, 2012 at 4:13 PM, Denys Vlasenko
[off-list ref] wrote:

quoted

quoted

+       /*
+        * Convert from rip-relative addressing to indirect addressing
+        * via a scratch register.  Change the r/m field from 0x5 (%rip)
+        * to 0x0 (%rax) or 0x1 (%rcx), and squeeze out the offset field.
+        */
+       reg = MODRM_REG(insn);
+       if (reg == 0) {
+               /*
+                * The register operand (if any) is either the A register
+                * (%rax, %eax, etc.) or (if the 0x4 bit is set in the
+                * REX prefix) %r8.  In any case, we know the C register
+                * is NOT the register operand, so we use %rcx (register
+                * #1) for the scratch register.
+                */
+               uprobe->arch_info.fixups = UPROBES_FIX_RIP_CX;
+               /* Change modrm from 00 000 101 to 00 000 001. */
+               *cursor = 0x1;

Hmm. I think we have a bug here.

What if this instruction has REX.B = 1? Granted, REX.B = 1 has no effect on
rip-relative addressing and therefore normally won't be generated by gcc/as,
but still. If you replace md and r/m fields as above, you are trying to convert
0x12345678(%rip) reference to (%rcx), but if REX.B = 1, then you in fact
converted it to (%r9)!

Right, thanks for finding :)
And %rax register reference encoding has same problem, doesn't it?

Yes.

The solution is trivial: "if (REX pfx exists) REX.B = 0;"

Also, I don't remember whether (%rip) addressing is
affected by 0x67 (address size) prefix. If it is, then
nothing needs to be done.

In Intel SDM vol.2, "2.2.1.6 RIP-Relative Addressing" explains
that as below;

---
RIP-relative addressing is enabled by 64-bit mode, not by a 64-bit address-size.
The use of the address-size prefix does not disable RIP-relative addressing. The
effect of the address-size prefix is to truncate and zero-extend the computed
effective address to 32 bits.
---

This means 0x67 pfx can be used with RIP-relative, and
effective address(EA) will be 32bit, as same as others.
So, I think it should work :)


Thank you!

-- 
Masami HIRAMATSU
Software Platform Research Dept. Linux Technology Center
Hitachi, Ltd., Yokohama Research Laboratory
E-mail: masami.hiramatsu.pt@hitachi.com

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>