[Bug ld/20868] ld relaxes TLS access erroneously for aarch64 in ilp32 mode

[ynorov at caviumnetworks dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=20868

--- Comment #4 from Yury Norov <ynorov at caviumnetworks dot com> ---
(In reply to Jiong Wang from comment #3)
> Hi Yury,
> 
> Some quick thoughts:
> 
> >   
> > But in executalbe binary ld relaxes tls access with direct address
> > calculation:
> > 279         register unsigned long __result asm ("w0");
> > 280         asm volatile ("adrp %0, :tlsgd:foo; \n"
> > 281   400618:       90000080        adrp    x0, 410000 
> > <__FRAME_END__+0xf828>
> > 282   40061c:       f948ec00        ldr     x0, [x0,#4568]
> > 283   400620:       d53bd041        mrs     x1, tpidr_el0
> > 284   400624:       8b000020        add     x0, x1, x0
> 
> The sequences is still loading tls offset from GOT table, then add it to the
> tp.  It seems to me be a GD->IE bug on ILP32.
> 
> Can you try to modify 0xf9400000 to 0xb9400000 in elfNN_aarch64_tls_relax
> when relaxing BFD_RELOC_AARCH64_TLSGD_ADD_LO12_NC, does it work for you?
> 
> The "ldr     x0, [x0,#4568]" looks suspicous to me.  the x0 implies 8bytes
> scale, while ILP32 relocations are doing 4bytes shift.

To me too.

This code replaces TLS access with GOT using "ldr w0, [x0, ...]" manually, and
it works:
        register unsigned long __result asm ("w0");
//      asm volatile ("adrp %0, :tlsgd:foo; \n"
//                      "add %w0, %w0, #:tlsgd_lo12:foo; \n"
//                      "bl my_tls_get_addr;\n"
//                      "nop\n"
//                      : "=r" (__result)
//                      :
//                      : "x1", "x2", "x3", "x4", "x5", "x6", "x7",
//                      "x8", "x9", "x10", "x11", "x12", "x13",
//                      "x14", "x15", "x16", "x17", "x18", "x30",
//                      "memory", "cc");
        asm volatile ("adrp %0, :gottprel:foo; \n"
                        "ldr %w0, [%0, #:gottprel_lo12:foo];\n"
                        "mrs x1, tpidr_el0;\n"
                        "add %w0, w1, %w0\n"
                        : "=r" (__result)
                        :
                        : "x1", "x2", "x3", "x4", "x5", "x6", "x7",
                        "x8", "x9", "x10", "x11", "x12", "x13",
                        "x14", "x15", "x16", "x17", "x18", "x30",
                        "memory", "cc");
        bp = (unsigned int *) __result;
        printf ("TLS_LD == %p\n", bp);
        printf ("*TLS_LD == %d\n", *bp);

I can prepare the patch for binutils, but I need some time to setup environment
and get into the code. If you find it urgent, you'd better do it by yourself

-- 
You are receiving this mail because:
You are on the CC list for the bug.