@ -10,11 +10,9 @@ xmlns:xl="http://www.w3.org/1999/xlink" version="5.0" xml:lang="en">
ELF header identification array, e_ident[EI_DATA], holds the value 2,
defined as data encoding ELFDATA2MSB. For a little-endian encoded ELF file,
it holds the value 1, defined as data encoding ELFDATA2LSB.</para>
<programlisting>
e_ident[EI_CLASS] ELFCLASS64 For all 64-bit implementations.
e_ident[EI_DATA] ELFDATA2MSB For all big-endian implementations.
e_ident[EI_DATA] ELFDATA2LSB For all little-endian implementations.
</programlisting>
<programlisting>e_ident[EI_CLASS] ELFCLASS64 For all 64-bit implementations.
e_ident[EI_DATA] ELFDATA2MSB For all big-endian implementations.
e_ident[EI_DATA] ELFDATA2LSB For all little-endian implementations.</programlisting>
<para>The ELF header's e_flags member holds bit flags associated with the
file. The 64-bit PowerPC processor family defines the following
flags.</para>
@ -57,9 +55,7 @@ xmlns:xl="http://www.w3.org/1999/xlink" version="5.0" xml:lang="en">
</informaltable>
<para>The ABI version to be used for the ELF header file is specified with
the .abiversion pseudo-op:</para>
<programlisting>
.abiversion 2
</programlisting>
<programlisting>.abiversion 2</programlisting>
<para>Processor identification resides in the ELF header's e_machine
member, and must have the value EM_PPC64, defined as the value 21.</para>
</section>
@ -253,18 +249,16 @@ xmlns:xl="http://www.w3.org/1999/xlink" version="5.0" xml:lang="en">
<para>The TOC may straddle the boundary between initialized and
uninitialized data in the data segment. The common order of sections in the
data segment, some of which may be empty, follows:</para>
<programlisting>
.rodata
.data
.data1
.got
.toc
.sdata
.sbss
.plt
.bss1
.bss
</programlisting>
<programlisting>.rodata
.data
.data1
.got
.toc
.sdata
.sbss
.plt
.bss1
.bss</programlisting>
<para>The medium code model is expected to provide a sufficiently large TOC
to provide all data addressing needs of a module with a single TOC.</para>
<para>Compilers may generate two-instruction medium code model references
@ -275,6 +269,7 @@ xmlns:xl="http://www.w3.org/1999/xlink" version="5.0" xml:lang="en">
instruction of the two instruction form with a nop and rewriting the second
instruction. Consequently, the TOC pointer must be live during the first
and second instruction of a two-instruction reference.)</para>
<para> </para>
<bridgehead>Modules Containing Multiple TOCs</bridgehead>
<para>The link editor may create multiple TOCs. In such a case, the
constituent .got, .toc, .sdata, and .sbss sections are conceptually
@ -436,16 +431,14 @@ xmlns:xl="http://www.w3.org/1999/xlink" version="5.0" xml:lang="en">
</informaltable>
<para>The local-entry-point handling field of st_other is generated with
the .localentry pseudo op:</para>
<programlisting>
.globl my_func
.type my_func, @function
my_func:
addis r2, r12, my_sym@ha(.TOC.-my_func)
addi r2, r2, my_sym@l(.TOC.-my_func)
.localentry my_func, .-my_func
... ; function definition
blr
</programlisting>
<programlisting revisionflag="changed"> .globl my_func
.type my_func, @function
my_func:
addis r2, r12, my_sym@ha(.TOC.-my_func)
addi r2, r2, my_sym@l(.TOC.-my_func)
.localentry my_func, .-my_func
... ; function definition
blr</programlisting>
<para>Functions called via symbols with an st_other value of 0 may be
called without a valid TOC pointer in r2. Symbols of functions that
require a local entry with a valid TOC pointer should generate a symbol
@ -2433,13 +2426,10 @@ xmlns:xl="http://www.w3.org/1999/xlink" version="5.0" xml:lang="en">
<tfoot>
<row>
<entry nameend="c4" namest="c1" align="left">
<para>
<note>
<para>Relocation values 8, 9, 12, 13, 18, 23, 32,
<para><emphasis role="bold">Note:</emphasis>Relocation values 8, 9, 12, 13, 18, 23, 32,
and 247 are not used. This is to maintain a
correspondence to the relocation values used by the
32-bit PowerPC ELF ABI.</para>
</note>
32-bit PowerPC ELF ABI.
</para>
</entry>
</row>
@ -4201,10 +4191,8 @@ xmlns:xl="http://www.w3.org/1999/xlink" version="5.0" xml:lang="en">
stored is given by the assembly syntax symbol@got. The value of the
symbol alone is the address of the variable named symbol.</para>
<para>For example:</para>
<programlisting>
addis r3, r2,x@got@ha
ld r3,x@got@l(r3)
</programlisting>
<programlisting>addis r3, r2,x@got@ha
ld r3,x@got@l(r3)</programlisting>
<para>Although the Power ISA only defines 16-bit displacements, many TOCs
(and hence a GOT) are larger then 64 KB but fit within 2 GB, which can be
addressed with 32-bit offsets from r2. Therefore, this ABI defines a
@ -4239,18 +4227,14 @@ xmlns:xl="http://www.w3.org/1999/xlink" version="5.0" xml:lang="en">
current object file.</para>
<para>The following code might appear in a PIC code setup sequence to
compute the distance from a function entry point to the TOC base:</para>
<programlisting>
addis 2,12,.TOC.-func@ha
addi 2,2,.TOC.-func@l
</programlisting>
<programlisting>addis 2,12,.TOC.-func@ha
addi 2,2,.TOC.-func@l</programlisting>
<para>The syntax
<emphasis>SYMBOL@localentry</emphasis> refers to the value of the local
entry point associated with a function symbol. It can be used to
initialize a memory word with the address of the local entry point as
follows:</para>
<programlisting>
.quad func@localentry
</programlisting>
<programlisting>.quad func@localentry</programlisting>
</section>
</section>
<section>
@ -4282,15 +4266,11 @@ xmlns:xl="http://www.w3.org/1999/xlink" version="5.0" xml:lang="en">
<emphasis>may</emphasis> optimize TOC reference code that consists of two
instructions with equivalent code when offset@ha is 0.</para>
<para>TOC reference code:</para>
<programlisting>
addis rt, r2, offset@ha
lwz rt, offset@l(rt)
</programlisting>
<programlisting>addis rt, r2, offset@ha
lwz rt, offset@l(rt)</programlisting>
<para>Equivalent code:</para>
<programlisting>
NOP
lwz rt, offset(r2)
</programlisting>
<programlisting>NOP
lwz rt, offset(r2)</programlisting>
<para>Compilers and programmers
<emphasis>must</emphasis> ensure that r2 is live at the actual data access
point associated with extended displacement addressing.</para>
@ -4308,32 +4288,30 @@ xmlns:xl="http://www.w3.org/1999/xlink" version="5.0" xml:lang="en">
disabling linker optimization. However, this behavior in support of
non-ABI-compliant code is not guaranteed to be portable and supported in
all systems.</para>
<para> </para>
<bridgehead>Compliant example</bridgehead>
<programlisting>
addis r4, r2, mysym@toc@ha
b target
<programlisting> addis r4, r2, mysym@toc@ha
b target
...
...
addis r4, r2, mysym@toc@ha
target:
addi r4, r4, mysym@toc@l
...
</programlisting>
addis r4, r2, mysym@toc@ha
target:
addi r4, r4, mysym@toc@l
...</programlisting>
<para> </para>
<bridgehead>Non-compliant example</bridgehead>
<programlisting>
li r4, 0 ; #d1
b target
<programlisting> li r4, 0 ; #d1
b target
...
...
addis r4, r2, mysym@toc@ha ; #d2
target:
addi r4, r4, mysym@toc@l ; incompatible definitions #d1 and #d2 reach this
...
</programlisting>
addis r4, r2, mysym@toc@ha ; #d2
target:
addi r4, r4, mysym@toc@l ; incompatible definitions #d1 and #d2 reach this
...</programlisting>
</section>
<section>
<title>Table Jump Sequences</title>
@ -4349,13 +4327,11 @@ xmlns:xl="http://www.w3.org/1999/xlink" version="5.0" xml:lang="en">
first an addis followed by a second instruction using a D form
instruction to create or load from a 32-bit offset from a register to
enable hardware fusion whenever possible:</para>
<programlisting>
addis r4, r3, upper
<lbz,lhz,lwz,ld> r4, lower(r4)
<programlisting>addis r4, r3, upper
<lbz,lhz,lwz,ld> r4, lower(r4)
addis r4, r3, upper
addi r4, r4, lower
</programlisting>
addis r4, r3, upper
addi r4, r4, lower</programlisting>
<para>It is encouraged that assemblers provide pseudo-ops to facilitate
such code generation with a single assembler mnemonic.</para>
</section>
@ -4408,59 +4384,53 @@ xmlns:xl="http://www.w3.org/1999/xlink" version="5.0" xml:lang="en">
computed as negative offsets from the TCB address. The fields must never
be rearranged for any reason.</para>
<para>The current glibc extended TCB is:</para>
<programlisting>
typedef struct {
/* Reservation for HWCAP data. */
unsigned int hwcap2;
unsigned int hwcap; /* not used in LE ABI */
<programlisting>typedef struct {
/* Reservation for HWCAP data. */
unsigned int hwcap2;
unsigned int hwcap; /* not used in LE ABI */
/* Indicate if HTM capable (ISA 2.07). */
int tm_capable;
int tm_pad;
/* Indicate if HTM capable (ISA 2.07). */
int tm_capable;
int tm_pad;
/* Reservation for dynamic system optimizer ABI. */
uintptr_t dso_slot2;
uintptr_t dso_slot1;
/* Reservation for dynamic system optimizer ABI. */
uintptr_t dso_slot2;
uintptr_t dso_slot1;
/* Reservation for tar register (ISA 2.07). */
uintptr_t tar_save;
/* Reservation for tar register (ISA 2.07). */
uintptr_t tar_save;
/* GCC split stack support. */
void *__private_ss;
/* GCC split stack support. */
void *__private_ss;
/* Reservation for the event-based branching ABI. */
uintptr_t ebb_handler;
uintptr_t ebb_ctx_pointer;
uintptr_t ebb_reserved1;
uintptr_t ebb_reserved2;
uintptr_t pointer_guard;
/* Reservation for the event-based branching ABI. */
uintptr_t ebb_handler;
uintptr_t ebb_ctx_pointer;
uintptr_t ebb_reserved1;
uintptr_t ebb_reserved2;
uintptr_t pointer_guard;
/* Reservation for stack guard */
uintptr_t stack_guard;
/* Reservation for stack guard */
uintptr_t stack_guard;
/* DTV pointer */
dtv_t *dtv;
} tcbhead_t;
</programlisting>
/* DTV pointer */
dtv_t *dtv;
} tcbhead_t;</programlisting>
<para>Modules that will not be unloaded will be present at startup time;
the TLS blocks for these are created consecutively and immediately follow
the TCB. The offset of the TLS block of an initially available module
from the TCB remains fixed after program start.</para>
<para>The tlsoffset(m) values for a module with index m, where m ranges 1
- M, M being the total number of modules, are computed as follows:</para>
<programlisting>
tlsoffset(1) = round(16, align(1))
tlsoffset(m + 1) = round(tlsoffset(m) + tlssize(m), align(m + 1))
</programlisting>
<programlisting>tlsoffset(1) = round(16, align(1))
tlsoffset(m + 1) = round(tlsoffset(m) + tlssize(m), align(m + 1))</programlisting>
<itemizedlist>
<listitem>
<para>The function round() returns its first argument rounded up to
the next multiple of its second argument:</para>
</listitem>
</itemizedlist>
<programlisting>
round(x, y) = y ×
</programlisting>
<programlisting>round(x, y) = y ×
<itemizedlist>
<listitem>
<para>The function ceiling() returns the smallest integer greater
@ -4468,24 +4438,20 @@ xmlns:xl="http://www.w3.org/1999/xlink" version="5.0" xml:lang="en">
1 < x ≤ n:</para>
</listitem>
</itemizedlist>
<programlisting>
ceiling(x) = n
</programlisting>
<programlisting>ceiling(x) = n</programlisting>
<para>In the case of dynamic shared objects (DSO), TLS blocks are
allocated on an as-needed basis, with the details of allocation
abstracted away by the __tls_get_addr() function, which is used to
retrieve the address of any TLS variable.</para>
<para>The prototype for the __tls_get_addr() function, is defined as
follows.</para>
<programlisting>
typedef struct
{
unsigned long int ti_module;
unsigned long int ti_offset;
} tls_index;
<programlisting>typedef struct
{
unsigned long int ti_module;
unsigned long int ti_offset;
} tls_index;
extern void *__tls_get_addr (tls_index *ti);
</programlisting>
extern void *__tls_get_addr (tls_index *ti);</programlisting>
<para>The thread pointer (TP) is held in r13 and is used to access the
TCB. The TP is initialized to point 0x7000 bytes past the end of the TCB.
The TP offset allows for efficient addressing of the TCB and up to 4 KB -
@ -4551,10 +4517,8 @@ xmlns:xl="http://www.w3.org/1999/xlink" version="5.0" xml:lang="en">
thread-local variable x, the __tls_get_addr() function is called with one
parameter. That parameter is a pointer to a data object of type
tls_index.</para>
<programlisting>
extern __thread unsigned int x;
&x;
</programlisting>
<programlisting>extern __thread unsigned int x;
&x;</programlisting>
<table frame="all" pgwide="1">
<title>General Dynamic Initial Relocations</title>
<tgroup cols="3">
@ -4702,14 +4666,12 @@ xmlns:xl="http://www.w3.org/1999/xlink" version="5.0" xml:lang="en">
sequences may be used, depending on the size of the thread storage block
offset to the variable. For the following code sequence, a different
relocation sequence is used for each variable.</para>
<programlisting>
static __thread unsigned int x1;
static __thread unsigned int x2;
static __thread unsigned int x3;
&x1;
&x2;
&x3;
</programlisting>
<programlisting>static __thread unsigned int x1;
static __thread unsigned int x2;
static __thread unsigned int x3;
&x1;
&x2;
&x3;</programlisting>
<table frame="all" pgwide="1" xml:id="dbdoclet.50655241_45768">
<title>Local Dynamic Initial Relocations</title>
<tgroup cols="3">
@ -5100,10 +5062,8 @@ xmlns:xl="http://www.w3.org/1999/xlink" version="5.0" xml:lang="en">
<para>Given the following code fragment, the relocation sequence in
<xref linkend="dbdoclet.50655241_17435" /> is used for the Initial Exec
TLS Model:</para>
<programlisting>
extern __thread unsigned int x;
&x;
</programlisting>
<programlisting>extern __thread unsigned int x;
&x;</programlisting>
<table frame="all" pgwide="1" xml:id="dbdoclet.50655241_17435">
<title>Initial Exec Initial Relocations</title>
<tgroup cols="3">
@ -5232,10 +5192,8 @@ xmlns:xl="http://www.w3.org/1999/xlink" version="5.0" xml:lang="en">
less than 2 GB + 28 KB relative to the end of the TCB. The third sequence
is identical to the Initial Exec sequence shown in
<xref linkend="dbdoclet.50655241_17435" />.</para>
<programlisting>
static __thread unsigned int x;
&x;
</programlisting>
<programlisting>static __thread unsigned int x;
&x;</programlisting>
<para><xref linkend="dbdoclet.50655241_51121" /> illustrates which sequence is
used.</para>
<para> </para>
@ -5765,12 +5723,10 @@ xmlns:xl="http://www.w3.org/1999/xlink" version="5.0" xml:lang="en">
<xref linkend="dbdoclet.50655241_16273" />, a linker may reschedule the
sequence to exploit fusion by generating a sequence that may be fused
by Power processors:</para>
<programlisting>
nop
addis r3, r13, x@tprel@ha
addi r3, r3, x@tprel@l
nop
</programlisting>
<programlisting>nop
addis r3, r13, x@tprel@ha
addi r3, r3, x@tprel@l
nop</programlisting>
</footnote></para>
<para> </para>
<table frame="all" pgwide="1">
@ -6752,6 +6708,7 @@ xmlns:xl="http://www.w3.org/1999/xlink" version="5.0" xml:lang="en">
<para>For more information, see
<xref linkend="dbdoclet.50655241_18894" />. For TLS relocations, see
<xref linkend="dbdoclet.50655241_47572" />.</para>
<para> </para>
<bridgehead>TLS Relocation Descriptions</bridgehead>
<para>The following marker relocations tie together instructions in TLS
code sequences. They allow the link editor to reliably optimize TLS code.
Jeff Scheel
8 years ago