[v1,16/17] loader: Add data swap option to load-elf

Commit Message

Peter Crosthwaite Jan. 18, 2016, 7:12 a.m. UTC

Some CPUs are of an opposite data-endianness to other components in the
system. Sometimes elfs have the data sections layed out with this CPU
data-endianess accounting for when loaded via the CPU, byte swaps
(relative to other system components) will occur.

The leading example, is ARM's BE32 mode, which is is basically LE with
address manipulation on half-word and byte accesses to access the
hw/byte reversed address. This means that word data is invariant
accross LE and BE32. This also means that instructions are still LE.
The expectation is that the elf will be loaded via the CPU in this
endianness scheme, which means the data in the elf is reversed at
compile time.

As QEMU loads via the system memory directly, rather than the CPU, we
need a mechanism to reverse elf data endianness to implement this
possibility.

Signed-off-by: Peter Crosthwaite <[e-mail]>
---

 hw/alpha/dp264.c               |  4 ++--
 hw/arm/armv7m.c                |  2 +-
 hw/arm/boot.c                  |  2 +-
 hw/core/loader.c               |  9 ++++++---
 hw/cris/boot.c                 |  2 +-
 hw/i386/multiboot.c            |  3 ++-
 hw/lm32/lm32_boards.c          |  4 ++--
 hw/lm32/milkymist.c            |  2 +-
 hw/m68k/an5206.c               |  2 +-
 hw/m68k/dummy_m68k.c           |  2 +-
 hw/m68k/mcf5208.c              |  2 +-
 hw/microblaze/boot.c           |  4 ++--
 hw/mips/mips_fulong2e.c        |  2 +-
 hw/mips/mips_malta.c           |  2 +-
 hw/mips/mips_mipssim.c         |  2 +-
 hw/mips/mips_r4k.c             |  2 +-
 hw/moxie/moxiesim.c            |  3 ++-
 hw/openrisc/openrisc_sim.c     |  3 ++-
 hw/pci-host/prep.c             |  2 +-
 hw/ppc/e500.c                  |  2 +-
 hw/ppc/mac_newworld.c          |  5 +++--
 hw/ppc/mac_oldworld.c          |  5 +++--
 hw/ppc/ppc440_bamboo.c         |  3 ++-
 hw/ppc/spapr.c                 |  6 ++++--
 hw/ppc/virtex_ml507.c          |  3 ++-
 hw/s390x/ipl.c                 |  4 ++--
 hw/sparc/leon3.c               |  2 +-
 hw/sparc/sun4m.c               |  4 ++--
 hw/sparc64/sun4u.c             |  4 ++--
 hw/tricore/tricore_testboard.c |  2 +-
 hw/xtensa/sim.c                |  4 ++--
 hw/xtensa/xtfpga.c             |  2 +-
 include/hw/elf_ops.h           | 22 +++++++++++++++++++++-
 include/hw/loader.h            |  2 +-
 34 files changed, 78 insertions(+), 46 deletions(-)

Comments

Peter Maydell Jan. 19, 2016, 5:53 p.m. UTC | #1

On 18 January 2016 at 07:12, Peter Crosthwaite
<[e-mail]> wrote:
> Some CPUs are of an opposite data-endianness to other components in the
> system. Sometimes elfs have the data sections layed out with this CPU
> data-endianess accounting for when loaded via the CPU, byte swaps
> (relative to other system components) will occur.
>
> The leading example, is ARM's BE32 mode, which is is basically LE with
> address manipulation on half-word and byte accesses to access the
> hw/byte reversed address. This means that word data is invariant
> accross LE and BE32. This also means that instructions are still LE.
> The expectation is that the elf will be loaded via the CPU in this
> endianness scheme, which means the data in the elf is reversed at
> compile time.
>
> As QEMU loads via the system memory directly, rather than the CPU, we
> need a mechanism to reverse elf data endianness to implement this
> possibility.
>
> Signed-off-by: Peter Crosthwaite <[e-mail]>

> diff --git a/include/hw/loader.h b/include/hw/loader.h
> index 33067f8..e542575 100644
> --- a/include/hw/loader.h
> +++ b/include/hw/loader.h
> @@ -35,7 +35,7 @@ const char *load_elf_strerror(int error);
>  int load_elf(const char *filename, uint64_t (*translate_fn)(void *, uint64_t),
>               void *translate_opaque, uint64_t *pentry, uint64_t *lowaddr,
>               uint64_t *highaddr, int big_endian, int elf_machine,
> -             int clear_lsb);
> +             int clear_lsb, int data_swab);
>  void load_elf_hdr(const char *filename, void *hdr, bool *is64, Error **errp);
>  int load_aout(const char *filename, hwaddr addr, int max_sz,
>                int bswap_needed, hwaddr target_page_size);

Can we have a doc comment so we have something that defines what
values data_swab accepts? (it's not just a bool).

Otherwise
Reviewed-by: Peter Maydell <[e-mail]>

thanks
-- PMM

Peter Crosthwaite Feb. 27, 2016, 11:14 p.m. UTC | #2

On Tue, Jan 19, 2016 at 9:53 AM, Peter Maydell <[e-mail]> wrote:
> On 18 January 2016 at 07:12, Peter Crosthwaite
> <[e-mail]> wrote:
>> Some CPUs are of an opposite data-endianness to other components in the
>> system. Sometimes elfs have the data sections layed out with this CPU
>> data-endianess accounting for when loaded via the CPU, byte swaps
>> (relative to other system components) will occur.
>>
>> The leading example, is ARM's BE32 mode, which is is basically LE with
>> address manipulation on half-word and byte accesses to access the
>> hw/byte reversed address. This means that word data is invariant
>> accross LE and BE32. This also means that instructions are still LE.
>> The expectation is that the elf will be loaded via the CPU in this
>> endianness scheme, which means the data in the elf is reversed at
>> compile time.
>>
>> As QEMU loads via the system memory directly, rather than the CPU, we
>> need a mechanism to reverse elf data endianness to implement this
>> possibility.
>>
>> Signed-off-by: Peter Crosthwaite <[e-mail]>
>
>> diff --git a/include/hw/loader.h b/include/hw/loader.h
>> index 33067f8..e542575 100644
>> --- a/include/hw/loader.h
>> +++ b/include/hw/loader.h
>> @@ -35,7 +35,7 @@ const char *load_elf_strerror(int error);
>>  int load_elf(const char *filename, uint64_t (*translate_fn)(void *, uint64_t),
>>               void *translate_opaque, uint64_t *pentry, uint64_t *lowaddr,
>>               uint64_t *highaddr, int big_endian, int elf_machine,
>> -             int clear_lsb);
>> +             int clear_lsb, int data_swab);
>>  void load_elf_hdr(const char *filename, void *hdr, bool *is64, Error **errp);
>>  int load_aout(const char *filename, hwaddr addr, int max_sz,
>>                int bswap_needed, hwaddr target_page_size);
>
> Can we have a doc comment so we have something that defines what
> values data_swab accepts? (it's not just a bool).
>

This is difficult to capture without writing to whole documentation
for load_elf. So here goes:

/** load_elf:
 * @filename: Path of ELF file
 * @translate_fn: optional function to translate load addresses
 * @translate_opaque: opaque data passed to @translate_fn
 * @pentry: Populated with program entry point. Ignored if NULL.
 * @lowaddr: Populated with lowest loaded address. Ignored if NULL.
 * @highaddr: Populated with highest loaded address. Ignored if NULL.
 * @bigendian: Expected ELF endianness. 0 for LE otherwise BE
 * @elf_machine: Expected ELF machine type
 * @clear_lsb: Set to mask off LSB of addresses (Some arch's use this
               for non-address data)
 * @data_swab: Set to order of byte swapping for data. 0 for no swap, 1
 *             for swapping bytes within halfwords, 2 for bytes within
 *             words and 3 for within doublewords.
 *
 * Load an ELF file's contents to the emulated systems address space.
 * Clients may optionally specify a callback to perform address
 * translations. @pentry, @lowaddr and @highaddr are optional pointers
 * which will be populated with various load information. @bigendian and
 * @elf_machine give the expected endianness and machine for the ELF the
 * load will fail it the target ELF does not match. Some architectures
 * have some arch-specific behaviours that come into effect when their
 * particular values for @elf_machine are set.
 */

Regards,
Peter

> Otherwise
> Reviewed-by: Peter Maydell <[e-mail]>
>
> thanks
> -- PMM

Peter Maydell Feb. 28, 2016, 3:28 p.m. UTC | #3

On 27 February 2016 at 23:14, Peter Crosthwaite
<[e-mail]> wrote:
> On Tue, Jan 19, 2016 at 9:53 AM, Peter Maydell <[e-mail]> wrote:
>> Can we have a doc comment so we have something that defines what
>> values data_swab accepts? (it's not just a bool).
>>
>
> This is difficult to capture without writing to whole documentation
> for load_elf. So here goes:

Thanks; a couple of typos below, but otherwise looks good.


> /** load_elf:
>  * @filename: Path of ELF file
>  * @translate_fn: optional function to translate load addresses
>  * @translate_opaque: opaque data passed to @translate_fn
>  * @pentry: Populated with program entry point. Ignored if NULL.
>  * @lowaddr: Populated with lowest loaded address. Ignored if NULL.
>  * @highaddr: Populated with highest loaded address. Ignored if NULL.
>  * @bigendian: Expected ELF endianness. 0 for LE otherwise BE
>  * @elf_machine: Expected ELF machine type
>  * @clear_lsb: Set to mask off LSB of addresses (Some arch's use this

Can we just write out "architectures" here?

>                for non-address data)
>  * @data_swab: Set to order of byte swapping for data. 0 for no swap, 1
>  *             for swapping bytes within halfwords, 2 for bytes within
>  *             words and 3 for within doublewords.
>  *
>  * Load an ELF file's contents to the emulated systems address space.

"system's"

>  * Clients may optionally specify a callback to perform address
>  * translations. @pentry, @lowaddr and @highaddr are optional pointers
>  * which will be populated with various load information. @bigendian and
>  * @elf_machine give the expected endianness and machine for the ELF the
>  * load will fail it the target ELF does not match. Some architectures

"if"

>  * have some arch-specific behaviours that come into effect when their

"architecture"

>  * particular values for @elf_machine are set.
>  */

thanks
-- PMM

Peter Crosthwaite Feb. 28, 2016, 8:16 p.m. UTC | #4

On Sun, Feb 28, 2016 at 7:28 AM, Peter Maydell <[e-mail]> wrote:
> On 27 February 2016 at 23:14, Peter Crosthwaite
> <[e-mail]> wrote:
>> On Tue, Jan 19, 2016 at 9:53 AM, Peter Maydell <[e-mail]> wrote:
>>> Can we have a doc comment so we have something that defines what
>>> values data_swab accepts? (it's not just a bool).
>>>
>>
>> This is difficult to capture without writing to whole documentation
>> for load_elf. So here goes:
>
> Thanks; a couple of typos below, but otherwise looks good.
>
>
>> /** load_elf:
>>  * @filename: Path of ELF file
>>  * @translate_fn: optional function to translate load addresses
>>  * @translate_opaque: opaque data passed to @translate_fn
>>  * @pentry: Populated with program entry point. Ignored if NULL.
>>  * @lowaddr: Populated with lowest loaded address. Ignored if NULL.
>>  * @highaddr: Populated with highest loaded address. Ignored if NULL.
>>  * @bigendian: Expected ELF endianness. 0 for LE otherwise BE
>>  * @elf_machine: Expected ELF machine type
>>  * @clear_lsb: Set to mask off LSB of addresses (Some arch's use this
>
> Can we just write out "architectures" here?
>
>>                for non-address data)
>>  * @data_swab: Set to order of byte swapping for data. 0 for no swap, 1
>>  *             for swapping bytes within halfwords, 2 for bytes within
>>  *             words and 3 for within doublewords.
>>  *
>>  * Load an ELF file's contents to the emulated systems address space.
>
> "system's"
>
>>  * Clients may optionally specify a callback to perform address
>>  * translations. @pentry, @lowaddr and @highaddr are optional pointers
>>  * which will be populated with various load information. @bigendian and
>>  * @elf_machine give the expected endianness and machine for the ELF the
>>  * load will fail it the target ELF does not match. Some architectures
>
> "if"
>
>>  * have some arch-specific behaviours that come into effect when their
>
> "architecture"
>

All fixed. Thanks for the pre-review.

Regards,
Peter

>>  * particular values for @elf_machine are set.
>>  */
>
> thanks
> -- PMM

Patch

diff --git a/hw/alpha/dp264.c b/hw/alpha/dp264.c
index 27bdaa1..df071fa 100644
--- a/hw/alpha/dp264.c
+++ b/hw/alpha/dp264.c
@@ -109,7 +109,7 @@  static void clipper_init(MachineState *machine)
     }
     size = load_elf(palcode_filename, cpu_alpha_superpage_to_phys,
                     NULL, &palcode_entry, &palcode_low, &palcode_high,
-                    0, EM_ALPHA, 0);
+                    0, EM_ALPHA, 0, 0);
     if (size < 0) {
         hw_error("could not load palcode '%s'\n", palcode_filename);
         exit(1);
@@ -129,7 +129,7 @@  static void clipper_init(MachineState *machine)
 
         size = load_elf(kernel_filename, cpu_alpha_superpage_to_phys,
                         NULL, &kernel_entry, &kernel_low, &kernel_high,
-                        0, EM_ALPHA, 0);
+                        0, EM_ALPHA, 0, 0);
         if (size < 0) {
             hw_error("could not load kernel '%s'\n", kernel_filename);
             exit(1);
diff --git a/hw/arm/armv7m.c b/hw/arm/armv7m.c
index a80d2ad..d721e5b 100644
--- a/hw/arm/armv7m.c
+++ b/hw/arm/armv7m.c
@@ -210,7 +210,7 @@  DeviceState *armv7m_init(MemoryRegion *system_memory, int mem_size, int num_irq,
 
     if (kernel_filename) {
         image_size = load_elf(kernel_filename, NULL, NULL, &entry, &lowaddr,
-                              NULL, big_endian, EM_ARM, 1);
+                              NULL, big_endian, EM_ARM, 1, 0);
         if (image_size < 0) {
             image_size = load_image_targphys(kernel_filename, 0, mem_size);
             lowaddr = 0;
diff --git a/hw/arm/boot.c b/hw/arm/boot.c
index 75f69bf..0de4269 100644
--- a/hw/arm/boot.c
+++ b/hw/arm/boot.c
@@ -700,7 +700,7 @@  static void arm_load_kernel_notify(Notifier *notifier, void *data)
     /* Assume that raw images are linux kernels, and ELF images are not.  */
     kernel_size = load_elf(info->kernel_filename, NULL, NULL, &elf_entry,
                            &elf_low_addr, &elf_high_addr, big_endian,
-                           elf_machine, 1);
+                           elf_machine, 1, 0);
     if (kernel_size > 0 && have_dtb(info)) {
         /* If there is still some room left at the base of RAM, try and put
          * the DTB there like we do for images loaded with -bios or -pflash.
diff --git a/hw/core/loader.c b/hw/core/loader.c
index 28da8e2..bee6cf5 100644
--- a/hw/core/loader.c
+++ b/hw/core/loader.c
@@ -382,7 +382,8 @@  fail:
 /* return < 0 if error, otherwise the number of bytes loaded in memory */
 int load_elf(const char *filename, uint64_t (*translate_fn)(void *, uint64_t),
              void *translate_opaque, uint64_t *pentry, uint64_t *lowaddr,
-             uint64_t *highaddr, int big_endian, int elf_machine, int clear_lsb)
+             uint64_t *highaddr, int big_endian, int elf_machine,
+             int clear_lsb, int data_swab)
 {
     int fd, data_order, target_data_order, must_swab, ret = ELF_LOAD_FAILED;
     uint8_t e_ident[EI_NIDENT];
@@ -421,10 +422,12 @@  int load_elf(const char *filename, uint64_t (*translate_fn)(void *, uint64_t),
     lseek(fd, 0, SEEK_SET);
     if (e_ident[EI_CLASS] == ELFCLASS64) {
         ret = load_elf64(filename, fd, translate_fn, translate_opaque, must_swab,
-                         pentry, lowaddr, highaddr, elf_machine, clear_lsb);
+                         pentry, lowaddr, highaddr, elf_machine, clear_lsb,
+                         data_swab);
     } else {
         ret = load_elf32(filename, fd, translate_fn, translate_opaque, must_swab,
-                         pentry, lowaddr, highaddr, elf_machine, clear_lsb);
+                         pentry, lowaddr, highaddr, elf_machine, clear_lsb,
+                         data_swab);
     }
 
  fail:
diff --git a/hw/cris/boot.c b/hw/cris/boot.c
index 1cfa339..d08b307 100644
--- a/hw/cris/boot.c
+++ b/hw/cris/boot.c
@@ -72,7 +72,7 @@  void cris_load_image(CRISCPU *cpu, struct cris_load_info *li)
     /* Boots a kernel elf binary, os/linux-2.6/vmlinux from the axis 
        devboard SDK.  */
     image_size = load_elf(li->image_filename, translate_kernel_address, NULL,
-                          &entry, NULL, &high, 0, EM_CRIS, 0);
+                          &entry, NULL, &high, 0, EM_CRIS, 0, 0);
     li->entry = entry;
     if (image_size < 0) {
         /* Takes a kimage from the axis devboard SDK.  */
diff --git a/hw/i386/multiboot.c b/hw/i386/multiboot.c
index 6774a19..b4435fa 100644
--- a/hw/i386/multiboot.c
+++ b/hw/i386/multiboot.c
@@ -195,7 +195,8 @@  int load_multiboot(FWCfgState *fw_cfg,
         }
 
         kernel_size = load_elf(kernel_filename, NULL, NULL, &elf_entry,
-                               &elf_low, &elf_high, 0, I386_ELF_MACHINE, 0);
+                               &elf_low, &elf_high, 0, I386_ELF_MACHINE,
+                               0, 0);
         if (kernel_size < 0) {
             fprintf(stderr, "Error while loading elf kernel\n");
             exit(1);
diff --git a/hw/lm32/lm32_boards.c b/hw/lm32/lm32_boards.c
index eb553a1..9b921c4 100644
--- a/hw/lm32/lm32_boards.c
+++ b/hw/lm32/lm32_boards.c
@@ -142,7 +142,7 @@  static void lm32_evr_init(MachineState *machine)
         int kernel_size;
 
         kernel_size = load_elf(kernel_filename, NULL, NULL, &entry, NULL, NULL,
-                               1, EM_LATTICEMICO32, 0);
+                               1, EM_LATTICEMICO32, 0, 0);
         reset_info->bootstrap_pc = entry;
 
         if (kernel_size < 0) {
@@ -244,7 +244,7 @@  static void lm32_uclinux_init(MachineState *machine)
         int kernel_size;
 
         kernel_size = load_elf(kernel_filename, NULL, NULL, &entry, NULL, NULL,
-                               1, EM_LATTICEMICO32, 0);
+                               1, EM_LATTICEMICO32, 0, 0);
         reset_info->bootstrap_pc = entry;
 
         if (kernel_size < 0) {
diff --git a/hw/lm32/milkymist.c b/hw/lm32/milkymist.c
index 13976b3..fe50d7e 100644
--- a/hw/lm32/milkymist.c
+++ b/hw/lm32/milkymist.c
@@ -176,7 +176,7 @@  milkymist_init(MachineState *machine)
 
         /* Boots a kernel elf binary.  */
         kernel_size = load_elf(kernel_filename, NULL, NULL, &entry, NULL, NULL,
-                               1, EM_LATTICEMICO32, 0);
+                               1, EM_LATTICEMICO32, 0, 0);
         reset_info->bootstrap_pc = entry;
 
         if (kernel_size < 0) {
diff --git a/hw/m68k/an5206.c b/hw/m68k/an5206.c
index c1dea17..c869f12 100644
--- a/hw/m68k/an5206.c
+++ b/hw/m68k/an5206.c
@@ -70,7 +70,7 @@  static void an5206_init(MachineState *machine)
     }
 
     kernel_size = load_elf(kernel_filename, NULL, NULL, &elf_entry,
-                           NULL, NULL, 1, EM_68K, 0);
+                           NULL, NULL, 1, EM_68K, 0, 0);
     entry = elf_entry;
     if (kernel_size < 0) {
         kernel_size = load_uimage(kernel_filename, &entry, NULL, NULL,
diff --git a/hw/m68k/dummy_m68k.c b/hw/m68k/dummy_m68k.c
index 8b3b775..9482309 100644
--- a/hw/m68k/dummy_m68k.c
+++ b/hw/m68k/dummy_m68k.c
@@ -49,7 +49,7 @@  static void dummy_m68k_init(MachineState *machine)
     /* Load kernel.  */
     if (kernel_filename) {
         kernel_size = load_elf(kernel_filename, NULL, NULL, &elf_entry,
-                               NULL, NULL, 1, EM_68K, 0);
+                               NULL, NULL, 1, EM_68K, 0, 0);
         entry = elf_entry;
         if (kernel_size < 0) {
             kernel_size = load_uimage(kernel_filename, &entry, NULL, NULL,
diff --git a/hw/m68k/mcf5208.c b/hw/m68k/mcf5208.c
index ddeccc5..42cf0f3 100644
--- a/hw/m68k/mcf5208.c
+++ b/hw/m68k/mcf5208.c
@@ -275,7 +275,7 @@  static void mcf5208evb_init(MachineState *machine)
     }
 
     kernel_size = load_elf(kernel_filename, NULL, NULL, &elf_entry,
-                           NULL, NULL, 1, EM_68K, 0);
+                           NULL, NULL, 1, EM_68K, 0, 0);
     entry = elf_entry;
     if (kernel_size < 0) {
         kernel_size = load_uimage(kernel_filename, &entry, NULL, NULL,
diff --git a/hw/microblaze/boot.c b/hw/microblaze/boot.c
index d7eaa1f..d2cd85a 100644
--- a/hw/microblaze/boot.c
+++ b/hw/microblaze/boot.c
@@ -141,12 +141,12 @@  void microblaze_load_kernel(MicroBlazeCPU *cpu, hwaddr ddr_base,
         /* Boots a kernel elf binary.  */
         kernel_size = load_elf(kernel_filename, NULL, NULL,
                                &entry, &low, &high,
-                               big_endian, EM_MICROBLAZE, 0);
+                               big_endian, EM_MICROBLAZE, 0, 0);
         base32 = entry;
         if (base32 == 0xc0000000) {
             kernel_size = load_elf(kernel_filename, translate_kernel_address,
                                    NULL, &entry, NULL, NULL,
-                                   big_endian, EM_MICROBLAZE, 0);
+                                   big_endian, EM_MICROBLAZE, 0, 0);
         }
         /* Always boot into physical ram.  */
         boot_info.bootstrap_pc = (uint32_t)entry;
diff --git a/hw/mips/mips_fulong2e.c b/hw/mips/mips_fulong2e.c
index 5988a88..5b29db0 100644
--- a/hw/mips/mips_fulong2e.c
+++ b/hw/mips/mips_fulong2e.c
@@ -116,7 +116,7 @@  static int64_t load_kernel (CPUMIPSState *env)
 
     if (load_elf(loaderparams.kernel_filename, cpu_mips_kseg0_to_phys, NULL,
                  (uint64_t *)&kernel_entry, (uint64_t *)&kernel_low,
-                 (uint64_t *)&kernel_high, 0, EM_MIPS, 1) < 0) {
+                 (uint64_t *)&kernel_high, 0, EM_MIPS, 1, 0) < 0) {
         fprintf(stderr, "qemu: could not load kernel '%s'\n",
                 loaderparams.kernel_filename);
         exit(1);
diff --git a/hw/mips/mips_malta.c b/hw/mips/mips_malta.c
index 91c36ba..25cf344 100644
--- a/hw/mips/mips_malta.c
+++ b/hw/mips/mips_malta.c
@@ -795,7 +795,7 @@  static int64_t load_kernel (void)
 
     if (load_elf(loaderparams.kernel_filename, cpu_mips_kseg0_to_phys, NULL,
                  (uint64_t *)&kernel_entry, NULL, (uint64_t *)&kernel_high,
-                 big_endian, EM_MIPS, 1) < 0) {
+                 big_endian, EM_MIPS, 1, 0) < 0) {
         fprintf(stderr, "qemu: could not load kernel '%s'\n",
                 loaderparams.kernel_filename);
         exit(1);
diff --git a/hw/mips/mips_mipssim.c b/hw/mips/mips_mipssim.c
index 23b35be..f08ceb7 100644
--- a/hw/mips/mips_mipssim.c
+++ b/hw/mips/mips_mipssim.c
@@ -69,7 +69,7 @@  static int64_t load_kernel(void)
     kernel_size = load_elf(loaderparams.kernel_filename, cpu_mips_kseg0_to_phys,
                            NULL, (uint64_t *)&entry, NULL,
                            (uint64_t *)&kernel_high, big_endian,
-                           EM_MIPS, 1);
+                           EM_MIPS, 1, 0);
     if (kernel_size >= 0) {
         if ((entry & ~0x7fffffffULL) == 0x80000000)
             entry = (int32_t)entry;
diff --git a/hw/mips/mips_r4k.c b/hw/mips/mips_r4k.c
index af10da1..71a3646 100644
--- a/hw/mips/mips_r4k.c
+++ b/hw/mips/mips_r4k.c
@@ -87,7 +87,7 @@  static int64_t load_kernel(void)
     kernel_size = load_elf(loaderparams.kernel_filename, cpu_mips_kseg0_to_phys,
                            NULL, (uint64_t *)&entry, NULL,
                            (uint64_t *)&kernel_high, big_endian,
-                           EM_MIPS, 1);
+                           EM_MIPS, 1, 0);
     if (kernel_size >= 0) {
         if ((entry & ~0x7fffffffULL) == 0x80000000)
             entry = (int32_t)entry;
diff --git a/hw/moxie/moxiesim.c b/hw/moxie/moxiesim.c
index ada3d58..fa28dac 100644
--- a/hw/moxie/moxiesim.c
+++ b/hw/moxie/moxiesim.c
@@ -53,7 +53,8 @@  static void load_kernel(MoxieCPU *cpu, LoaderParams *loader_params)
     ram_addr_t initrd_offset;
 
     kernel_size = load_elf(loader_params->kernel_filename,  NULL, NULL,
-                           &entry, &kernel_low, &kernel_high, 1, EM_MOXIE, 0);
+                           &entry, &kernel_low, &kernel_high, 1, EM_MOXIE,
+                           0, 0);
 
     if (kernel_size <= 0) {
         fprintf(stderr, "qemu: could not load kernel '%s'\n",
diff --git a/hw/openrisc/openrisc_sim.c b/hw/openrisc/openrisc_sim.c
index be6c9b5..a99be45 100644
--- a/hw/openrisc/openrisc_sim.c
+++ b/hw/openrisc/openrisc_sim.c
@@ -68,7 +68,8 @@  static void cpu_openrisc_load_kernel(ram_addr_t ram_size,
 
     if (kernel_filename && !qtest_enabled()) {
         kernel_size = load_elf(kernel_filename, NULL, NULL,
-                               &elf_entry, NULL, NULL, 1, EM_OPENRISC, 1);
+                               &elf_entry, NULL, NULL, 1, EM_OPENRISC,
+                               1, 0);
         entry = elf_entry;
         if (kernel_size < 0) {
             kernel_size = load_uimage(kernel_filename,
diff --git a/hw/pci-host/prep.c b/hw/pci-host/prep.c
index da88cb3..da1c7da 100644
--- a/hw/pci-host/prep.c
+++ b/hw/pci-host/prep.c
@@ -312,7 +312,7 @@  static void raven_realize(PCIDevice *d, Error **errp)
         if (filename) {
             if (s->elf_machine != EM_NONE) {
                 bios_size = load_elf(filename, NULL, NULL, NULL,
-                                     NULL, NULL, 1, s->elf_machine, 0);
+                                     NULL, NULL, 1, s->elf_machine, 0, 0);
             }
             if (bios_size < 0) {
                 bios_size = get_image_size(filename);
diff --git a/hw/ppc/e500.c b/hw/ppc/e500.c
index b3418db..c73c699 100644
--- a/hw/ppc/e500.c
+++ b/hw/ppc/e500.c
@@ -1017,7 +1017,7 @@  void ppce500_init(MachineState *machine, PPCE500Params *params)
     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
 
     bios_size = load_elf(filename, NULL, NULL, &bios_entry, &loadaddr, NULL,
-                         1, PPC_ELF_MACHINE, 0);
+                         1, PPC_ELF_MACHINE, 0, 0);
     if (bios_size < 0) {
         /*
          * Hrm. No ELF image? Try a uImage, maybe someone is giving us an
diff --git a/hw/ppc/mac_newworld.c b/hw/ppc/mac_newworld.c
index 1b9a573..4cc0668 100644
--- a/hw/ppc/mac_newworld.c
+++ b/hw/ppc/mac_newworld.c
@@ -219,7 +219,7 @@  static void ppc_core99_init(MachineState *machine)
     /* Load OpenBIOS (ELF) */
     if (filename) {
         bios_size = load_elf(filename, NULL, NULL, NULL,
-                             NULL, NULL, 1, PPC_ELF_MACHINE, 0);
+                             NULL, NULL, 1, PPC_ELF_MACHINE, 0, 0);
 
         g_free(filename);
     } else {
@@ -242,7 +242,8 @@  static void ppc_core99_init(MachineState *machine)
         kernel_base = KERNEL_LOAD_ADDR;
 
         kernel_size = load_elf(kernel_filename, translate_kernel_address, NULL,
-                               NULL, &lowaddr, NULL, 1, PPC_ELF_MACHINE, 0);
+                               NULL, &lowaddr, NULL, 1, PPC_ELF_MACHINE,
+                               0, 0);
         if (kernel_size < 0)
             kernel_size = load_aout(kernel_filename, kernel_base,
                                     ram_size - kernel_base, bswap_needed,
diff --git a/hw/ppc/mac_oldworld.c b/hw/ppc/mac_oldworld.c
index 21eaf0e..1717cea 100644
--- a/hw/ppc/mac_oldworld.c
+++ b/hw/ppc/mac_oldworld.c
@@ -147,7 +147,7 @@  static void ppc_heathrow_init(MachineState *machine)
     /* Load OpenBIOS (ELF) */
     if (filename) {
         bios_size = load_elf(filename, 0, NULL, NULL, NULL, NULL,
-                             1, PPC_ELF_MACHINE, 0);
+                             1, PPC_ELF_MACHINE, 0, 0);
         g_free(filename);
     } else {
         bios_size = -1;
@@ -168,7 +168,8 @@  static void ppc_heathrow_init(MachineState *machine)
 #endif
         kernel_base = KERNEL_LOAD_ADDR;
         kernel_size = load_elf(kernel_filename, translate_kernel_address, NULL,
-                               NULL, &lowaddr, NULL, 1, PPC_ELF_MACHINE, 0);
+                               NULL, &lowaddr, NULL, 1, PPC_ELF_MACHINE,
+                               0, 0);
         if (kernel_size < 0)
             kernel_size = load_aout(kernel_filename, kernel_base,
                                     ram_size - kernel_base, bswap_needed,
diff --git a/hw/ppc/ppc440_bamboo.c b/hw/ppc/ppc440_bamboo.c
index b66c113..3fb1bac 100644
--- a/hw/ppc/ppc440_bamboo.c
+++ b/hw/ppc/ppc440_bamboo.c
@@ -256,7 +256,8 @@  static void bamboo_init(MachineState *machine)
                               NULL, NULL);
         if (success < 0) {
             success = load_elf(kernel_filename, NULL, NULL, &elf_entry,
-                               &elf_lowaddr, NULL, 1, PPC_ELF_MACHINE, 0);
+                               &elf_lowaddr, NULL, 1, PPC_ELF_MACHINE,
+                               0, 0);
             entry = elf_entry;
             loadaddr = elf_lowaddr;
         }
diff --git a/hw/ppc/spapr.c b/hw/ppc/spapr.c
index 6bfb908..f6f0af3 100644
--- a/hw/ppc/spapr.c
+++ b/hw/ppc/spapr.c
@@ -1934,11 +1934,13 @@  static void ppc_spapr_init(MachineState *machine)
         uint64_t lowaddr = 0;
 
         kernel_size = load_elf(kernel_filename, translate_kernel_address, NULL,
-                               NULL, &lowaddr, NULL, 1, PPC_ELF_MACHINE, 0);
+                               NULL, &lowaddr, NULL, 1, PPC_ELF_MACHINE,
+                               0, 0);
         if (kernel_size == ELF_LOAD_WRONG_ENDIAN) {
             kernel_size = load_elf(kernel_filename,
                                    translate_kernel_address, NULL,
-                                   NULL, &lowaddr, NULL, 0, PPC_ELF_MACHINE, 0);
+                                   NULL, &lowaddr, NULL, 0, PPC_ELF_MACHINE,
+                                   0, 0);
             kernel_le = kernel_size > 0;
         }
         if (kernel_size < 0) {
diff --git a/hw/ppc/virtex_ml507.c b/hw/ppc/virtex_ml507.c
index c2b5e44..12a6681 100644
--- a/hw/ppc/virtex_ml507.c
+++ b/hw/ppc/virtex_ml507.c
@@ -257,7 +257,8 @@  static void virtex_init(MachineState *machine)
 
         /* Boots a kernel elf binary.  */
         kernel_size = load_elf(kernel_filename, NULL, NULL,
-                               &entry, &low, &high, 1, PPC_ELF_MACHINE, 0);
+                               &entry, &low, &high, 1, PPC_ELF_MACHINE,
+                               0, 0);
         boot_info.bootstrap_pc = entry & 0x00ffffff;
 
         if (kernel_size < 0) {
diff --git a/hw/s390x/ipl.c b/hw/s390x/ipl.c
index b91fcc6..2446fa7 100644
--- a/hw/s390x/ipl.c
+++ b/hw/s390x/ipl.c
@@ -100,7 +100,7 @@  static void s390_ipl_realize(DeviceState *dev, Error **errp)
 
         bios_size = load_elf(bios_filename, bios_translate_addr, &fwbase,
                              &ipl->bios_start_addr, NULL, NULL, 1,
-                             EM_S390, 0);
+                             EM_S390, 0, 0);
         if (bios_size > 0) {
             /* Adjust ELF start address to final location */
             ipl->bios_start_addr += fwbase;
@@ -123,7 +123,7 @@  static void s390_ipl_realize(DeviceState *dev, Error **errp)
 
     if (ipl->kernel) {
         kernel_size = load_elf(ipl->kernel, NULL, NULL, &pentry, NULL,
-                               NULL, 1, EM_S390, 0);
+                               NULL, 1, EM_S390, 0, 0);
         if (kernel_size < 0) {
             kernel_size = load_image_targphys(ipl->kernel, 0, ram_size);
         }
diff --git a/hw/sparc/leon3.c b/hw/sparc/leon3.c
index 22d1e7e..e10b2a5 100644
--- a/hw/sparc/leon3.c
+++ b/hw/sparc/leon3.c
@@ -193,7 +193,7 @@  static void leon3_generic_hw_init(MachineState *machine)
         uint64_t entry;
 
         kernel_size = load_elf(kernel_filename, NULL, NULL, &entry, NULL, NULL,
-                               1 /* big endian */, EM_SPARC, 0);
+                               1 /* big endian */, EM_SPARC, 0, 0);
         if (kernel_size < 0) {
             fprintf(stderr, "qemu: could not load kernel '%s'\n",
                     kernel_filename);
diff --git a/hw/sparc/sun4m.c b/hw/sparc/sun4m.c
index 230dac9..0203bd8 100644
--- a/hw/sparc/sun4m.c
+++ b/hw/sparc/sun4m.c
@@ -300,7 +300,7 @@  static unsigned long sun4m_load_kernel(const char *kernel_filename,
         bswap_needed = 0;
 #endif
         kernel_size = load_elf(kernel_filename, translate_kernel_address, NULL,
-                               NULL, NULL, NULL, 1, EM_SPARC, 0);
+                               NULL, NULL, NULL, 1, EM_SPARC, 0, 0);
         if (kernel_size < 0)
             kernel_size = load_aout(kernel_filename, KERNEL_LOAD_ADDR,
                                     RAM_size - KERNEL_LOAD_ADDR, bswap_needed,
@@ -744,7 +744,7 @@  static void prom_init(hwaddr addr, const char *bios_name)
     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
     if (filename) {
         ret = load_elf(filename, translate_prom_address, &addr, NULL,
-                       NULL, NULL, 1, EM_SPARC, 0);
+                       NULL, NULL, 1, EM_SPARC, 0, 0);
         if (ret < 0 || ret > PROM_SIZE_MAX) {
             ret = load_image_targphys(filename, addr, PROM_SIZE_MAX);
         }
diff --git a/hw/sparc64/sun4u.c b/hw/sparc64/sun4u.c
index d6b929c..683605c 100644
--- a/hw/sparc64/sun4u.c
+++ b/hw/sparc64/sun4u.c
@@ -208,7 +208,7 @@  static uint64_t sun4u_load_kernel(const char *kernel_filename,
         bswap_needed = 0;
 #endif
         kernel_size = load_elf(kernel_filename, NULL, NULL, kernel_entry,
-                               kernel_addr, &kernel_top, 1, EM_SPARCV9, 0);
+                               kernel_addr, &kernel_top, 1, EM_SPARCV9, 0, 0);
         if (kernel_size < 0) {
             *kernel_addr = KERNEL_LOAD_ADDR;
             *kernel_entry = KERNEL_LOAD_ADDR;
@@ -671,7 +671,7 @@  static void prom_init(hwaddr addr, const char *bios_name)
     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
     if (filename) {
         ret = load_elf(filename, translate_prom_address, &addr,
-                       NULL, NULL, NULL, 1, EM_SPARCV9, 0);
+                       NULL, NULL, NULL, 1, EM_SPARCV9, 0, 0);
         if (ret < 0 || ret > PROM_SIZE_MAX) {
             ret = load_image_targphys(filename, addr, PROM_SIZE_MAX);
         }
diff --git a/hw/tricore/tricore_testboard.c b/hw/tricore/tricore_testboard.c
index 4ff5e7b..a1ed338 100644
--- a/hw/tricore/tricore_testboard.c
+++ b/hw/tricore/tricore_testboard.c
@@ -44,7 +44,7 @@  static void tricore_load_kernel(CPUTriCoreState *env)
     kernel_size = load_elf(tricoretb_binfo.kernel_filename, NULL,
                            NULL, (uint64_t *)&entry, NULL,
                            NULL, 0,
-                           EM_TRICORE, 1);
+                           EM_TRICORE, 1, 0);
     if (kernel_size <= 0) {
         error_report("qemu: no kernel file '%s'",
                 tricoretb_binfo.kernel_filename);
diff --git a/hw/xtensa/sim.c b/hw/xtensa/sim.c
index 6266b8d..6008eee 100644
--- a/hw/xtensa/sim.c
+++ b/hw/xtensa/sim.c
@@ -93,10 +93,10 @@  static void xtensa_sim_init(MachineState *machine)
         uint64_t elf_lowaddr;
 #ifdef TARGET_WORDS_BIGENDIAN
         int success = load_elf(kernel_filename, translate_phys_addr, cpu,
-                &elf_entry, &elf_lowaddr, NULL, 1, EM_XTENSA, 0);
+                &elf_entry, &elf_lowaddr, NULL, 1, EM_XTENSA, 0, 0);
 #else
         int success = load_elf(kernel_filename, translate_phys_addr, cpu,
-                &elf_entry, &elf_lowaddr, NULL, 0, EM_XTENSA, 0);
+                &elf_entry, &elf_lowaddr, NULL, 0, EM_XTENSA, 0, 0);
 #endif
         if (success > 0) {
             env->pc = elf_entry;
diff --git a/hw/xtensa/xtfpga.c b/hw/xtensa/xtfpga.c
index c1bc5ae..0bd0555 100644
--- a/hw/xtensa/xtfpga.c
+++ b/hw/xtensa/xtfpga.c
@@ -354,7 +354,7 @@  static void lx_init(const LxBoardDesc *board, MachineState *machine)
         uint64_t elf_entry;
         uint64_t elf_lowaddr;
         int success = load_elf(kernel_filename, translate_phys_addr, cpu,
-                &elf_entry, &elf_lowaddr, NULL, be, EM_XTENSA, 0);
+                &elf_entry, &elf_lowaddr, NULL, be, EM_XTENSA, 0, 0);
         if (success > 0) {
             entry_point = elf_entry;
         } else {
diff --git a/include/hw/elf_ops.h b/include/hw/elf_ops.h
index 0010c44..f510e7e 100644
--- a/include/hw/elf_ops.h
+++ b/include/hw/elf_ops.h
@@ -263,7 +263,7 @@  static int glue(load_elf, SZ)(const char *name, int fd,
                               void *translate_opaque,
                               int must_swab, uint64_t *pentry,
                               uint64_t *lowaddr, uint64_t *highaddr,
-                              int elf_machine, int clear_lsb)
+                              int elf_machine, int clear_lsb, int data_swab)
 {
     struct elfhdr ehdr;
     struct elf_phdr *phdr = NULL, *ph;
@@ -366,6 +366,26 @@  static int glue(load_elf, SZ)(const char *name, int fd,
                 addr = ph->p_paddr;
             }
 
+            if (data_swab) {
+                int j;
+                for (j = 0; j < file_size; j += (1 << data_swab)) {
+                    uint8_t *dp = data + j;
+                    switch (data_swab) {
+                    case (1):
+                        *(uint16_t *)dp = bswap16(*(uint16_t *)dp);
+                        break;
+                    case (2):
+                        *(uint32_t *)dp = bswap32(*(uint32_t *)dp);
+                        break;
+                    case (3):
+                        *(uint64_t *)dp = bswap64(*(uint64_t *)dp);
+                        break;
+                    default:
+                        g_assert_not_reached();
+                    }
+                }
+            }
+
             /* the entry pointer in the ELF header is a virtual
              * address, if the text segments paddr and vaddr differ
              * we need to adjust the entry */
diff --git a/include/hw/loader.h b/include/hw/loader.h
index 33067f8..e542575 100644
--- a/include/hw/loader.h
+++ b/include/hw/loader.h
@@ -35,7 +35,7 @@  const char *load_elf_strerror(int error);
 int load_elf(const char *filename, uint64_t (*translate_fn)(void *, uint64_t),
              void *translate_opaque, uint64_t *pentry, uint64_t *lowaddr,
              uint64_t *highaddr, int big_endian, int elf_machine,
-             int clear_lsb);
+             int clear_lsb, int data_swab);
 void load_elf_hdr(const char *filename, void *hdr, bool *is64, Error **errp);
 int load_aout(const char *filename, hwaddr addr, int max_sz,
               int bswap_needed, hwaddr target_page_size);