On Fri, Jan 24, 2020 at 09:24:16AM +0100, Clément Leger wrote:
> Hi Mathieu,
>
> ----- On 24 Jan, 2020, at 01:53, Mathieu Poirier mathieu.poirier@xxxxxxxxxx wrote:
>
> > Hi Clement,
> >
> > On Fri, Oct 04, 2019 at 08:42:20PM +0200, Clement Leger wrote:
> >> elf32 and elf64 mainly differ by their types. In order to avoid
> >> copy/pasting the whole loader code, generate static inline functions
> >> which will access values according to the elf class. It allows to keep a
> >> common loader basis.
> >> In order to accomodate both elf types sizes, the maximum size for a
> >> elf header member is chosen using the maximum value of both elf class.
> >>
> >> Signed-off-by: Clement Leger <cleger@xxxxxxxxx>
> >> Tested-by: Arnaud POULIQUEN <arnaud.pouliquen@xxxxxx>
> >> ---
> >> Changes in v2:
> >> - Add ELF64 support in documentation
> >>
> >
> > First and foremost please address the complaints from checkpatch.
>
> I fixed one typo in accommodate. The other checkpatch complaint is
> about missing parenthesis for macros with complex values which is
> unfortunately not possible since I'm generating inline functions.
>
> Did you have any other one ?
No
>
> >
> >> ---
> >> Documentation/remoteproc.txt | 2 +-
> >> drivers/remoteproc/remoteproc_elf_loader.c | 135 ++++++++++++++++++-----------
> >> drivers/remoteproc/remoteproc_elf_loader.h | 69 +++++++++++++++
> >> drivers/remoteproc/remoteproc_internal.h | 2 +-
> >> drivers/remoteproc/st_remoteproc.c | 2 +-
> >> include/linux/remoteproc.h | 4 +-
> >> 6 files changed, 157 insertions(+), 57 deletions(-)
> >> create mode 100644 drivers/remoteproc/remoteproc_elf_loader.h
> >>
> >> diff --git a/Documentation/remoteproc.txt b/Documentation/remoteproc.txt
> >> index 77fb03acdbb4..bf4f0c41ec4e 100644
> >> --- a/Documentation/remoteproc.txt
> >> +++ b/Documentation/remoteproc.txt
> >> @@ -230,7 +230,7 @@ in the used rings.
> >> Binary Firmware Structure
> >> =========================
> >>
> >> -At this point remoteproc only supports ELF32 firmware binaries. However,
> >> +At this point remoteproc supports ELF32 and ELF64 firmware binaries. However,
> >> it is quite expected that other platforms/devices which we'd want to
> >> support with this framework will be based on different binary formats.
> >>
> >> diff --git a/drivers/remoteproc/remoteproc_elf_loader.c
> >> b/drivers/remoteproc/remoteproc_elf_loader.c
> >> index b17d72ec8603..6a2d31d6092c 100644
> >> --- a/drivers/remoteproc/remoteproc_elf_loader.c
> >> +++ b/drivers/remoteproc/remoteproc_elf_loader.c
> >> @@ -31,6 +31,7 @@
> >> #include <linux/elf.h>
> >>
> >> #include "remoteproc_internal.h"
> >> +#include "remoteproc_elf_loader.h"
> >>
> >> /**
> >> * rproc_elf_sanity_check() - Sanity Check ELF firmware image
> >> @@ -43,8 +44,16 @@ int rproc_elf_sanity_check(struct rproc *rproc, const struct
> >> firmware *fw)
> >> {
> >> const char *name = rproc->firmware;
> >> struct device *dev = &rproc->dev;
> >> + /*
> >> + * Elf files are beginning with the same structure. Thus, to simplify
> >> + * header parsing, we can use the elf32_hdr one for both elf64 and
> >> + * elf32.
> >> + */
> >> struct elf32_hdr *ehdr;
> >> + u32 elf_shdr_size;
> >> + u64 phoff, shoff;
> >> char class;
> >> + u16 phnum;
> >>
> >> if (!fw) {
> >> dev_err(dev, "failed to load %s\n", name);
> >> @@ -58,9 +67,13 @@ int rproc_elf_sanity_check(struct rproc *rproc, const struct
> >> firmware *fw)
> >
> > In the current code [1] fw->size is compared against the size of an elf32_hdr.
> > If support for elf64 is added that code needs to be modified to check for the
> > right header size using fw_elf_get_class().
>
> Actually, the elf32 header is smaller than the elf64 one so this check is
> there to ensure that we have at least a minimal elf header (elf32).
> And since the class is derived from the header, you better have to check the
> header size before accessing it.
This is a chicken-and-egg issue... We can't derive the class from the header if
we don't check the header and we can't check the header if we don't have the
class. Since ehdr->e_ident[] is the same for both, it is probably safe to
access e_ident[EI_CLASS] without further checks.
>
> To be more clear, I could compare it to min(sizeof(struct elf32_hdr),
> sizeof(struct elf64_hdr)) or add a comment at least stating that since
> elf header contains the same fields for identification, we can use the
> elf32 one.
I am concerned about cases where we deal with an elf64 and fw->size is >=
sizeof(struct elf32_hdr) and < sizeof(struct elf64_hdr). With the current code
we won't catch the error. It might be that we do something like this:
size_t header_size;
header_size = (fw_elf_get_class(fw) == ELFCLASS32 ?
sizeof(structelf32_hdr):
sizeof(structelf64_hdr);
if (fw->size < header_size ) {
...
}
>
> >
> > [1]
> > https://elixir.bootlin.com/linux/v5.5-rc6/source/drivers/remoteproc/remoteproc_elf_loader.c#L46
> >
> >>
> >> ehdr = (struct elf32_hdr *)fw->data;
> >>
> >> - /* We only support ELF32 at this point */
> >> + if (memcmp(ehdr->e_ident, ELFMAG, SELFMAG)) {
> >> + dev_err(dev, "Image is corrupted (bad magic)\n");
> >> + return -EINVAL;
> >> + }
> >> +
> >
> > Is there a reason to move this check up where? If not please bring it back to
> > its original location, that is below:
> >
> > "if (fw->size < ehdr->e_shoff + sizeof(structelf32_shdr))"
> >
>
> This is because the new check for size uses elf_shdr_size which is derived from
> the class. And since the class is extracted from the elf header, we need to check
> the header to be correct first.
>
> >> class = ehdr->e_ident[EI_CLASS];
> >> - if (class != ELFCLASS32) {
> >> + if (class != ELFCLASS32 && class != ELFCLASS64) {
> >> dev_err(dev, "Unsupported class: %d\n", class);
> >> return -EINVAL;
> >> }
> >> @@ -75,26 +88,29 @@ int rproc_elf_sanity_check(struct rproc *rproc, const struct
> >> firmware *fw)
> >> return -EINVAL;
> >> }
> >>
> >> - if (fw->size < ehdr->e_shoff + sizeof(struct elf32_shdr)) {
> >> - dev_err(dev, "Image is too small\n");
> >> - return -EINVAL;
> >> - }
> >> + phoff = elf_hdr_e_phoff(class, fw->data);
> >> + shoff = elf_hdr_e_shoff(class, fw->data);
> >> + phnum = elf_hdr_e_phnum(class, fw->data);
> >> + elf_shdr_size = elf_size_of_shdr(class);
> >>
> >> - if (memcmp(ehdr->e_ident, ELFMAG, SELFMAG)) {
> >> - dev_err(dev, "Image is corrupted (bad magic)\n");
> >> + if (fw->size < shoff + elf_shdr_size) {
> >> + dev_err(dev, "Image is too small\n");
> >> return -EINVAL;
> >> }
> >>
> >> - if (ehdr->e_phnum == 0) {
> >> + if (phnum == 0) {
> >> dev_err(dev, "No loadable segments\n");
> >> return -EINVAL;
> >> }
> >>
> >> - if (ehdr->e_phoff > fw->size) {
> >> + if (phoff > fw->size) {
> >> dev_err(dev, "Firmware size is too small\n");
> >> return -EINVAL;
> >> }
> >>
> >> + dev_dbg(dev, "Firmware is an elf%d file\n",
> >> + class == ELFCLASS32 ? 32 : 64);
> >> +
> >
> > Yes, this is useful.
> >
> >> return 0;
> >> }
> >> EXPORT_SYMBOL(rproc_elf_sanity_check);
> >> @@ -110,11 +126,9 @@ EXPORT_SYMBOL(rproc_elf_sanity_check);
> >> * Note that the boot address is not a configurable property of all remote
> >> * processors. Some will always boot at a specific hard-coded address.
> >> */
> >> -u32 rproc_elf_get_boot_addr(struct rproc *rproc, const struct firmware *fw)
> >> +u64 rproc_elf_get_boot_addr(struct rproc *rproc, const struct firmware *fw)
> >> {
> >> - struct elf32_hdr *ehdr = (struct elf32_hdr *)fw->data;
> >> -
> >> - return ehdr->e_entry;
> >> + return elf_hdr_e_entry(fw_elf_get_class(fw), fw->data);
> >> }
> >> EXPORT_SYMBOL(rproc_elf_get_boot_addr);
> >>
> >> @@ -145,37 +159,41 @@ EXPORT_SYMBOL(rproc_elf_get_boot_addr);
> >> int rproc_elf_load_segments(struct rproc *rproc, const struct firmware *fw)
> >> {
> >> struct device *dev = &rproc->dev;
> >> - struct elf32_hdr *ehdr;
> >> - struct elf32_phdr *phdr;
> >> + const void *ehdr, *phdr;
> >> int i, ret = 0;
> >> + u16 phnum;
> >> const u8 *elf_data = fw->data;
> >> + u8 class = fw_elf_get_class(fw);
> >> + u32 elf_phdr_size = elf_size_of_phdr(class);
> >>
> >> - ehdr = (struct elf32_hdr *)elf_data;
> >> - phdr = (struct elf32_phdr *)(elf_data + ehdr->e_phoff);
> >> + ehdr = elf_data;
> >> + phnum = elf_hdr_e_phnum(class, ehdr);
> >> + phdr = elf_data + elf_hdr_e_phoff(class, ehdr);
> >>
> >> /* go through the available ELF segments */
> >> - for (i = 0; i < ehdr->e_phnum; i++, phdr++) {
> >> - u32 da = phdr->p_paddr;
> >> - u32 memsz = phdr->p_memsz;
> >> - u32 filesz = phdr->p_filesz;
> >> - u32 offset = phdr->p_offset;
> >> + for (i = 0; i < phnum; i++, phdr += elf_phdr_size) {
> >> + u64 da = elf_phdr_p_paddr(class, phdr);
> >> + u64 memsz = elf_phdr_p_memsz(class, phdr);
> >> + u64 filesz = elf_phdr_p_filesz(class, phdr);
> >> + u64 offset = elf_phdr_p_offset(class, phdr);
> >> + u32 type = elf_phdr_p_type(class, phdr);
> >> void *ptr;
> >>
> >> - if (phdr->p_type != PT_LOAD)
> >> + if (type != PT_LOAD)
> >> continue;
> >>
> >> - dev_dbg(dev, "phdr: type %d da 0x%x memsz 0x%x filesz 0x%x\n",
> >> - phdr->p_type, da, memsz, filesz);
> >> + dev_dbg(dev, "phdr: type %d da 0x%llx memsz 0x%llx filesz 0x%llx\n",
> >> + type, da, memsz, filesz);
> >>
> >> if (filesz > memsz) {
> >> - dev_err(dev, "bad phdr filesz 0x%x memsz 0x%x\n",
> >> + dev_err(dev, "bad phdr filesz 0x%llx memsz 0x%llx\n",
> >> filesz, memsz);
> >> ret = -EINVAL;
> >> break;
> >> }
> >>
> >> if (offset + filesz > fw->size) {
> >> - dev_err(dev, "truncated fw: need 0x%x avail 0x%zx\n",
> >> + dev_err(dev, "truncated fw: need 0x%llx avail 0x%zx\n",
> >> offset + filesz, fw->size);
> >> ret = -EINVAL;
> >> break;
> >> @@ -184,14 +202,15 @@ int rproc_elf_load_segments(struct rproc *rproc, const
> >> struct firmware *fw)
> >> /* grab the kernel address for this device address */
> >> ptr = rproc_da_to_va(rproc, da, memsz);
> >> if (!ptr) {
> >> - dev_err(dev, "bad phdr da 0x%x mem 0x%x\n", da, memsz);
> >> + dev_err(dev, "bad phdr da 0x%llx mem 0x%llx\n", da,
> >> + memsz);
> >> ret = -EINVAL;
> >> break;
> >> }
> >>
> >> /* put the segment where the remote processor expects it */
> >> - if (phdr->p_filesz)
> >> - memcpy(ptr, elf_data + phdr->p_offset, filesz);
> >> + if (filesz)
> >> + memcpy(ptr, elf_data + offset, filesz);
> >>
> >> /*
> >> * Zero out remaining memory for this segment.
> >> @@ -208,24 +227,32 @@ int rproc_elf_load_segments(struct rproc *rproc, const
> >> struct firmware *fw)
> >> }
> >> EXPORT_SYMBOL(rproc_elf_load_segments);
> >>
> >> -static struct elf32_shdr *
> >> -find_table(struct device *dev, struct elf32_hdr *ehdr, size_t fw_size)
> >> +static const void *
> >
> > Not sure I understand the "const" - was the compiler complaining?
>
> It's actually caused by the fact I used a const void* shdr in the caller.
> I will check if this is mandatory.
>
> >
> >> +find_table(struct device *dev, const struct firmware *fw)
> >> {
> >> - struct elf32_shdr *shdr;
> >> + const void *shdr, *name_table_shdr;
> >> int i;
> >> const char *name_table;
> >> struct resource_table *table = NULL;
> >> - const u8 *elf_data = (void *)ehdr;
> >> + const u8 *elf_data = (void *)fw->data;
> >> + u8 class = fw_elf_get_class(fw);
> >> + size_t fw_size = fw->size;
> >> + const void *ehdr = elf_data;
> >> + u16 shnum = elf_hdr_e_shnum(class, ehdr);
> >> + u32 elf_shdr_size = elf_size_of_shdr(class);
> >> + u16 shstrndx = elf_hdr_e_shstrndx(class, ehdr);
> >>
> >> /* look for the resource table and handle it */
> >> - shdr = (struct elf32_shdr *)(elf_data + ehdr->e_shoff);
> >> - name_table = elf_data + shdr[ehdr->e_shstrndx].sh_offset;
> >> + shdr = elf_data + elf_hdr_e_shoff(class, ehdr);
> >> + name_table_shdr = shdr + (shstrndx * elf_shdr_size);
> >> + name_table = elf_data + elf_shdr_sh_offset(class, name_table_shdr);
> >
> > It took me a while but I figured out what is happening here. To save me (and
> > other people) from going through the same mental process every time I look at
> > this code, please add a comment for each of the above 3 lines.
>
> Indeed.
>
> >
> >>
> >> - for (i = 0; i < ehdr->e_shnum; i++, shdr++) {
> >> - u32 size = shdr->sh_size;
> >> - u32 offset = shdr->sh_offset;
> >> + for (i = 0; i < shnum; i++, shdr += elf_shdr_size) {
> >> + u64 size = elf_shdr_sh_size(class, shdr);
> >> + u64 offset = elf_shdr_sh_offset(class, shdr);
> >> + u32 name = elf_shdr_sh_name(class, shdr);
> >>
> >> - if (strcmp(name_table + shdr->sh_name, ".resource_table"))
> >> + if (strcmp(name_table + name, ".resource_table"))
> >> continue;
> >>
> >> table = (struct resource_table *)(elf_data + offset);
> >> @@ -279,21 +306,21 @@ find_table(struct device *dev, struct elf32_hdr *ehdr,
> >> size_t fw_size)
> >> */
> >> int rproc_elf_load_rsc_table(struct rproc *rproc, const struct firmware *fw)
> >> {
> >> - struct elf32_hdr *ehdr;
> >> - struct elf32_shdr *shdr;
> >> + const void *shdr;
> >> struct device *dev = &rproc->dev;
> >> struct resource_table *table = NULL;
> >> const u8 *elf_data = fw->data;
> >> size_t tablesz;
> >> + u8 class = fw_elf_get_class(fw);
> >> + u64 sh_offset;
> >>
> >> - ehdr = (struct elf32_hdr *)elf_data;
> >> -
> >> - shdr = find_table(dev, ehdr, fw->size);
> >> + shdr = find_table(dev, fw);
> >> if (!shdr)
> >> return -EINVAL;
> >>
> >> - table = (struct resource_table *)(elf_data + shdr->sh_offset);
> >> - tablesz = shdr->sh_size;
> >> + sh_offset = elf_shdr_sh_offset(class, shdr);
> >> + table = (struct resource_table *)(elf_data + sh_offset);
> >> + tablesz = elf_shdr_sh_size(class, shdr);
> >>
> >> /*
> >> * Create a copy of the resource table. When a virtio device starts
> >> @@ -326,13 +353,17 @@ EXPORT_SYMBOL(rproc_elf_load_rsc_table);
> >> struct resource_table *rproc_elf_find_loaded_rsc_table(struct rproc *rproc,
> >> const struct firmware *fw)
> >> {
> >> - struct elf32_hdr *ehdr = (struct elf32_hdr *)fw->data;
> >> - struct elf32_shdr *shdr;
> >> + const void *shdr;
> >> + u64 sh_addr, sh_size;
> >> + u8 class = fw_elf_get_class(fw);
> >>
> >> - shdr = find_table(&rproc->dev, ehdr, fw->size);
> >> + shdr = find_table(&rproc->dev, fw);
> >> if (!shdr)
> >> return NULL;
> >>
> >> - return rproc_da_to_va(rproc, shdr->sh_addr, shdr->sh_size);
> >> + sh_addr = elf_shdr_sh_addr(class, shdr);
> >> + sh_size = elf_shdr_sh_size(class, shdr);
> >> +
> >> + return rproc_da_to_va(rproc, sh_addr, sh_size);
> >
> > The prototype for the above is as follow:
> >
> > void *rproc_da_to_va(struct rproc *rproc, u64 da, int len)
> >
> > But sh_size is a u64, which will cause problem that are hard to debug. I think
> > it is better to move 'len' to an 8 byte type along with the refactoring of the
> > existing code that is implied. I suggest to split this work in a preparatory
> > patch (that will still be part of this set).
>
> Nice catch ! I will do that.
>
> >
> >> }
> >> EXPORT_SYMBOL(rproc_elf_find_loaded_rsc_table);
> >> diff --git a/drivers/remoteproc/remoteproc_elf_loader.h
> >> b/drivers/remoteproc/remoteproc_elf_loader.h
> >> new file mode 100644
> >> index 000000000000..fac3565734f9
> >> --- /dev/null
> >> +++ b/drivers/remoteproc/remoteproc_elf_loader.h
> >> @@ -0,0 +1,69 @@
> >> +/* SPDX-License-Identifier: GPL-2.0 */
> >> +/*
> >> + * Remote processor elf loader defines
> >> + *
> >> + * Copyright (C) 2019 Kalray, Inc.
> >> + */
> >> +
> >> +#ifndef REMOTEPROC_ELF_LOADER_H
> >> +#define REMOTEPROC_ELF_LOADER_H
> >> +
> >> +#include <linux/elf.h>
> >> +#include <linux/types.h>
> >> +
> >> +/**
> >> + * fw_elf_get_class - Get elf class
> >> + * @fw: the ELF firmware image
> >> + *
> >> + * Note that we use and elf32_hdr to access the class since the start of the
> >> + * struct is the same for both elf class
> >> + *
> >> + * Return: elf class of the firmware
> >> + */
> >> +static inline u8 fw_elf_get_class(const struct firmware *fw)
> >> +{
> >> + struct elf32_hdr *ehdr = (struct elf32_hdr *)fw->data;
> >> +
> >> + return ehdr->e_ident[EI_CLASS];
> >> +}
> >> +
> >> +#define ELF_GET_FIELD(__s, __field, __type) \
> >> +static inline __type elf_##__s##_##__field(u8 class, const void *arg) \
> >> +{ \
> >> + if (class == ELFCLASS32) \
> >> + return (__type) ((const struct elf32_##__s *) arg)->__field; \
> >> + else \
> >> + return (__type) ((const struct elf64_##__s *) arg)->__field; \
> >> +}
> >> +
> >> +ELF_GET_FIELD(hdr, e_entry, u64)
> >> +ELF_GET_FIELD(hdr, e_phnum, u16)
> >> +ELF_GET_FIELD(hdr, e_shnum, u16)
> >> +ELF_GET_FIELD(hdr, e_phoff, u64)
> >> +ELF_GET_FIELD(hdr, e_shoff, u64)
> >> +ELF_GET_FIELD(hdr, e_shstrndx, u16)
> >> +
> >> +ELF_GET_FIELD(phdr, p_paddr, u64)
> >> +ELF_GET_FIELD(phdr, p_filesz, u64)
> >> +ELF_GET_FIELD(phdr, p_memsz, u64)
> >> +ELF_GET_FIELD(phdr, p_type, u32)
> >> +ELF_GET_FIELD(phdr, p_offset, u64)
> >> +
> >> +ELF_GET_FIELD(shdr, sh_size, u64)
> >> +ELF_GET_FIELD(shdr, sh_offset, u64)
> >> +ELF_GET_FIELD(shdr, sh_name, u32)
> >> +ELF_GET_FIELD(shdr, sh_addr, u64)
> >
> > I like how you did this.
> >
> >> +
> >> +#define ELF_STRUCT_SIZE(__s) \
> >> +static inline unsigned long elf_size_of_##__s(u8 class) \
> >> +{ \
> >> + if (class == ELFCLASS32)\
> >> + return sizeof(struct elf32_##__s); \
> >> + else \
> >> + return sizeof(struct elf64_##__s); \
> >> +}
> >> +
> >> +ELF_STRUCT_SIZE(shdr)
> >> +ELF_STRUCT_SIZE(phdr)
> >> +
> >> +#endif /* REMOTEPROC_ELF_LOADER_H */
> >> diff --git a/drivers/remoteproc/remoteproc_internal.h
> >> b/drivers/remoteproc/remoteproc_internal.h
> >> index 45ff76a06c72..4ef745e3a1bc 100644
> >> --- a/drivers/remoteproc/remoteproc_internal.h
> >> +++ b/drivers/remoteproc/remoteproc_internal.h
> >> @@ -63,7 +63,7 @@ phys_addr_t rproc_va_to_pa(void *cpu_addr);
> >> int rproc_trigger_recovery(struct rproc *rproc);
> >>
> >> int rproc_elf_sanity_check(struct rproc *rproc, const struct firmware *fw);
> >> -u32 rproc_elf_get_boot_addr(struct rproc *rproc, const struct firmware *fw);
> >> +u64 rproc_elf_get_boot_addr(struct rproc *rproc, const struct firmware *fw);
> >> int rproc_elf_load_segments(struct rproc *rproc, const struct firmware *fw);
> >> int rproc_elf_load_rsc_table(struct rproc *rproc, const struct firmware *fw);
> >> struct resource_table *rproc_elf_find_loaded_rsc_table(struct rproc *rproc,
> >> diff --git a/drivers/remoteproc/st_remoteproc.c
> >> b/drivers/remoteproc/st_remoteproc.c
> >> index 51049d17b1e5..e23abd8a96b0 100644
> >> --- a/drivers/remoteproc/st_remoteproc.c
> >> +++ b/drivers/remoteproc/st_remoteproc.c
> >> @@ -193,7 +193,7 @@ static int st_rproc_start(struct rproc *rproc)
> >> }
> >> }
> >>
> >> - dev_info(&rproc->dev, "Started from 0x%x\n", rproc->bootaddr);
> >> + dev_info(&rproc->dev, "Started from 0x%llx\n", rproc->bootaddr);
> >>
> >> return 0;
> >>
> >> diff --git a/include/linux/remoteproc.h b/include/linux/remoteproc.h
> >> index 04d04709f2bd..512de9a2590c 100644
> >> --- a/include/linux/remoteproc.h
> >> +++ b/include/linux/remoteproc.h
> >> @@ -362,7 +362,7 @@ struct rproc_ops {
> >> struct rproc *rproc, const struct firmware *fw);
> >> int (*load)(struct rproc *rproc, const struct firmware *fw);
> >> int (*sanity_check)(struct rproc *rproc, const struct firmware *fw);
> >> - u32 (*get_boot_addr)(struct rproc *rproc, const struct firmware *fw);
> >> + u64 (*get_boot_addr)(struct rproc *rproc, const struct firmware *fw);
> >> };
> >>
> >> /**
> >> @@ -478,7 +478,7 @@ struct rproc {
> >> int num_traces;
> >> struct list_head carveouts;
> >> struct list_head mappings;
> >> - u32 bootaddr;
> >> + u64 bootaddr;
> >> struct list_head rvdevs;
> >> struct list_head subdevs;
> >> struct idr notifyids;
> >
> > Please hold off before doing another respin of this patch. While doing
> > something completely different I noticed TI also did some work in this area.
> > I'd like to take some time to look at their implementation and see if they carry
> > features that haven't been included here. I intend to do this tomorrow.
>
> Ok,
>
> Thanks for your review,
>
> Clément
>
> >
> > Thanks,
> > Mathieu
> >
> >> --
> >> 2.15.0.276.g89ea799
