| Message ID | 20170410145247.6023-1-d-gerlach@ti.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
* Dave Gerlach <d-gerlach@ti.com> [170410 07:55]: > Currently the sram-exec functionality, which allows allocation of > executable memory and provides an API to move code to it, is only > selected in configs for the ARM architecture. Based on commit > 5756e9dd0de6 ("ARM: 6640/1: Thumb-2: Symbol manipulation macros for > function body copying") simply copying a C function pointer address > using memcpy without consideration of alignment and Thumb is unsafe on > ARM platforms. > > The aforementioned patch introduces the fncpy macro which is a safe way > to copy executable code on ARM platforms, so let's make use of that here > rather than the unsafe plain memcpy that was previously used by > sram_exec_copy. Now sram_exec_copy will move the code to "dst" and > return an address that is guaranteed to be safely callable. > > In the future, architectures hoping to make use of the sram-exec > functionality must define an fncpy macro just as ARM has done to > guarantee or check for safe copying to executable memory before allowing > the arch to select CONFIG_SRAM_EXEC. Looks good to me: Acked-by: Tony Lindgren <tony@atomide.com>
On Mon, Apr 10, 2017 at 09:52:47AM -0500, Dave Gerlach wrote: > Currently the sram-exec functionality, which allows allocation of > executable memory and provides an API to move code to it, is only > selected in configs for the ARM architecture. Based on commit > 5756e9dd0de6 ("ARM: 6640/1: Thumb-2: Symbol manipulation macros for > function body copying") simply copying a C function pointer address > using memcpy without consideration of alignment and Thumb is unsafe on > ARM platforms. > > The aforementioned patch introduces the fncpy macro which is a safe way > to copy executable code on ARM platforms, so let's make use of that here > rather than the unsafe plain memcpy that was previously used by > sram_exec_copy. Now sram_exec_copy will move the code to "dst" and > return an address that is guaranteed to be safely callable. > > In the future, architectures hoping to make use of the sram-exec > functionality must define an fncpy macro just as ARM has done to > guarantee or check for safe copying to executable memory before allowing > the arch to select CONFIG_SRAM_EXEC. > > Signed-off-by: Dave Gerlach <d-gerlach@ti.com> Looks a lot saner, thanks. It's just a bit sad that we lose the type checking. Acked-by: Russell King <rmk+kernel@armlinux.org.uk> > --- > > v1: http://www.spinics.net/lists/linux-omap/msg136517.html > > v2 changes: Return value of fncpy, as the returned address is the safely > executable one, and add supporting docs in comments. > > drivers/misc/sram-exec.c | 27 ++++++++++++++++++++------- > include/linux/sram.h | 8 ++++---- > 2 files changed, 24 insertions(+), 11 deletions(-) > > diff --git a/drivers/misc/sram-exec.c b/drivers/misc/sram-exec.c > index ac522417c462..9d54e14e8360 100644 > --- a/drivers/misc/sram-exec.c > +++ b/drivers/misc/sram-exec.c > @@ -19,6 +19,7 @@ > #include <linux/sram.h> > > #include <asm/cacheflush.h> > +#include <asm/fncpy.h> > > #include "sram.h" > > @@ -57,20 +58,32 @@ int sram_add_protect_exec(struct sram_partition *part) > * @src: Source address for the data to copy > * @size: Size of copy to perform, which starting from dst, must reside in pool > * > + * Return: Address for copied data that can safely be called through function > + * pointer, or NULL if problem. > + * > * This helper function allows sram driver to act as central control location > * of 'protect-exec' pools which are normal sram pools but are always set > * read-only and executable except when copying data to them, at which point > * they are set to read-write non-executable, to make sure no memory is > * writeable and executable at the same time. This region must be page-aligned > * and is checked during probe, otherwise page attribute manipulation would > - * not be possible. > + * not be possible. Care must be taken to only call the returned address as > + * dst address is not guaranteed to be safely callable. > + * > + * NOTE: This function uses the fncpy macro to move code to the executable > + * region. Some architectures have strict requirements for relocating > + * executable code, so fncpy is a macro that must be defined by any arch > + * making use of this functionality that guarantees a safe copy of exec > + * data and returns a safe address that can be called as a C function > + * pointer. > */ > -int sram_exec_copy(struct gen_pool *pool, void *dst, void *src, > - size_t size) > +void *sram_exec_copy(struct gen_pool *pool, void *dst, void *src, > + size_t size) > { > struct sram_partition *part = NULL, *p; > unsigned long base; > int pages; > + void *dst_cpy; > > mutex_lock(&exec_pool_list_mutex); > list_for_each_entry(p, &exec_pool_list, list) { > @@ -80,10 +93,10 @@ int sram_exec_copy(struct gen_pool *pool, void *dst, void *src, > mutex_unlock(&exec_pool_list_mutex); > > if (!part) > - return -EINVAL; > + return NULL; > > if (!addr_in_gen_pool(pool, (unsigned long)dst, size)) > - return -EINVAL; > + return NULL; > > base = (unsigned long)part->base; > pages = PAGE_ALIGN(size) / PAGE_SIZE; > @@ -93,13 +106,13 @@ int sram_exec_copy(struct gen_pool *pool, void *dst, void *src, > set_memory_nx((unsigned long)base, pages); > set_memory_rw((unsigned long)base, pages); > > - memcpy(dst, src, size); > + dst_cpy = fncpy(dst, src, size); > > set_memory_ro((unsigned long)base, pages); > set_memory_x((unsigned long)base, pages); > > mutex_unlock(&part->lock); > > - return 0; > + return dst_cpy; > } > EXPORT_SYMBOL_GPL(sram_exec_copy); > diff --git a/include/linux/sram.h b/include/linux/sram.h > index c97dcbe8ce25..4fb405fb0480 100644 > --- a/include/linux/sram.h > +++ b/include/linux/sram.h > @@ -16,12 +16,12 @@ > struct gen_pool; > > #ifdef CONFIG_SRAM_EXEC > -int sram_exec_copy(struct gen_pool *pool, void *dst, void *src, size_t size); > +void *sram_exec_copy(struct gen_pool *pool, void *dst, void *src, size_t size); > #else > -static inline int sram_exec_copy(struct gen_pool *pool, void *dst, void *src, > - size_t size) > +static inline void *sram_exec_copy(struct gen_pool *pool, void *dst, void *src, > + size_t size) > { > - return -ENODEV; > + return NULL; > } > #endif /* CONFIG_SRAM_EXEC */ > #endif /* __LINUX_SRAM_H__ */ > -- > 2.11.0 >
On 10/04/2017 at 09:52:47 -0500, Dave Gerlach wrote: > Currently the sram-exec functionality, which allows allocation of > executable memory and provides an API to move code to it, is only > selected in configs for the ARM architecture. Based on commit > 5756e9dd0de6 ("ARM: 6640/1: Thumb-2: Symbol manipulation macros for > function body copying") simply copying a C function pointer address > using memcpy without consideration of alignment and Thumb is unsafe on > ARM platforms. > > The aforementioned patch introduces the fncpy macro which is a safe way > to copy executable code on ARM platforms, so let's make use of that here > rather than the unsafe plain memcpy that was previously used by > sram_exec_copy. Now sram_exec_copy will move the code to "dst" and > return an address that is guaranteed to be safely callable. > > In the future, architectures hoping to make use of the sram-exec > functionality must define an fncpy macro just as ARM has done to > guarantee or check for safe copying to executable memory before allowing > the arch to select CONFIG_SRAM_EXEC. > > Signed-off-by: Dave Gerlach <d-gerlach@ti.com> Reviewed-by: Alexandre Belloni <alexandre.belloni@free-electrons.com> > --- > > v1: http://www.spinics.net/lists/linux-omap/msg136517.html > > v2 changes: Return value of fncpy, as the returned address is the safely > executable one, and add supporting docs in comments. > > drivers/misc/sram-exec.c | 27 ++++++++++++++++++++------- > include/linux/sram.h | 8 ++++---- > 2 files changed, 24 insertions(+), 11 deletions(-) > > diff --git a/drivers/misc/sram-exec.c b/drivers/misc/sram-exec.c > index ac522417c462..9d54e14e8360 100644 > --- a/drivers/misc/sram-exec.c > +++ b/drivers/misc/sram-exec.c > @@ -19,6 +19,7 @@ > #include <linux/sram.h> > > #include <asm/cacheflush.h> > +#include <asm/fncpy.h> > > #include "sram.h" > > @@ -57,20 +58,32 @@ int sram_add_protect_exec(struct sram_partition *part) > * @src: Source address for the data to copy > * @size: Size of copy to perform, which starting from dst, must reside in pool > * > + * Return: Address for copied data that can safely be called through function > + * pointer, or NULL if problem. > + * > * This helper function allows sram driver to act as central control location > * of 'protect-exec' pools which are normal sram pools but are always set > * read-only and executable except when copying data to them, at which point > * they are set to read-write non-executable, to make sure no memory is > * writeable and executable at the same time. This region must be page-aligned > * and is checked during probe, otherwise page attribute manipulation would > - * not be possible. > + * not be possible. Care must be taken to only call the returned address as > + * dst address is not guaranteed to be safely callable. > + * > + * NOTE: This function uses the fncpy macro to move code to the executable > + * region. Some architectures have strict requirements for relocating > + * executable code, so fncpy is a macro that must be defined by any arch > + * making use of this functionality that guarantees a safe copy of exec > + * data and returns a safe address that can be called as a C function > + * pointer. > */ > -int sram_exec_copy(struct gen_pool *pool, void *dst, void *src, > - size_t size) > +void *sram_exec_copy(struct gen_pool *pool, void *dst, void *src, > + size_t size) > { > struct sram_partition *part = NULL, *p; > unsigned long base; > int pages; > + void *dst_cpy; > > mutex_lock(&exec_pool_list_mutex); > list_for_each_entry(p, &exec_pool_list, list) { > @@ -80,10 +93,10 @@ int sram_exec_copy(struct gen_pool *pool, void *dst, void *src, > mutex_unlock(&exec_pool_list_mutex); > > if (!part) > - return -EINVAL; > + return NULL; > > if (!addr_in_gen_pool(pool, (unsigned long)dst, size)) > - return -EINVAL; > + return NULL; > > base = (unsigned long)part->base; > pages = PAGE_ALIGN(size) / PAGE_SIZE; > @@ -93,13 +106,13 @@ int sram_exec_copy(struct gen_pool *pool, void *dst, void *src, > set_memory_nx((unsigned long)base, pages); > set_memory_rw((unsigned long)base, pages); > > - memcpy(dst, src, size); > + dst_cpy = fncpy(dst, src, size); > > set_memory_ro((unsigned long)base, pages); > set_memory_x((unsigned long)base, pages); > > mutex_unlock(&part->lock); > > - return 0; > + return dst_cpy; > } > EXPORT_SYMBOL_GPL(sram_exec_copy); > diff --git a/include/linux/sram.h b/include/linux/sram.h > index c97dcbe8ce25..4fb405fb0480 100644 > --- a/include/linux/sram.h > +++ b/include/linux/sram.h > @@ -16,12 +16,12 @@ > struct gen_pool; > > #ifdef CONFIG_SRAM_EXEC > -int sram_exec_copy(struct gen_pool *pool, void *dst, void *src, size_t size); > +void *sram_exec_copy(struct gen_pool *pool, void *dst, void *src, size_t size); > #else > -static inline int sram_exec_copy(struct gen_pool *pool, void *dst, void *src, > - size_t size) > +static inline void *sram_exec_copy(struct gen_pool *pool, void *dst, void *src, > + size_t size) > { > - return -ENODEV; > + return NULL; > } > #endif /* CONFIG_SRAM_EXEC */ > #endif /* __LINUX_SRAM_H__ */ > -- > 2.11.0 >
* Russell King - ARM Linux <linux@armlinux.org.uk> [170503 11:58]: > On Mon, Apr 10, 2017 at 09:52:47AM -0500, Dave Gerlach wrote: > > Currently the sram-exec functionality, which allows allocation of > > executable memory and provides an API to move code to it, is only > > selected in configs for the ARM architecture. Based on commit > > 5756e9dd0de6 ("ARM: 6640/1: Thumb-2: Symbol manipulation macros for > > function body copying") simply copying a C function pointer address > > using memcpy without consideration of alignment and Thumb is unsafe on > > ARM platforms. > > > > The aforementioned patch introduces the fncpy macro which is a safe way > > to copy executable code on ARM platforms, so let's make use of that here > > rather than the unsafe plain memcpy that was previously used by > > sram_exec_copy. Now sram_exec_copy will move the code to "dst" and > > return an address that is guaranteed to be safely callable. > > > > In the future, architectures hoping to make use of the sram-exec > > functionality must define an fncpy macro just as ARM has done to > > guarantee or check for safe copying to executable memory before allowing > > the arch to select CONFIG_SRAM_EXEC. > > > > Signed-off-by: Dave Gerlach <d-gerlach@ti.com> > > Looks a lot saner, thanks. It's just a bit sad that we lose the type > checking. > > Acked-by: Russell King <rmk+kernel@armlinux.org.uk> Looks like this is still pending so I'll add it into omap-for-v4.12/fixes so we can get this out of the way. Regards, Tony
On Tue, May 16, 2017 at 09:01:27AM -0700, Tony Lindgren wrote: > * Russell King - ARM Linux <linux@armlinux.org.uk> [170503 11:58]: > > On Mon, Apr 10, 2017 at 09:52:47AM -0500, Dave Gerlach wrote: > > > Currently the sram-exec functionality, which allows allocation of > > > executable memory and provides an API to move code to it, is only > > > selected in configs for the ARM architecture. Based on commit > > > 5756e9dd0de6 ("ARM: 6640/1: Thumb-2: Symbol manipulation macros for > > > function body copying") simply copying a C function pointer address > > > using memcpy without consideration of alignment and Thumb is unsafe on > > > ARM platforms. > > > > > > The aforementioned patch introduces the fncpy macro which is a safe way > > > to copy executable code on ARM platforms, so let's make use of that here > > > rather than the unsafe plain memcpy that was previously used by > > > sram_exec_copy. Now sram_exec_copy will move the code to "dst" and > > > return an address that is guaranteed to be safely callable. > > > > > > In the future, architectures hoping to make use of the sram-exec > > > functionality must define an fncpy macro just as ARM has done to > > > guarantee or check for safe copying to executable memory before allowing > > > the arch to select CONFIG_SRAM_EXEC. > > > > > > Signed-off-by: Dave Gerlach <d-gerlach@ti.com> > > > > Looks a lot saner, thanks. It's just a bit sad that we lose the type > > checking. > > > > Acked-by: Russell King <rmk+kernel@armlinux.org.uk> > > Looks like this is still pending so I'll add it into > omap-for-v4.12/fixes so we can get this out of the way. It's a "fix"? Looked to be a 4.13 issue, sorry for the delay, otherwise I would have queued it up earlier. No objection for you to take this through your tree. greg k-h
On Wed, May 17, 2017 at 11:13:17AM +0200, Greg Kroah-Hartman wrote: > On Tue, May 16, 2017 at 09:01:27AM -0700, Tony Lindgren wrote: > > * Russell King - ARM Linux <linux@armlinux.org.uk> [170503 11:58]: > > > On Mon, Apr 10, 2017 at 09:52:47AM -0500, Dave Gerlach wrote: > > > > Currently the sram-exec functionality, which allows allocation of > > > > executable memory and provides an API to move code to it, is only > > > > selected in configs for the ARM architecture. Based on commit > > > > 5756e9dd0de6 ("ARM: 6640/1: Thumb-2: Symbol manipulation macros for > > > > function body copying") simply copying a C function pointer address > > > > using memcpy without consideration of alignment and Thumb is unsafe on > > > > ARM platforms. > > > > > > > > The aforementioned patch introduces the fncpy macro which is a safe way > > > > to copy executable code on ARM platforms, so let's make use of that here > > > > rather than the unsafe plain memcpy that was previously used by > > > > sram_exec_copy. Now sram_exec_copy will move the code to "dst" and > > > > return an address that is guaranteed to be safely callable. > > > > > > > > In the future, architectures hoping to make use of the sram-exec > > > > functionality must define an fncpy macro just as ARM has done to > > > > guarantee or check for safe copying to executable memory before allowing > > > > the arch to select CONFIG_SRAM_EXEC. > > > > > > > > Signed-off-by: Dave Gerlach <d-gerlach@ti.com> > > > > > > Looks a lot saner, thanks. It's just a bit sad that we lose the type > > > checking. > > > > > > Acked-by: Russell King <rmk+kernel@armlinux.org.uk> > > > > Looks like this is still pending so I'll add it into > > omap-for-v4.12/fixes so we can get this out of the way. > > It's a "fix"? Looked to be a 4.13 issue, sorry for the delay, otherwise > I would have queued it up earlier. Technically, it is a fix, but my greps for "sram_exec_copy" indicate that the code does not yet have any in-tree users. So I don't think there's any urgency to picking this up, and I think no need to back port to stable trees.
* Russell King - ARM Linux <linux@armlinux.org.uk> [170517 04:46]: > On Wed, May 17, 2017 at 11:13:17AM +0200, Greg Kroah-Hartman wrote: > > On Tue, May 16, 2017 at 09:01:27AM -0700, Tony Lindgren wrote: > > > * Russell King - ARM Linux <linux@armlinux.org.uk> [170503 11:58]: > > > > On Mon, Apr 10, 2017 at 09:52:47AM -0500, Dave Gerlach wrote: > > > > > Currently the sram-exec functionality, which allows allocation of > > > > > executable memory and provides an API to move code to it, is only > > > > > selected in configs for the ARM architecture. Based on commit > > > > > 5756e9dd0de6 ("ARM: 6640/1: Thumb-2: Symbol manipulation macros for > > > > > function body copying") simply copying a C function pointer address > > > > > using memcpy without consideration of alignment and Thumb is unsafe on > > > > > ARM platforms. > > > > > > > > > > The aforementioned patch introduces the fncpy macro which is a safe way > > > > > to copy executable code on ARM platforms, so let's make use of that here > > > > > rather than the unsafe plain memcpy that was previously used by > > > > > sram_exec_copy. Now sram_exec_copy will move the code to "dst" and > > > > > return an address that is guaranteed to be safely callable. > > > > > > > > > > In the future, architectures hoping to make use of the sram-exec > > > > > functionality must define an fncpy macro just as ARM has done to > > > > > guarantee or check for safe copying to executable memory before allowing > > > > > the arch to select CONFIG_SRAM_EXEC. > > > > > > > > > > Signed-off-by: Dave Gerlach <d-gerlach@ti.com> > > > > > > > > Looks a lot saner, thanks. It's just a bit sad that we lose the type > > > > checking. > > > > > > > > Acked-by: Russell King <rmk+kernel@armlinux.org.uk> > > > > > > Looks like this is still pending so I'll add it into > > > omap-for-v4.12/fixes so we can get this out of the way. > > > > It's a "fix"? Looked to be a 4.13 issue, sorry for the delay, otherwise > > I would have queued it up earlier. > > Technically, it is a fix, but my greps for "sram_exec_copy" indicate > that the code does not yet have any in-tree users. So I don't think > there's any urgency to picking this up, and I think no need to back > port to stable trees. OK fine, I'll drop it today from my fixes (and for-next) no problem. I did add a fixes tag to it which would then create confusion later on too with stable trees. Dave, probably best to resend the patch to Greg in few days with acks added and rebased against v4.12-rc1 because it won't apply without a merge because of the header changes. Regards, Tony
On 05/17/2017 08:47 AM, Tony Lindgren wrote: > * Russell King - ARM Linux <linux@armlinux.org.uk> [170517 04:46]: >> On Wed, May 17, 2017 at 11:13:17AM +0200, Greg Kroah-Hartman wrote: >>> On Tue, May 16, 2017 at 09:01:27AM -0700, Tony Lindgren wrote: >>>> * Russell King - ARM Linux <linux@armlinux.org.uk> [170503 11:58]: >>>>> On Mon, Apr 10, 2017 at 09:52:47AM -0500, Dave Gerlach wrote: >>>>>> Currently the sram-exec functionality, which allows allocation of >>>>>> executable memory and provides an API to move code to it, is only >>>>>> selected in configs for the ARM architecture. Based on commit >>>>>> 5756e9dd0de6 ("ARM: 6640/1: Thumb-2: Symbol manipulation macros for >>>>>> function body copying") simply copying a C function pointer address >>>>>> using memcpy without consideration of alignment and Thumb is unsafe on >>>>>> ARM platforms. >>>>>> >>>>>> The aforementioned patch introduces the fncpy macro which is a safe way >>>>>> to copy executable code on ARM platforms, so let's make use of that here >>>>>> rather than the unsafe plain memcpy that was previously used by >>>>>> sram_exec_copy. Now sram_exec_copy will move the code to "dst" and >>>>>> return an address that is guaranteed to be safely callable. >>>>>> >>>>>> In the future, architectures hoping to make use of the sram-exec >>>>>> functionality must define an fncpy macro just as ARM has done to >>>>>> guarantee or check for safe copying to executable memory before allowing >>>>>> the arch to select CONFIG_SRAM_EXEC. >>>>>> >>>>>> Signed-off-by: Dave Gerlach <d-gerlach@ti.com> >>>>> >>>>> Looks a lot saner, thanks. It's just a bit sad that we lose the type >>>>> checking. >>>>> >>>>> Acked-by: Russell King <rmk+kernel@armlinux.org.uk> >>>> >>>> Looks like this is still pending so I'll add it into >>>> omap-for-v4.12/fixes so we can get this out of the way. >>> >>> It's a "fix"? Looked to be a 4.13 issue, sorry for the delay, otherwise >>> I would have queued it up earlier. >> >> Technically, it is a fix, but my greps for "sram_exec_copy" indicate >> that the code does not yet have any in-tree users. So I don't think >> there's any urgency to picking this up, and I think no need to back >> port to stable trees. > > OK fine, I'll drop it today from my fixes (and for-next) no problem. > I did add a fixes tag to it which would then create confusion later > on too with stable trees. > > Dave, probably best to resend the patch to Greg in few days with acks > added and rebased against v4.12-rc1 because it won't apply without a > merge because of the header changes. Yes, there are no users yet. Was planning on resending this anyway, so I'll do that before I send my patches that make use of this. Regards, Dave > > Regards, > > Tony > > > > > >
On Mon, Apr 10, 2017 at 09:52:47AM -0500, Dave Gerlach wrote: > Currently the sram-exec functionality, which allows allocation of > executable memory and provides an API to move code to it, is only > selected in configs for the ARM architecture. Based on commit > 5756e9dd0de6 ("ARM: 6640/1: Thumb-2: Symbol manipulation macros for > function body copying") simply copying a C function pointer address > using memcpy without consideration of alignment and Thumb is unsafe on > ARM platforms. > > The aforementioned patch introduces the fncpy macro which is a safe way > to copy executable code on ARM platforms, so let's make use of that here > rather than the unsafe plain memcpy that was previously used by > sram_exec_copy. Now sram_exec_copy will move the code to "dst" and > return an address that is guaranteed to be safely callable. > > In the future, architectures hoping to make use of the sram-exec > functionality must define an fncpy macro just as ARM has done to > guarantee or check for safe copying to executable memory before allowing > the arch to select CONFIG_SRAM_EXEC. > > Signed-off-by: Dave Gerlach <d-gerlach@ti.com> > Acked-by: Russell King <rmk+kernel@armlinux.org.uk> > Reviewed-by: Alexandre Belloni <alexandre.belloni@free-electrons.com> > Acked-by: Tony Lindgren <tony@atomide.com> Can you rebase this and resend as it doesn't apply to the tree right now :( thanks, greg k-h
On 05/18/2017 10:01 AM, Greg Kroah-Hartman wrote: > On Mon, Apr 10, 2017 at 09:52:47AM -0500, Dave Gerlach wrote: >> Currently the sram-exec functionality, which allows allocation of >> executable memory and provides an API to move code to it, is only >> selected in configs for the ARM architecture. Based on commit >> 5756e9dd0de6 ("ARM: 6640/1: Thumb-2: Symbol manipulation macros for >> function body copying") simply copying a C function pointer address >> using memcpy without consideration of alignment and Thumb is unsafe on >> ARM platforms. >> >> The aforementioned patch introduces the fncpy macro which is a safe way >> to copy executable code on ARM platforms, so let's make use of that here >> rather than the unsafe plain memcpy that was previously used by >> sram_exec_copy. Now sram_exec_copy will move the code to "dst" and >> return an address that is guaranteed to be safely callable. >> >> In the future, architectures hoping to make use of the sram-exec >> functionality must define an fncpy macro just as ARM has done to >> guarantee or check for safe copying to executable memory before allowing >> the arch to select CONFIG_SRAM_EXEC. >> >> Signed-off-by: Dave Gerlach <d-gerlach@ti.com> >> Acked-by: Russell King <rmk+kernel@armlinux.org.uk> >> Reviewed-by: Alexandre Belloni <alexandre.belloni@free-electrons.com> >> Acked-by: Tony Lindgren <tony@atomide.com> > > Can you rebase this and resend as it doesn't apply to the tree right now > :( > Resent as v3 based on v4.12-rc1 and with Acks from v2 added, thanks. Regards, Dave > thanks, > > greg k-h >
diff --git a/drivers/misc/sram-exec.c b/drivers/misc/sram-exec.c index ac522417c462..9d54e14e8360 100644 --- a/drivers/misc/sram-exec.c +++ b/drivers/misc/sram-exec.c @@ -19,6 +19,7 @@ #include <linux/sram.h> #include <asm/cacheflush.h> +#include <asm/fncpy.h> #include "sram.h" @@ -57,20 +58,32 @@ int sram_add_protect_exec(struct sram_partition *part) * @src: Source address for the data to copy * @size: Size of copy to perform, which starting from dst, must reside in pool * + * Return: Address for copied data that can safely be called through function + * pointer, or NULL if problem. + * * This helper function allows sram driver to act as central control location * of 'protect-exec' pools which are normal sram pools but are always set * read-only and executable except when copying data to them, at which point * they are set to read-write non-executable, to make sure no memory is * writeable and executable at the same time. This region must be page-aligned * and is checked during probe, otherwise page attribute manipulation would - * not be possible. + * not be possible. Care must be taken to only call the returned address as + * dst address is not guaranteed to be safely callable. + * + * NOTE: This function uses the fncpy macro to move code to the executable + * region. Some architectures have strict requirements for relocating + * executable code, so fncpy is a macro that must be defined by any arch + * making use of this functionality that guarantees a safe copy of exec + * data and returns a safe address that can be called as a C function + * pointer. */ -int sram_exec_copy(struct gen_pool *pool, void *dst, void *src, - size_t size) +void *sram_exec_copy(struct gen_pool *pool, void *dst, void *src, + size_t size) { struct sram_partition *part = NULL, *p; unsigned long base; int pages; + void *dst_cpy; mutex_lock(&exec_pool_list_mutex); list_for_each_entry(p, &exec_pool_list, list) { @@ -80,10 +93,10 @@ int sram_exec_copy(struct gen_pool *pool, void *dst, void *src, mutex_unlock(&exec_pool_list_mutex); if (!part) - return -EINVAL; + return NULL; if (!addr_in_gen_pool(pool, (unsigned long)dst, size)) - return -EINVAL; + return NULL; base = (unsigned long)part->base; pages = PAGE_ALIGN(size) / PAGE_SIZE; @@ -93,13 +106,13 @@ int sram_exec_copy(struct gen_pool *pool, void *dst, void *src, set_memory_nx((unsigned long)base, pages); set_memory_rw((unsigned long)base, pages); - memcpy(dst, src, size); + dst_cpy = fncpy(dst, src, size); set_memory_ro((unsigned long)base, pages); set_memory_x((unsigned long)base, pages); mutex_unlock(&part->lock); - return 0; + return dst_cpy; } EXPORT_SYMBOL_GPL(sram_exec_copy); diff --git a/include/linux/sram.h b/include/linux/sram.h index c97dcbe8ce25..4fb405fb0480 100644 --- a/include/linux/sram.h +++ b/include/linux/sram.h @@ -16,12 +16,12 @@ struct gen_pool; #ifdef CONFIG_SRAM_EXEC -int sram_exec_copy(struct gen_pool *pool, void *dst, void *src, size_t size); +void *sram_exec_copy(struct gen_pool *pool, void *dst, void *src, size_t size); #else -static inline int sram_exec_copy(struct gen_pool *pool, void *dst, void *src, - size_t size) +static inline void *sram_exec_copy(struct gen_pool *pool, void *dst, void *src, + size_t size) { - return -ENODEV; + return NULL; } #endif /* CONFIG_SRAM_EXEC */ #endif /* __LINUX_SRAM_H__ */
Currently the sram-exec functionality, which allows allocation of executable memory and provides an API to move code to it, is only selected in configs for the ARM architecture. Based on commit 5756e9dd0de6 ("ARM: 6640/1: Thumb-2: Symbol manipulation macros for function body copying") simply copying a C function pointer address using memcpy without consideration of alignment and Thumb is unsafe on ARM platforms. The aforementioned patch introduces the fncpy macro which is a safe way to copy executable code on ARM platforms, so let's make use of that here rather than the unsafe plain memcpy that was previously used by sram_exec_copy. Now sram_exec_copy will move the code to "dst" and return an address that is guaranteed to be safely callable. In the future, architectures hoping to make use of the sram-exec functionality must define an fncpy macro just as ARM has done to guarantee or check for safe copying to executable memory before allowing the arch to select CONFIG_SRAM_EXEC. Signed-off-by: Dave Gerlach <d-gerlach@ti.com> --- v1: http://www.spinics.net/lists/linux-omap/msg136517.html v2 changes: Return value of fncpy, as the returned address is the safely executable one, and add supporting docs in comments. drivers/misc/sram-exec.c | 27 ++++++++++++++++++++------- include/linux/sram.h | 8 ++++---- 2 files changed, 24 insertions(+), 11 deletions(-)