[U-Boot] [PATCH] core: of_addr: Correct the size type of of_get_address to fdt_size_t

Currently the size parameter is defined as u64 type. Correct the size type of of_get_address to fdt_size_t so that both 64 bit and 32 bit architectures are taken care of.
The initial bug report: https://patchwork.ozlabs.org/patch/1090094/#2212555
Fixes: e679d03b08fb ("core: ofnode: Add ofnode_get_addr_size_index") Reported-by: Eugeniu Rosca rosca.eugeniu@gmail.com Tested-by: Eugeniu Rosca rosca.eugeniu@gmail.com Signed-off-by: Keerthy j-keerthy@ti.com ---
Changes from RFT:
* Fixed a typo in the commit log. * Added Reported-by: Eugeniu Rosca rosca.eugeniu@gmail.com Tested-by: Eugeniu Rosca rosca.eugeniu@gmail.com
drivers/core/of_addr.c | 4 ++-- drivers/core/ofnode.c | 2 +- include/dm/of_addr.h | 2 +- 3 files changed, 4 insertions(+), 4 deletions(-)
diff --git a/drivers/core/of_addr.c b/drivers/core/of_addr.c index 4e256d9926..812a400b19 100644 --- a/drivers/core/of_addr.c +++ b/drivers/core/of_addr.c @@ -122,7 +122,7 @@ static void dev_count_cells(const struct device_node *np, int *nap, int *nsp) }
const __be32 *of_get_address(const struct device_node *dev, int index, - u64 *size, unsigned int *flags) + fdt_size_t *size, unsigned int *flags) { const __be32 *prop; int psize; @@ -347,7 +347,7 @@ int of_address_to_resource(const struct device_node *dev, int index, struct resource *r) { const __be32 *addrp; - u64 size; + fdt_size_t size; unsigned int flags; const char *name = NULL;
diff --git a/drivers/core/ofnode.c b/drivers/core/ofnode.c index 2ac73af934..b21b5183a3 100644 --- a/drivers/core/ofnode.c +++ b/drivers/core/ofnode.c @@ -260,7 +260,7 @@ fdt_addr_t ofnode_get_addr_size_index(ofnode node, int index, fdt_size_t *size) uint flags;
prop_val = of_get_address(ofnode_to_np(node), index, - (u64 *)size, &flags); + size, &flags); if (!prop_val) return FDT_ADDR_T_NONE;
diff --git a/include/dm/of_addr.h b/include/dm/of_addr.h index 3fa1ffce81..52443ef296 100644 --- a/include/dm/of_addr.h +++ b/include/dm/of_addr.h @@ -58,7 +58,7 @@ u64 of_translate_dma_address(const struct device_node *no, const __be32 *in_addr * @return pointer to address which can be read */ const __be32 *of_get_address(const struct device_node *no, int index, - u64 *size, unsigned int *flags); + fdt_size_t *size, unsigned int *flags);
struct resource;

Hi Keerthy,
On Wed, Aug 14, 2019 at 03:56:48PM +0530, Keerthy wrote:
Currently the size parameter is defined as u64 type. Correct the size type of of_get_address to fdt_size_t so that both 64 bit and 32 bit architectures are taken care of.
The initial bug report: https://patchwork.ozlabs.org/patch/1090094/#2212555
Fixes: e679d03b08fb ("core: ofnode: Add ofnode_get_addr_size_index") Reported-by: Eugeniu Rosca rosca.eugeniu@gmail.com Tested-by: Eugeniu Rosca rosca.eugeniu@gmail.com Signed-off-by: Keerthy j-keerthy@ti.com
Changes from RFT:
- Fixed a typo in the commit log.
- Added Reported-by: Eugeniu Rosca rosca.eugeniu@gmail.com Tested-by: Eugeniu Rosca rosca.eugeniu@gmail.com
drivers/core/of_addr.c | 4 ++-- drivers/core/ofnode.c | 2 +- include/dm/of_addr.h | 2 +- 3 files changed, 4 insertions(+), 4 deletions(-)
diff --git a/drivers/core/of_addr.c b/drivers/core/of_addr.c index 4e256d9926..812a400b19 100644 --- a/drivers/core/of_addr.c +++ b/drivers/core/of_addr.c @@ -122,7 +122,7 @@ static void dev_count_cells(const struct device_node *np, int *nap, int *nsp) }
const __be32 *of_get_address(const struct device_node *dev, int index,
u64 *size, unsigned int *flags)
fdt_size_t *size, unsigned int *flags)
I took some time to also review the changes in addition to testing.
I can see that, since its inception in Linux [1], of_get_address() used 'u64*' type for its 'size' argument. That's still valid in v5.3-rc5. So, it looks to me that with this patch we diverge from Linux.
I would barely think that the ASAN issue being fixed in this patch exists in Linux, since the latter receives much more KASAN-enabled testing on regular basis.
Do you foresee any alternative fix w/o diverging from Linux?
[1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?i...

On 19/08/19 3:54 PM, Eugeniu Rosca wrote:
Hi Keerthy,
On Wed, Aug 14, 2019 at 03:56:48PM +0530, Keerthy wrote:
Currently the size parameter is defined as u64 type. Correct the size type of of_get_address to fdt_size_t so that both 64 bit and 32 bit architectures are taken care of.
The initial bug report: https://patchwork.ozlabs.org/patch/1090094/#2212555
Fixes: e679d03b08fb ("core: ofnode: Add ofnode_get_addr_size_index") Reported-by: Eugeniu Rosca rosca.eugeniu@gmail.com Tested-by: Eugeniu Rosca rosca.eugeniu@gmail.com Signed-off-by: Keerthy j-keerthy@ti.com
Changes from RFT:
- Fixed a typo in the commit log.
- Added Reported-by: Eugeniu Rosca rosca.eugeniu@gmail.com Tested-by: Eugeniu Rosca rosca.eugeniu@gmail.com
drivers/core/of_addr.c | 4 ++-- drivers/core/ofnode.c | 2 +- include/dm/of_addr.h | 2 +- 3 files changed, 4 insertions(+), 4 deletions(-)
diff --git a/drivers/core/of_addr.c b/drivers/core/of_addr.c index 4e256d9926..812a400b19 100644 --- a/drivers/core/of_addr.c +++ b/drivers/core/of_addr.c @@ -122,7 +122,7 @@ static void dev_count_cells(const struct device_node *np, int *nap, int *nsp) }
const __be32 *of_get_address(const struct device_node *dev, int index,
u64 *size, unsigned int *flags)
fdt_size_t *size, unsigned int *flags)
I took some time to also review the changes in addition to testing.
I can see that, since its inception in Linux [1], of_get_address() used 'u64*' type for its 'size' argument. That's still valid in v5.3-rc5. So, it looks to me that with this patch we diverge from Linux.
I would barely think that the ASAN issue being fixed in this patch exists in Linux, since the latter receives much more KASAN-enabled testing on regular basis.
Do you foresee any alternative fix w/o diverging from Linux?
I am afraid No but isn't fdt_size_t also right type to represent size?
[1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?i...

Hi Keerthy, Hi Simon, cc: Tom
On Tue, Aug 20, 2019 at 09:39:32AM +0530, Keerthy wrote:
On 19/08/19 3:54 PM, Eugeniu Rosca wrote:
[..]
I took some time to also review the changes in addition to testing.
I can see that, since its inception in Linux [1], of_get_address() used 'u64*' type for its 'size' argument. That's still valid in v5.3-rc5. So, it looks to me that with this patch we diverge from Linux.
I would barely think that the ASAN issue being fixed in this patch exists in Linux, since the latter receives much more KASAN-enabled testing on regular basis.
Do you foresee any alternative fix w/o diverging from Linux?
I am afraid No but isn't fdt_size_t also right type to represent size?
'fdt_size_t' is a U-Boot-ism, i.e. it exists nowhere else but in U-Boot. Just for the record, it appears to be added by Simon's v2013.04 commit 4397a2a80baef ("fdt: Add fdtdec_get_addr_size() to read reg properties")
IMHO injecting a U-Boot-specific data type into the prototype of a Linux-backported function has below major drawbacks:
- It is a non-upstream-able solution. Linux will unlikely accept 'fdt_size_t' as a new type simply b/c Linux OF code existed over a decade and it didn't need this type so far. - Mutilating Linux upstream code will make further U-Boot backports painful, time consuming and error-prone.
Are we on the same page here?
[1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?i...

Hi,
On Mon, 26 Aug 2019 at 11:16, Eugeniu Rosca erosca@de.adit-jv.com wrote:
Hi Keerthy, Hi Simon, cc: Tom
On Tue, Aug 20, 2019 at 09:39:32AM +0530, Keerthy wrote:
On 19/08/19 3:54 PM, Eugeniu Rosca wrote:
[..]
I took some time to also review the changes in addition to testing.
I can see that, since its inception in Linux [1], of_get_address() used 'u64*' type for its 'size' argument. That's still valid in v5.3-rc5. So, it looks to me that with this patch we diverge from Linux.
I would barely think that the ASAN issue being fixed in this patch exists in Linux, since the latter receives much more KASAN-enabled testing on regular basis.
Do you foresee any alternative fix w/o diverging from Linux?
I am afraid No but isn't fdt_size_t also right type to represent size?
'fdt_size_t' is a U-Boot-ism, i.e. it exists nowhere else but in U-Boot. Just for the record, it appears to be added by Simon's v2013.04 commit 4397a2a80baef ("fdt: Add fdtdec_get_addr_size() to read reg properties")
IMHO injecting a U-Boot-specific data type into the prototype of a Linux-backported function has below major drawbacks:
- It is a non-upstream-able solution. Linux will unlikely accept 'fdt_size_t' as a new type simply b/c Linux OF code existed over a decade and it didn't need this type so far.
- Mutilating Linux upstream code will make further U-Boot backports painful, time consuming and error-prone.
Are we on the same page here?
[1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?i...
Yes I agree.
What do people think about this approach?
https://gitlab.denx.de/u-boot/custodians/u-boot-dm/commit/82c65a88284c27a02d...
Regards, SImon
participants (3)
-
Eugeniu Rosca
-
Keerthy
-
Simon Glass