[U-Boot] [PATCH] ML: SPL: NOR: Add CONFIG_SPL_NOR_COPY_ENTIRE_IMAGE define to enable whole image copy from NOR
This define gives the possibility to copy entire image (including header) from NOR parallel memory to e.g. SDRAM.
The legacy behavior is preserved, since other board don't enabled this option.
Signed-off-by: Lukasz Majewski l.majewski@majess.pl --- Kconfig | 10 ++++++++++ README | 4 ++++ common/spl/spl_nor.c | 12 +++++++++--- 3 files changed, 23 insertions(+), 3 deletions(-)
diff --git a/Kconfig b/Kconfig index e443490..473a099 100644 --- a/Kconfig +++ b/Kconfig @@ -159,6 +159,16 @@ config SPL_NOR_SUPPORT help This option enables SPL NOR driver.
+config SPL_NOR_COPY_ENTIRE_IMAGE + bool + depends on SPL_NOR_SUPPORT + prompt "Copy entire image from NOR memory" + default n + help + By default the SPL NOR driver supports copying only payload to + destination address. Say Y if you want to copy entire image (including + its image header). + config SPL_SYS_MALLOC_SIMPLE bool depends on SPL diff --git a/README b/README index f41a6af..e9a3c5b 100644 --- a/README +++ b/README @@ -3661,6 +3661,10 @@ FIT uImage format: It conflicts with SPL env from storage medium specified by CONFIG_ENV_IS_xxx but CONFIG_ENV_IS_NOWHERE
+ CONFIG_SPL_NOR_COPY_ENTIRE_IMAGE + Allow NOR stored u-boot image to be copied by SPL with header to + SDRAM memory and executed from it + CONFIG_SPL_PAD_TO Image offset to which the SPL should be padded before appending the SPL payload. By default, this is defined as diff --git a/common/spl/spl_nor.c b/common/spl/spl_nor.c index 8ea874c..1e0f739 100644 --- a/common/spl/spl_nor.c +++ b/common/spl/spl_nor.c @@ -9,13 +9,15 @@
int spl_nor_load_image(void) { + void *img_src; int ret; +#ifndef CONFIG_SPL_NOR_COPY_ENTIRE_IMAGE /* * Loading of the payload to SDRAM is done with skipping of * the mkimage header in this SPL NOR driver */ spl_image.flags |= SPL_COPY_PAYLOAD_ONLY; - +#endif #ifdef CONFIG_SPL_OS_BOOT if (!spl_start_uboot()) { const struct image_header *header; @@ -64,9 +66,13 @@ int spl_nor_load_image(void) if (ret) return ret;
+ img_src = (void *)CONFIG_SYS_UBOOT_BASE; +#ifndef CONFIG_SPL_NOR_COPY_ENTIRE_IMAGE + img_src += sizeof(struct image_header)); +#endif + memcpy((void *)(unsigned long)spl_image.load_addr, - (void *)(CONFIG_SYS_UBOOT_BASE + sizeof(struct image_header)), - spl_image.size); + img_src, spl_image.size);
return 0; }
On Fri, Sep 16, 2016 at 10:43:51AM +0200, Lukasz Majewski wrote:
This define gives the possibility to copy entire image (including header) from NOR parallel memory to e.g. SDRAM.
The legacy behavior is preserved, since other board don't enabled this option.
What's the use case?
Signed-off-by: Lukasz Majewski l.majewski@majess.pl
Kconfig | 10 ++++++++++
This should now get moved to common/spl/Kconfig
README | 4 ++++
This can be dropped.
[snip]
diff --git a/common/spl/spl_nor.c b/common/spl/spl_nor.c index 8ea874c..1e0f739 100644 --- a/common/spl/spl_nor.c +++ b/common/spl/spl_nor.c @@ -9,13 +9,15 @@
int spl_nor_load_image(void) {
- void *img_src; int ret;
+#ifndef CONFIG_SPL_NOR_COPY_ENTIRE_IMAGE /* * Loading of the payload to SDRAM is done with skipping of * the mkimage header in this SPL NOR driver */ spl_image.flags |= SPL_COPY_PAYLOAD_ONLY;
+#endif #ifdef CONFIG_SPL_OS_BOOT if (!spl_start_uboot()) { const struct image_header *header; @@ -64,9 +66,13 @@ int spl_nor_load_image(void) if (ret) return ret;
We should have a comment here about how we're calculating the base of where to copy from.
- img_src = (void *)CONFIG_SYS_UBOOT_BASE;
+#ifndef CONFIG_SPL_NOR_COPY_ENTIRE_IMAGE
- img_src += sizeof(struct image_header));
+#endif
- memcpy((void *)(unsigned long)spl_image.load_addr,
(void *)(CONFIG_SYS_UBOOT_BASE + sizeof(struct image_header)),spl_image.size);
img_src, spl_image.size);return 0;
}
Thanks!
Hi Tom,
On Fri, Sep 16, 2016 at 10:43:51AM +0200, Lukasz Majewski wrote:
This define gives the possibility to copy entire image (including header) from NOR parallel memory to e.g. SDRAM.
The legacy behavior is preserved, since other board don't enabled this option.
What's the use case?
This code is used on some boards (ppc [1]) in which we copy only payload of the image (i.e. we drop the package metadata).
In my particular case I would like to copy the image with its meta data (in similar way works spl_nand.c and spl_mmc.c).
Since, I wanted to avoid code duplication, I've decided to add this flag.
Signed-off-by: Lukasz Majewski l.majewski@majess.pl
Kconfig | 10 ++++++++++
This should now get moved to common/spl/Kconfig
OK.
As a side note - I do have some other patches which move some SPL_NOR options to Kconfig.
However, I omitted them when I saw great work done by Simon to move a lot of SPL options to Kconfig. I hope them will be merged soon :-)
README | 4 ++++
This can be dropped.
So, we don't add any new entries to README and now rely solely on Kconfig "help" sections?
[snip]
diff --git a/common/spl/spl_nor.c b/common/spl/spl_nor.c index 8ea874c..1e0f739 100644 --- a/common/spl/spl_nor.c +++ b/common/spl/spl_nor.c @@ -9,13 +9,15 @@
int spl_nor_load_image(void) {
- void *img_src; int ret;
+#ifndef CONFIG_SPL_NOR_COPY_ENTIRE_IMAGE /* * Loading of the payload to SDRAM is done with skipping of * the mkimage header in this SPL NOR driver */ spl_image.flags |= SPL_COPY_PAYLOAD_ONLY;
+#endif #ifdef CONFIG_SPL_OS_BOOT if (!spl_start_uboot()) { const struct image_header *header; @@ -64,9 +66,13 @@ int spl_nor_load_image(void) if (ret) return ret;
We should have a comment here about how we're calculating the base of where to copy from.
I will be more verbose with v2 :-).
Thanks for review.
- img_src = (void *)CONFIG_SYS_UBOOT_BASE;
+#ifndef CONFIG_SPL_NOR_COPY_ENTIRE_IMAGE
- img_src += sizeof(struct image_header));
+#endif
- memcpy((void *)(unsigned long)spl_image.load_addr,
(void *)(CONFIG_SYS_UBOOT_BASE + sizeof(structimage_header)),
spl_image.size);
img_src, spl_image.size);return 0;
}
Thanks!
Best regards, Łukasz Majewski
Notes: [1] - Extract from source code (spl_nor.c):
/* * On some system (e.g. powerpc), the load-address and * entry-point is located at address 0. We can't load * to 0-0x40. So skip header in this case. */
On Sun, Sep 18, 2016 at 09:10:34PM +0200, Lukasz Majewski wrote:
Hi Tom,
On Fri, Sep 16, 2016 at 10:43:51AM +0200, Lukasz Majewski wrote:
This define gives the possibility to copy entire image (including header) from NOR parallel memory to e.g. SDRAM.
The legacy behavior is preserved, since other board don't enabled this option.
What's the use case?
This code is used on some boards (ppc [1]) in which we copy only payload of the image (i.e. we drop the package metadata).
In my particular case I would like to copy the image with its meta data (in similar way works spl_nand.c and spl_mmc.c).
Since, I wanted to avoid code duplication, I've decided to add this flag.
So another way of saying this, if I follow correctly, is that today SPL NOR doesn't work with u-boot.img but only u-boot.bin ?
Signed-off-by: Lukasz Majewski l.majewski@majess.pl
Kconfig | 10 ++++++++++
This should now get moved to common/spl/Kconfig
OK.
As a side note - I do have some other patches which move some SPL_NOR options to Kconfig.
However, I omitted them when I saw great work done by Simon to move a lot of SPL options to Kconfig. I hope them will be merged soon :-)
It's in now :)
README | 4 ++++
This can be dropped.
So, we don't add any new entries to README and now rely solely on Kconfig "help" sections?
Correct.
[snip]
diff --git a/common/spl/spl_nor.c b/common/spl/spl_nor.c index 8ea874c..1e0f739 100644 --- a/common/spl/spl_nor.c +++ b/common/spl/spl_nor.c @@ -9,13 +9,15 @@
int spl_nor_load_image(void) {
- void *img_src; int ret;
+#ifndef CONFIG_SPL_NOR_COPY_ENTIRE_IMAGE /* * Loading of the payload to SDRAM is done with skipping of * the mkimage header in this SPL NOR driver */ spl_image.flags |= SPL_COPY_PAYLOAD_ONLY;
+#endif #ifdef CONFIG_SPL_OS_BOOT if (!spl_start_uboot()) { const struct image_header *header; @@ -64,9 +66,13 @@ int spl_nor_load_image(void) if (ret) return ret;
We should have a comment here about how we're calculating the base of where to copy from.
I will be more verbose with v2 :-).
Thanks for review.
No problem, thank you!
Hi Tom,
On Sun, Sep 18, 2016 at 09:10:34PM +0200, Lukasz Majewski wrote:
Hi Tom,
On Fri, Sep 16, 2016 at 10:43:51AM +0200, Lukasz Majewski wrote:
This define gives the possibility to copy entire image (including header) from NOR parallel memory to e.g. SDRAM.
The legacy behavior is preserved, since other board don't enabled this option.
What's the use case?
This code is used on some boards (ppc [1]) in which we copy only payload of the image (i.e. we drop the package metadata).
In my particular case I would like to copy the image with its meta data (in similar way works spl_nand.c and spl_mmc.c).
Since, I wanted to avoid code duplication, I've decided to add this flag.
So another way of saying this, if I follow correctly, is that today SPL NOR doesn't work with u-boot.img but only u-boot.bin ?
Yes, exactly.
Signed-off-by: Lukasz Majewski l.majewski@majess.pl
Kconfig | 10 ++++++++++
This should now get moved to common/spl/Kconfig
OK.
As a side note - I do have some other patches which move some SPL_NOR options to Kconfig.
However, I omitted them when I saw great work done by Simon to move a lot of SPL options to Kconfig. I hope them will be merged soon :-)
It's in now :)
Great :-)
README | 4 ++++
This can be dropped.
So, we don't add any new entries to README and now rely solely on Kconfig "help" sections?
Correct.
OK.
[snip]
diff --git a/common/spl/spl_nor.c b/common/spl/spl_nor.c index 8ea874c..1e0f739 100644 --- a/common/spl/spl_nor.c +++ b/common/spl/spl_nor.c @@ -9,13 +9,15 @@
int spl_nor_load_image(void) {
- void *img_src; int ret;
+#ifndef CONFIG_SPL_NOR_COPY_ENTIRE_IMAGE /* * Loading of the payload to SDRAM is done with skipping of * the mkimage header in this SPL NOR driver */ spl_image.flags |= SPL_COPY_PAYLOAD_ONLY;
+#endif #ifdef CONFIG_SPL_OS_BOOT if (!spl_start_uboot()) { const struct image_header *header; @@ -64,9 +66,13 @@ int spl_nor_load_image(void) if (ret) return ret;
We should have a comment here about how we're calculating the base of where to copy from.
I will be more verbose with v2 :-).
Thanks for review.
No problem, thank you!
Best regards, Łukasz Majewski
On Fri, Sep 16, 2016 at 10:43:51AM +0200, Lukasz Majewski wrote:
This define gives the possibility to copy entire image (including header) from NOR parallel memory to e.g. SDRAM.
The legacy behavior is preserved, since other board don't enabled this option.
Signed-off-by: Lukasz Majewski l.majewski@majess.pl
Please re-base this on top of master, thanks!
This define gives the possibility to copy entire image (including header) from NOR parallel memory to e.g. SDRAM.
The legacy behavior is preserved, since other board don't enabled this option.
Signed-off-by: Lukasz Majewski l.majewski@majess.pl --- common/spl/Kconfig | 10 ++++++++++ common/spl/spl_nor.c | 12 +++++++++--- 2 files changed, 19 insertions(+), 3 deletions(-)
diff --git a/common/spl/Kconfig b/common/spl/Kconfig index df9e0ce..d31b26d 100644 --- a/common/spl/Kconfig +++ b/common/spl/Kconfig @@ -399,6 +399,16 @@ config SPL_NOR_SUPPORT a memory-mapped device makes it very easy to access. Loading from NOR is typically achieved with just a memcpy().
+config SPL_NOR_COPY_ENTIRE_IMAGE + bool + depends on SPL_NOR_SUPPORT + prompt "Copy entire image from NOR memory" + default n + help + By default the SPL NOR driver supports copying only payload to + destination address. Say Y if you want to copy entire image (including + its image header). + config SPL_ONENAND_SUPPORT bool "Support OneNAND flash" depends on SPL diff --git a/common/spl/spl_nor.c b/common/spl/spl_nor.c index 6bfa399..44f3e99 100644 --- a/common/spl/spl_nor.c +++ b/common/spl/spl_nor.c @@ -10,13 +10,15 @@ static int spl_nor_load_image(struct spl_image_info *spl_image, struct spl_boot_device *bootdev) { + void *img_src; int ret; +#ifndef CONFIG_SPL_NOR_COPY_ENTIRE_IMAGE /* * Loading of the payload to SDRAM is done with skipping of * the mkimage header in this SPL NOR driver */ spl_image->flags |= SPL_COPY_PAYLOAD_ONLY; - +#endif #ifdef CONFIG_SPL_OS_BOOT if (!spl_start_uboot()) { const struct image_header *header; @@ -65,9 +67,13 @@ static int spl_nor_load_image(struct spl_image_info *spl_image, if (ret) return ret;
+ img_src = (void *)CONFIG_SYS_UBOOT_BASE; +#ifndef CONFIG_SPL_NOR_COPY_ENTIRE_IMAGE + img_src += sizeof(struct image_header)); +#endif + memcpy((void *)(unsigned long)spl_image->load_addr, - (void *)(CONFIG_SYS_UBOOT_BASE + sizeof(struct image_header)), - spl_image->size); + img_src, spl_image->size);
return 0; }
On 11/28/2016 07:27 AM, Lukasz Majewski wrote:
This define gives the possibility to copy entire image (including header) from NOR parallel memory to e.g. SDRAM.
The legacy behavior is preserved, since other board don't enabled this option.
Signed-off-by: Lukasz Majewski l.majewski@majess.pl
What is the usecase ?
common/spl/Kconfig | 10 ++++++++++ common/spl/spl_nor.c | 12 +++++++++--- 2 files changed, 19 insertions(+), 3 deletions(-)
diff --git a/common/spl/Kconfig b/common/spl/Kconfig index df9e0ce..d31b26d 100644 --- a/common/spl/Kconfig +++ b/common/spl/Kconfig @@ -399,6 +399,16 @@ config SPL_NOR_SUPPORT a memory-mapped device makes it very easy to access. Loading from NOR is typically achieved with just a memcpy().
+config SPL_NOR_COPY_ENTIRE_IMAGE
- bool
- depends on SPL_NOR_SUPPORT
- prompt "Copy entire image from NOR memory"
- default n
- help
By default the SPL NOR driver supports copying only payload todestination address. Say Y if you want to copy entire image (includingits image header).config SPL_ONENAND_SUPPORT bool "Support OneNAND flash" depends on SPL diff --git a/common/spl/spl_nor.c b/common/spl/spl_nor.c index 6bfa399..44f3e99 100644 --- a/common/spl/spl_nor.c +++ b/common/spl/spl_nor.c @@ -10,13 +10,15 @@ static int spl_nor_load_image(struct spl_image_info *spl_image, struct spl_boot_device *bootdev) {
- void *img_src; int ret;
+#ifndef CONFIG_SPL_NOR_COPY_ENTIRE_IMAGE /* * Loading of the payload to SDRAM is done with skipping of * the mkimage header in this SPL NOR driver */ spl_image->flags |= SPL_COPY_PAYLOAD_ONLY;
+#endif #ifdef CONFIG_SPL_OS_BOOT if (!spl_start_uboot()) { const struct image_header *header; @@ -65,9 +67,13 @@ static int spl_nor_load_image(struct spl_image_info *spl_image, if (ret) return ret;
- img_src = (void *)CONFIG_SYS_UBOOT_BASE;
+#ifndef CONFIG_SPL_NOR_COPY_ENTIRE_IMAGE
- img_src += sizeof(struct image_header));
+#endif
- memcpy((void *)(unsigned long)spl_image->load_addr,
(void *)(CONFIG_SYS_UBOOT_BASE + sizeof(struct image_header)),spl_image->size);
img_src, spl_image->size);return 0;
}
On Mon, Nov 28, 2016 at 01:43:44PM +0100, Marek Vasut wrote:
On 11/28/2016 07:27 AM, Lukasz Majewski wrote:
This define gives the possibility to copy entire image (including header) from NOR parallel memory to e.g. SDRAM.
The legacy behavior is preserved, since other board don't enabled this option.
Signed-off-by: Lukasz Majewski l.majewski@majess.pl
What is the usecase ?
This is actually a v2, and the answer from v1 is at http://lists.denx.de/pipermail/u-boot/2016-September/267019.html and should be more prominent in the commit message / help text (so, v3 please!). Without this option u-boot.img doesn't work from NOR, only u-boot.bin
On 11/28/2016 05:50 PM, Tom Rini wrote:
On Mon, Nov 28, 2016 at 01:43:44PM +0100, Marek Vasut wrote:
On 11/28/2016 07:27 AM, Lukasz Majewski wrote:
This define gives the possibility to copy entire image (including header) from NOR parallel memory to e.g. SDRAM.
The legacy behavior is preserved, since other board don't enabled this option.
Signed-off-by: Lukasz Majewski l.majewski@majess.pl
What is the usecase ?
This is actually a v2, and the answer from v1 is at http://lists.denx.de/pipermail/u-boot/2016-September/267019.html and should be more prominent in the commit message / help text (so, v3 please!). Without this option u-boot.img doesn't work from NOR, only u-boot.bin
Ah, I recall this issue, thanks!
This define gives the possibility to copy entire image (including header - e.g. u-boot.img) from NOR parallel memory to e.g. SDRAM. The current code only supports loading the raw binary image (the u-boot.bin).
The legacy behavior is preserved, since other board don't enabled this option.
Signed-off-by: Lukasz Majewski l.majewski@majess.pl --- Changes for v2: - Update to code to apply on v2016.11+ Changes for v3: - Write better commit mesage to explain the problem
--- common/spl/Kconfig | 10 ++++++++++ common/spl/spl_nor.c | 12 +++++++++--- 2 files changed, 19 insertions(+), 3 deletions(-)
diff --git a/common/spl/Kconfig b/common/spl/Kconfig index df9e0ce..d31b26d 100644 --- a/common/spl/Kconfig +++ b/common/spl/Kconfig @@ -399,6 +399,16 @@ config SPL_NOR_SUPPORT a memory-mapped device makes it very easy to access. Loading from NOR is typically achieved with just a memcpy().
+config SPL_NOR_COPY_ENTIRE_IMAGE + bool + depends on SPL_NOR_SUPPORT + prompt "Copy entire image from NOR memory" + default n + help + By default the SPL NOR driver supports copying only payload to + destination address. Say Y if you want to copy entire image (including + its image header). + config SPL_ONENAND_SUPPORT bool "Support OneNAND flash" depends on SPL diff --git a/common/spl/spl_nor.c b/common/spl/spl_nor.c index 6bfa399..44f3e99 100644 --- a/common/spl/spl_nor.c +++ b/common/spl/spl_nor.c @@ -10,13 +10,15 @@ static int spl_nor_load_image(struct spl_image_info *spl_image, struct spl_boot_device *bootdev) { + void *img_src; int ret; +#ifndef CONFIG_SPL_NOR_COPY_ENTIRE_IMAGE /* * Loading of the payload to SDRAM is done with skipping of * the mkimage header in this SPL NOR driver */ spl_image->flags |= SPL_COPY_PAYLOAD_ONLY; - +#endif #ifdef CONFIG_SPL_OS_BOOT if (!spl_start_uboot()) { const struct image_header *header; @@ -65,9 +67,13 @@ static int spl_nor_load_image(struct spl_image_info *spl_image, if (ret) return ret;
+ img_src = (void *)CONFIG_SYS_UBOOT_BASE; +#ifndef CONFIG_SPL_NOR_COPY_ENTIRE_IMAGE + img_src += sizeof(struct image_header)); +#endif + memcpy((void *)(unsigned long)spl_image->load_addr, - (void *)(CONFIG_SYS_UBOOT_BASE + sizeof(struct image_header)), - spl_image->size); + img_src, spl_image->size);
return 0; }
On 11/28/2016 10:09 PM, Lukasz Majewski wrote:
This define gives the possibility to copy entire image (including header - e.g. u-boot.img) from NOR parallel memory to e.g. SDRAM. The current code only supports loading the raw binary image (the u-boot.bin).
The legacy behavior is preserved, since other board don't enabled this option.
Sooooo, what's the usecase again ? ;-) I still miss it being documented.
Signed-off-by: Lukasz Majewski l.majewski@majess.pl
Changes for v2:
- Update to code to apply on v2016.11+
Changes for v3:
- Write better commit mesage to explain the problem
common/spl/Kconfig | 10 ++++++++++ common/spl/spl_nor.c | 12 +++++++++--- 2 files changed, 19 insertions(+), 3 deletions(-)
diff --git a/common/spl/Kconfig b/common/spl/Kconfig index df9e0ce..d31b26d 100644 --- a/common/spl/Kconfig +++ b/common/spl/Kconfig @@ -399,6 +399,16 @@ config SPL_NOR_SUPPORT a memory-mapped device makes it very easy to access. Loading from NOR is typically achieved with just a memcpy().
+config SPL_NOR_COPY_ENTIRE_IMAGE
- bool
- depends on SPL_NOR_SUPPORT
- prompt "Copy entire image from NOR memory"
- default n
- help
By default the SPL NOR driver supports copying only payload todestination address. Say Y if you want to copy entire image (includingits image header).config SPL_ONENAND_SUPPORT bool "Support OneNAND flash" depends on SPL diff --git a/common/spl/spl_nor.c b/common/spl/spl_nor.c index 6bfa399..44f3e99 100644 --- a/common/spl/spl_nor.c +++ b/common/spl/spl_nor.c @@ -10,13 +10,15 @@ static int spl_nor_load_image(struct spl_image_info *spl_image, struct spl_boot_device *bootdev) {
- void *img_src; int ret;
+#ifndef CONFIG_SPL_NOR_COPY_ENTIRE_IMAGE /* * Loading of the payload to SDRAM is done with skipping of * the mkimage header in this SPL NOR driver */ spl_image->flags |= SPL_COPY_PAYLOAD_ONLY;
+#endif #ifdef CONFIG_SPL_OS_BOOT if (!spl_start_uboot()) { const struct image_header *header; @@ -65,9 +67,13 @@ static int spl_nor_load_image(struct spl_image_info *spl_image, if (ret) return ret;
- img_src = (void *)CONFIG_SYS_UBOOT_BASE;
+#ifndef CONFIG_SPL_NOR_COPY_ENTIRE_IMAGE
- img_src += sizeof(struct image_header));
+#endif
- memcpy((void *)(unsigned long)spl_image->load_addr,
(void *)(CONFIG_SYS_UBOOT_BASE + sizeof(struct image_header)),spl_image->size);
img_src, spl_image->size);return 0;
}
Hi Marek,
On 11/28/2016 10:09 PM, Lukasz Majewski wrote:
This define gives the possibility to copy entire image (including header - e.g. u-boot.img) from NOR parallel memory to e.g. SDRAM. The current code only supports loading the raw binary image (the u-boot.bin).
The legacy behavior is preserved, since other board don't enabled this option.
Sooooo, what's the usecase again ? ;-)
:-)
The use case is to allow u-boot.img being loaded from Parallel NOR. The current code only supports u-boot.bin.
This code is crucial for boards booting from NOR, as a board which support will be sent soon to u-boot ML :-).
Best regards, Łukasz Majewski
I still miss it being documented.
Signed-off-by: Lukasz Majewski l.majewski@majess.pl
Changes for v2:
- Update to code to apply on v2016.11+
Changes for v3:
- Write better commit mesage to explain the problem
common/spl/Kconfig | 10 ++++++++++ common/spl/spl_nor.c | 12 +++++++++--- 2 files changed, 19 insertions(+), 3 deletions(-)
diff --git a/common/spl/Kconfig b/common/spl/Kconfig index df9e0ce..d31b26d 100644 --- a/common/spl/Kconfig +++ b/common/spl/Kconfig @@ -399,6 +399,16 @@ config SPL_NOR_SUPPORT a memory-mapped device makes it very easy to access. Loading from NOR is typically achieved with just a memcpy().
+config SPL_NOR_COPY_ENTIRE_IMAGE
- bool
- depends on SPL_NOR_SUPPORT
- prompt "Copy entire image from NOR memory"
- default n
- help
By default the SPL NOR driver supports copying onlypayload to
destination address. Say Y if you want to copy entireimage (including
its image header).config SPL_ONENAND_SUPPORT bool "Support OneNAND flash" depends on SPL diff --git a/common/spl/spl_nor.c b/common/spl/spl_nor.c index 6bfa399..44f3e99 100644 --- a/common/spl/spl_nor.c +++ b/common/spl/spl_nor.c @@ -10,13 +10,15 @@ static int spl_nor_load_image(struct spl_image_info *spl_image, struct spl_boot_device *bootdev) {
- void *img_src; int ret;
+#ifndef CONFIG_SPL_NOR_COPY_ENTIRE_IMAGE /* * Loading of the payload to SDRAM is done with skipping of * the mkimage header in this SPL NOR driver */ spl_image->flags |= SPL_COPY_PAYLOAD_ONLY;
+#endif #ifdef CONFIG_SPL_OS_BOOT if (!spl_start_uboot()) { const struct image_header *header; @@ -65,9 +67,13 @@ static int spl_nor_load_image(struct spl_image_info *spl_image, if (ret) return ret;
- img_src = (void *)CONFIG_SYS_UBOOT_BASE;
+#ifndef CONFIG_SPL_NOR_COPY_ENTIRE_IMAGE
- img_src += sizeof(struct image_header));
+#endif
- memcpy((void *)(unsigned long)spl_image->load_addr,
(void *)(CONFIG_SYS_UBOOT_BASE + sizeof(structimage_header)),
spl_image->size);
img_src, spl_image->size);return 0;
}
On 11/29/2016 10:11 AM, Lukasz Majewski wrote:
Hi Marek,
On 11/28/2016 10:09 PM, Lukasz Majewski wrote:
This define gives the possibility to copy entire image (including header - e.g. u-boot.img) from NOR parallel memory to e.g. SDRAM. The current code only supports loading the raw binary image (the u-boot.bin).
The legacy behavior is preserved, since other board don't enabled this option.
Sooooo, what's the usecase again ? ;-)
:-)
The use case is to allow u-boot.img being loaded from Parallel NOR. The current code only supports u-boot.bin.
Why is u-boot.bin (or the payload) not sufficient ? Why do you need the header ?
This code is crucial for boards booting from NOR, as a board which support will be sent soon to u-boot ML :-).
Best regards, Łukasz Majewski
I still miss it being documented.
[...]
On Tue, Nov 29, 2016 at 11:50:34AM +0100, Marek Vasut wrote:
On 11/29/2016 10:11 AM, Lukasz Majewski wrote:
Hi Marek,
On 11/28/2016 10:09 PM, Lukasz Majewski wrote:
This define gives the possibility to copy entire image (including header - e.g. u-boot.img) from NOR parallel memory to e.g. SDRAM. The current code only supports loading the raw binary image (the u-boot.bin).
The legacy behavior is preserved, since other board don't enabled this option.
Sooooo, what's the usecase again ? ;-)
:-)
The use case is to allow u-boot.img being loaded from Parallel NOR. The current code only supports u-boot.bin.
Why is u-boot.bin (or the payload) not sufficient ? Why do you need the header ?
Well, the general use-case and code flow is that we load u-boot.img (or a FIT image) and if all else fails, fall back to assuming a .bin and a known address).
On 11/29/2016 07:18 PM, Tom Rini wrote:
On Tue, Nov 29, 2016 at 11:50:34AM +0100, Marek Vasut wrote:
On 11/29/2016 10:11 AM, Lukasz Majewski wrote:
Hi Marek,
On 11/28/2016 10:09 PM, Lukasz Majewski wrote:
This define gives the possibility to copy entire image (including header - e.g. u-boot.img) from NOR parallel memory to e.g. SDRAM. The current code only supports loading the raw binary image (the u-boot.bin).
The legacy behavior is preserved, since other board don't enabled this option.
Sooooo, what's the usecase again ? ;-)
:-)
The use case is to allow u-boot.img being loaded from Parallel NOR. The current code only supports u-boot.bin.
Why is u-boot.bin (or the payload) not sufficient ? Why do you need the header ?
Well, the general use-case and code flow is that we load u-boot.img (or a FIT image) and if all else fails, fall back to assuming a .bin and a known address).
And exactly how is that whole image useful in RAM ? Sorry, I still do not see it, usually you just need the executable payload, although even that can be left in flash most of the time.
Hi Marek,
On 11/29/2016 07:18 PM, Tom Rini wrote:
On Tue, Nov 29, 2016 at 11:50:34AM +0100, Marek Vasut wrote:
On 11/29/2016 10:11 AM, Lukasz Majewski wrote:
Hi Marek,
On 11/28/2016 10:09 PM, Lukasz Majewski wrote:
This define gives the possibility to copy entire image (including header - e.g. u-boot.img) from NOR parallel memory to e.g. SDRAM. The current code only supports loading the raw binary image (the u-boot.bin).
The legacy behavior is preserved, since other board don't enabled this option.
Sooooo, what's the usecase again ? ;-)
:-)
The use case is to allow u-boot.img being loaded from Parallel NOR. The current code only supports u-boot.bin.
Why is u-boot.bin (or the payload) not sufficient ? Why do you need the header ?
Well, the general use-case and code flow is that we load u-boot.img (or a FIT image) and if all else fails, fall back to assuming a .bin and a known address).
And exactly how is that whole image useful in RAM ? Sorry, I still do not see it, usually you just need the executable payload, although even that can be left in flash most of the time.
The use case is that I do want to boot from SD card/eMMC and NOR with using u-boot.img.
I would like to avoid situation when for NOR I must use u-boot.bin and for eMMC u-boot.img.
Such approach keeps things as simple as possible :-)
Best regards, Łukasz Majewski
On 12/26/2016 05:36 PM, Lukasz Majewski wrote:
Hi Marek,
On 11/29/2016 07:18 PM, Tom Rini wrote:
On Tue, Nov 29, 2016 at 11:50:34AM +0100, Marek Vasut wrote:
On 11/29/2016 10:11 AM, Lukasz Majewski wrote:
Hi Marek,
On 11/28/2016 10:09 PM, Lukasz Majewski wrote: > This define gives the possibility to copy entire image > (including header - e.g. u-boot.img) from NOR parallel memory > to e.g. SDRAM. The current code only supports loading the raw > binary image (the u-boot.bin). > > The legacy behavior is preserved, since other board don't > enabled this option.
Sooooo, what's the usecase again ? ;-)
:-)
The use case is to allow u-boot.img being loaded from Parallel NOR. The current code only supports u-boot.bin.
Why is u-boot.bin (or the payload) not sufficient ? Why do you need the header ?
Well, the general use-case and code flow is that we load u-boot.img (or a FIT image) and if all else fails, fall back to assuming a .bin and a known address).
And exactly how is that whole image useful in RAM ? Sorry, I still do not see it, usually you just need the executable payload, although even that can be left in flash most of the time.
The use case is that I do want to boot from SD card/eMMC and NOR with using u-boot.img.
I would like to avoid situation when for NOR I must use u-boot.bin and for eMMC u-boot.img.
Such approach keeps things as simple as possible :-)
Oh, so it allows you to detect bitrot for the content in SPI NOR ? It's a bit strange we had to use u-boot.bin with SPL there.
Hi Marek,
On 12/26/2016 05:36 PM, Lukasz Majewski wrote:
Hi Marek,
On 11/29/2016 07:18 PM, Tom Rini wrote:
On Tue, Nov 29, 2016 at 11:50:34AM +0100, Marek Vasut wrote:
On 11/29/2016 10:11 AM, Lukasz Majewski wrote:
Hi Marek,
> On 11/28/2016 10:09 PM, Lukasz Majewski wrote: >> This define gives the possibility to copy entire image >> (including header - e.g. u-boot.img) from NOR parallel memory >> to e.g. SDRAM. The current code only supports loading the raw >> binary image (the u-boot.bin). >> >> The legacy behavior is preserved, since other board don't >> enabled this option. > > Sooooo, what's the usecase again ? ;-)
:-)
The use case is to allow u-boot.img being loaded from Parallel NOR. The current code only supports u-boot.bin.
Why is u-boot.bin (or the payload) not sufficient ? Why do you need the header ?
Well, the general use-case and code flow is that we load u-boot.img (or a FIT image) and if all else fails, fall back to assuming a .bin and a known address).
And exactly how is that whole image useful in RAM ? Sorry, I still do not see it, usually you just need the executable payload, although even that can be left in flash most of the time.
The use case is that I do want to boot from SD card/eMMC and NOR with using u-boot.img.
I would like to avoid situation when for NOR I must use u-boot.bin and for eMMC u-boot.img.
Such approach keeps things as simple as possible :-)
Oh, so it allows you to detect bitrot for the content in SPI NOR ?
I do not use SPI NOR, it is parallel NOR.
It's a bit strange we had to use u-boot.bin with SPL there.
This is how the legacy system behaves. It uses (by default) Parallel NOR for booting (with advised/provided NOR memory timings). After doing some measurements, it turned out that for "tunned" u-boot/SPL there would be the best way to copy it to ram and execute it from there (just like eMMC).
Hence, I would like to use u-boot.img in both booting scenarios.
Best regards, Łukasz Majewski
On 12/28/2016 09:52 AM, Lukasz Majewski wrote:
Hi Marek,
On 12/26/2016 05:36 PM, Lukasz Majewski wrote:
Hi Marek,
On 11/29/2016 07:18 PM, Tom Rini wrote:
On Tue, Nov 29, 2016 at 11:50:34AM +0100, Marek Vasut wrote:
On 11/29/2016 10:11 AM, Lukasz Majewski wrote: > Hi Marek, > >> On 11/28/2016 10:09 PM, Lukasz Majewski wrote: >>> This define gives the possibility to copy entire image >>> (including header - e.g. u-boot.img) from NOR parallel memory >>> to e.g. SDRAM. The current code only supports loading the raw >>> binary image (the u-boot.bin). >>> >>> The legacy behavior is preserved, since other board don't >>> enabled this option. >> >> Sooooo, what's the usecase again ? ;-) > > :-) > > The use case is to allow u-boot.img being loaded from Parallel > NOR. The current code only supports u-boot.bin.
Why is u-boot.bin (or the payload) not sufficient ? Why do you need the header ?
Well, the general use-case and code flow is that we load u-boot.img (or a FIT image) and if all else fails, fall back to assuming a .bin and a known address).
And exactly how is that whole image useful in RAM ? Sorry, I still do not see it, usually you just need the executable payload, although even that can be left in flash most of the time.
The use case is that I do want to boot from SD card/eMMC and NOR with using u-boot.img.
I would like to avoid situation when for NOR I must use u-boot.bin and for eMMC u-boot.img.
Such approach keeps things as simple as possible :-)
Oh, so it allows you to detect bitrot for the content in SPI NOR ?
I do not use SPI NOR, it is parallel NOR.
Sorry, I meant parallel NOR of course.
It's a bit strange we had to use u-boot.bin with SPL there.
This is how the legacy system behaves. It uses (by default) Parallel NOR for booting (with advised/provided NOR memory timings). After doing some measurements, it turned out that for "tunned" u-boot/SPL there would be the best way to copy it to ram and execute it from there (just like eMMC).
Hence, I would like to use u-boot.img in both booting scenarios.
I think I was mistaken yesterday, I don't think I understand why copying the image including the header into RAM has any benefit compared to copying just the image payload to RAM (and yes, we're getting back to my original question).
On Thu, Dec 29, 2016 at 12:41:06AM +0100, Marek Vasut wrote:
On 12/28/2016 09:52 AM, Lukasz Majewski wrote:
Hi Marek,
On 12/26/2016 05:36 PM, Lukasz Majewski wrote:
Hi Marek,
On 11/29/2016 07:18 PM, Tom Rini wrote:
On Tue, Nov 29, 2016 at 11:50:34AM +0100, Marek Vasut wrote: > On 11/29/2016 10:11 AM, Lukasz Majewski wrote: >> Hi Marek, >> >>> On 11/28/2016 10:09 PM, Lukasz Majewski wrote: >>>> This define gives the possibility to copy entire image >>>> (including header - e.g. u-boot.img) from NOR parallel memory >>>> to e.g. SDRAM. The current code only supports loading the raw >>>> binary image (the u-boot.bin). >>>> >>>> The legacy behavior is preserved, since other board don't >>>> enabled this option. >>> >>> Sooooo, what's the usecase again ? ;-) >> >> :-) >> >> The use case is to allow u-boot.img being loaded from Parallel >> NOR. The current code only supports u-boot.bin. > > Why is u-boot.bin (or the payload) not sufficient ? Why do you > need the header ?
Well, the general use-case and code flow is that we load u-boot.img (or a FIT image) and if all else fails, fall back to assuming a .bin and a known address).
And exactly how is that whole image useful in RAM ? Sorry, I still do not see it, usually you just need the executable payload, although even that can be left in flash most of the time.
The use case is that I do want to boot from SD card/eMMC and NOR with using u-boot.img.
I would like to avoid situation when for NOR I must use u-boot.bin and for eMMC u-boot.img.
Such approach keeps things as simple as possible :-)
Oh, so it allows you to detect bitrot for the content in SPI NOR ?
I do not use SPI NOR, it is parallel NOR.
Sorry, I meant parallel NOR of course.
It's a bit strange we had to use u-boot.bin with SPL there.
This is how the legacy system behaves. It uses (by default) Parallel NOR for booting (with advised/provided NOR memory timings). After doing some measurements, it turned out that for "tunned" u-boot/SPL there would be the best way to copy it to ram and execute it from there (just like eMMC).
Hence, I would like to use u-boot.img in both booting scenarios.
I think I was mistaken yesterday, I don't think I understand why copying the image including the header into RAM has any benefit compared to copying just the image payload to RAM (and yes, we're getting back to my original question).
Code complexity and forward compatibility? The general case in the SPL framework is that we have either a "legacy" image or a FIT image and we fall back to "well, just run it!".
On 12/29/2016 04:26 PM, Tom Rini wrote:
On Thu, Dec 29, 2016 at 12:41:06AM +0100, Marek Vasut wrote:
On 12/28/2016 09:52 AM, Lukasz Majewski wrote:
Hi Marek,
On 12/26/2016 05:36 PM, Lukasz Majewski wrote:
Hi Marek,
On 11/29/2016 07:18 PM, Tom Rini wrote: > On Tue, Nov 29, 2016 at 11:50:34AM +0100, Marek Vasut wrote: >> On 11/29/2016 10:11 AM, Lukasz Majewski wrote: >>> Hi Marek, >>> >>>> On 11/28/2016 10:09 PM, Lukasz Majewski wrote: >>>>> This define gives the possibility to copy entire image >>>>> (including header - e.g. u-boot.img) from NOR parallel memory >>>>> to e.g. SDRAM. The current code only supports loading the raw >>>>> binary image (the u-boot.bin). >>>>> >>>>> The legacy behavior is preserved, since other board don't >>>>> enabled this option. >>>> >>>> Sooooo, what's the usecase again ? ;-) >>> >>> :-) >>> >>> The use case is to allow u-boot.img being loaded from Parallel >>> NOR. The current code only supports u-boot.bin. >> >> Why is u-boot.bin (or the payload) not sufficient ? Why do you >> need the header ? > > Well, the general use-case and code flow is that we load > u-boot.img (or a FIT image) and if all else fails, fall back to > assuming a .bin and a known address). > And exactly how is that whole image useful in RAM ? Sorry, I still do not see it, usually you just need the executable payload, although even that can be left in flash most of the time.
The use case is that I do want to boot from SD card/eMMC and NOR with using u-boot.img.
I would like to avoid situation when for NOR I must use u-boot.bin and for eMMC u-boot.img.
Such approach keeps things as simple as possible :-)
Oh, so it allows you to detect bitrot for the content in SPI NOR ?
I do not use SPI NOR, it is parallel NOR.
Sorry, I meant parallel NOR of course.
It's a bit strange we had to use u-boot.bin with SPL there.
This is how the legacy system behaves. It uses (by default) Parallel NOR for booting (with advised/provided NOR memory timings). After doing some measurements, it turned out that for "tunned" u-boot/SPL there would be the best way to copy it to ram and execute it from there (just like eMMC).
Hence, I would like to use u-boot.img in both booting scenarios.
I think I was mistaken yesterday, I don't think I understand why copying the image including the header into RAM has any benefit compared to copying just the image payload to RAM (and yes, we're getting back to my original question).
Code complexity and forward compatibility?
This is adding code complexity, but this is not my point.
The general case in the SPL framework is that we have either a "legacy" image or a FIT image and we fall back to "well, just run it!".
Well, this doesn't answer my question, because if I understand this patch correctly, it copies the entire legacy image (incl. header) into RAM instead of copying just the image payload (which we already do). I don't really understand why we want to do this. Or do I misunderstand something ?
Hi Marek,
On 12/29/2016 04:26 PM, Tom Rini wrote:
On Thu, Dec 29, 2016 at 12:41:06AM +0100, Marek Vasut wrote:
On 12/28/2016 09:52 AM, Lukasz Majewski wrote:
Hi Marek,
On 12/26/2016 05:36 PM, Lukasz Majewski wrote:
Hi Marek,
> On 11/29/2016 07:18 PM, Tom Rini wrote: >> On Tue, Nov 29, 2016 at 11:50:34AM +0100, Marek Vasut wrote: >>> On 11/29/2016 10:11 AM, Lukasz Majewski wrote: >>>> Hi Marek, >>>> >>>>> On 11/28/2016 10:09 PM, Lukasz Majewski wrote: >>>>>> This define gives the possibility to copy entire image >>>>>> (including header - e.g. u-boot.img) from NOR parallel >>>>>> memory to e.g. SDRAM. The current code only supports >>>>>> loading the raw binary image (the u-boot.bin). >>>>>> >>>>>> The legacy behavior is preserved, since other board don't >>>>>> enabled this option. >>>>> >>>>> Sooooo, what's the usecase again ? ;-) >>>> >>>> :-) >>>> >>>> The use case is to allow u-boot.img being loaded from >>>> Parallel NOR. The current code only supports u-boot.bin. >>> >>> Why is u-boot.bin (or the payload) not sufficient ? Why do >>> you need the header ? >> >> Well, the general use-case and code flow is that we load >> u-boot.img (or a FIT image) and if all else fails, fall back >> to assuming a .bin and a known address). >> > And exactly how is that whole image useful in RAM ? Sorry, I > still do not see it, usually you just need the executable > payload, although even that can be left in flash most of the > time.
The use case is that I do want to boot from SD card/eMMC and NOR with using u-boot.img.
I would like to avoid situation when for NOR I must use u-boot.bin and for eMMC u-boot.img.
Such approach keeps things as simple as possible :-)
Oh, so it allows you to detect bitrot for the content in SPI NOR ?
I do not use SPI NOR, it is parallel NOR.
Sorry, I meant parallel NOR of course.
It's a bit strange we had to use u-boot.bin with SPL there.
This is how the legacy system behaves. It uses (by default) Parallel NOR for booting (with advised/provided NOR memory timings). After doing some measurements, it turned out that for "tunned" u-boot/SPL there would be the best way to copy it to ram and execute it from there (just like eMMC).
Hence, I would like to use u-boot.img in both booting scenarios.
I think I was mistaken yesterday, I don't think I understand why copying the image including the header into RAM has any benefit compared to copying just the image payload to RAM (and yes, we're getting back to my original question).
Code complexity and forward compatibility?
This is adding code complexity, but this is not my point.
The general case in the SPL framework is that we have either a "legacy" image or a FIT image and we fall back to "well, just run it!".
Well, this doesn't answer my question, because if I understand this patch correctly, it copies the entire legacy image (incl. header) into RAM instead of copying just the image payload (which we already do). I don't really understand why we want to do this. Or do I misunderstand something ?
No, you understood everything correctly. After some thoughts, I think that only payload should be copied.
I'll prepare test setup and double check this patch usability by Monday.
Best regards, Łukasz Majewski
On 12/30/2016 10:28 PM, Lukasz Majewski wrote:
Hi Marek,
On 12/29/2016 04:26 PM, Tom Rini wrote:
On Thu, Dec 29, 2016 at 12:41:06AM +0100, Marek Vasut wrote:
On 12/28/2016 09:52 AM, Lukasz Majewski wrote:
Hi Marek,
On 12/26/2016 05:36 PM, Lukasz Majewski wrote: > Hi Marek, > >> On 11/29/2016 07:18 PM, Tom Rini wrote: >>> On Tue, Nov 29, 2016 at 11:50:34AM +0100, Marek Vasut wrote: >>>> On 11/29/2016 10:11 AM, Lukasz Majewski wrote: >>>>> Hi Marek, >>>>> >>>>>> On 11/28/2016 10:09 PM, Lukasz Majewski wrote: >>>>>>> This define gives the possibility to copy entire image >>>>>>> (including header - e.g. u-boot.img) from NOR parallel >>>>>>> memory to e.g. SDRAM. The current code only supports >>>>>>> loading the raw binary image (the u-boot.bin). >>>>>>> >>>>>>> The legacy behavior is preserved, since other board don't >>>>>>> enabled this option. >>>>>> >>>>>> Sooooo, what's the usecase again ? ;-) >>>>> >>>>> :-) >>>>> >>>>> The use case is to allow u-boot.img being loaded from >>>>> Parallel NOR. The current code only supports u-boot.bin. >>>> >>>> Why is u-boot.bin (or the payload) not sufficient ? Why do >>>> you need the header ? >>> >>> Well, the general use-case and code flow is that we load >>> u-boot.img (or a FIT image) and if all else fails, fall back >>> to assuming a .bin and a known address). >>> >> And exactly how is that whole image useful in RAM ? Sorry, I >> still do not see it, usually you just need the executable >> payload, although even that can be left in flash most of the >> time. > > The use case is that I do want to boot from SD card/eMMC and NOR > with using u-boot.img. > > I would like to avoid situation when for NOR I must use > u-boot.bin and for eMMC u-boot.img. > > Such approach keeps things as simple as possible :-)
Oh, so it allows you to detect bitrot for the content in SPI NOR ?
I do not use SPI NOR, it is parallel NOR.
Sorry, I meant parallel NOR of course.
It's a bit strange we had to use u-boot.bin with SPL there.
This is how the legacy system behaves. It uses (by default) Parallel NOR for booting (with advised/provided NOR memory timings). After doing some measurements, it turned out that for "tunned" u-boot/SPL there would be the best way to copy it to ram and execute it from there (just like eMMC).
Hence, I would like to use u-boot.img in both booting scenarios.
I think I was mistaken yesterday, I don't think I understand why copying the image including the header into RAM has any benefit compared to copying just the image payload to RAM (and yes, we're getting back to my original question).
Code complexity and forward compatibility?
This is adding code complexity, but this is not my point.
The general case in the SPL framework is that we have either a "legacy" image or a FIT image and we fall back to "well, just run it!".
Well, this doesn't answer my question, because if I understand this patch correctly, it copies the entire legacy image (incl. header) into RAM instead of copying just the image payload (which we already do). I don't really understand why we want to do this. Or do I misunderstand something ?
No, you understood everything correctly. After some thoughts, I think that only payload should be copied.
But that's what we already do, no ? So what is the point of this patch ?
I'll prepare test setup and double check this patch usability by Monday.
OK
Hi Marek,
On 12/30/2016 10:28 PM, Lukasz Majewski wrote:
Hi Marek,
On 12/29/2016 04:26 PM, Tom Rini wrote:
On Thu, Dec 29, 2016 at 12:41:06AM +0100, Marek Vasut wrote:
On 12/28/2016 09:52 AM, Lukasz Majewski wrote:
Hi Marek,
> On 12/26/2016 05:36 PM, Lukasz Majewski wrote: >> Hi Marek, >> >>> On 11/29/2016 07:18 PM, Tom Rini wrote: >>>> On Tue, Nov 29, 2016 at 11:50:34AM +0100, Marek Vasut wrote: >>>>> On 11/29/2016 10:11 AM, Lukasz Majewski wrote: >>>>>> Hi Marek, >>>>>> >>>>>>> On 11/28/2016 10:09 PM, Lukasz Majewski wrote: >>>>>>>> This define gives the possibility to copy entire image >>>>>>>> (including header - e.g. u-boot.img) from NOR parallel >>>>>>>> memory to e.g. SDRAM. The current code only supports >>>>>>>> loading the raw binary image (the u-boot.bin). >>>>>>>> >>>>>>>> The legacy behavior is preserved, since other board >>>>>>>> don't enabled this option. >>>>>>> >>>>>>> Sooooo, what's the usecase again ? ;-) >>>>>> >>>>>> :-) >>>>>> >>>>>> The use case is to allow u-boot.img being loaded from >>>>>> Parallel NOR. The current code only supports u-boot.bin. >>>>> >>>>> Why is u-boot.bin (or the payload) not sufficient ? Why do >>>>> you need the header ? >>>> >>>> Well, the general use-case and code flow is that we load >>>> u-boot.img (or a FIT image) and if all else fails, fall back >>>> to assuming a .bin and a known address). >>>> >>> And exactly how is that whole image useful in RAM ? Sorry, I >>> still do not see it, usually you just need the executable >>> payload, although even that can be left in flash most of the >>> time. >> >> The use case is that I do want to boot from SD card/eMMC and >> NOR with using u-boot.img. >> >> I would like to avoid situation when for NOR I must use >> u-boot.bin and for eMMC u-boot.img. >> >> Such approach keeps things as simple as possible :-) > > Oh, so it allows you to detect bitrot for the content in SPI > NOR ?
I do not use SPI NOR, it is parallel NOR.
Sorry, I meant parallel NOR of course.
> It's a bit strange we had to use u-boot.bin with SPL there. >
This is how the legacy system behaves. It uses (by default) Parallel NOR for booting (with advised/provided NOR memory timings). After doing some measurements, it turned out that for "tunned" u-boot/SPL there would be the best way to copy it to ram and execute it from there (just like eMMC).
Hence, I would like to use u-boot.img in both booting scenarios.
I think I was mistaken yesterday, I don't think I understand why copying the image including the header into RAM has any benefit compared to copying just the image payload to RAM (and yes, we're getting back to my original question).
Code complexity and forward compatibility?
This is adding code complexity, but this is not my point.
The general case in the SPL framework is that we have either a "legacy" image or a FIT image and we fall back to "well, just run it!".
Well, this doesn't answer my question, because if I understand this patch correctly, it copies the entire legacy image (incl. header) into RAM instead of copying just the image payload (which we already do). I don't really understand why we want to do this. Or do I misunderstand something ?
No, you understood everything correctly. After some thoughts, I think that only payload should be copied.
But that's what we already do, no ? So what is the point of this patch ?
That is why I need to double check it :-)
I'll prepare test setup and double check this patch usability by Monday.
OK
Best regards, Łukasz Majewski
Hi Marek,
On 12/30/2016 10:28 PM, Lukasz Majewski wrote:
Hi Marek,
On 12/29/2016 04:26 PM, Tom Rini wrote:
On Thu, Dec 29, 2016 at 12:41:06AM +0100, Marek Vasut wrote:
On 12/28/2016 09:52 AM, Lukasz Majewski wrote:
Hi Marek,
> On 12/26/2016 05:36 PM, Lukasz Majewski wrote: >> Hi Marek, >> >>> On 11/29/2016 07:18 PM, Tom Rini wrote: >>>> On Tue, Nov 29, 2016 at 11:50:34AM +0100, Marek Vasut wrote: >>>>> On 11/29/2016 10:11 AM, Lukasz Majewski wrote: >>>>>> Hi Marek, >>>>>> >>>>>>> On 11/28/2016 10:09 PM, Lukasz Majewski wrote: >>>>>>>> This define gives the possibility to copy entire image >>>>>>>> (including header - e.g. u-boot.img) from NOR parallel >>>>>>>> memory to e.g. SDRAM. The current code only supports >>>>>>>> loading the raw binary image (the u-boot.bin). >>>>>>>> >>>>>>>> The legacy behavior is preserved, since other board >>>>>>>> don't enabled this option. >>>>>>> >>>>>>> Sooooo, what's the usecase again ? ;-) >>>>>> >>>>>> :-) >>>>>> >>>>>> The use case is to allow u-boot.img being loaded from >>>>>> Parallel NOR. The current code only supports u-boot.bin. >>>>> >>>>> Why is u-boot.bin (or the payload) not sufficient ? Why do >>>>> you need the header ? >>>> >>>> Well, the general use-case and code flow is that we load >>>> u-boot.img (or a FIT image) and if all else fails, fall back >>>> to assuming a .bin and a known address). >>>> >>> And exactly how is that whole image useful in RAM ? Sorry, I >>> still do not see it, usually you just need the executable >>> payload, although even that can be left in flash most of the >>> time. >> >> The use case is that I do want to boot from SD card/eMMC and >> NOR with using u-boot.img. >> >> I would like to avoid situation when for NOR I must use >> u-boot.bin and for eMMC u-boot.img. >> >> Such approach keeps things as simple as possible :-) > > Oh, so it allows you to detect bitrot for the content in SPI > NOR ?
I do not use SPI NOR, it is parallel NOR.
Sorry, I meant parallel NOR of course.
> It's a bit strange we had to use u-boot.bin with SPL there. >
This is how the legacy system behaves. It uses (by default) Parallel NOR for booting (with advised/provided NOR memory timings). After doing some measurements, it turned out that for "tunned" u-boot/SPL there would be the best way to copy it to ram and execute it from there (just like eMMC).
Hence, I would like to use u-boot.img in both booting scenarios.
I think I was mistaken yesterday, I don't think I understand why copying the image including the header into RAM has any benefit compared to copying just the image payload to RAM (and yes, we're getting back to my original question).
Code complexity and forward compatibility?
This is adding code complexity, but this is not my point.
The general case in the SPL framework is that we have either a "legacy" image or a FIT image and we fall back to "well, just run it!".
Well, this doesn't answer my question, because if I understand this patch correctly, it copies the entire legacy image (incl. header) into RAM instead of copying just the image payload (which we already do). I don't really understand why we want to do this. Or do I misunderstand something ?
No, you understood everything correctly. After some thoughts, I think that only payload should be copied.
But that's what we already do, no ? So what is the point of this patch ?
So now I do know a bit more ...
Let's start with ./common/spl/spl.c - spl_parse_image_header()
With SPL_COPY_PAYLOAD_ONLY flag set (@ common/spl/spl_nor.c) we go to:
if (spl_image->flags & SPL_COPY_PAYLOAD_ONLY) { /* * On some system (e.g. powerpc), the load-address and * entry-point is located at address 0. We can't load * to 0-0x40. So skip header in this case. */ spl_image->load_addr = image_get_load(header); spl_image->entry_point = image_get_ep(header); ^^^^^^^^^^^^^^^^^^^^^^^- here we set it to 0x0 by default (which is not true for our setup - we expect to boot from 0x17800000)
spl_image->size = image_get_data_size(header); }
My patch:
1. Do not set SPL_COPY_PAYLOAD_ONLY flag, so we go to else clause, which according to comment: "/* Load including the header */" and performs some address manipulation.
2. In my patch I undo those address calculations to load only payload.
Conclusion :-) --------------
And most of all ........ after thinking for a few minutes it is just enough to add:
#define CONFIG_SYS_UBOOT_START CONFIG_SYS_TEXT_BASE
to my config file and this patch can be dropped :-)
Big thanks Marek to be vigilant :-)
I'll prepare test setup and double check this patch usability by Monday.
OK
Best regards,
Lukasz Majewski
--
DENX Software Engineering GmbH, Managing Director: Wolfgang Denk HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd@denx.de
On 01/02/2017 01:07 AM, Lukasz Majewski wrote:
Hi Marek,
On 12/30/2016 10:28 PM, Lukasz Majewski wrote:
Hi Marek,
On 12/29/2016 04:26 PM, Tom Rini wrote:
On Thu, Dec 29, 2016 at 12:41:06AM +0100, Marek Vasut wrote:
On 12/28/2016 09:52 AM, Lukasz Majewski wrote: > Hi Marek, > >> On 12/26/2016 05:36 PM, Lukasz Majewski wrote: >>> Hi Marek, >>> >>>> On 11/29/2016 07:18 PM, Tom Rini wrote: >>>>> On Tue, Nov 29, 2016 at 11:50:34AM +0100, Marek Vasut wrote: >>>>>> On 11/29/2016 10:11 AM, Lukasz Majewski wrote: >>>>>>> Hi Marek, >>>>>>> >>>>>>>> On 11/28/2016 10:09 PM, Lukasz Majewski wrote: >>>>>>>>> This define gives the possibility to copy entire image >>>>>>>>> (including header - e.g. u-boot.img) from NOR parallel >>>>>>>>> memory to e.g. SDRAM. The current code only supports >>>>>>>>> loading the raw binary image (the u-boot.bin). >>>>>>>>> >>>>>>>>> The legacy behavior is preserved, since other board >>>>>>>>> don't enabled this option. >>>>>>>> >>>>>>>> Sooooo, what's the usecase again ? ;-) >>>>>>> >>>>>>> :-) >>>>>>> >>>>>>> The use case is to allow u-boot.img being loaded from >>>>>>> Parallel NOR. The current code only supports u-boot.bin. >>>>>> >>>>>> Why is u-boot.bin (or the payload) not sufficient ? Why do >>>>>> you need the header ? >>>>> >>>>> Well, the general use-case and code flow is that we load >>>>> u-boot.img (or a FIT image) and if all else fails, fall back >>>>> to assuming a .bin and a known address). >>>>> >>>> And exactly how is that whole image useful in RAM ? Sorry, I >>>> still do not see it, usually you just need the executable >>>> payload, although even that can be left in flash most of the >>>> time. >>> >>> The use case is that I do want to boot from SD card/eMMC and >>> NOR with using u-boot.img. >>> >>> I would like to avoid situation when for NOR I must use >>> u-boot.bin and for eMMC u-boot.img. >>> >>> Such approach keeps things as simple as possible :-) >> >> Oh, so it allows you to detect bitrot for the content in SPI >> NOR ? > > I do not use SPI NOR, it is parallel NOR.
Sorry, I meant parallel NOR of course.
>> It's a bit strange we had to use u-boot.bin with SPL there. >> > > This is how the legacy system behaves. It uses (by default) > Parallel NOR for booting (with advised/provided NOR memory > timings). After doing some measurements, it turned out that for > "tunned" u-boot/SPL there would be the best way to copy it to > ram and execute it from there (just like eMMC). > > Hence, I would like to use u-boot.img in both booting scenarios.
I think I was mistaken yesterday, I don't think I understand why copying the image including the header into RAM has any benefit compared to copying just the image payload to RAM (and yes, we're getting back to my original question).
Code complexity and forward compatibility?
This is adding code complexity, but this is not my point.
The general case in the SPL framework is that we have either a "legacy" image or a FIT image and we fall back to "well, just run it!".
Well, this doesn't answer my question, because if I understand this patch correctly, it copies the entire legacy image (incl. header) into RAM instead of copying just the image payload (which we already do). I don't really understand why we want to do this. Or do I misunderstand something ?
No, you understood everything correctly. After some thoughts, I think that only payload should be copied.
But that's what we already do, no ? So what is the point of this patch ?
So now I do know a bit more ...
Let's start with ./common/spl/spl.c - spl_parse_image_header()
With SPL_COPY_PAYLOAD_ONLY flag set (@ common/spl/spl_nor.c) we go to:
if (spl_image->flags & SPL_COPY_PAYLOAD_ONLY) { /* * On some system (e.g. powerpc), the load-address and * entry-point is located at address 0. We can't load * to 0-0x40. So skip header in this case. */ spl_image->load_addr = image_get_load(header); spl_image->entry_point = image_get_ep(header); ^^^^^^^^^^^^^^^^^^^^^^^- here we set it to 0x0 by default (which is not true for our setup - we expect to boot from 0x17800000)
spl_image->size = image_get_data_size(header); }
My patch:
- Do not set SPL_COPY_PAYLOAD_ONLY flag, so we go to else clause,
which according to comment: "/* Load including the header */" and performs some address manipulation.
- In my patch I undo those address calculations to load only payload.
Conclusion :-)
And most of all ........ after thinking for a few minutes it is just enough to add:
#define CONFIG_SYS_UBOOT_START CONFIG_SYS_TEXT_BASE
to my config file and this patch can be dropped :-)
Great, I like this solution.
Big thanks Marek to be vigilant :-)
You're welcome ;-)
participants (4)
-
Lukasz Majewski -
Lukasz Majewski -
Marek Vasut -
Tom Rini