[U-Boot] Master branch broken for omap4_panda (SPL_BUILD)

[U-Boot] Master branch broken for omap4_panda (SPL_BUILD)

[Aguirre, Sergio]

Hi all,

I'm trying to build this commit:

http://git.denx.de/?p=u-boot.git;a=commit;h=7b8ffea2ac44097ed1c99ba70b8c6a4cf12ba0b4

with:

make omap4_panda_config
make

But it fails for spl build, while trying to link libfat.o with
malloc()/free() functions.

Git-bisect throws this commit to blame:

http://git.denx.de/?p=u-boot.git;a=commit;h=ac4977719e157bcb3c45c70d9dd781164727530d

Has anyone seen this?

Regards,
Sergio

[U-Boot] Master branch broken for omap4_panda (SPL_BUILD)

[Aneesh V]

Hi Sergio,

On Tuesday 04 October 2011 12:20 AM, Aguirre, Sergio wrote:
> Hi all,
>
> I'm trying to build this commit:
>
> http://git.denx.de/?p=u-boot.git;a=commit;h=7b8ffea2ac44097ed1c99ba70b8c6a4cf12ba0b4
>
> with:
>
> make omap4_panda_config
> make
>
> But it fails for spl build, while trying to link libfat.o with
> malloc()/free() functions.
>
> Git-bisect throws this commit to blame:
>
> http://git.denx.de/?p=u-boot.git;a=commit;h=ac4977719e157bcb3c45c70d9dd781164727530d
>
> Has anyone seen this?

Yes, I tried this just now and I am also facing this issue. I can fix
the build issue using the following change. But that doesn't solve the
problem. We do not have heap in SPL. So, this is a real serious problem
for any SPL that has FAT support. I will try to setup a heap in SDRAM
as soon as I get some time to work on this.

---
diff --git a/common/Makefile b/common/Makefile
index 2edbd71..b424beb 100644
--- a/common/Makefile
+++ b/common/Makefile
@@ -29,7 +29,6 @@ LIB	= $(obj)libcommon.o
  ifndef CONFIG_SPL_BUILD
  COBJS-y += main.o
  COBJS-y += command.o
-COBJS-y += dlmalloc.o
  COBJS-y += exports.o
  COBJS-$(CONFIG_SYS_HUSH_PARSER) += hush.o
  COBJS-y += image.o
@@ -176,6 +175,7 @@ COBJS-$(CONFIG_USB_KEYBOARD) += usb_kbd.o
  endif

  COBJS-y += console.o
+COBJS-y += dlmalloc.o
  COBJS-y += memsize.o
  COBJS-y += stdio.o



>
> Regards,
> Sergio
> _______________________________________________
> U-Boot mailing list
> U-Boot at lists.denx.de
> http://lists.denx.de/mailman/listinfo/u-boot

[U-Boot] Master branch broken for omap4_panda (SPL_BUILD)

[Wolfgang Denk]

Dear Aneesh V,

In message <4E8B1508.7020706@ti.com> you wrote:
> 
> Yes, I tried this just now and I am also facing this issue. I can fix
> the build issue using the following change. But that doesn't solve the
> problem. We do not have heap in SPL. So, this is a real serious problem
> for any SPL that has FAT support. I will try to setup a heap in SDRAM
> as soon as I get some time to work on this.

All file system code has been designed and implemented to run AFTER
relocation.  Access to a FAT file system is not supposed to happen
from SPL code.

Best regards,

Wolfgang Denk

-- 
DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd at denx.de
Landing: a controlled mid-air collision with a planet.

[U-Boot] [PATCH] omap: spl: fix build break due to changes in FAT

[Aneesh V]

From: a0393566local (none) <a0393566local@ula0393566>

FAT library now uses malloc() and free(). But SPL doesn't
have heap until now. Setup a heap in SDRAM to fix this issue.

However this increases SPL footprint beyond the available SRAM
budget. So, compile out some fancy features in the SDARM init
bring back footprint under control

CC: Sandeep Paulraj <s-paulraj@ti.com>
CC: Wolfgang Denk <wd@denx.de>
Signed-off-by: Aneesh V <aneesh@ti.com>
---
---
 arch/arm/cpu/armv7/omap-common/spl.c |    4 ++++
 common/Makefile                      |    2 +-
 include/configs/omap4_common.h       |   19 +++++++++++--------
 3 files changed, 16 insertions(+), 9 deletions(-)

diff --git a/arch/arm/cpu/armv7/omap-common/spl.c b/arch/arm/cpu/armv7/omap-common/spl.c
index c76fea6..46a1413 100644
--- a/arch/arm/cpu/armv7/omap-common/spl.c
+++ b/arch/arm/cpu/armv7/omap-common/spl.c
@@ -35,6 +35,7 @@
 #include <asm/arch/mmc_host_def.h>
 #include <i2c.h>
 #include <image.h>
+#include <malloc.h>
 
 DECLARE_GLOBAL_DATA_PTR;
 
@@ -106,6 +107,9 @@ void board_init_r(gd_t *id, ulong dummy)
 	u32 boot_device;
 	debug(">>spl:board_init_r()\n");
 
+	mem_malloc_init(CONFIG_SYS_SPL_MALLOC_START,
+			CONFIG_SYS_SPL_MALLOC_SIZE);
+
 	timer_init();
 	i2c_init(CONFIG_SYS_I2C_SPEED, CONFIG_SYS_I2C_SLAVE);
 
diff --git a/common/Makefile b/common/Makefile
index 2edbd71..b424beb 100644
--- a/common/Makefile
+++ b/common/Makefile
@@ -29,7 +29,6 @@ LIB	= $(obj)libcommon.o
 ifndef CONFIG_SPL_BUILD
 COBJS-y += main.o
 COBJS-y += command.o
-COBJS-y += dlmalloc.o
 COBJS-y += exports.o
 COBJS-$(CONFIG_SYS_HUSH_PARSER) += hush.o
 COBJS-y += image.o
@@ -176,6 +175,7 @@ COBJS-$(CONFIG_USB_KEYBOARD) += usb_kbd.o
 endif
 
 COBJS-y += console.o
+COBJS-y += dlmalloc.o
 COBJS-y += memsize.o
 COBJS-y += stdio.o
 
diff --git a/include/configs/omap4_common.h b/include/configs/omap4_common.h
index 228eac5..f438fc2 100644
--- a/include/configs/omap4_common.h
+++ b/include/configs/omap4_common.h
@@ -238,6 +238,7 @@
 #endif
 
 /* Defines for SDRAM init */
+#define CONFIG_SYS_EMIF_PRECALCULATED_TIMING_REGS
 #ifndef CONFIG_SYS_EMIF_PRECALCULATED_TIMING_REGS
 #define CONFIG_SYS_AUTOMATIC_SDRAM_DETECTION
 #define CONFIG_SYS_DEFAULT_LPDDR2_TIMINGS
@@ -251,6 +252,16 @@
 
 #define CONFIG_SPL_BSS_START_ADDR	0x80000000
 #define CONFIG_SPL_BSS_MAX_SIZE		0x80000		/* 512 KB */
+/*
+ * 1MB into the SDRAM to allow for SPL's bss at the beginning of SDRAM
+ * 64 bytes before this address should be set aside for u-boot.img's
+ * header. That is 0x800FFFC0--0x80100000 should not be used for any
+ * other needs.
+ */
+#define CONFIG_SYS_TEXT_BASE		0x80100000
+#define CONFIG_SYS_SPL_MALLOC_START	0x80200000
+#define CONFIG_SYS_SPL_MALLOC_SIZE	0x100000	/* 1 MB */
+
 
 #define CONFIG_SYS_MMCSD_RAW_MODE_U_BOOT_SECTOR	0x300 /* address 0x60000 */
 #define CONFIG_SYS_U_BOOT_MAX_SIZE_SECTORS	0x200 /* 256 KB */
@@ -266,12 +277,4 @@
 #define CONFIG_SPL_SERIAL_SUPPORT
 #define CONFIG_SPL_LDSCRIPT "arch/arm/cpu/armv7/omap-common/u-boot-spl.lds"
 
-/*
- * 1MB into the SDRAM to allow for SPL's bss at the beginning of SDRAM
- * 64 bytes before this address should be set aside for u-boot.img's
- * header. That is 0x800FFFC0--0x80100000 should not be used for any
- * other needs.
- */
-#define CONFIG_SYS_TEXT_BASE		0x80100000
-
 #endif /* __CONFIG_OMAP4_COMMON_H */
-- 
1.7.1

[U-Boot] Master branch broken for omap4_panda (SPL_BUILD)

[Aneesh V]

Dear Wolfgang,

On Tuesday 04 October 2011 08:21 PM, Wolfgang Denk wrote:
> Dear Aneesh V,
>
> In message<4E8B1508.7020706@ti.com>  you wrote:
>>
>> Yes, I tried this just now and I am also facing this issue. I can fix
>> the build issue using the following change. But that doesn't solve the
>> problem. We do not have heap in SPL. So, this is a real serious problem
>> for any SPL that has FAT support. I will try to setup a heap in SDRAM
>> as soon as I get some time to work on this.
>
> All file system code has been designed and implemented to run AFTER
> relocation.  Access to a FAT file system is not supposed to happen
> from SPL code.

SPL does call relocate_code() but defeats it by passing the relocation
target same as the SPL_TEXT_BASE.

OMAP has traditionally supported booting from the FAT partition of an
external SD card. The ROM code can pick up the SPL from the FAT
partition and SPL should pick up u-boot from there too. Some of our
boards like OMAP4 Panda have only an SD slot for non-volatile memory
and no flash memory installed. Of course we could use the SD card in
raw mode. But the FAT mode is much more convenient for many users. If
FAT is not supported in SPL, it will be a major regression compared to
x-loader and a show-stopper for us.

I have submitted a patch to fix the problem reported in this thread and
FAT boot is working fine again.

However, I am worried about the increasing footprint of SPL. The main
contributors are MMC and FAT. x-loader, with all it's short-comings,
does the same job with a smaller footprint. NAND has a custom driver
for SPL while MMC and FAT doesn't have. I don't think it's practical
either. However, I wonder if these drivers could be scaled down in size
for SPL with some '#ifdef CONFIG_SPL_BUILD's. I wish if somebody could
look into this.

best regards,
Aneesh

[U-Boot] [PATCH] omap: spl: fix build break due to changes in FAT

[Wolfgang Denk]

Dear Aneesh V,

In message <1317741066-31121-1-git-send-email-aneesh@ti.com> you wrote:
> From: a0393566local (none) <a0393566local@ula0393566>
> 
> FAT library now uses malloc() and free(). But SPL doesn't
> have heap until now. Setup a heap in SDRAM to fix this issue.
> 
> However this increases SPL footprint beyond the available SRAM
> budget. So, compile out some fancy features in the SDARM init
> bring back footprint under control
> 
> CC: Sandeep Paulraj <s-paulraj@ti.com>
> CC: Wolfgang Denk <wd@denx.de>
> Signed-off-by: Aneesh V <aneesh@ti.com>

I think it is a bad idea to go this way.

We should face the fact that SPL code is running before proper
relocation to RAM, and thus there are certain limitations.

Certain parts of the code, including file system code, has not been
written with such limitations in mind.  It makes use of functions that
are not available in SPL code, or of features that are not available
in SPL code (like a heap, or a virtually unlimited stack).

You may be lucky here, and your test cases with the FFAT code may
actually work.  But I would not bet on it.


U-Boot has not been designed to run complex code like file systemes
before relocation, and SPL code _is_ before relocation.

Best regards,

Wolfgang Denk

-- 
DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd at denx.de
Lead me not into temptation... I can find it myself.

[U-Boot] [PATCH] omap: spl: fix build break due to changes in FAT

[Aneesh V]

Dear Wolfgang,

On Tuesday 04 October 2011 09:08 PM, Wolfgang Denk wrote:
> Dear Aneesh V,
>
> In message<1317741066-31121-1-git-send-email-aneesh@ti.com>  you wrote:
>> From: a0393566local (none)<a0393566local@ula0393566>
>>
>> FAT library now uses malloc() and free(). But SPL doesn't
>> have heap until now. Setup a heap in SDRAM to fix this issue.
>>
>> However this increases SPL footprint beyond the available SRAM
>> budget. So, compile out some fancy features in the SDARM init
>> bring back footprint under control
>>
>> CC: Sandeep Paulraj<s-paulraj@ti.com>
>> CC: Wolfgang Denk<wd@denx.de>
>> Signed-off-by: Aneesh V<aneesh@ti.com>
>
> I think it is a bad idea to go this way.
>
> We should face the fact that SPL code is running before proper
> relocation to RAM, and thus there are certain limitations.
>
> Certain parts of the code, including file system code, has not been
> written with such limitations in mind.  It makes use of functions that
> are not available in SPL code, or of features that are not available
> in SPL code (like a heap, or a virtually unlimited stack).
>
> You may be lucky here, and your test cases with the FFAT code may
> actually work.  But I would not bet on it.
>

I have mentioned why FAT is inevitable for us in a separate mail in
this thread.

>
> U-Boot has not been designed to run complex code like file systemes
> before relocation, and SPL code _is_ before relocation.

Please note that SPL is not as restricted as it seems(at least the way
it is now). For instance, OMAP SPL has
1. gd in SRAM
2. .bss in SDRAM and
3. heap in SDRAM(with this patch)
So, we have everything that takes for any driver to function as long as
the .text fits in the SRAM.

I think the same principles apply to all platforms not just OMAP. These
complex drivers are likely to be used only after SDRAM is initialized.
After SDRAM is initialized you have plenty of space for .bss and heap.
So the only limitation for an SPL should be .text and .data size.

best regards,
Aneesh

[U-Boot] [PATCH] omap: spl: fix build break due to changes in FAT

[Scott Wood]

On 10/04/2011 10:38 AM, Wolfgang Denk wrote:
> I think it is a bad idea to go this way.
> 
> We should face the fact that SPL code is running before proper
> relocation to RAM, and thus there are certain limitations.
> 
> Certain parts of the code, including file system code, has not been
> written with such limitations in mind.  It makes use of functions that
> are not available in SPL code, or of features that are not available
> in SPL code (like a heap, or a virtually unlimited stack).
> 
> You may be lucky here, and your test cases with the FFAT code may
> actually work.  But I would not bet on it.
> 
> 
> U-Boot has not been designed to run complex code like file systemes
> before relocation, and SPL code _is_ before relocation.

SPL has its own relocation.  If it were under the same restrictions as
normal pre-relocation U-Boot, where would it load an image to?

All SPL really is, is some makefile infrastructure to produce a second,
special-purpose U-Boot image.  The degree to which it is stripped down
depends on the requirements of the particular target.

-Scott

[U-Boot] Master branch broken for omap4_panda (SPL_BUILD)

[Wolfgang Denk]

Dear Aneesh V,

In message <4E8B282A.8040301@ti.com> you wrote:
>
> > All file system code has been designed and implemented to run AFTER
> > relocation.  Access to a FAT file system is not supposed to happen
> > from SPL code.
...
> OMAP has traditionally supported booting from the FAT partition of an
> external SD card. The ROM code can pick up the SPL from the FAT
> partition and SPL should pick up u-boot from there too. Some of our
> boards like OMAP4 Panda have only an SD slot for non-volatile memory
> and no flash memory installed. Of course we could use the SD card in
> raw mode. But the FAT mode is much more convenient for many users. If
> FAT is not supported in SPL, it will be a major regression compared to
> x-loader and a show-stopper for us.

I understand this, and you may have noted that I did not NAK the
patch.  I just want to point out that what you are doing is pretty
dangerous, and there are no seat-belts.

> I have submitted a patch to fix the problem reported in this thread and
> FAT boot is working fine again.

For now.  But please expect that the next breakage may happen any
time, and there is no guarantee that it can be fixed as easily.

> However, I am worried about the increasing footprint of SPL. The main
> contributors are MMC and FAT. x-loader, with all it's short-comings,
> does the same job with a smaller footprint. ...

Why is this the case?  I doubt they use other FAT code, or do they?
What about MMC?

>                                         ... NAND has a custom driver
> for SPL while MMC and FAT doesn't have. I don't think it's practical
> either. However, I wonder if these drivers could be scaled down in size
> for SPL with some '#ifdef CONFIG_SPL_BUILD's. I wish if somebody could
> look into this.

Do you have exact numbers of the sizes for the MMC, FAT and NAND code
in x-loader compared to SPL U-Boot?  And what are for example the
restrictions of (V)FAT support in x-loader?

Best regards,

Wolfgang Denk

-- 
DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd at denx.de
You can fool some of the people all of the time, and You can fool all
of the people some of the time, but You can't fool mom.

[U-Boot] Master branch broken for omap4_panda (SPL_BUILD)

[Aneesh V]

Dear Wolfgang,

On Tuesday 04 October 2011 11:35 PM, Wolfgang Denk wrote:
> Dear Aneesh V,
>
> In message<4E8B282A.8040301@ti.com>  you wrote:
>>
>>> All file system code has been designed and implemented to run AFTER
>>> relocation.  Access to a FAT file system is not supposed to happen
>>> from SPL code.
> ...
>> OMAP has traditionally supported booting from the FAT partition of an
>> external SD card. The ROM code can pick up the SPL from the FAT
>> partition and SPL should pick up u-boot from there too. Some of our
>> boards like OMAP4 Panda have only an SD slot for non-volatile memory
>> and no flash memory installed. Of course we could use the SD card in
>> raw mode. But the FAT mode is much more convenient for many users. If
>> FAT is not supported in SPL, it will be a major regression compared to
>> x-loader and a show-stopper for us.
>
> I understand this, and you may have noted that I did not NAK the
> patch.  I just want to point out that what you are doing is pretty
> dangerous, and there are no seat-belts.
>

Thank you. I realize that maintaining our SPL will not be easy,
particularly given the fact that most boards do not have SPL, much less
FAT support in it. So, we are bound to see issues inadvertently
introduced by others. But we have no other choice.

>> I have submitted a patch to fix the problem reported in this thread and
>> FAT boot is working fine again.
>
> For now.  But please expect that the next breakage may happen any
> time, and there is no guarantee that it can be fixed as easily.
>
>> However, I am worried about the increasing footprint of SPL. The main
>> contributors are MMC and FAT. x-loader, with all it's short-comings,
>> does the same job with a smaller footprint. ...
>
> Why is this the case?  I doubt they use other FAT code, or do they?
> What about MMC?
>

X-loader, as far as I know, is based on a very old fork of u-boot. MMC
driver is a custom monolithic driver for OMAP4(not the generic MMC
framework in the current mainline). We didn't take a closer look at the
code differences yet.

>>                                          ... NAND has a custom driver
>> for SPL while MMC and FAT doesn't have. I don't think it's practical
>> either. However, I wonder if these drivers could be scaled down in size
>> for SPL with some '#ifdef CONFIG_SPL_BUILD's. I wish if somebody could
>> look into this.
>
> Do you have exact numbers of the sizes for the MMC, FAT and NAND code
> in x-loader compared to SPL U-Boot?  And what are for example the
> restrictions of (V)FAT support in x-loader?

Yes, we did a crude measurement by comparing the image size before and
after commenting out the calls to the respective libraries(Pls note
that we have -gc-sections enabled in both the cases). Here are the
results.

		x-loader	SPL
MMC		3724		6900
FAT		2464		9340

Together the two makes a difference of 10052 Bytes, and far outweighs
the gains made in sdram init and clock init.

best regards,
Aneesh


>
> Best regards,
>
> Wolfgang Denk
>

[U-Boot] Master branch broken for omap4_panda (SPL_BUILD)

[Tom Rini]

On Tue, Oct 4, 2011 at 7:51 AM, Wolfgang Denk <wd@denx.de> wrote:
> Dear Aneesh V,
>
> In message <4E8B1508.7020706@ti.com> you wrote:
>>
>> Yes, I tried this just now and I am also facing this issue. I can fix
>> the build issue using the following change. But that doesn't solve the
>> problem. We do not have heap in SPL. So, this is a real serious problem
>> for any SPL that has FAT support. I will try to setup a heap in SDRAM
>> as soon as I get some time to work on this.
>
> All file system code has been designed and implemented to run AFTER
> relocation. ?Access to a FAT file system is not supposed to happen
> from SPL code.

Since this wasn't a NAK just a warning about possible future problems,
are you expecting this to come in via Sandeep's tree?  Thanks!

-- 
Tom

[U-Boot] Master branch broken for omap4_panda (SPL_BUILD)

[Wolfgang Denk]

Dear Tom Rini,

In message <CA+M6bX=GQEJ0nccRYkgc-qz1u3cMYXEmmhKL0QvZkZiy2VWXag@mail.gmail.com> you wrote:
>
> > All file system code has been designed and implemented to run AFTER
> > relocation. =A0Access to a FAT file system is not supposed to happen
> > from SPL code.
> 
> Since this wasn't a NAK just a warning about possible future problems,
> are you expecting this to come in via Sandeep's tree?  Thanks!

Yes: I'm not happy about it, but I will not block it.

Best regards,

Wolfgang Denk

-- 
DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd at denx.de
A committee is a life form with six or more legs and no brain.
                              -- Lazarus Long, "Time Enough For Love"