Uncompress debug for multiplatform

Spam Status: Spamassassin 0% probability of being spam.

Full report:
No, score=-2.6 required=5.0 tests=BAYES_00,RCVD_IN_DNSWL_LOW, UNPARSEABLE_RELAY autolearn=unavailable version=3.3.1 Following the approach handling DEBUG_LL inclusion, the patch creates
a Kconfig symbol CONFIG_UNCOMPRESS_INCLUDE for choosing the correct
uncompress header. For traditional build, mach/uncompress.h will be
included in arch/arm/boot/compressed/misc.c. For multiplatform build,
debug/uncompress.h which contains a suite of empty functions will be
used. In this way, a platform with particular uncompress.h
implementation could choose its own uncompress.h with this Kconfig
option.

Signed-off-by Shawn Guo <[email protected]>

 arch/arm/boot/compressed/Makefile   |    3 +++
 arch/arm/boot/compressed/debug.S    |    3 +++
 arch/arm/include/debug/highbank.S   |    3 +++
 arch/arm/include/debug/imx.S        |    3 +++
 arch/arm/include/debug/mvebu.S      |    3 +++
 arch/arm/include/debug/picoxcell.S  |    3 +++
 arch/arm/include/debug/socfpga.S    |    3 +++
 arch/arm/include/debug/sunxi.S      |    3 +++
 arch/arm/include/debug/tegra.S      |    3 +++
 arch/arm/include/debug/uncompress.h |    8 +++++++-
 arch/arm/include/debug/vexpress.S   |    3 +++
 arch/arm/include/debug/zynq.S       |    3 +++
 12 files changed, 40 insertions(+), 1 deletion(-)
 create mode 100644 arch/arm/boot/compressed/debug.S

arch/arm/boot/compressed/debug.S new file +3 -0

new file mode 100644
index 0000000..7283d5c
@@ -0,0 +1,3 @@
+#define UNCOMPRESS_DEBUG
+
+#include "../../kernel/debug.S"

arch/arm/include/debug/highbank.S +0 -0

index 8cad432..13056cf 100644

However, there is a problem with above approach. DEBUG_LL routines will check MMU enable bit to decide whether physical or virtual address should be used to access debug port. In case virtual address needs to be used, the address returned by addruart will not work for decompressor who uses a different mapping from what kernel uses. Fortunately, decompressor uses a flat mapping (same physical and virtual addresses). So we can easily work it around by asking platform addruart return physical address as virtual address when it runs in decompressor. The macro UNCOMPRESS_DEBUG is defined for this use.

Quoted 9 lines · Shawn Guo

However, there is a problem with above approach. DEBUG_LL routines will check MMU enable bit to decide whether physical or virtual address should be used to access debug port. In case virtual address needs to be used, the address returned by addruart will not work for decompressor who uses a different mapping from what kernel uses. Fortunately, decompressor uses a flat mapping (same physical and virtual addresses). So we can easily work it around by asking platform addruart return physical address as virtual address when it runs in decompressor. The macro UNCOMPRESS_DEBUG is defined for this use.

Can't you just create a new copy of kernel/debug.S in boot/compressed/ that provides only a putc() function and uses the physical address unconditionally?

arch/arm/boot/compressed/debug.S line 11

Quoted 7 lines · Shawn Guo

arch/arm/boot/compressed/debug.S new file +7 -0

+ENTRY(putc)
+       addruart r1, r2, r3
+       waituart r3, r1
+       senduart r0, r1
+       busyuart r3, r1
+       mov      pc, lr
+ENDPROC(putc)

Ah, so it actually worked? I was expecting at least some part of my code to be wrong ;-) My assembler skills are very much lacking and I had not tried it.

Upon closer inspection, it seems that the CR/LF logic from the printascii function is not here, and it probably should be.

arch/arm/boot/compressed/debug.S line 11

Quoted 7 lines · Shawn Guo

arch/arm/boot/compressed/debug.S new file +7 -0

+ENTRY(putc)
+       addruart r1, r2, r3
+       waituart r3, r1
+       senduart r0, r1
+       busyuart r3, r1
+       mov      pc, lr
+ENDPROC(putc)

Ah, so it actually worked? I was expecting at least some part of my code to be wrong ;-) My assembler skills are very much lacking and I had not tried it.

Upon closer inspection, it seems that the CR/LF logic from the printascii function is not here, and it probably should be.

Instead of giving zero support of uncompress debug for multiplatform build, the patch turns uncompress debug into one part of DEBUG_LL support. When DEBUG_LL is turned on for a particular platform, uncompress debug works too for that platform.

It reuses the platform DEBUG_LL macros by creating a simple arch/arm/boot/compressed/debug.S with CONFIG_DEBUG_LL_INCLUDE included there, and implements a generic putc() using those macros.

Signed-off-by Shawn Guo <[email protected]>

Instead of giving zero support of uncompress debug for multiplatform build, the patch turns uncompress debug into one part of DEBUG_LL support. When DEBUG_LL is turned on for a particular platform, uncompress debug works too for that platform.

It reuses the platform DEBUG_LL macros by creating a simple arch/arm/boot/compressed/debug.S with CONFIG_DEBUG_LL_INCLUDE included there, and implements a generic putc() using those macros.

Quoted 8 lines · Shawn Guo

Instead of giving zero support of uncompress debug for multiplatform build, the patch turns uncompress debug into one part of DEBUG_LL support. When DEBUG_LL is turned on for a particular platform, uncompress debug works too for that platform.

It reuses the platform DEBUG_LL macros by creating a simple arch/arm/boot/compressed/debug.S with CONFIG_DEBUG_LL_INCLUDE included there, and implements a generic putc() using those macros.

Ok, I've applied this on the previso that _no one_ in future whinges if the debug infrastructure doesn't quite meet their expectation. The debug infrastructure remains first and foremost that: a simple debug infrastructure suitable for use in the early assembly and the like.

That is its primary concern and trumps any requirements from consoles, early printk, decompressor output, and the like.

Quoted 15 lines · Russell King - ARM Linux

Quoted 8 lines · Shawn Guo

Instead of giving zero support of uncompress debug for multiplatform build, the patch turns uncompress debug into one part of DEBUG_LL support. When DEBUG_LL is turned on for a particular platform, uncompress debug works too for that platform.

It reuses the platform DEBUG_LL macros by creating a simple arch/arm/boot/compressed/debug.S with CONFIG_DEBUG_LL_INCLUDE included there, and implements a generic putc() using those macros.

Ok, I've applied this on the previso that _no one_ in future whinges if the debug infrastructure doesn't quite meet their expectation. The debug infrastructure remains first and foremost that: a simple debug infrastructure suitable for use in the early assembly and the like.

Quoted 2 lines

That is its primary concern and trumps any requirements from consoles, early printk, decompressor output, and the like.

... and now I've dropped the two patches because it causes build failures for all OMAP and PXA platforms.

Quoted 19 lines · Russell King - ARM Linux

Quoted 17 lines · Russell King - ARM Linux

Quoted 8 lines · Shawn Guo

Instead of giving zero support of uncompress debug for multiplatform build, the patch turns uncompress debug into one part of DEBUG_LL support. When DEBUG_LL is turned on for a particular platform, uncompress debug works too for that platform.

It reuses the platform DEBUG_LL macros by creating a simple arch/arm/boot/compressed/debug.S with CONFIG_DEBUG_LL_INCLUDE included there, and implements a generic putc() using those macros.

Ok, I've applied this on the previso that _no one_ in future whinges if the debug infrastructure doesn't quite meet their expectation. The debug infrastructure remains first and foremost that: a simple debug infrastructure suitable for use in the early assembly and the like.

That is its primary concern and trumps any requirements from consoles, early printk, decompressor output, and the like.

Quoted 2 lines

... and now I've dropped the two patches because it causes build failures for all OMAP and PXA platforms.

It also breaks tegra_defconfig. For reference, the (or perhaps just a) reason here is that arch/arm/include/debug/tegra.S references symbol tegra_uart_config, which is declared in arch/arm/mach-tegra/common.c, which isn't part of the decompressor build. Tegra's uncompress.h doesn't touch this symbol, hence avoids this problem.

From what I see, the patch breaks omap, pxa and tegra build by

I'd guess OMAP is broken for similar reasons, since when I created the tegra_uart_config symbol I was inspired by the OMAP DEBUG_LL code.

Perhaps the patch can be re-cast to only affect multi-platform kernels, and leave unconverted platforms using uncompress.h (at least, I assume that must be the problem).

Quoted 20 lines · Stephen Warren

Quoted 19 lines · Russell King - ARM Linux

Quoted 17 lines · Russell King - ARM Linux

Quoted 8 lines · Shawn Guo

Instead of giving zero support of uncompress debug for multiplatform build, the patch turns uncompress debug into one part of DEBUG_LL support. When DEBUG_LL is turned on for a particular platform, uncompress debug works too for that platform.

It reuses the platform DEBUG_LL macros by creating a simple arch/arm/boot/compressed/debug.S with CONFIG_DEBUG_LL_INCLUDE included there, and implements a generic putc() using those macros.

Ok, I've applied this on the previso that _no one_ in future whinges if the debug infrastructure doesn't quite meet their expectation. The debug infrastructure remains first and foremost that: a simple debug infrastructure suitable for use in the early assembly and the like.

That is its primary concern and trumps any requirements from consoles, early printk, decompressor output, and the like.

Quoted 9 lines

Quoted 2 lines

... and now I've dropped the two patches because it causes build failures for all OMAP and PXA platforms.

It also breaks tegra_defconfig. For reference, the (or perhaps just a) reason here is that arch/arm/include/debug/tegra.S references symbol tegra_uart_config, which is declared in arch/arm/mach-tegra/common.c, which isn't part of the decompressor build. Tegra's uncompress.h doesn't touch this symbol, hence avoids this problem.

Thanks for the info, Stephen.

From what I see, the patch breaks omap, pxa and tegra build by different causes.

Quoted 3 lines

Perhaps the patch can be re-cast to only affect multi-platform kernels, and leave unconverted platforms using uncompress.h (at least, I assume that must be the problem).

Indeed. The arch/arm/boot/compressed/debug.S shouldn't be part of traditional build but only multiplatform. Let's force that with the change below.

8<---

arch/arm/boot/compressed/Makefile +0 -0

index c9865f6..13bdd10 100644
@@ -24,9 +24,11 @@ endif
 AFLAGS_head.o += -DTEXT_OFFSET=$(TEXT_OFFSET)
 HEAD   = head.o

=== omap ===

AS arch/arm/boot/compressed/debug.o
LD arch/arm/boot/compressed/vmlinux
.data' referenced in section .text' of arch/arm/boot/compressed/debug.o: defined in discarded section `.data' of arch/arm/boot/compressed/debug.o make[3]: *** [arch/arm/boot/compressed/vmlinux] Error 1

The following change moving the variables out of stack section seems fixing the failure, but I'm not quite sure if it works.

8<---

arch/arm/mach-omap2/include/mach/debug-macro.S +0 -0

index cfaed13..7b2877e 100644

Instead of giving zero support of uncompress debug for multiplatform build, the patch turns uncompress debug into one part of DEBUG_LL support. When DEBUG_LL is turned on for a particular platform, uncompress debug works too for that platform.

It reuses the platform DEBUG_LL macros by creating a simple arch/arm/boot/compressed/debug.S with CONFIG_DEBUG_LL_INCLUDE included there, and implements a generic putc() using those macros.

Signed-off-by Shawn Guo <[email protected]>

Acked-by Olof Johansson <[email protected]>

---
Changes since v2:
* Build arch/arm/boot/compressed/debug.S only for multiplatform kernel to fix the breakage on traditional platform build
* Include asm/assembler.h in arch/arm/boot/compressed/debug.S, as it may commonly be needed by platforms to be converted to multiplatform

Russell,

I will put it into patch system if I do not receive any comment tomorrow.

Quoted 18 lines · Shawn Guo

Instead of giving zero support of uncompress debug for multiplatform build, the patch turns uncompress debug into one part of DEBUG_LL support. When DEBUG_LL is turned on for a particular platform, uncompress debug works too for that platform.

It reuses the platform DEBUG_LL macros by creating a simple arch/arm/boot/compressed/debug.S with CONFIG_DEBUG_LL_INCLUDE included there, and implements a generic putc() using those macros.

Signed-off-by Shawn Guo <[email protected]>

Acked-by Olof Johansson <[email protected]>

---
Changes since v2:
* Build arch/arm/boot/compressed/debug.S only for multiplatform kernel to fix the breakage on traditional platform build
* Include asm/assembler.h in arch/arm/boot/compressed/debug.S, as it may commonly be needed by platforms to be converted to multiplatform

Quoted 4 lines

Russell,

I will put it into patch system if I do not receive any comment tomorrow.

This doesn't address the problem with OMAP, which is that the OMAP debug code places data into the .data section, and that is not allowed in the decompressor.

Quoted 7 lines · Shawn Guo

It's not about running single zImage on multiple platforms with uncompress debug output, but having uncompress debug work for platform that DEBUG_LL is enabled for. But still it's better than nothing.

Shawn Guo (2):
ARM: use Kconfig to select uncompress.h
ARM: use DEBUG_LL infrastructural for multiplatform uncompress debug

Just put them into patch tracker as #7633 and #7634.

Shawn