MIPS: relocation issue

MIPS: relocation issue

[Antony Pavlov]

Hi!

Since 28ed10d6a6 ("MIPS: relocation: add relocation support")
iomem output on memory regions is invalid,
all the code that rely on registered sdram memory regions
after relocation can fail. E.g. the memtest commands fails
on barebox.

How to reproduce the problem ov v2020.01.0:

$ wget https://www.barebox.org/download/barebox-2020.01.0.tar.bz2
...
$ tar fx barebox-2020.01.0.tar.bz2
$ cd barebox-2020.01.0
$ export ARCH=mips
$ export CROSS_COMPILE=mips-linux-gnu-
$ make qemu-malta_defconfig
$ sed -i "s/^.*CONFIG_CMD_MEMTEST.*/CONFIG_CMD_MEMTEST=y/" .config
$ make
$ qemu-system-mips -nodefaults -M malta -m 256 -nographic -serial stdio -monitor null -bios barebox-flash-image
...
barebox@qemu malta:/ memtest
Start iteration 1 of 1.
Testing memory space: 0xa0865000 -> 0xafffffff:
Starting data line test.
Check for address bits stuck high.
Check for address bits stuck low or shorted.
Starting moving inversions test of RAM:
Fill with address, compare, fill with inverted address, compare again
	[##########################################                       ]
FAILURE (read/write): expected 0x03cca9f2, actual 0xa08126c4 at address 0xafb8f7c4.

Ooops, TLB miss on load or ifetch!

$ 0   : 00000000 00000000 fffffffb a085c900
$ 4   : a0407e90 0000000a b80003f8 00000001
$ 8   : a0858f28 00000000 00000053 00000008
$12   : 00000000 00000000 0000005c 00000001
$16   : 03ccab3c 03ccab3d a0825c48 00000000
$20   : 00000000 00000001 a0853518 a085352c
$24   : 00000018 a08124b8                  
$28   : 00000000 8fb8fcf8 a0860000 03ccab3e
Hi    : 0000000a
Lo    : 00000000
epc   : 03ccab3e
ra    : 03ccab3e
Status: 00000002
Cause : 00000408
Config: 80008482

### ERROR ### Please RESET the board ###


memtest for qemu-malta can be repaired by disabling relocation, e.g.

diff --git a/arch/mips/boot/start.S b/arch/mips/boot/start.S
index c1cd2d9dd5..95b6596dde 100644
--- a/arch/mips/boot/start.S
+++ b/arch/mips/boot/start.S
@@ -34,7 +34,7 @@ EXPORT(_start)
        move    a0, s0
        move    a1, s1
        move    a2, s2
-       la      v0, relocate_code
+       la      v0, main_entry
        jal     v0
         nop


Additional testing on real hardware has to be done.

-- 
Best regards,
  Antony Pavlov

_______________________________________________
barebox mailing list
barebox@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/barebox

Re: MIPS: relocation issue

[Oleksij Rempel]

Hi Antony,

thank you for your bug report. Can you please test my patch "[PATCH v1] MIPS: remove .bss 
to __rel_start overlay" with MMU enabled.

If it works, then we need to decide what should be done with MMU dependency.

On 28.01.20 07:46, Antony Pavlov wrote:
> Hi!
> 
> Since 28ed10d6a6 ("MIPS: relocation: add relocation support")
> iomem output on memory regions is invalid,
> all the code that rely on registered sdram memory regions
> after relocation can fail. E.g. the memtest commands fails
> on barebox.
> 
> How to reproduce the problem ov v2020.01.0:
> 
> $ wget https://www.barebox.org/download/barebox-2020.01.0.tar.bz2
> ...
> $ tar fx barebox-2020.01.0.tar.bz2
> $ cd barebox-2020.01.0
> $ export ARCH=mips
> $ export CROSS_COMPILE=mips-linux-gnu-
> $ make qemu-malta_defconfig
> $ sed -i "s/^.*CONFIG_CMD_MEMTEST.*/CONFIG_CMD_MEMTEST=y/" .config
> $ make
> $ qemu-system-mips -nodefaults -M malta -m 256 -nographic -serial stdio -monitor null -bios barebox-flash-image
> ...
> barebox@qemu malta:/ memtest
> Start iteration 1 of 1.
> Testing memory space: 0xa0865000 -> 0xafffffff:
> Starting data line test.
> Check for address bits stuck high.
> Check for address bits stuck low or shorted.
> Starting moving inversions test of RAM:
> Fill with address, compare, fill with inverted address, compare again
> 	[##########################################                       ]
> FAILURE (read/write): expected 0x03cca9f2, actual 0xa08126c4 at address 0xafb8f7c4.
> 
> Ooops, TLB miss on load or ifetch!
> 
> $ 0   : 00000000 00000000 fffffffb a085c900
> $ 4   : a0407e90 0000000a b80003f8 00000001
> $ 8   : a0858f28 00000000 00000053 00000008
> $12   : 00000000 00000000 0000005c 00000001
> $16   : 03ccab3c 03ccab3d a0825c48 00000000
> $20   : 00000000 00000001 a0853518 a085352c
> $24   : 00000018 a08124b8
> $28   : 00000000 8fb8fcf8 a0860000 03ccab3e
> Hi    : 0000000a
> Lo    : 00000000
> epc   : 03ccab3e
> ra    : 03ccab3e
> Status: 00000002
> Cause : 00000408
> Config: 80008482
> 
> ### ERROR ### Please RESET the board ###
> 
> 
> memtest for qemu-malta can be repaired by disabling relocation, e.g.
> 
> diff --git a/arch/mips/boot/start.S b/arch/mips/boot/start.S
> index c1cd2d9dd5..95b6596dde 100644
> --- a/arch/mips/boot/start.S
> +++ b/arch/mips/boot/start.S
> @@ -34,7 +34,7 @@ EXPORT(_start)
>          move    a0, s0
>          move    a1, s1
>          move    a2, s2
> -       la      v0, relocate_code
> +       la      v0, main_entry
>          jal     v0
>           nop
> 
> 
> Additional testing on real hardware has to be done.
> 

Kind regards,
Oleksij Rempel

-- 
Pengutronix e.K.                           |                             |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |
Peiner Str. 6-8, 31137 Hildesheim, Germany | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |

_______________________________________________
barebox mailing list
barebox@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/barebox

RE: MIPS: relocation issue

[Peter Mamonov]

Hi Oleksij,

Below you will find the patch, which adds the `whereami` command. The command
prints several addresses belonging to code, data, bss, heap and stack areas.
I've run it on qemu/malta with relocations enabled and disabled as per Antony's
mail. Output comparison shows that only the stack is relocated. Looks like 
relocation doesn't work as expected. Also note that the stack was relocated 
from the uncached KSEG1 to the cached KSEG0 segment, which is inconsistent at 
least.

Relocation enabled:

	barebox@qemu malta:/ whereami
	code   @ a0829bcc (a0829bc0)
	data   @ a085f3c0
	bss    @ a0861ae0
	heap   @ a0428c38
	stack  @ 8fb8fd50
	barebox@qemu malta:/ version

	barebox 2020.01.0-00104-gfe12ba2776 #3 Tue Jan 28 13:34:06 MSK 2020


Relocation disabled:

	barebox@qemu malta:/ whereami
	code   @ a0829bcc (a0829bc0)
	data   @ a085f1e0
	bss    @ a0861900
	heap   @ a0428c38
	stack  @ a03ffd50

Regards,
Peter
---
 commands/Kconfig    |  5 +++++
 commands/Makefile   |  1 +
 commands/whereami.c | 45 +++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 51 insertions(+)
 create mode 100644 commands/whereami.c

diff --git a/commands/Kconfig b/commands/Kconfig
index add0d26ee9..a660e09db4 100644
--- a/commands/Kconfig
+++ b/commands/Kconfig
@@ -2217,6 +2217,11 @@ config CMD_UBSAN
 	  This is a test command for the undefined behavior sanitizer.
 	  It triggers various undefined behavior, and detect it.
 
+config CMD_WHEREAMI
+	tristate "whereami"
+	depends on MIPS
+	default y
+
 # end Miscellaneous commands
 endmenu
 
diff --git a/commands/Makefile b/commands/Makefile
index 8e8c449bc1..dfa3d6e447 100644
--- a/commands/Makefile
+++ b/commands/Makefile
@@ -128,6 +128,7 @@ obj-$(CONFIG_CMD_NAND_BITFLIP)	+= nand-bitflip.o
 obj-$(CONFIG_CMD_SEED)		+= seed.o
 obj-$(CONFIG_CMD_IP_ROUTE_GET)  += ip-route-get.o
 obj-$(CONFIG_CMD_UBSAN)		+= ubsan.o
+obj-$(CONFIG_CMD_WHEREAMI)	+= whereami.o
 
 UBSAN_SANITIZE_ubsan.o := y
 
diff --git a/commands/whereami.c b/commands/whereami.c
new file mode 100644
index 0000000000..3edb070fe8
--- /dev/null
+++ b/commands/whereami.c
@@ -0,0 +1,45 @@
+#include <common.h>
+#include <command.h>
+#include <complete.h>
+
+static volatile int data = 1;
+static volatile int bss;
+
+static void* code();
+static noinline void* code()
+{
+	void *ret;
+	__asm__ __volatile__(
+		".set	push\n"
+		".set	noreorder\n"
+		".set	noat\n"
+		"	move	$1, $31\n"
+		"	bal	255f\n"
+		"	nop\n"
+		"255:	move	%0, $31\n"
+		"	move	$31, $1\n"
+		".set	pop\n"
+		:"=r" (ret)
+	);
+	return ret;
+}
+
+static int do_whereami(int argc, char *argv[])
+{
+	volatile int stack;
+	int *heapp = malloc(sizeof(int));
+
+	printf("code   @ %p (%p)\n", code(), &code);
+	printf("data   @ %p\n", &data);
+	printf("bss    @ %p\n", &bss);
+	printf("heap   @ %p\n", heapp);
+	printf("stack  @ %p\n", &stack);
+	return 0;
+}
+
+BAREBOX_CMD_START(whereami)
+	.cmd		= do_whereami,
+	BAREBOX_CMD_DESC("print addresses from code, stack, heap and data sections")
+	BAREBOX_CMD_GROUP(CMD_GRP_INFO)
+	BAREBOX_CMD_COMPLETE(empty_complete)
+BAREBOX_CMD_END
-- 
2.24.0


_______________________________________________
barebox mailing list
barebox@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/barebox

Re: MIPS: relocation issue

[Oleksij Rempel]

On 28.01.20 11:49, Peter Mamonov wrote:
> Hi Oleksij,
> 
> Below you will find the patch, which adds the `whereami` command. The command
> prints several addresses belonging to code, data, bss, heap and stack areas.
> I've run it on qemu/malta with relocations enabled and disabled as per Antony's
> mail. Output comparison shows that only the stack is relocated. Looks like
> relocation doesn't work as expected. Also note that the stack was relocated
> from the uncached KSEG1 to the cached KSEG0 segment, which is inconsistent at
> least.
>

ACK. The patch "MIPS: remove .bss to __rel_start overlay" should fix it. Please test :)

> Relocation enabled:
> 
> 	barebox@qemu malta:/ whereami
> 	code   @ a0829bcc (a0829bc0)
> 	data   @ a085f3c0
> 	bss    @ a0861ae0
> 	heap   @ a0428c38
> 	stack  @ 8fb8fd50
> 	barebox@qemu malta:/ version
> 
> 	barebox 2020.01.0-00104-gfe12ba2776 #3 Tue Jan 28 13:34:06 MSK 2020
> 
> 
> Relocation disabled:
> 
> 	barebox@qemu malta:/ whereami
> 	code   @ a0829bcc (a0829bc0)
> 	data   @ a085f1e0
> 	bss    @ a0861900
> 	heap   @ a0428c38
> 	stack  @ a03ffd50

This functionality is available with command iomem.

> Regards,
> Peter
> ---
>   commands/Kconfig    |  5 +++++
>   commands/Makefile   |  1 +
>   commands/whereami.c | 45 +++++++++++++++++++++++++++++++++++++++++++++
>   3 files changed, 51 insertions(+)
>   create mode 100644 commands/whereami.c
> 
> diff --git a/commands/Kconfig b/commands/Kconfig
> index add0d26ee9..a660e09db4 100644
> --- a/commands/Kconfig
> +++ b/commands/Kconfig
> @@ -2217,6 +2217,11 @@ config CMD_UBSAN
>   	  This is a test command for the undefined behavior sanitizer.
>   	  It triggers various undefined behavior, and detect it.
>   
> +config CMD_WHEREAMI
> +	tristate "whereami"
> +	depends on MIPS
> +	default y
> +
>   # end Miscellaneous commands
>   endmenu
>   
> diff --git a/commands/Makefile b/commands/Makefile
> index 8e8c449bc1..dfa3d6e447 100644
> --- a/commands/Makefile
> +++ b/commands/Makefile
> @@ -128,6 +128,7 @@ obj-$(CONFIG_CMD_NAND_BITFLIP)	+= nand-bitflip.o
>   obj-$(CONFIG_CMD_SEED)		+= seed.o
>   obj-$(CONFIG_CMD_IP_ROUTE_GET)  += ip-route-get.o
>   obj-$(CONFIG_CMD_UBSAN)		+= ubsan.o
> +obj-$(CONFIG_CMD_WHEREAMI)	+= whereami.o
>   
>   UBSAN_SANITIZE_ubsan.o := y
>   
> diff --git a/commands/whereami.c b/commands/whereami.c
> new file mode 100644
> index 0000000000..3edb070fe8
> --- /dev/null
> +++ b/commands/whereami.c
> @@ -0,0 +1,45 @@
> +#include <common.h>
> +#include <command.h>
> +#include <complete.h>
> +
> +static volatile int data = 1;
> +static volatile int bss;
> +
> +static void* code();
> +static noinline void* code()
> +{
> +	void *ret;
> +	__asm__ __volatile__(
> +		".set	push\n"
> +		".set	noreorder\n"
> +		".set	noat\n"
> +		"	move	$1, $31\n"
> +		"	bal	255f\n"
> +		"	nop\n"
> +		"255:	move	%0, $31\n"
> +		"	move	$31, $1\n"
> +		".set	pop\n"
> +		:"=r" (ret)
> +	);
> +	return ret;
> +}
> +
> +static int do_whereami(int argc, char *argv[])
> +{
> +	volatile int stack;
> +	int *heapp = malloc(sizeof(int));
> +
> +	printf("code   @ %p (%p)\n", code(), &code);
> +	printf("data   @ %p\n", &data);
> +	printf("bss    @ %p\n", &bss);
> +	printf("heap   @ %p\n", heapp);
> +	printf("stack  @ %p\n", &stack);
> +	return 0;
> +}
> +
> +BAREBOX_CMD_START(whereami)
> +	.cmd		= do_whereami,
> +	BAREBOX_CMD_DESC("print addresses from code, stack, heap and data sections")
> +	BAREBOX_CMD_GROUP(CMD_GRP_INFO)
> +	BAREBOX_CMD_COMPLETE(empty_complete)
> +BAREBOX_CMD_END
> 

Kind regards,
Oleksij Rempel

-- 
Pengutronix e.K.                           |                             |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |
Peiner Str. 6-8, 31137 Hildesheim, Germany | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |

_______________________________________________
barebox mailing list
barebox@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/barebox

Re: MIPS: relocation issue

[Peter Mamonov]

On Tue, Jan 28, 2020 at 12:03:35PM +0100, Oleksij Rempel wrote:
> 
> 
> On 28.01.20 11:49, Peter Mamonov wrote:
> > Hi Oleksij,
> > 
> > Below you will find the patch, which adds the `whereami` command. The command
> > prints several addresses belonging to code, data, bss, heap and stack areas.
> > I've run it on qemu/malta with relocations enabled and disabled as per Antony's
> > mail. Output comparison shows that only the stack is relocated. Looks like
> > relocation doesn't work as expected. Also note that the stack was relocated
> > from the uncached KSEG1 to the cached KSEG0 segment, which is inconsistent at
> > least.
> > 
> 
> ACK. The patch "MIPS: remove .bss to __rel_start overlay" should fix it. Please test :)

Ok, will report back soon.

> 
> > Relocation enabled:
> > 
> > 	barebox@qemu malta:/ whereami
> > 	code   @ a0829bcc (a0829bc0)
> > 	data   @ a085f3c0
> > 	bss    @ a0861ae0
> > 	heap   @ a0428c38
> > 	stack  @ 8fb8fd50
> > 	barebox@qemu malta:/ version
> > 
> > 	barebox 2020.01.0-00104-gfe12ba2776 #3 Tue Jan 28 13:34:06 MSK 2020
> > 
> > 
> > Relocation disabled:
> > 
> > 	barebox@qemu malta:/ whereami
> > 	code   @ a0829bcc (a0829bc0)
> > 	data   @ a085f1e0
> > 	bss    @ a0861900
> > 	heap   @ a0428c38
> > 	stack  @ a03ffd50
> 
> This functionality is available with command iomem.

At the moment iomem's output is inconsistent in case of the stack at least:

barebox@qemu malta:/ iomem
...
  0xa0000000 - 0xafffffff (size 0x10000000) kseg1_ram0
    0xa03f9000 - 0xa03fffff (size 0x00007000) stack <<<<<<<<<<<<<<<<<<< !!!!
    0xa0400000 - 0xa07fffff (size 0x00400000) malloc space
    0xa0800000 - 0xa085ae9f (size 0x0005aea0) barebox
    0xa085aea0 - 0xa085fadf (size 0x00004c40) barebox data
    0xa085fae0 - 0xa0863bdf (size 0x00004100) bss
barebox@qemu malta:/ whereami
code   @ a0829bcc (a0829bc0)
data   @ a085f3c0
bss    @ a0861ae0
heap   @ a0429518
stack  @ 8fb8fd50  <<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<< !!!
barebox@qemu malta:/ 

Regards,
Peter

> > ---
> >   commands/Kconfig    |  5 +++++
> >   commands/Makefile   |  1 +
> >   commands/whereami.c | 45 +++++++++++++++++++++++++++++++++++++++++++++
> >   3 files changed, 51 insertions(+)
> >   create mode 100644 commands/whereami.c
> > 
> > diff --git a/commands/Kconfig b/commands/Kconfig
> > index add0d26ee9..a660e09db4 100644
> > --- a/commands/Kconfig
> > +++ b/commands/Kconfig
> > @@ -2217,6 +2217,11 @@ config CMD_UBSAN
> >   	  This is a test command for the undefined behavior sanitizer.
> >   	  It triggers various undefined behavior, and detect it.
> > +config CMD_WHEREAMI
> > +	tristate "whereami"
> > +	depends on MIPS
> > +	default y
> > +
> >   # end Miscellaneous commands
> >   endmenu
> > diff --git a/commands/Makefile b/commands/Makefile
> > index 8e8c449bc1..dfa3d6e447 100644
> > --- a/commands/Makefile
> > +++ b/commands/Makefile
> > @@ -128,6 +128,7 @@ obj-$(CONFIG_CMD_NAND_BITFLIP)	+= nand-bitflip.o
> >   obj-$(CONFIG_CMD_SEED)		+= seed.o
> >   obj-$(CONFIG_CMD_IP_ROUTE_GET)  += ip-route-get.o
> >   obj-$(CONFIG_CMD_UBSAN)		+= ubsan.o
> > +obj-$(CONFIG_CMD_WHEREAMI)	+= whereami.o
> >   UBSAN_SANITIZE_ubsan.o := y
> > diff --git a/commands/whereami.c b/commands/whereami.c
> > new file mode 100644
> > index 0000000000..3edb070fe8
> > --- /dev/null
> > +++ b/commands/whereami.c
> > @@ -0,0 +1,45 @@
> > +#include <common.h>
> > +#include <command.h>
> > +#include <complete.h>
> > +
> > +static volatile int data = 1;
> > +static volatile int bss;
> > +
> > +static void* code();
> > +static noinline void* code()
> > +{
> > +	void *ret;
> > +	__asm__ __volatile__(
> > +		".set	push\n"
> > +		".set	noreorder\n"
> > +		".set	noat\n"
> > +		"	move	$1, $31\n"
> > +		"	bal	255f\n"
> > +		"	nop\n"
> > +		"255:	move	%0, $31\n"
> > +		"	move	$31, $1\n"
> > +		".set	pop\n"
> > +		:"=r" (ret)
> > +	);
> > +	return ret;
> > +}
> > +
> > +static int do_whereami(int argc, char *argv[])
> > +{
> > +	volatile int stack;
> > +	int *heapp = malloc(sizeof(int));
> > +
> > +	printf("code   @ %p (%p)\n", code(), &code);
> > +	printf("data   @ %p\n", &data);
> > +	printf("bss    @ %p\n", &bss);
> > +	printf("heap   @ %p\n", heapp);
> > +	printf("stack  @ %p\n", &stack);
> > +	return 0;
> > +}
> > +
> > +BAREBOX_CMD_START(whereami)
> > +	.cmd		= do_whereami,
> > +	BAREBOX_CMD_DESC("print addresses from code, stack, heap and data sections")
> > +	BAREBOX_CMD_GROUP(CMD_GRP_INFO)
> > +	BAREBOX_CMD_COMPLETE(empty_complete)
> > +BAREBOX_CMD_END
> > 
> 
> Kind regards,
> Oleksij Rempel
> 
> -- 
> Pengutronix e.K.                           |                             |
> Industrial Linux Solutions                 | http://www.pengutronix.de/  |
> Peiner Str. 6-8, 31137 Hildesheim, Germany | Phone: +49-5121-206917-0    |
> Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |

_______________________________________________
barebox mailing list
barebox@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/barebox

maeling list issue

[Oleksij Rempel]

Hi all,

starting from 24.01.2020 i do not get barebox mails to my current mail address. Current impact and
reason are unknown. Do any one else is affected by this?

I included to cc at two devs.. since, affect won't be able to respond anyway... :)

I do not know if it is a good idea to ask not affected receivers to pong on this ping.

--
Regards,
Oleksij

_______________________________________________
barebox mailing list
barebox@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/barebox

Re: maeling list issue

[Sascha Hauer]

On Thu, Jan 30, 2020 at 09:26:17AM +0100, Oleksij Rempel wrote:
> Hi all,
> 
> starting from 24.01.2020 i do not get barebox mails to my current mail address. Current impact and
> reason are unknown. Do any one else is affected by this?
> 
> I included to cc at two devs.. since, affect won't be able to respond anyway... :)
> 
> I do not know if it is a good idea to ask not affected receivers to pong on this ping.

Your status was set to "nomail" due to exceesive bounces. I enabled
mails for you again.

Sascha


-- 
Pengutronix e.K.                           |                             |
Steuerwalder Str. 21                       | http://www.pengutronix.de/  |
31137 Hildesheim, Germany                  | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |

_______________________________________________
barebox mailing list
barebox@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/barebox