This is the mail archive of the
gdb-patches@sourceware.org
mailing list for the GDB project.
Re: MSP430 SIM: Add support for hardware multiply
- From: Nick Clifton <nickc at redhat dot com>
- To: vapier at gentoo dot org
- Cc: gdb-patches at sourceware dot org, binutils at sourceware dot org
- Date: Tue, 5 Jan 2016 16:43:34 +0000
- Subject: Re: MSP430 SIM: Add support for hardware multiply
- Authentication-results: sourceware.org; auth=none
- References: <87zjislnam dot fsf at redhat dot com> <20151225002204 dot GQ25803 at vapier dot lan>
Hi Mike,
I am applying the attached patch to add support for simulating the
hardware multiply feature found on some MSP430 MCUs.
this broke the testsuite:
FAIL: msp430 add.s (execution)
Actually, as far as I can see this test never should have worked in the
first place. The problem is that the test harness code (in
testutils.inc) assumes that the VMA and LMA addresses of the .data
section are the same. This is not the case (at least for the msp430),
and without runtime code to copy data from the LMA address to the VMA
address all the data addresses in the test executables are wrong.
So, I am checking in the following two patches. The first patch updates
the linker's built in link script so that it does not place data into
the msp430's hardware address range, and so that that it provides a
symbol (__romdatastart) with the LMA address of the .data section.
[Note - as far as I can see only the sim tests actually use the linker's
built in linker script. Everyone else uses scripts explicitly provided
on the linker command line].
The second patch updates the testutils.inc file so that it will use the
LMA addresses of the strings in the .data section not the VMA addresses.
This makes the tests work, although it is a bit of a hack. A proper
solution would be to provide start-up code that copies the data from the
LMA address to the VMA address before the test proper runs, but this
seems like overkill to me.
Cheers
Nick
ld/ChangeLog
2016-01-05 Nick Clifton <nickc@redhat.com>
* emulparams/msp430elf.sh (RAM_START): Move to 0x500 - above the
MSP430 hardware multiply address range.
* scripttempl/elf32msp430.sc (__romdatastart): Define.
(__romdatacopysize): Define.
* scripttempl/elf32msp430_3.sc: Likewise.
sim/testsuite/sim/msp430/ChangeLog
2016-01-05 Nick Clifton <nickc@redhat.com>
* testutils.inc (__pass): Use the LMA addresses of the _passmsg
symbol.
(__fail): Likewise.
diff --git a/sim/testsuite/sim/msp430/testutils.inc b/sim/testsuite/sim/msp430/testutils.inc
index 6c540b1..1ddef23 100644
--- a/sim/testsuite/sim/msp430/testutils.inc
+++ b/sim/testsuite/sim/msp430/testutils.inc
@@ -9,13 +9,43 @@
.global __pass
.type __pass, function
__pass:
- write 1, _passmsg, 5
+ # Note - we cannot just invoke:
+ #
+ # write 1, _passmsg, 5
+ #
+ # here because _passmsg contains the run-time (VMA) address of
+ # the pass string (probably 0x500) not the load-time (LMA)
+ # address (probably 0x804c). Normally using the VMA address
+ # would be the correct thing to do - *if* there was some start
+ # up code which copied data from LMA to VMA. But we have no
+ # start up code, so the data still resides at the LMA
+ # address. Hence we use __romdatastart instead.
+ #
+ # Note - we are cheating because the address that we pass to
+ # "write" should actually be:
+ #
+ # __romdatastart + (_passmsg - __datastart)
+ #
+ # but the assembler cannot cope with this expression. So we
+ # cheat and use the fact that we know that _passmsg is the
+ # first string in the .data section and so (_passmsg -
+ # __datastart) evaluates to zero.
+
+ write 1, __romdatastart, 5
exit 0
.global __fail
.type __fail, function
__fail:
- write 1, _failmsg, 5
+ # Note - see above.
+ #
+ # write 1, _failmsg, 5
+ #
+ # This time we use the fact that _passmsg is aligned to a
+ # 16 byte boundary to work out that (_failmsg - __datastart)
+ # evaluates to 0x10.
+
+ write 1, __romdatastart + 0x10, 5
exit 1
.data
diff --git a/ld/emulparams/msp430elf.sh b/ld/emulparams/msp430elf.sh
index a76e9ee..e9d0237 100644
--- a/ld/emulparams/msp430elf.sh
+++ b/ld/emulparams/msp430elf.sh
@@ -13,6 +13,6 @@ EMBEDDED=yes
ARCH=msp:14
ROM_START=0x8000
ROM_SIZE=0x7fe0
-RAM_START=0x0200
+RAM_START=0x0500
RAM_SIZE=1K
STACK=0x600
diff --git a/ld/scripttempl/elf32msp430.sc b/ld/scripttempl/elf32msp430.sc
index 78c7c12..50b7ddb 100644
--- a/ld/scripttempl/elf32msp430.sc
+++ b/ld/scripttempl/elf32msp430.sc
@@ -269,6 +269,9 @@ SECTIONS
${RELOCATING+ _edata = . ; }
} ${RELOCATING+ > data ${RELOCATING+AT> text}}
+ __romdatastart = LOADADDR(.data);
+ __romdatacopysize = SIZEOF(.data);
+
.bss ${RELOCATING+ SIZEOF(.data) + ADDR(.data)} :
{
${RELOCATING+. = ALIGN(2);}
diff --git a/ld/scripttempl/elf32msp430_3.sc b/ld/scripttempl/elf32msp430_3.sc
index 7a13081..7ad04e1 100644
--- a/ld/scripttempl/elf32msp430_3.sc
+++ b/ld/scripttempl/elf32msp430_3.sc
@@ -147,6 +147,9 @@ SECTIONS
${RELOCATING+ _edata = . ; }
} ${RELOCATING+ > data ${RELOCATING+AT> text}}
+ __romdatastart = LOADADDR(.data);
+ __romdatacopysize = SIZEOF(.data);
+
.bss ${RELOCATING+ SIZEOF(.data) + ADDR(.data)} :
{
${RELOCATING+. = ALIGN(2);}