anyone using Itanium B-step support?
The kernel still has an ITANIUM_BSTEP_SPECIFIC configure option, as of 2.6.8.1, which enables use of the gcc -mb-step option. As far as I know, this is the only user of the gcc/gas B-step support. I am skeptical that this support still works, and suggest that it be removed, so that I can remove the gcc/gas B-step support.
The reason why this comes up now is because gas has support to emit warnings in cases where extra nops must be emitted to avoid B-step errata. This is enabled by default. At the time, this was useful to ensure backwards compatibility, but now all it does is confuse end users who have no idea what an Itanium B-step processor is. To fix this, I need to add an option to enable it, which means I need to fix gcc to pass the option to gas, and I also need to fix the kernel to add the option to aflags when CONFIG_ITANIUM_BSTEP_SPECIFIC.
However, rather than add more ITANIUM_BSTEP_SPECIFIC support to the kernel, I think it makes more sense to obsolete it and remove it.
Signature · 1 line
Jim Wilson, GNU Tools Support, http://www.SpecifixInc.com
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · James E Wilson
Jim> However, rather than add more ITANIUM_BSTEP_SPECIFIC support to Jim> the kernel, I think it makes more sense to obsolete it and Jim> remove it.
I agree. I happened to run into the warning just recently myself and came to the same conclusion.
We probably still have some B-step Merceds in our lab but IIRC, it was so rare to trigger the erratum that even on those boxes we can live without the workaround just fine. And it's not likely that anyone is using such old prototype hardware for anything mission critical...
--david
Quoted 2 lines · James E Wilson
However, rather than add more
ITANIUM_BSTEP_SPECIFICsupport to the kernel, I think it makes more sense to obsolete it and remove it.
Here is the change I am suggesting.
*** Kconfig.orig Sat Aug 14 03:56:22 2004
--- Kconfig Thu Jan 27 17:16:51 2005
*************** config IA64_BRL_EMU
*** 134,147 ****
depends on ITANIUM
default y
- config ITANIUM_BSTEP_SPECIFIC
- bool "Itanium B-step specific code"
- depends on ITANIUM
- help
- Select this option to build a kernel for an Itanium prototype system - with a B-step CPU. You have a B-step CPU if the "revision" field in - /proc/cpuinfo has a value in the range from 1 to 4. - # align cache-sensitive data to 128 bytes config IA64_L1_CACHE_SHIFT int --- 134,139 ---- *** Makefile.orig Sat Aug 14 03:54:48 2004 --- Makefile Thu Jan 27 17:17:03 2005 *************** ifeq ($(GCC_VERSION),3) *** 49,56 **** endif endif
- cflags-$(CONFIG_ITANIUM_BSTEP_SPECIFIC) += -mb-step
-
CFLAGS += $(cflags-y)
head-y := arch/ia64/kernel/head.o arch/ia64/kernel/init_task.o
--- 49,54 ----
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · James E Wilson
Quoted 3 lines · James E Wilson
Quoted 3 lines
However, rather than add more
ITANIUM_BSTEP_SPECIFICsupport to the kernel, I think it makes more sense to obsolete it and remove it.
James> Here is the change I am suggesting.
Looks fine to me. I found one more reference to CONFIG_ITANIUM_BSTEP_SPECIFIC in arch/ia64/configs/bigsur_defconfig. Not that it really matters, but from a cleanliness perspective, we may want to get rid of that at the same time.
I would have done it myself but there was no "Signed-off-by" trailer in your patch so I think you may need to resend the patch anyhow before Tony will take it (unless he judges the patch to be trivial enough not to need this).
--david
Quoted 2 lines · David Mosberger
Looks fine to me. I found one more reference to
CONFIG_ITANIUM_BSTEP_SPECIFICinarch/ia64/configs/bigsur_defconfig.
I see no such file in my copy of linux-2.6.8.1 downloaded from kernels.org, but yes, that should be fixed too.
Quoted 2 lines
I would have done it myself but there was no "Signed-off-by" trailer in your patch so I think you may need to resend the patch anyhow
I know nothing about submitting linux kernel patches. My patch is completely untested, by the way, as I doubt that I can build a working kernel without help. It was more a suggestion than an actual patch.
Quoted 5 lines
Quoted 5 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line · James E Wilson
I would have done it myself but there was no "Signed-off-by" trailer in your patch so I think you may need to resend the patch anyhow
James> I know nothing about submitting linux kernel patches. My James> patch is completely untested, by the way, as I doubt that I James> can build a working kernel without help. It was more a James> suggestion than an actual patch.
Ah, OK, in that case, here is a (compile-)tested patch.
Tony, please apply.
--david
ia64: Remove Merced B-step support
Remove left-over support for Merced B-step CPUs as suggested by Jim Wilson.
=== arch/ia64/Kconfig 1.85 vs edited ==--- 1.85/arch/ia64/Kconfig 2005-01-24 13:41:13 -08:00
+++ edited/arch/ia64/Kconfig 2005-01-27 22:05:34 -08:00
@@ -158,14 +158,6 @@
depends on ITANIUM
default y
-config ITANIUM_BSTEP_SPECIFIC
- bool "Itanium B-step specific code"
- depends on ITANIUM
- help
- Select this option to build a kernel for an Itanium prototype system - with a B-step CPU. You have a B-step CPU if the "revision" field in - /proc/cpuinfo has a value in the range from 1 to 4. - # align cache-sensitive data to 128 bytes config IA64_L1_CACHE_SHIFT int === arch/ia64/Makefile 1.74 vs edited ==--- 1.74/arch/ia64/Makefile 2005-01-14 15:52:52 -08:00 +++ edited/arch/ia64/Makefile 2005-01-27 22:05:43 -08:00 @@ -46,8 +46,6 @@ cflags-$(CONFIG_MCKINLEY) += -mtune=mckinley endif
-cflags-$(CONFIG_ITANIUM_BSTEP_SPECIFIC) += -mb-step
-
CFLAGS += $(cflags-y)
head-y := arch/ia64/kernel/head.o arch/ia64/kernel/init_task.o
=== arch/ia64/configs/bigsur_defconfig 1.2 vs edited ==--- 1.2/arch/ia64/configs/bigsur_defconfig 2004-11-30 08:48:16 -08:00
+++ edited/arch/ia64/configs/bigsur_defconfig 2005-01-27 22:05:20 -08:00
@@ -73,7 +73,6 @@
CONFIG_IA64_PAGE_SIZE_16KB=y
# CONFIG_IA64_PAGE_SIZE_64KB is not set
CONFIG_IA64_BRL_EMU=y
-# CONFIG_ITANIUM_BSTEP_SPECIFIC is not set
CONFIG_IA64_L1_CACHE_SHIFT=6
# CONFIG_NUMA is not set
# CONFIG_VIRTUAL_MEM_MAP is not set