| Message ID | 1343683803-15507-1-git-send-email-sboyd@codeaurora.org (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Mon, Jul 30, 2012 at 02:30:03PM -0700, Stephen Boyd wrote: > Add the ARM machine identifier to sortextable and select the > config option so that we can sort the exception table at compile > time. sortextable relies on a section named __ex_table existing > in the vmlinux, but ARM's linker script places the exception > table in the data section. Give the exception table its own > section so that sortextable can find it. > > This allows us to skip the runtime sorting step during boot. > > Signed-off-by: Stephen Boyd <sboyd@codeaurora.org> > Cc: David Daney <david.daney@cavium.com> > --- > > I can't find any information on why the exception table lives in the data > section. If there's a good reason for that, I'll look into changing > sortextable to look for the __start___ex_table symbol. > > arch/arm/Kconfig | 1 + > arch/arm/kernel/vmlinux.lds.S | 19 +++++++++---------- > scripts/sortextable.c | 1 + > 3 files changed, 11 insertions(+), 10 deletions(-) > > diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig > index b25c9d3..2af95e6 100644 > --- a/arch/arm/Kconfig > +++ b/arch/arm/Kconfig > @@ -1,6 +1,7 @@ > config ARM > bool > default y > + select BUILDTIME_EXTABLE_SORT if MMU > select ARCH_HAVE_CUSTOM_GPIO_H > select HAVE_AOUT > select HAVE_DMA_API_DEBUG > diff --git a/arch/arm/kernel/vmlinux.lds.S b/arch/arm/kernel/vmlinux.lds.S > index 36ff15b..0e3e8b4 100644 > --- a/arch/arm/kernel/vmlinux.lds.S > +++ b/arch/arm/kernel/vmlinux.lds.S > @@ -220,16 +220,6 @@ SECTIONS > READ_MOSTLY_DATA(L1_CACHE_BYTES) > > /* > - * The exception fixup table (might need resorting at runtime) > - */ > - . = ALIGN(4); > - __start___ex_table = .; > -#ifdef CONFIG_MMU > - *(__ex_table) > -#endif > - __stop___ex_table = .; > - > - /* > * and the usual data section > */ > DATA_DATA > @@ -239,6 +229,15 @@ SECTIONS > } > _edata_loc = __data_loc + SIZEOF(.data); > > + . = ALIGN(4); > + __ex_table : AT(ADDR(__ex_table) - LOAD_OFFSET) { > + __start___ex_table = .; > +#ifdef CONFIG_MMU > + *(__ex_table) > +#endif > + __stop___ex_table = .; > + } > + Here you should be able to use the macro EXCEPTION_TABLE from vmlinux.lds.h. I cannot see why the ifdef for not NON-MMU case is needed, but if it is needed the macro is not good... Sam
On Mon, Jul 30, 2012 at 02:30:03PM -0700, Stephen Boyd wrote: > Add the ARM machine identifier to sortextable and select the > config option so that we can sort the exception table at compile > time. sortextable relies on a section named __ex_table existing > in the vmlinux, but ARM's linker script places the exception > table in the data section. Give the exception table its own > section so that sortextable can find it. > > This allows us to skip the runtime sorting step during boot. > > Signed-off-by: Stephen Boyd <sboyd@codeaurora.org> > Cc: David Daney <david.daney@cavium.com> > --- > > I can't find any information on why the exception table lives in the data > section. If there's a good reason for that, I'll look into changing > sortextable to look for the __start___ex_table symbol. Be careful about the placement of this, especially with XIP.
On Mon, Jul 30, 2012 at 11:51:18PM +0200, Sam Ravnborg wrote: > Here you should be able to use the macro EXCEPTION_TABLE from > vmlinux.lds.h. > I cannot see why the ifdef for not NON-MMU case is needed, > but if it is needed the macro is not good... Because you can not mention the same input section in two different places and end up with predictable output from the linker. We discard the __ex_table for noMMU, but I think the start/stop symbols are still referenced somewhere. Dunno, I don't have much to do with noMMU ARM, and the only platform I'd be interested in never got merged.
On 07/30/12 14:56, Russell King - ARM Linux wrote: > On Mon, Jul 30, 2012 at 02:30:03PM -0700, Stephen Boyd wrote: >> Add the ARM machine identifier to sortextable and select the >> config option so that we can sort the exception table at compile >> time. sortextable relies on a section named __ex_table existing >> in the vmlinux, but ARM's linker script places the exception >> table in the data section. Give the exception table its own >> section so that sortextable can find it. >> >> This allows us to skip the runtime sorting step during boot. >> >> Signed-off-by: Stephen Boyd <sboyd@codeaurora.org> >> Cc: David Daney <david.daney@cavium.com> >> --- >> >> I can't find any information on why the exception table lives in the data >> section. If there's a good reason for that, I'll look into changing >> sortextable to look for the __start___ex_table symbol. > Be careful about the placement of this, especially with XIP. > Thanks for the hint. I'm unfamiliar with how XIP works so I'll take a closer look there.
On 07/30/12 15:19, Russell King - ARM Linux wrote: > On Mon, Jul 30, 2012 at 11:51:18PM +0200, Sam Ravnborg wrote: >> Here you should be able to use the macro EXCEPTION_TABLE from >> vmlinux.lds.h. >> I cannot see why the ifdef for not NON-MMU case is needed, >> but if it is needed the macro is not good... > Because you can not mention the same input section in two different > places and end up with predictable output from the linker. > > We discard the __ex_table for noMMU, but I think the start/stop > symbols are still referenced somewhere. Dunno, I don't have much to > do with noMMU ARM, and the only platform I'd be interested in never > got merged. I was thinking, perhaps we can ifdef out the exception fixup sections in the places where they're added? Then we can just use the EXCEPTION_TABLE macro from vmlinux.lds.h knowing that there are no __ex_table sections in the input object files?
On 07/30/12 15:38, Stephen Boyd wrote: > On 07/30/12 14:56, Russell King - ARM Linux wrote: >> On Mon, Jul 30, 2012 at 02:30:03PM -0700, Stephen Boyd wrote: >>> Add the ARM machine identifier to sortextable and select the >>> config option so that we can sort the exception table at compile >>> time. sortextable relies on a section named __ex_table existing >>> in the vmlinux, but ARM's linker script places the exception >>> table in the data section. Give the exception table its own >>> section so that sortextable can find it. >>> >>> This allows us to skip the runtime sorting step during boot. >>> >>> Signed-off-by: Stephen Boyd <sboyd@codeaurora.org> >>> Cc: David Daney <david.daney@cavium.com> >>> --- >>> >>> I can't find any information on why the exception table lives in the data >>> section. If there's a good reason for that, I'll look into changing >>> sortextable to look for the __start___ex_table symbol. >> Be careful about the placement of this, especially with XIP. >> > Thanks for the hint. I'm unfamiliar with how XIP works so I'll take a > closer look there. > Ok. It looks like the exception table is placed in the data section so that XIP kernels can boot up and sort the exception table (otherwise the exception table would be in read-only flash). Now that the exception table is sorted at compile time, we should be able to place the exception table in the rodata area, essentially placing the table in the non-volatile storage on XIP kernels. I propose we move the exception table right after the rodata. Other arches (x86/mips) may actually want to put the exception table into the rodata, but that looks like a larger change.
diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig index b25c9d3..2af95e6 100644 --- a/arch/arm/Kconfig +++ b/arch/arm/Kconfig @@ -1,6 +1,7 @@ config ARM bool default y + select BUILDTIME_EXTABLE_SORT if MMU select ARCH_HAVE_CUSTOM_GPIO_H select HAVE_AOUT select HAVE_DMA_API_DEBUG diff --git a/arch/arm/kernel/vmlinux.lds.S b/arch/arm/kernel/vmlinux.lds.S index 36ff15b..0e3e8b4 100644 --- a/arch/arm/kernel/vmlinux.lds.S +++ b/arch/arm/kernel/vmlinux.lds.S @@ -220,16 +220,6 @@ SECTIONS READ_MOSTLY_DATA(L1_CACHE_BYTES) /* - * The exception fixup table (might need resorting at runtime) - */ - . = ALIGN(4); - __start___ex_table = .; -#ifdef CONFIG_MMU - *(__ex_table) -#endif - __stop___ex_table = .; - - /* * and the usual data section */ DATA_DATA @@ -239,6 +229,15 @@ SECTIONS } _edata_loc = __data_loc + SIZEOF(.data); + . = ALIGN(4); + __ex_table : AT(ADDR(__ex_table) - LOAD_OFFSET) { + __start___ex_table = .; +#ifdef CONFIG_MMU + *(__ex_table) +#endif + __stop___ex_table = .; + } + #ifdef CONFIG_HAVE_TCM /* * We align everything to a page boundary so we can diff --git a/scripts/sortextable.c b/scripts/sortextable.c index 1ca9ceb..591acb3 100644 --- a/scripts/sortextable.c +++ b/scripts/sortextable.c @@ -248,6 +248,7 @@ do_file(char const *const fname) custom_sort = sort_x86_table; break; case EM_MIPS: + case EM_ARM: break; } /* end switch */
Add the ARM machine identifier to sortextable and select the config option so that we can sort the exception table at compile time. sortextable relies on a section named __ex_table existing in the vmlinux, but ARM's linker script places the exception table in the data section. Give the exception table its own section so that sortextable can find it. This allows us to skip the runtime sorting step during boot. Signed-off-by: Stephen Boyd <sboyd@codeaurora.org> Cc: David Daney <david.daney@cavium.com> --- I can't find any information on why the exception table lives in the data section. If there's a good reason for that, I'll look into changing sortextable to look for the __start___ex_table symbol. arch/arm/Kconfig | 1 + arch/arm/kernel/vmlinux.lds.S | 19 +++++++++---------- scripts/sortextable.c | 1 + 3 files changed, 11 insertions(+), 10 deletions(-)