On Thu, Feb 07, 2019 at 11:47:59AM +0000, Dave Martin wrote:

On Wed, Feb 06, 2019 at 01:31:05PM +0000, Andrew Murray wrote:

quoted

The introduction of elf_hwcap2 introduced accessors which ensure that
features are set/tested in the appropriate elf_hwcapX variable.

Let's now mandate access to elf_hwcapX via these accessors by making
elf_hwcapX static within cpufeature.c.

Signed-off-by: Andrew Murray <redacted>
---
 arch/arm64/include/asm/cpufeature.h | 19 ++++---------------
 arch/arm64/include/asm/hwcap.h      |  6 +++---
 arch/arm64/kernel/cpufeature.c      | 33 ++++++++++++++++++++++++++++-----
 3 files changed, 35 insertions(+), 23 deletions(-)

[...]

quoted

diff --git a/arch/arm64/include/asm/hwcap.h b/arch/arm64/include/asm/hwcap.h
index 05ee9b9..ced87ad 100644
--- a/arch/arm64/include/asm/hwcap.h
+++ b/arch/arm64/include/asm/hwcap.h

[...]

quoted

@@ -97,6 +98,5 @@ enum {
 #endif
 };
 
-extern unsigned long elf_hwcap, elf_hwcap2;
 #endif
 #endif

diff --git a/arch/arm64/kernel/cpufeature.c b/arch/arm64/kernel/cpufeature.c
index d10a455..5b9620d 100644
--- a/arch/arm64/kernel/cpufeature.c
+++ b/arch/arm64/kernel/cpufeature.c

@@ -35,11 +35,8 @@
 #include <asm/traps.h>
 #include <asm/virt.h>
 
-unsigned long elf_hwcap __read_mostly;
-EXPORT_SYMBOL_GPL(elf_hwcap);
-

Will this affect out-of-tree modules?

I also note an out-of-arch/ use in
drivers/clocksource/arm_arch_timer.c, which this series doesn't appear
to address (or I missed it).

It was in "arm64: HWCAP: add support for ELF_HWCAP2" - though due to it
being used by arm32 code I had to use #ifdef CONFIG_ARM64 as
cpu_set_feature doesn't exist there.

We are allowed to break EXPORT_SYMBOL_GPL()s in general so long as it's
not done without any meaningful reason, so maybe this is not a huge
concern so long as we catch all the in-tree users.

The benefit of removing the EXPORT_SYMBOL_GPL is that we can encapsulate
the elf_hwcap variables with getters/setters that ensure they are modified
correctly. This is only relevant now as we split bits between elf_hwcap and
elf_hwcap2.

My suggestion is to remove the EXPORT_SYMBOL_GPL and then also stick with
using just one elf_hwcap variable. As you previously suggested we can then
update the ELF_HWCAP{,2} marcos to produce the relevant bit fields. This
also has the added benefit of simplifying cpu_{have,set}_feature functions
as the number of hwcaps grow.

Thanks,

Andrew Murray

[...]

Cheers
---Dave

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel