| Message ID | 20210302090118.30666-3-ardb@kernel.org |
|---|---|
| State | Accepted |
| Commit | 4c4dcd3541f83d216f2e403cb83dd431e09759b1 |
| Headers | show |
| Series | running kernel mode SIMD with softirqs disabled | expand |
On Tue, Mar 02, 2021 at 10:01:11AM +0100, Ard Biesheuvel wrote: > The AArch64 asm syntax has this slightly tedious property that the names > used in mnemonics to refer to registers depend on whether the opcode in > question targets the entire 64-bits (xN), or only the least significant > 8, 16 or 32 bits (wN). When writing parameterized code such as macros, > this can be annoying, as macro arguments don't lend themselves to > indexed lookups, and so generating a reference to wN in a macro that > receives xN as an argument is problematic. > > For instance, an upcoming patch that modifies the implementation of the > cond_yield macro to be able to refer to 32-bit registers would need to > modify invocations such as > > cond_yield 3f, x8 > > to > > cond_yield 3f, 8 > > so that the second argument can be token pasted after x or w to emit the > correct register reference. Unfortunately, this interferes with the self > documenting nature of the first example, where the second argument is > obviously a register, whereas in the second example, one would need to > go and look at the code to find out what '8' means. > > So let's fix this by defining wxN aliases for all xN registers, which > resolve to the 32-bit alias of each respective 64-bit register. This > allows the macro implementation to paste the xN reference after a w to > obtain the correct register name. > > Signed-off-by: Ard Biesheuvel <ardb@kernel.org> > --- > arch/arm64/include/asm/assembler.h | 8 ++++++++ > 1 file changed, 8 insertions(+) > > diff --git a/arch/arm64/include/asm/assembler.h b/arch/arm64/include/asm/assembler.h > index e0fc1d424f9b..7b076ccd1a54 100644 > --- a/arch/arm64/include/asm/assembler.h > +++ b/arch/arm64/include/asm/assembler.h > @@ -23,6 +23,14 @@ > #include <asm/ptrace.h> > #include <asm/thread_info.h> > > + /* > + * Provide a wxN alias for each wN register so what we can paste a xN > + * reference after a 'w' to obtain the 32-bit version. > + */ > + .irp n,0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30 > + wx\n .req w\n > + .endr That's a pretty neat hack! I remember seeing code elsewhere which would benefit from this, so might be worth a look at our other macros as I'm sure I got annoyed by one the other day... ah yes, the SVE macros in fpsimdmacros.h Acked-by: Will Deacon <will@kernel.org> Will
diff --git a/arch/arm64/include/asm/assembler.h b/arch/arm64/include/asm/assembler.h index e0fc1d424f9b..7b076ccd1a54 100644 --- a/arch/arm64/include/asm/assembler.h +++ b/arch/arm64/include/asm/assembler.h @@ -23,6 +23,14 @@ #include <asm/ptrace.h> #include <asm/thread_info.h> + /* + * Provide a wxN alias for each wN register so what we can paste a xN + * reference after a 'w' to obtain the 32-bit version. + */ + .irp n,0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30 + wx\n .req w\n + .endr + .macro save_and_disable_daif, flags mrs \flags, daif msr daifset, #0xf
The AArch64 asm syntax has this slightly tedious property that the names used in mnemonics to refer to registers depend on whether the opcode in question targets the entire 64-bits (xN), or only the least significant 8, 16 or 32 bits (wN). When writing parameterized code such as macros, this can be annoying, as macro arguments don't lend themselves to indexed lookups, and so generating a reference to wN in a macro that receives xN as an argument is problematic. For instance, an upcoming patch that modifies the implementation of the cond_yield macro to be able to refer to 32-bit registers would need to modify invocations such as cond_yield 3f, x8 to cond_yield 3f, 8 so that the second argument can be token pasted after x or w to emit the correct register reference. Unfortunately, this interferes with the self documenting nature of the first example, where the second argument is obviously a register, whereas in the second example, one would need to go and look at the code to find out what '8' means. So let's fix this by defining wxN aliases for all xN registers, which resolve to the 32-bit alias of each respective 64-bit register. This allows the macro implementation to paste the xN reference after a w to obtain the correct register name. Signed-off-by: Ard Biesheuvel <ardb@kernel.org> --- arch/arm64/include/asm/assembler.h | 8 ++++++++ 1 file changed, 8 insertions(+)