2022-11-21 01:47:30

by chenlifu

Subject: [PATCH -next] RISC-V: KVM: optimize kvm_arch_hardware_enable()

The values of CSR_HEDELEG and CSR_HIDELEG registers are constants,
so change them from variables to macros.

Signed-off-by: Chen Lifu <[email protected]>
---
arch/riscv/kvm/main.c | 31 +++++++++++++------------------
1 file changed, 13 insertions(+), 18 deletions(-)

diff --git a/arch/riscv/kvm/main.c b/arch/riscv/kvm/main.c
index df2d8716851f..70196b03b6f9 100644
--- a/arch/riscv/kvm/main.c
+++ b/arch/riscv/kvm/main.c
@@ -12,10 +12,21 @@
#include <linux/kvm_host.h>
#include <asm/csr.h>
#include <asm/hwcap.h>
#include <asm/sbi.h>

+#define EXC_HEDELEG ((1UL << EXC_INST_MISALIGNED) | \
+ (1UL << EXC_BREAKPOINT) | \
+ (1UL << EXC_SYSCALL) | \
+ (1UL << EXC_INST_PAGE_FAULT) | \
+ (1UL << EXC_LOAD_PAGE_FAULT) | \
+ (1UL << EXC_STORE_PAGE_FAULT))
+
+#define IRQ_HIDELEG ((1UL << IRQ_VS_SOFT) | \
+ (1UL << IRQ_VS_TIMER) | \
+ (1UL << IRQ_VS_EXT))
+
long kvm_arch_dev_ioctl(struct file *filp,
unsigned int ioctl, unsigned long arg)
{
return -EINVAL;
}
@@ -30,29 +41,13 @@ int kvm_arch_hardware_setup(void *opaque)
return 0;
}

int kvm_arch_hardware_enable(void)
{
- unsigned long hideleg, hedeleg;
-
- hedeleg = 0;
- hedeleg |= (1UL << EXC_INST_MISALIGNED);
- hedeleg |= (1UL << EXC_BREAKPOINT);
- hedeleg |= (1UL << EXC_SYSCALL);
- hedeleg |= (1UL << EXC_INST_PAGE_FAULT);
- hedeleg |= (1UL << EXC_LOAD_PAGE_FAULT);
- hedeleg |= (1UL << EXC_STORE_PAGE_FAULT);
- csr_write(CSR_HEDELEG, hedeleg);
-
- hideleg = 0;
- hideleg |= (1UL << IRQ_VS_SOFT);
- hideleg |= (1UL << IRQ_VS_TIMER);
- hideleg |= (1UL << IRQ_VS_EXT);
- csr_write(CSR_HIDELEG, hideleg);
-
+ csr_write(CSR_HEDELEG, EXC_HEDELEG);
+ csr_write(CSR_HIDELEG, IRQ_HIDELEG);
csr_write(CSR_HCOUNTEREN, -1UL);
-
csr_write(CSR_HVIP, 0);

return 0;
}

--
2.37.1

2022-11-21 11:26:17

by Andrew Jones

Subject: Re: [PATCH -next] RISC-V: KVM: optimize kvm_arch_hardware_enable()

On Mon, Nov 21, 2022 at 08:39:15AM +0800, Chen Lifu wrote:
> The values of CSR_HEDELEG and CSR_HIDELEG registers are constants,
> so change them from variables to macros.
>
> Signed-off-by: Chen Lifu <[email protected]>
> ---
> arch/riscv/kvm/main.c | 31 +++++++++++++------------------
> 1 file changed, 13 insertions(+), 18 deletions(-)
>
> diff --git a/arch/riscv/kvm/main.c b/arch/riscv/kvm/main.c
> index df2d8716851f..70196b03b6f9 100644
> --- a/arch/riscv/kvm/main.c
> +++ b/arch/riscv/kvm/main.c
> @@ -12,10 +12,21 @@
> #include <linux/kvm_host.h>
> #include <asm/csr.h>
> #include <asm/hwcap.h>
> #include <asm/sbi.h>
>
> +#define EXC_HEDELEG ((1UL << EXC_INST_MISALIGNED) | \
> + (1UL << EXC_BREAKPOINT) | \
> + (1UL << EXC_SYSCALL) | \
> + (1UL << EXC_INST_PAGE_FAULT) | \
> + (1UL << EXC_LOAD_PAGE_FAULT) | \
> + (1UL << EXC_STORE_PAGE_FAULT))
> +
> +#define IRQ_HIDELEG ((1UL << IRQ_VS_SOFT) | \
> + (1UL << IRQ_VS_TIMER) | \
> + (1UL << IRQ_VS_EXT))
> +
> long kvm_arch_dev_ioctl(struct file *filp,
> unsigned int ioctl, unsigned long arg)
> {
> return -EINVAL;
> }
> @@ -30,29 +41,13 @@ int kvm_arch_hardware_setup(void *opaque)
> return 0;
> }
>
> int kvm_arch_hardware_enable(void)
> {
> - unsigned long hideleg, hedeleg;
> -
> - hedeleg = 0;
> - hedeleg |= (1UL << EXC_INST_MISALIGNED);
> - hedeleg |= (1UL << EXC_BREAKPOINT);
> - hedeleg |= (1UL << EXC_SYSCALL);
> - hedeleg |= (1UL << EXC_INST_PAGE_FAULT);
> - hedeleg |= (1UL << EXC_LOAD_PAGE_FAULT);
> - hedeleg |= (1UL << EXC_STORE_PAGE_FAULT);
> - csr_write(CSR_HEDELEG, hedeleg);
> -
> - hideleg = 0;
> - hideleg |= (1UL << IRQ_VS_SOFT);
> - hideleg |= (1UL << IRQ_VS_TIMER);
> - hideleg |= (1UL << IRQ_VS_EXT);
> - csr_write(CSR_HIDELEG, hideleg);
> -
> + csr_write(CSR_HEDELEG, EXC_HEDELEG);
> + csr_write(CSR_HIDELEG, IRQ_HIDELEG);
> csr_write(CSR_HCOUNTEREN, -1UL);
> -
> csr_write(CSR_HVIP, 0);
>
> return 0;
> }
>
> --
> 2.37.1
>

I don't think this optimizes anything. I'm pretty sure the compiler will
load the input to csr_write() in the most efficient way it can, regardless
of using a constant or building the input out of constants. This could
maybe be considered a code cleanup, but, in this case, it's really in the
eye of the beholder.

Thanks,
drew

2022-11-21 13:50:25

by chenlifu

Subject: Re: [PATCH -next] RISC-V: KVM: optimize kvm_arch_hardware_enable()

在 2022/11/21 18:53, Andrew Jones 写道:
> On Mon, Nov 21, 2022 at 08:39:15AM +0800, Chen Lifu wrote:
>> The values of CSR_HEDELEG and CSR_HIDELEG registers are constants,
>> so change them from variables to macros.
>>
>> Signed-off-by: Chen Lifu <[email protected]>
>> ---
>> arch/riscv/kvm/main.c | 31 +++++++++++++------------------
>> 1 file changed, 13 insertions(+), 18 deletions(-)
>>
>> diff --git a/arch/riscv/kvm/main.c b/arch/riscv/kvm/main.c
>> index df2d8716851f..70196b03b6f9 100644
>> --- a/arch/riscv/kvm/main.c
>> +++ b/arch/riscv/kvm/main.c
>> @@ -12,10 +12,21 @@
>> #include <linux/kvm_host.h>
>> #include <asm/csr.h>
>> #include <asm/hwcap.h>
>> #include <asm/sbi.h>
>>
>> +#define EXC_HEDELEG ((1UL << EXC_INST_MISALIGNED) | \
>> + (1UL << EXC_BREAKPOINT) | \
>> + (1UL << EXC_SYSCALL) | \
>> + (1UL << EXC_INST_PAGE_FAULT) | \
>> + (1UL << EXC_LOAD_PAGE_FAULT) | \
>> + (1UL << EXC_STORE_PAGE_FAULT))
>> +
>> +#define IRQ_HIDELEG ((1UL << IRQ_VS_SOFT) | \
>> + (1UL << IRQ_VS_TIMER) | \
>> + (1UL << IRQ_VS_EXT))
>> +
>> long kvm_arch_dev_ioctl(struct file *filp,
>> unsigned int ioctl, unsigned long arg)
>> {
>> return -EINVAL;
>> }
>> @@ -30,29 +41,13 @@ int kvm_arch_hardware_setup(void *opaque)
>> return 0;
>> }
>>
>> int kvm_arch_hardware_enable(void)
>> {
>> - unsigned long hideleg, hedeleg;
>> -
>> - hedeleg = 0;
>> - hedeleg |= (1UL << EXC_INST_MISALIGNED);
>> - hedeleg |= (1UL << EXC_BREAKPOINT);
>> - hedeleg |= (1UL << EXC_SYSCALL);
>> - hedeleg |= (1UL << EXC_INST_PAGE_FAULT);
>> - hedeleg |= (1UL << EXC_LOAD_PAGE_FAULT);
>> - hedeleg |= (1UL << EXC_STORE_PAGE_FAULT);
>> - csr_write(CSR_HEDELEG, hedeleg);
>> -
>> - hideleg = 0;
>> - hideleg |= (1UL << IRQ_VS_SOFT);
>> - hideleg |= (1UL << IRQ_VS_TIMER);
>> - hideleg |= (1UL << IRQ_VS_EXT);
>> - csr_write(CSR_HIDELEG, hideleg);
>> -
>> + csr_write(CSR_HEDELEG, EXC_HEDELEG);
>> + csr_write(CSR_HIDELEG, IRQ_HIDELEG);
>> csr_write(CSR_HCOUNTEREN, -1UL);
>> -
>> csr_write(CSR_HVIP, 0);
>>
>> return 0;
>> }
>>
>> --
>> 2.37.1
>>
>
> I don't think this optimizes anything. I'm pretty sure the compiler will
> load the input to csr_write() in the most efficient way it can, regardless
> of using a constant or building the input out of constants. This could
> maybe be considered a code cleanup, but, in this case, it's really in the
> eye of the beholder.
>
> Thanks,
> drew

Thanks drew，you're right.
I compared two version arch/riscv/kvm/main.o files, they are same.