2024-04-11 01:05:50

by Tiezhu Yang

Subject: [PATCH v3 3/4] LoongArch: Save and restore PERCPU_BASE_KS for ACPI S3 state

Currently, per_cpu_offset(x) is defined as __per_cpu_offset[x])
only under CONFIG_SMP in include/asm-generic/percpu.h, that is
to say, the implementation of loongarch_common_resume() which
calls per_cpu_offset(0) is not suitable for the case of non-SMP,
so do not write per_cpu_offset(0) to PERCPU_BASE_KS when resume,
just save the value of PERCPU_BASE_KS when suspend and restore
it when resume to make it work well for both SMP and non-SMP.

Tested with the command "rtcwake -d rtc1 -s 20 -m mem -v", dmesg
shows that "PM: suspend entry (deep)" and "PM: suspend exit".

Signed-off-by: Tiezhu Yang <[email protected]>
---
arch/loongarch/power/suspend.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/arch/loongarch/power/suspend.c b/arch/loongarch/power/suspend.c
index 166d9e06a64b..7a97949ddade 100644
--- a/arch/loongarch/power/suspend.c
+++ b/arch/loongarch/power/suspend.c
@@ -16,6 +16,7 @@
#include <asm/tlbflush.h>

u64 loongarch_suspend_addr;
+static u64 saved_pcpu_base;

struct saved_registers {
u32 ecfg;
@@ -36,6 +37,7 @@ void loongarch_common_suspend(void)
saved_regs.pwctl1 = csr_read32(LOONGARCH_CSR_PWCTL1);
saved_regs.ecfg = csr_read32(LOONGARCH_CSR_ECFG);
saved_regs.euen = csr_read32(LOONGARCH_CSR_EUEN);
+ saved_pcpu_base = csr_read64(PERCPU_BASE_KS);

loongarch_suspend_addr = loongson_sysconf.suspend_addr;
}
@@ -44,7 +46,7 @@ void loongarch_common_resume(void)
{
sync_counter();
local_flush_tlb_all();
- csr_write64(per_cpu_offset(0), PERCPU_BASE_KS);
+ csr_write64(saved_pcpu_base, PERCPU_BASE_KS);
csr_write64(eentry, LOONGARCH_CSR_EENTRY);
csr_write64(eentry, LOONGARCH_CSR_MERRENTRY);
csr_write64(tlbrentry, LOONGARCH_CSR_TLBRENTRY);
--
2.42.0

2024-04-12 04:19:03

by Huacai Chen

Subject: Re: [PATCH v3 3/4] LoongArch: Save and restore PERCPU_BASE_KS for ACPI S3 state

Hi, Tiezhu,

On Thu, Apr 11, 2024 at 9:05 AM Tiezhu Yang <[email protected]> wrote:
>
> Currently, per_cpu_offset(x) is defined as __per_cpu_offset[x])
> only under CONFIG_SMP in include/asm-generic/percpu.h, that is
> to say, the implementation of loongarch_common_resume() which
> calls per_cpu_offset(0) is not suitable for the case of non-SMP,
> so do not write per_cpu_offset(0) to PERCPU_BASE_KS when resume,
> just save the value of PERCPU_BASE_KS when suspend and restore
> it when resume to make it work well for both SMP and non-SMP.
For non-SMP you need PERCPU_BASE_KS to do what?

Huacai

>
> Tested with the command "rtcwake -d rtc1 -s 20 -m mem -v", dmesg
> shows that "PM: suspend entry (deep)" and "PM: suspend exit".
>
> Signed-off-by: Tiezhu Yang <[email protected]>
> ---
> arch/loongarch/power/suspend.c | 4 +++-
> 1 file changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/arch/loongarch/power/suspend.c b/arch/loongarch/power/suspend.c
> index 166d9e06a64b..7a97949ddade 100644
> --- a/arch/loongarch/power/suspend.c
> +++ b/arch/loongarch/power/suspend.c
> @@ -16,6 +16,7 @@
> #include <asm/tlbflush.h>
>
> u64 loongarch_suspend_addr;
> +static u64 saved_pcpu_base;
>
> struct saved_registers {
> u32 ecfg;
> @@ -36,6 +37,7 @@ void loongarch_common_suspend(void)
> saved_regs.pwctl1 = csr_read32(LOONGARCH_CSR_PWCTL1);
> saved_regs.ecfg = csr_read32(LOONGARCH_CSR_ECFG);
> saved_regs.euen = csr_read32(LOONGARCH_CSR_EUEN);
> + saved_pcpu_base = csr_read64(PERCPU_BASE_KS);
>
> loongarch_suspend_addr = loongson_sysconf.suspend_addr;
> }
> @@ -44,7 +46,7 @@ void loongarch_common_resume(void)
> {
> sync_counter();
> local_flush_tlb_all();
> - csr_write64(per_cpu_offset(0), PERCPU_BASE_KS);
> + csr_write64(saved_pcpu_base, PERCPU_BASE_KS);
> csr_write64(eentry, LOONGARCH_CSR_EENTRY);
> csr_write64(eentry, LOONGARCH_CSR_MERRENTRY);
> csr_write64(tlbrentry, LOONGARCH_CSR_TLBRENTRY);
> --
> 2.42.0
>
>

2024-04-12 09:30:16

by Tiezhu Yang

Subject: Re: [PATCH v3 3/4] LoongArch: Save and restore PERCPU_BASE_KS for ACPI S3 state

On 04/12/2024 12:18 PM, Huacai Chen wrote:
> Hi, Tiezhu,
>
> On Thu, Apr 11, 2024 at 9:05 AM Tiezhu Yang <[email protected]> wrote:
>>
>> Currently, per_cpu_offset(x) is defined as __per_cpu_offset[x])
>> only under CONFIG_SMP in include/asm-generic/percpu.h, that is
>> to say, the implementation of loongarch_common_resume() which
>> calls per_cpu_offset(0) is not suitable for the case of non-SMP,
>> so do not write per_cpu_offset(0) to PERCPU_BASE_KS when resume,
>> just save the value of PERCPU_BASE_KS when suspend and restore
>> it when resume to make it work well for both SMP and non-SMP.
> For non-SMP you need PERCPU_BASE_KS to do what?

The initial aim is to avoid build error under !CONFIG_SMP
and it works well on both !CONFIG_SMP and CONFIG_SMP, the
changes are similar with the code in hibernate.c.

An alternative way is to do the following simple change,
but it seems a little ugly due to the ifdef, let me know
what is your preference.

diff --git a/arch/loongarch/power/suspend.c b/arch/loongarch/power/suspend.c
index 166d9e06a64b..35191afefcda 100644
--- a/arch/loongarch/power/suspend.c
+++ b/arch/loongarch/power/suspend.c
@@ -44,7 +44,9 @@ void loongarch_common_resume(void)
{
sync_counter();
local_flush_tlb_all();
+#ifdef
csr_write64(per_cpu_offset(0), PERCPU_BASE_KS);
+#endif
csr_write64(eentry, LOONGARCH_CSR_EENTRY);
csr_write64(eentry, LOONGARCH_CSR_MERRENTRY);
csr_write64(tlbrentry, LOONGARCH_CSR_TLBRENTRY);

Thanks,
Tiezhu

2024-04-13 07:23:08

by Huacai Chen

Subject: Re: [PATCH v3 3/4] LoongArch: Save and restore PERCPU_BASE_KS for ACPI S3 state

On Fri, Apr 12, 2024 at 5:30 PM Tiezhu Yang <[email protected]> wrote:
>
>
>
> On 04/12/2024 12:18 PM, Huacai Chen wrote:
> > Hi, Tiezhu,
> >
> > On Thu, Apr 11, 2024 at 9:05 AM Tiezhu Yang <[email protected]> wrote:
> >>
> >> Currently, per_cpu_offset(x) is defined as __per_cpu_offset[x])
> >> only under CONFIG_SMP in include/asm-generic/percpu.h, that is
> >> to say, the implementation of loongarch_common_resume() which
> >> calls per_cpu_offset(0) is not suitable for the case of non-SMP,
> >> so do not write per_cpu_offset(0) to PERCPU_BASE_KS when resume,
> >> just save the value of PERCPU_BASE_KS when suspend and restore
> >> it when resume to make it work well for both SMP and non-SMP.
> > For non-SMP you need PERCPU_BASE_KS to do what?
>
> The initial aim is to avoid build error under !CONFIG_SMP
> and it works well on both !CONFIG_SMP and CONFIG_SMP, the
> changes are similar with the code in hibernate.c.
>
> An alternative way is to do the following simple change,
> but it seems a little ugly due to the ifdef, let me know
> what is your preference.
Add "u64 pcpu_base" at the last of saved_registers is better than this
patch, and can avoid ifdef.

Huacai

>
> diff --git a/arch/loongarch/power/suspend.c b/arch/loongarch/power/suspend.c
> index 166d9e06a64b..35191afefcda 100644
> --- a/arch/loongarch/power/suspend.c
> +++ b/arch/loongarch/power/suspend.c
> @@ -44,7 +44,9 @@ void loongarch_common_resume(void)
> {
> sync_counter();
> local_flush_tlb_all();
> +#ifdef
> csr_write64(per_cpu_offset(0), PERCPU_BASE_KS);
> +#endif
> csr_write64(eentry, LOONGARCH_CSR_EENTRY);
> csr_write64(eentry, LOONGARCH_CSR_MERRENTRY);
> csr_write64(tlbrentry, LOONGARCH_CSR_TLBRENTRY);
>
> Thanks,
> Tiezhu
>
>