Re: [RFC V2 3/8] powerpc: introduce arch_enter_from_user_mode

[Mukesh Kumar Chaurasiya <mkchauras@linux.ibm.com>]

On 9/14/25 2:32 PM, Shrikanth Hegde wrote:
>
>
> On 9/9/25 2:32 AM, Mukesh Kumar Chaurasiya wrote:
>> - Implement the hook arch_enter_from_user_mode for syscall entry.
>
> nit: for generic syscall infra.
Cool, will change this.
>
>> - Move booke_load_dbcr0 from interrupt.c to interrupt.h
>>
>> No functional change intended.
>>
>> Signed-off-by: Mukesh Kumar Chaurasiya <mchauras@linux.ibm.com>
>> ---
>>   arch/powerpc/include/asm/entry-common.h | 96 +++++++++++++++++++++++++
>>   arch/powerpc/include/asm/interrupt.h    | 23 ++++++
>>   arch/powerpc/kernel/interrupt.c         | 22 ------
>>   3 files changed, 119 insertions(+), 22 deletions(-)
>>
>> diff --git a/arch/powerpc/include/asm/entry-common.h 
>> b/arch/powerpc/include/asm/entry-common.h
>> index 3af16d821d07e..49607292bf5a5 100644
>> --- a/arch/powerpc/include/asm/entry-common.h
>> +++ b/arch/powerpc/include/asm/entry-common.h
>> @@ -5,7 +5,103 @@
>>     #ifdef CONFIG_GENERIC_IRQ_ENTRY
>>   +#include <asm/cputime.h>
>> +#include <asm/interrupt.h>
>>   #include <asm/stacktrace.h>
>> +#include <asm/tm.h>
>> +
>> +static __always_inline void arch_enter_from_user_mode(struct pt_regs 
>> *regs)
>> +{
>> +    if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG))
>> +        BUG_ON(irq_soft_mask_return() != IRQS_ALL_DISABLED);
>> +
>> +    BUG_ON(regs_is_unrecoverable(regs));
>> +    BUG_ON(!user_mode(regs));
>> +    BUG_ON(regs_irqs_disabled(regs));
>> +
>> +#ifdef CONFIG_PPC_PKEY
>> +    if (mmu_has_feature(MMU_FTR_PKEY)) {
>> +        unsigned long amr, iamr;
>> +        bool flush_needed = false;
>> +        /*
>> +         * When entering from userspace we mostly have the AMR/IAMR
>> +         * different from kernel default values. Hence don't compare.
>> +         */
>> +        amr = mfspr(SPRN_AMR);
>> +        iamr = mfspr(SPRN_IAMR);
>> +        regs->amr  = amr;
>> +        regs->iamr = iamr;
>> +        if (mmu_has_feature(MMU_FTR_KUAP)) {
>> +            mtspr(SPRN_AMR, AMR_KUAP_BLOCKED);
>> +            flush_needed = true;
>> +        }
>> +        if (mmu_has_feature(MMU_FTR_BOOK3S_KUEP)) {
>> +            mtspr(SPRN_IAMR, AMR_KUEP_BLOCKED);
>> +            flush_needed = true;
>> +        }
>> +        if (flush_needed)
>> +            isync();
>> +    } else
>> +#endif
>> +        kuap_assert_locked();
>> +
>> +    booke_restore_dbcr0();
>> +
>> +    account_cpu_user_entry();
>> +
>> +    account_stolen_time();
>> +
>> +    /*
>> +     * This is not required for the syscall exit path, but makes the
>> +     * stack frame look nicer. If this was initialised in the first 
>> stack
>> +     * frame, or if the unwinder was taught the first stack frame 
>> always
>> +     * returns to user with IRQS_ENABLED, this store could be avoided!
>> +     */
>> +    irq_soft_mask_regs_set_state(regs, IRQS_ENABLED);
>> +
>> +    /*
>> +     * If system call is called with TM active, set _TIF_RESTOREALL to
>> +     * prevent RFSCV being used to return to userspace, because POWER9
>> +     * TM implementation has problems with this instruction 
>> returning to
>> +     * transactional state. Final register values are not relevant 
>> because
>> +     * the transaction will be aborted upon return anyway. Or in the 
>> case
>> +     * of unsupported_scv SIGILL fault, the return state does not much
>> +     * matter because it's an edge case.
>> +     */
>> +    if (IS_ENABLED(CONFIG_PPC_TRANSACTIONAL_MEM) &&
>> +            unlikely(MSR_TM_TRANSACTIONAL(regs->msr)))
>> +        set_bits(_TIF_RESTOREALL, &current_thread_info()->flags);
>> +
>> +    /*
>> +     * If the system call was made with a transaction active, doom 
>> it and
>> +     * return without performing the system call. Unless it was an
>> +     * unsupported scv vector, in which case it's treated like an 
>> illegal
>> +     * instruction.
>> +     */
>> +#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
>> +    if (unlikely(MSR_TM_TRANSACTIONAL(regs->msr)) &&
>> +        !trap_is_unsupported_scv(regs)) {
>> +        /* Enable TM in the kernel, and disable EE (for scv) */
>> +        hard_irq_disable();
>> +        mtmsr(mfmsr() | MSR_TM);
>> +
>> +        /* tabort, this dooms the transaction, nothing else */
>> +        asm volatile(".long 0x7c00071d | ((%0) << 16)"
>> +                :: "r"(TM_CAUSE_SYSCALL|TM_CAUSE_PERSISTENT));
>> +
>> +        /*
>> +         * Userspace will never see the return value. Execution will
>> +         * resume after the tbegin. of the aborted transaction with the
>> +         * checkpointed register state. A context switch could occur
>> +         * or signal delivered to the process before resuming the
>> +         * doomed transaction context, but that should all be handled
>> +         * as expected.
>> +         */
>> +        return;
>> +    }
>> +#endif // CONFIG_PPC_TRANSACTIONAL_MEM
>
> nit: Better to follow standard comment practices.
>         /* CONFIG_PPC_TRANSACTIONAL_MEM */
>
Sure.
>> +}
>> +#define arch_enter_from_user_mode arch_enter_from_user_mode
>>     #endif /* CONFIG_GENERIC_IRQ_ENTRY */
>>   #endif /* _ASM_PPC_ENTRY_COMMON_H */
>> diff --git a/arch/powerpc/include/asm/interrupt.h 
>> b/arch/powerpc/include/asm/interrupt.h
>> index 56bc8113b8cde..6edf064a0fea2 100644
>> --- a/arch/powerpc/include/asm/interrupt.h
>> +++ b/arch/powerpc/include/asm/interrupt.h
>> @@ -138,6 +138,29 @@ static inline void nap_adjust_return(struct 
>> pt_regs *regs)
>>   #endif
>>   }
>>   +static inline void booke_load_dbcr0(void)
>> +{
>> +#ifdef CONFIG_PPC_ADV_DEBUG_REGS
>> +       unsigned long dbcr0 = current->thread.debug.dbcr0;
>> +
>> +       if (likely(!(dbcr0 & DBCR0_IDM)))
>> +               return;
>> +
>> +       /*
>> +        * Check to see if the dbcr0 register is set up to debug.
>> +        * Use the internal debug mode bit to do this.
>> +        */
>> +       mtmsr(mfmsr() & ~MSR_DE);
>> +       if (IS_ENABLED(CONFIG_PPC32)) {
>> +               isync();
>> +               global_dbcr0[smp_processor_id()] = mfspr(SPRN_DBCR0);
>> +       }
>> +       mtspr(SPRN_DBCR0, dbcr0);
>> +       mtspr(SPRN_DBSR, -1);
>> +#endif
>> +}
>> +
>
> Please run checkpatch.pl --strict on the series and fix the simple
> ones such as need to using tabs, spaces and alignments, extra lines etc.
>
Sure, will fix these.

Thanks,

Mukesh

>> +
>>   static inline void booke_restore_dbcr0(void)
>>   {
>>   #ifdef CONFIG_PPC_ADV_DEBUG_REGS
>> diff --git a/arch/powerpc/kernel/interrupt.c 
>> b/arch/powerpc/kernel/interrupt.c
>> index 0d8fd47049a19..2a09ac5dabd62 100644
>> --- a/arch/powerpc/kernel/interrupt.c
>> +++ b/arch/powerpc/kernel/interrupt.c
>> @@ -78,28 +78,6 @@ static notrace __always_inline bool 
>> prep_irq_for_enabled_exit(bool restartable)
>>       return true;
>>   }
>>   -static notrace void booke_load_dbcr0(void)
>> -{
>> -#ifdef CONFIG_PPC_ADV_DEBUG_REGS
>> -    unsigned long dbcr0 = current->thread.debug.dbcr0;
>> -
>> -    if (likely(!(dbcr0 & DBCR0_IDM)))
>> -        return;
>> -
>> -    /*
>> -     * Check to see if the dbcr0 register is set up to debug.
>> -     * Use the internal debug mode bit to do this.
>> -     */
>> -    mtmsr(mfmsr() & ~MSR_DE);
>> -    if (IS_ENABLED(CONFIG_PPC32)) {
>> -        isync();
>> -        global_dbcr0[smp_processor_id()] = mfspr(SPRN_DBCR0);
>> -    }
>> -    mtspr(SPRN_DBCR0, dbcr0);
>> -    mtspr(SPRN_DBSR, -1);
>> -#endif
>> -}
>> -
>>   static notrace void check_return_regs_valid(struct pt_regs *regs)
>>   {
>>   #ifdef CONFIG_PPC_BOOK3S_64
>