2020-02-03 18:56:25

by Joakim Tjernlund

[permalink] [raw]
Subject: Re: [PATCH] powerpc/32s: Slenderize _tlbia() for powerpc 603/603e

On Mon, 2020-02-03 at 16:47 +0000, Christophe Leroy wrote:
> CAUTION: This email originated from outside of the organization. Do not click links or open attachments unless you recognize the sender and know the content is safe.
>
>
> _tlbia() is a function used only on 603/603e core, ie on CPUs which
> don't have a hash table.
>
> _tlbia() uses the tlbia macro which implements a loop of 1024 tlbie.
>
> On the 603/603e core, flushing the entire TLB requires no more than
> 32 tlbie.
>
> Replace tlbia by a loop of 32 tlbie.
>
> Signed-off-by: Christophe Leroy <[email protected]>
> ---
> arch/powerpc/mm/book3s32/hash_low.S | 13 ++++++++-----
> 1 file changed, 8 insertions(+), 5 deletions(-)
>
> diff --git a/arch/powerpc/mm/book3s32/hash_low.S b/arch/powerpc/mm/book3s32/hash_low.S
> index c11b0a005196..a5039ad10429 100644
> --- a/arch/powerpc/mm/book3s32/hash_low.S
> +++ b/arch/powerpc/mm/book3s32/hash_low.S
> @@ -696,18 +696,21 @@ _GLOBAL(_tlbia)
> bne- 10b
> stwcx. r8,0,r9
> bne- 10b
> +#endif /* CONFIG_SMP */
> + li r5, 32
> + lis r4, KERNELBASE@h
> + mtctr r5
> sync
> - tlbia
> +0: tlbie r4
> + addi r4, r4, 0x1000

Is page size always 4096 here or does it not matter ?

> + bdnz 0b
> sync
> +#ifdef CONFIG_SMP
> TLBSYNC
> li r0,0
> stw r0,0(r9) /* clear mmu_hash_lock */
> mtmsr r10
> SYNC_601
> isync
> -#else /* CONFIG_SMP */
> - sync
> - tlbia
> - sync
> #endif /* CONFIG_SMP */
> blr
> --
> 2.25.0
>


2020-02-03 19:08:13

by Christophe Leroy

[permalink] [raw]
Subject: Re: [PATCH] powerpc/32s: Slenderize _tlbia() for powerpc 603/603e



Le 03/02/2020 à 17:57, Joakim Tjernlund a écrit :
> On Mon, 2020-02-03 at 16:47 +0000, Christophe Leroy wrote:
>> CAUTION: This email originated from outside of the organization. Do not click links or open attachments unless you recognize the sender and know the content is safe.
>>
>>
>> _tlbia() is a function used only on 603/603e core, ie on CPUs which
>> don't have a hash table.
>>
>> _tlbia() uses the tlbia macro which implements a loop of 1024 tlbie.
>>
>> On the 603/603e core, flushing the entire TLB requires no more than
>> 32 tlbie.
>>
>> Replace tlbia by a loop of 32 tlbie.
>>
>> Signed-off-by: Christophe Leroy <[email protected]>
>> ---
>> arch/powerpc/mm/book3s32/hash_low.S | 13 ++++++++-----
>> 1 file changed, 8 insertions(+), 5 deletions(-)
>>
>> diff --git a/arch/powerpc/mm/book3s32/hash_low.S b/arch/powerpc/mm/book3s32/hash_low.S
>> index c11b0a005196..a5039ad10429 100644
>> --- a/arch/powerpc/mm/book3s32/hash_low.S
>> +++ b/arch/powerpc/mm/book3s32/hash_low.S
>> @@ -696,18 +696,21 @@ _GLOBAL(_tlbia)
>> bne- 10b
>> stwcx. r8,0,r9
>> bne- 10b
>> +#endif /* CONFIG_SMP */
>> + li r5, 32
>> + lis r4, KERNELBASE@h
>> + mtctr r5
>> sync
>> - tlbia
>> +0: tlbie r4
>> + addi r4, r4, 0x1000
>
> Is page size always 4096 here or does it not matter ?

603 and its derivatives (G2, e300, ...) only support 4k pages.

And regardless, the reference manual says:

The tlbia instruction is not implemented on the MPC603e and when its
opcode is encountered, an illegal instruction program exception is
generated. To invalidate all entries of both TLBs, 32 tlbie instructions
must be executed, incrementing the value in EA[15–19] by 1 each time

Christophe