Re: [PATCH RFC 1/2] m68k: io_mm.h: conditionalize ISA address translation on Atari

[Michael Schmitz <schmitzmic@gmail.com>]

Hi Finn,

Thanks for the quick review!

Am 06.06.2021 um 16:53 schrieb Finn Thain:
> On Sun, 6 Jun 2021, Michael Schmitz wrote:
>
>>
>> This patch to io_mm.h (on top of my current patch), plus setting isa_type to
>> ISA_TPYE_AG100 using a module parameter, should do the trick:
>>
>> diff --git a/arch/m68k/include/asm/io_mm.h b/arch/m68k/include/asm/io_mm.h
>> index f6b487b..6f79a5e 100644
>> --- a/arch/m68k/include/asm/io_mm.h
>> +++ b/arch/m68k/include/asm/io_mm.h
>> @@ -102,6 +102,11 @@
>>  #define ISA_TYPE_AG   (2)
>>  #define ISA_TYPE_ENEC (3)
>>
>> +#if defined(CONFIG_AMIGA_PCMCIA_100)
>> +#define ISA_TYPE_AG100 (4)     /* for 100 MBit APNE card */
>
> IMHO this would be simpler if that #define was unconditional like the
> others.

Yep, that one could be unconditional (and we could rearrange the defs to 
keep the new type between AG and ENEC).

>
>> +#define MULTI_ISA 1
>
> Hmm...
>

That one I didn't want to make unconditional, in order to allow the 
optimization below.

>> +#endif
>> +
>>  #if defined(CONFIG_Q40) && !defined(MULTI_ISA)
>>  #define ISA_TYPE ISA_TYPE_Q40
>>  #define ISA_SEX  0
>> @@ -135,6 +140,9 @@ static inline u8 __iomem *isa_itb(unsigned long addr)
>>  #ifdef CONFIG_Q40
>>      case ISA_TYPE_Q40: return (u8 __iomem *)Q40_ISA_IO_B(addr);
>>  #endif
>> +#if defined(CONFIG_AMIGA_PCMCIA_100)
>> +    case ISA_TYPE_AG100: fallthrough;
>> +#endif
>
> I wonder whether that 'fallthrough' is needed. One would hope not...

It won't be, but it ought to shut up compiler warnings, no?

>
>>  #ifdef CONFIG_AMIGA_PCMCIA
>>      case ISA_TYPE_AG: return (u8 __iomem *)AG_ISA_IO_B(addr);
>>  #endif
>> @@ -153,6 +161,9 @@ static inline u16 __iomem *isa_itw(unsigned long addr)
>>  #ifdef CONFIG_Q40
>>      case ISA_TYPE_Q40: return (u16 __iomem *)Q40_ISA_IO_W(addr);
>>  #endif
>> +#if defined(CONFIG_AMIGA_PCMCIA_100)
>> +    case ISA_TYPE_AG100: fallthrough;
>> +#endif
>>  #ifdef CONFIG_AMIGA_PCMCIA
>>      case ISA_TYPE_AG: return (u16 __iomem *)AG_ISA_IO_W(addr);
>>  #endif
>> @@ -168,6 +179,9 @@ static inline u32 __iomem *isa_itl(unsigned long addr)
>>  {
>>    switch(ISA_TYPE)
>>      {
>> +#if defined(CONFIG_AMIGA_PCMCIA_100)
>> +    case ISA_TYPE_AG100: fallthrough;
>> +#endif
>>  #ifdef CONFIG_AMIGA_PCMCIA
>>      case ISA_TYPE_AG: return (u32 __iomem *)AG_ISA_IO_W(addr);
>>  #endif
>> @@ -181,6 +195,9 @@ static inline u8 __iomem *isa_mtb(unsigned long addr)
>>  #ifdef CONFIG_Q40
>>      case ISA_TYPE_Q40: return (u8 __iomem *)Q40_ISA_MEM_B(addr);
>>  #endif
>> +#if defined(CONFIG_AMIGA_PCMCIA_100)
>> +    case ISA_TYPE_AG100: fallthrough;
>> +#endif
>>  #ifdef CONFIG_AMIGA_PCMCIA
>>      case ISA_TYPE_AG: return (u8 __iomem *)addr;
>>  #endif
>> @@ -199,6 +216,9 @@ static inline u16 __iomem *isa_mtw(unsigned long addr)
>>  #ifdef CONFIG_Q40
>>      case ISA_TYPE_Q40: return (u16 __iomem *)Q40_ISA_MEM_W(addr);
>>  #endif
>> +#if defined(CONFIG_AMIGA_PCMCIA_100)
>> +    case ISA_TYPE_AG100: fallthrough;
>> +#endif
>>  #ifdef CONFIG_AMIGA_PCMCIA
>>      case ISA_TYPE_AG: return (u16 __iomem *)addr;
>>  #endif
>> @@ -219,6 +239,11 @@ static inline u16 __iomem *isa_mtw(unsigned long addr)
>>  #define isa_outw(val,port) (ISA_SEX ? out_be16(isa_itw(port),(val)) :
>> out_le16(isa_itw(port),(val)))
>>  #define isa_outl(val,port) (ISA_SEX ? out_be32(isa_itl(port),(val)) :
>> out_le32(isa_itl(port),(val)))
>>
>> +#if defined(CONFIG_AMIGA_PCMCIA_100)
>> +#undef isa_inb
>> +#define isa_inb(port)      ((ISA_TYPE == ISA_TYPE_AG100) ? ((port) & 1 ?
>> isa_inw((port) - 1) & 0xff : isa_inw(port) >> 8) : in_8(isa_itb(port))
>> +#endif
>> +
>
> Would (port & ~1) be faster than (port - 1) here?

Perhaps it would - I'd hope the compiler will pick the most efficient 
solution here.

>
> Also, I don't think you need '#if defined' here. When
> !defined(CONFIG_AMIGA_PCMCIA_100), I think ISA_TYPE should be a
> compile-time constant 0, and the dead code will get tossed out anyway.

Right, that's a good point.

>>  #define isa_readb(p)       in_8(isa_mtb((unsigned long)(p)))
>>  #define isa_readw(p)       \
>>         (ISA_SEX ? in_be16(isa_mtw((unsigned long)(p))) \
>>
>> (linebreak-mangled, sorry).
>>
>> The card reset patch hunk from Alex' patch can probably go into the APNE
>> driver regardless?
>>
>> It's been quite a while - can you still try and build/test this change, Alex?
>
> Note that isa_type is never assigned to ISA_TYPE_AG100 in
> arch/m68k/kernel/setup_mm.c which means (IIUC) this patch won't take
> effect with MULTI_ISA == 1.

In order to make this useful, a Kconfig option is needed, and a module 
parameter will set isa_type to ISA_TYPE_AG100 in the APNE probe routine. 
I'll send the complete set as RFC soon. Will include your suggestions 
before doing that, of course.

Cheers,

	Michael



>