[PATCH v2] i.MX31 and i.MX35 : fix errate TLSbo65953 and ENGcm09472

[Uwe Kleine-König]

On Tue, Oct 05, 2010 at 09:31:27PM +0200, Eric Bénard wrote:
> Hi Uwe,
>
> Le 05/10/2010 20:33, Uwe Kleine-König a écrit :
>>> +			/* invalidate I cache */
>>> +			"mov %0, #0\n"
>>> +			"mcr p15, 0, %0, c7, c5, 0\n"
>>> +			/* clear and invalidate D cache */
>>> +			"mov %0, #0\n"
>> mcr doesn't change the value of %0, does it?  Then there's no need to
>> set it to 0 once more.
>>
>>> +			"mcr p15, 0, %0, c7, c14, 0\n"
>>> +			/* WFI */
>>> +			"mov %0, #0\n"
>> ditto
>>
>>> +			"mcr p15, 0, %0, c7, c0, 4\n"
>>> +			"nop\n" "nop\n" "nop\n" "nop\n"
>>> +			"nop\n" "nop\n" "nop\n"
>>> +			/* enable I and D cache */
>>> +			"mrc p15, 0, %0, c1, c0, 0\n"
>> If you spend two registers there is no need to reread this register.
>>
>>> +			"orr %0, %0, #0x00001000\n"
>>> +			"orr %0, %0, #0x00000004\n"
>>> +			"mcr p15, 0, %0, c1, c0, 0\n"
>>> +			:: "r" (reg));
>> ... and the s/:: "/: "=/ as I suggested earlier.
>>
> well, this code comes from Freescale and if everything was perfect it  
> wouldn't be necessary as this is an errata fix so I'm not sure this can  
> be considered as a code to optimize.
>
> Moreover it follows exactly what is suggested in the workaround  
> described on page 8 of this PDF :  
> http://cache.freescale.com/files/32bit/doc/errata/MCIMX31CE.pdf so I'm  
> not sure we should modify it.

> The other function they sent me had even more duplicated lines than this  
> one so it's difficult to know what is necessary and what isn't.
>
> I can test it with your suggestions but the extra mov & mrc may be there  
> to add instructions cycle between each mcr to be sure the workaround  
> does its job.
hmm, it's a copy of the code suggested there, yes.  The text before the
example doesn't suggest that timing is an issue for the work around.  I
interpret it that everything is fine when the caches are off before
executing the wfi instruction.  But OK, it wouldn't be the first time
hardware documentation isn't exact to the point.

Hmm, when the caches are off before entering wfi, does that mean that
all interrupt (and fiq) handlers run with the caches off, too?  That's
very bad, isn't it?

Best regards
Uwe

-- 
Pengutronix e.K.                           | Uwe Kleine-König            |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |