flush_icache_range in the sticore driver

flush_icache_range in the sticore driver

[Christoph Hellwig]

Hi all,

flush_icache_range is supposed to flush the instruction cache, which is
something no driver should be doing.  It was added in commit 03b18f1b2afe
("[PARISC] Clean up sti_flush") but the explanation in there looks odd.

Can someone shed a light what flushes this tries to flush and why it
can't be done behind a proper API?

Re: flush_icache_range in the sticore driver

[James Bottomley]

On Tue, 2022-07-05 at 18:46 +0200, Christoph Hellwig wrote:
> Hi all,
> 
> flush_icache_range is supposed to flush the instruction cache, which
> is something no driver should be doing.  It was added in commit
> 03b18f1b2afe ("[PARISC] Clean up sti_flush") but the explanation in
> there looks odd.
> 
> Can someone shed a light what flushes this tries to flush and why it
> can't be done behind a proper API?

What's wrong with the explanation:

"sti_flush is supposed to flush the caches so we can execute the STI
rom we copied to memory."

Which is in that commit?

The STI driver is taking the executable STI ROM code, which is in a
very slow to access part of the system and copying it to RAM so it can
be executed.  After the driver has done the copy it needs to flush the
icache just in case there's anything stale in there before executing
the newly copied code.

If you grep the drive for sti_call, you'll see all the points we call
into the copied rom code.

James