Re: [PATCH] radeonfb: (#2) Implement proper workarounds for PLL accesses

[Benjamin Herrenschmidt <benh@kernel.crashing.org>]

On Sat, 2005-04-09 at 18:24 +0200, Andreas Schwab wrote:
> Benjamin Herrenschmidt <benh@kernel.crashing.org> writes:
> 
> > Can you redo the counting of the workarounds with the patch ?
> 
> Switching from X to console:
> 
> radeon_write_pll_regs: INPLL
> radeon_write_pll_regs: INPLL
> radeon_write_mode: OUTPLL
> radeonfb_engine_reset: INPLL
> radeonfb_engine_reset: OUTPLL
> radeonfb_engine_reset: OUTPLL
> radeonfb_setcmap: INPLL
> radeonfb_setcmap: OUTPLL
> radeonfb_setcmap: OUTPLL
> radeon_write_pll_regs: INPLL
> radeon_write_pll_regs: INPLL
> radeon_write_mode: OUTPLL
> radeonfb_engine_reset: INPLL
> radeonfb_engine_reset: OUTPLL
> radeonfb_engine_reset: OUTPLL
> radeonfb_setcmap: INPLL
> radeonfb_setcmap: OUTPLL
> radeonfb_setcmap: OUTPLL
> radeonfb_setcmap: INPLL
> radeonfb_setcmap: OUTPLL
> radeonfb_setcmap: OUTPLL
> radeonfb_setcmap: INPLL
> radeonfb_setcmap: OUTPLL
> radeonfb_setcmap: OUTPLL
> radeonfb_setcmap: INPLL
> radeonfb_setcmap: OUTPLL

Ok, so the above is interesting, something is callign setcmap way too
much here I would say... Besides, it looks like set_par is called twice,
which is wrong too.

> Switching from console to X:
> 
> radeonfb_setcmap: OUTPLL
> radeon_write_pll_regs: INPLL
> radeon_write_pll_regs: INPLL
> radeon_write_mode: OUTPLL
> radeonfb_engine_reset: INPLL
> radeonfb_engine_reset: OUTPLL
> radeonfb_engine_reset: OUTPLL
> radeonfb_setcmap: INPLL
> radeonfb_setcmap: OUTPLL
> radeonfb_setcmap: OUTPLL
> agpgart: Putting AGP V2 device at 0000:00:0b.0 into 1x mode
> agpgart: Putting AGP V2 device at 0000:00:10.0 into 1x mode
> radeonfb_setcolreg: INPLL
> radeonfb_setcolreg: OUTPLL
> radeonfb_setcolreg: OUTPLL
> ... last three lines repeated 63 times

Hrm... the last (serie of 64 setcolreg) are probably X beeing extremely
dumb, and calling the ioctl 64 times to set each palette entry instead
of doing a single call for the whole palette...

Anyway. Except for maybe the double set-par on switch from X to console,
there isn't much more we can do here. We might be able to improve X but
there is a significant lag between a fix done to X.org HEAD appears in
any distro. The fact is, according to ATI, there is a HW bug on M6 taht
can cause lockups of the chip, and this 5ms workaround is necessary to
avoid it... 

Ben.