TLB on R10k

TLB on R10k

[Stanislaw Skowronek]

Why, in tlb-andes.c, all exception handlers are prefixed with an
ampersand (&) when copying them to main memory, only the r10k fill handler
isn't? I'm getting a blackhole-style crash (no messages, no panic,
interrupts dead as a doornail, nobody knows what is happening) as soon as
I try to jump to usermode.

Stanislaw Skowronek

--<=>--
  Paranoid: one who is truly in touch with reality.

Re: TLB on R10k

[Thiemo Seufer]

Stanislaw Skowronek wrote:
> Why, in tlb-andes.c, all exception handlers are prefixed with an
> ampersand (&) when copying them to main memory, only the r10k fill handler
> isn't?

Sloppiness, I guess. They resolve to function pointers anyway. But the
flush_icache_range should be fixed to cover all memcpy'ed memory.


Thiemo

Re: TLB on R10k

[Stanislaw Skowronek]

> Sloppiness, I guess. They resolve to function pointers anyway. But the
> flush_icache_range should be fixed to cover all memcpy'ed memory.

Yes, I've done it, too. Still gets hung. I wonder why, I have looked at
all exception code and it is 64-bit-clean (except for some bit that is
only compiled on SMP, so I don't really care).

/s

Re: TLB on R10k

[Ralf Baechle]

On Tue, Apr 27, 2004 at 08:07:23PM +0200, Stanislaw Skowronek wrote:

> Why, in tlb-andes.c, all exception handlers are prefixed with an
> ampersand (&) when copying them to main memory, only the r10k fill handler
> isn't? I'm getting a blackhole-style crash (no messages, no panic,
> interrupts dead as a doornail, nobody knows what is happening) as soon as
> I try to jump to usermode.

Have you verified that the UX bit is set correctly by your kernel?  BEV
also plays a role but since you survive BogoMIPS it should be right.

  Ralf

Re: TLB on R10k

[Stanislaw Skowronek]

> Have you verified that the UX bit is set correctly by your kernel?  BEV
> also plays a role but since you survive BogoMIPS it should be right.

From what I remember, the UX bit is fixed set. I have made the machine
print a '*' (no, I didn't use printk, but since it's my own console
driver I'm pretty sure it can work in interrupts - all it does is
hardware writes) whenever it gets a TLB refill and flushed the TLB before
entering usermode. Guess what, I didn't get a single '*' after ERET. To
verify my method, I've made a single read from the usermode PC from the
kernel, and the '*' appeared. I don't know what's up.

Stanislaw

Re: TLB on R10k

[Stanislaw Skowronek]

> Have you verified that the UX bit is set correctly by your kernel?  BEV
> also plays a role but since you survive BogoMIPS it should be right.

Re: TLB on R10k

[Ralf Baechle]

On Wed, Apr 28, 2004 at 07:37:06AM +0200, Stanislaw Skowronek wrote:

> > Have you verified that the UX bit is set correctly by your kernel?  BEV
> > also plays a role but since you survive BogoMIPS it should be right.
> 
> >From what I remember, the UX bit is fixed set. I have made the machine
> print a '*' (no, I didn't use printk, but since it's my own console
> driver I'm pretty sure it can work in interrupts - all it does is
> hardware writes) whenever it gets a TLB refill and flushed the TLB before
> entering usermode. Guess what, I didn't get a single '*' after ERET. To
> verify my method, I've made a single read from the usermode PC from the
> kernel, and the '*' appeared. I don't know what's up.

printk can be used in exceptions.

Be careful when changing the TLB exception handlers.  They occupy slots
of just 128 bytes or 32 instructions.  Maybe your debugging code just
extended it beyond that maximum?

  Ralf

Re: TLB on R10k

[Alex Deucher]

This thread may be ARM specific, but it sounds like a similar problem. 
if it's of no use, please ignore.

http://lists.arm.linux.org.uk/pipermail/linux-arm-kernel/2003-November/018303.html
http://lists.arm.linux.org.uk/pipermail/linux-arm-kernel/2003-November/018409.html

Alex

-----------------------------

> Have you verified that the UX bit is set correctly by your kernel? 
BEV
> also plays a role but since you survive BogoMIPS it should be right.

From what I remember, the UX bit is fixed set. I have made the machine
print a '*' (no, I didn't use printk, but since it's my own console
driver I'm pretty sure it can work in interrupts - all it does is
hardware writes) whenever it gets a TLB refill and flushed the TLB
before
entering usermode. Guess what, I didn't get a single '*' after ERET. To
verify my method, I've made a single read from the usermode PC from the
kernel, and the '*' appeared. I don't know what's up.

Stanislaw


	
		
__________________________________
Do you Yahoo!?
Win a $20,000 Career Makeover at Yahoo! HotJobs  
http://hotjobs.sweepstakes.yahoo.com/careermakeover 

Re: TLB on R10k

[Alex Deucher]

This thread may be ARM specific, but it sounds like a similar problem. 
if it's of no use, please ignore.

http://lists.arm.linux.org.uk/pipermail/linux-arm-kernel/2003-November/018303.html
http://lists.arm.linux.org.uk/pipermail/linux-arm-kernel/2003-November/018409.html

Alex

-----------------------------

> Have you verified that the UX bit is set correctly by your kernel? 
BEV
> also plays a role but since you survive BogoMIPS it should be right.