Re: Status of 2.5.X kernels on PPC Linux, anyone working with NPTL?

quoted

quoted

quoted

quoted

Kevin Hendricks writes:

Kevin> Also, is anyone here actively working on Ulrich Drepper's New Posix
Kevin> Thread Library (NPTL) to add in support for ppc Linux?  Has anyone
Kevin> decided on what register will be used for Thread Local Storage on PPC
Kevin> (32bit)?  (Is this r13 on ppc64?).  Have the necessary modifications to
Kevin> the ABI/compiler been made to support a dedicated thread local storage
Kevin> register?

	There still is a question of what the final version of thread
support will look like in Glibc.  TLS support has not been added to
PowerPC GCC yet.  GPR 13 is reserved for the thread ID.  I do not know
whether that also can be used as the TLS base pointer.

David


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Hi David,

Thanks for responding.

	There still is a question of what the final version of thread
support will look like in Glibc.  TLS support has not been added to
PowerPC GCC yet.  GPR 13 is reserved for the thread ID.  I do not know
whether that also can be used as the TLS base pointer.

Is IBM's NGPT the alternative being considered or just changes in how
NPTL is being implemented?

I did not realize that two registers were needed (a thread id register
and a tls base pointer).
Are you and Geoff the people that will make the decision of which
registers will be allocated and when?

Thanks,

Kevin


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

quoted

quoted

quoted

quoted

"David" == David Edelsohn [off-list ref] writes:

quoted

quoted

quoted

quoted

Kevin Hendricks writes:

 Kevin> Also, is anyone here actively working on Ulrich Drepper's New Posix
 Kevin> Thread Library (NPTL) to add in support for ppc Linux?  Has anyone
 Kevin> decided on what register will be used for Thread Local Storage on PPC
 Kevin> (32bit)?  (Is this r13 on ppc64?).  Have the necessary modifications to
 Kevin> the ABI/compiler been made to support a dedicated thread local storage
 Kevin> register?

 > 	There still is a question of what the final version of thread
 > support will look like in Glibc.  TLS support has not been added to
 > PowerPC GCC yet.  GPR 13 is reserved for the thread ID.  I do not know
 > whether that also can be used as the TLS base pointer.

Funny I was just talking to Richard Henderson on the plane about this
last week.

David, I was thinking on working on TLS for PPC (gcc side of things),
but Kevin is right, we need to get a register and change the ABI, else
things can get really spooky.  I was thinking we could use r13 (?)
like PPC64 does.

I also vaguely recall that Geoff has a list of ABI changes pending.
Perhaps we should do all these at the same time.

Anywhoooo, we should probably discuss this on the gcc lis instead.

Aldy

** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Hi,

I also vaguely recall that Geoff has a list of ABI changes pending.
Perhaps we should do all these at the same time.

Current violations of the abi for ppc linux include:

- missing support for small structure return values in register pairs
r3, r4

- missing passing floats as doubles on the outgoing parameter stack
aligned to 8
   (I think that is how the abi should be interpreted but)

But changing either of these will necessitate rewrite of the libffi
code used by gcj,
the JDK code (sys_invokeNative and etc), the mozilla code, the gcc
bridge code in
OpenOffice, and I am sure there are a number of other projects that use
some form of call
on the fly stuff for components.

So if these are the abi changes Geoff has waiting, we should really
think about a
simultaneous push of patches to many of these projects to help them
maintain
support for ppc linux.

My 2 cents.

Kevin


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

On Sun, Nov 03, 2002 at 01:45:18PM -0500, Kevin B. Hendricks wrote:

Hi,

quoted

I also vaguely recall that Geoff has a list of ABI changes pending.
Perhaps we should do all these at the same time.

Current violations of the abi for ppc linux include:

- missing support for small structure return values in register pairs
r3, r4

I thought support for this already in place, but gcc had the wrong
endianness in the registers.  If this is so, I already have a patch
for this (not commited though ;-)).

think about a
simultaneous push of patches to many of these projects to help them
maintain
support for ppc linux.

But but... I've been hacking gcc so long, I can only write in K&R :-).

Aldy

** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Hi,

quoted

- missing support for small structure return values in register pairs
r3, r4

I thought support for this already in place, but gcc had the wrong
endianness in the registers.  If this is so, I already have a patch
for this (not commited though ;-)).

According to my tests small structure returns in registers are not
currently being used for PPC Linux (32 bit) for gcc 3.2.1 and earlier
(and never have been AFAIK) but I can not speak for
the 3.3 tree and later.

Kevin


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/