public inbox for libc-ports@sourceware.org
 help / color / mirror / Atom feed
* e500 port: getcontext / setcontext / swapcontext
@ 2013-09-18 17:29 Joseph S. Myers
  2013-09-18 18:25 ` Steven Munroe
  2013-09-19 15:52 ` Ryan S. Arnold
  0 siblings, 2 replies; 5+ messages in thread
From: Joseph S. Myers @ 2013-09-18 17:29 UTC (permalink / raw)
  To: libc-alpha, libc-ports; +Cc: Ryan S. Arnold

This patch adds e500 support to powerpc getcontext / setcontext /
swapcontext.  Like setjmp/longjmp, it's compile-time conditional
rather than using HWCAPs.  Unlike setjmp/longjmp, this is following an
external ABI: the context layout used by the Linux kernel.  Nothing is
done about the ACC register; as it's call-clobbered and not used for
argument passing, only the signal context case needs to handle it and
that does it via a syscall.  (But as is generally usual for context
code in glibc, this code doesn't otherwise try to limit GPR saving /
restoring to avoid call-clobbered registers for which it isn't
actually needed.)

To avoid duplication of the code sequences between getcontext /
setcontext and swapcontext, they are put in assembler macros defined
in a single place, and those macros are then used as needed under the
__CONTEXT_ENABLE_E500 conditional.

Tested with a build for e500.

2013-09-18  Joseph Myers  <joseph@codesourcery.com>

	* sysdeps/unix/sysv/linux/powerpc/powerpc32/getcontext-common.S
	(__CONTEXT_FUNC_NAME) [__CONTEXT_ENABLE_E500]: Use
	getcontext_e500.
	* sysdeps/unix/sysv/linux/powerpc/powerpc32/setcontext-common.S
	(__CONTEXT_FUNC_NAME) [__CONTEXT_ENABLE_E500]: Use
	setcontext_e500.
	* sysdeps/unix/sysv/linux/powerpc/powerpc32/swapcontext-common.S
	(__CONTEXT_FUNC_NAME) [__CONTEXT_ENABLE_E500]: Use getcontext_e500
	and setcontext_e500.

ports/ChangeLog.powerpc:
2013-09-18  Joseph Myers  <joseph@codesourcery.com>

	* sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/context-e500.h:
	New file.
	* sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/getcontext.S:
	Include <context-e500.h>.
	* sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/setcontext.S:
	Likewise.
	* sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/swapcontext.S:
	Likewise.

diff --git a/ports/sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/context-e500.h b/ports/sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/context-e500.h
new file mode 100644
index 0000000..9eb1a95
--- /dev/null
+++ b/ports/sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/context-e500.h
@@ -0,0 +1,144 @@
+/* getcontext/setcontext/makecontext support for e500 high parts of registers.
+   Copyright (C) 2006-2013 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <http://www.gnu.org/licenses/>.  */
+
+#ifndef _CONTEXT_E500_H
+#define _CONTEXT_E500_H 1
+
+#if defined __SPE__ || (defined __NO_FPRS__ && !defined _SOFT_FLOAT)
+
+# define __CONTEXT_ENABLE_E500 1
+
+/* We follow the kernel's layout, which saves the high parts of the
+   SPE registers in the vregs area, immediately followed by the ACC
+   value (call-clobbered, not handled here) and the SPEFSCR value.  */
+
+.macro getcontext_e500
+	la	r10,(_UC_VREGS)(r3)
+	evstwwe	r0,(0*4)(r10)
+	evstwwe	r1,(1*4)(r10)
+	evstwwe	r2,(2*4)(r10)
+	evstwwe	r3,(3*4)(r10)
+	evstwwe	r4,(4*4)(r10)
+	evstwwe	r5,(5*4)(r10)
+	evstwwe	r6,(6*4)(r10)
+	evstwwe	r7,(7*4)(r10)
+	evstwwe	r8,(8*4)(r10)
+	evstwwe	r9,(9*4)(r10)
+	evstwwe	r10,(10*4)(r10)
+	evstwwe	r11,(11*4)(r10)
+	evstwwe	r12,(12*4)(r10)
+	evstwwe	r13,(13*4)(r10)
+	evstwwe	r14,(14*4)(r10)
+	evstwwe	r15,(15*4)(r10)
+	evstwwe	r16,(16*4)(r10)
+	evstwwe	r17,(17*4)(r10)
+	evstwwe	r18,(18*4)(r10)
+	evstwwe	r19,(19*4)(r10)
+	evstwwe	r20,(20*4)(r10)
+	evstwwe	r21,(21*4)(r10)
+	evstwwe	r22,(22*4)(r10)
+	evstwwe	r23,(23*4)(r10)
+	evstwwe	r24,(24*4)(r10)
+	evstwwe	r25,(25*4)(r10)
+	evstwwe	r26,(26*4)(r10)
+	evstwwe	r27,(27*4)(r10)
+	evstwwe	r28,(28*4)(r10)
+	evstwwe	r29,(29*4)(r10)
+	evstwwe	r30,(30*4)(r10)
+	evstwwe	r31,(31*4)(r10)
+	mfspefscr	r9
+	stw	r9,(34*4)(r10)
+.endm
+
+.macro setcontext_e500
+	lwz	r3,_UC_VREGS+(0*4)(r31)
+	evmergelo	r0,r3,r0
+	lwz	r3,_UC_VREGS+(1*4)(r31)
+	evmergelo	r1,r3,r1
+	lwz	r3,_UC_VREGS+(2*4)(r31)
+	evmergelo	r2,r3,r2
+	lwz	r3,_UC_VREGS+(1*4)(r31)
+	evmergelo	r1,r3,r1
+	lwz	r3,_UC_VREGS+(2*4)(r31)
+	evmergelo	r2,r3,r2
+	lwz	r3,_UC_VREGS+(3*4)(r31)
+	evmergelo	r3,r3,r3
+	lwz	r3,_UC_VREGS+(4*4)(r31)
+	evmergelo	r4,r3,r4
+	lwz	r3,_UC_VREGS+(5*4)(r31)
+	evmergelo	r5,r3,r5
+	lwz	r3,_UC_VREGS+(6*4)(r31)
+	evmergelo	r6,r3,r6
+	lwz	r3,_UC_VREGS+(7*4)(r31)
+	evmergelo	r7,r3,r7
+	lwz	r3,_UC_VREGS+(8*4)(r31)
+	evmergelo	r8,r3,r8
+	lwz	r3,_UC_VREGS+(9*4)(r31)
+	evmergelo	r9,r3,r9
+	lwz	r3,_UC_VREGS+(10*4)(r31)
+	evmergelo	r10,r3,r10
+	lwz	r3,_UC_VREGS+(11*4)(r31)
+	evmergelo	r11,r3,r11
+	lwz	r3,_UC_VREGS+(12*4)(r31)
+	evmergelo	r12,r3,r12
+	lwz	r3,_UC_VREGS+(13*4)(r31)
+	evmergelo	r13,r3,r13
+	lwz	r3,_UC_VREGS+(14*4)(r31)
+	evmergelo	r14,r3,r14
+	lwz	r3,_UC_VREGS+(15*4)(r31)
+	evmergelo	r15,r3,r15
+	lwz	r3,_UC_VREGS+(16*4)(r31)
+	evmergelo	r16,r3,r16
+	lwz	r3,_UC_VREGS+(17*4)(r31)
+	evmergelo	r17,r3,r17
+	lwz	r3,_UC_VREGS+(18*4)(r31)
+	evmergelo	r18,r3,r18
+	lwz	r3,_UC_VREGS+(19*4)(r31)
+	evmergelo	r19,r3,r19
+	lwz	r3,_UC_VREGS+(20*4)(r31)
+	evmergelo	r20,r3,r20
+	lwz	r3,_UC_VREGS+(21*4)(r31)
+	evmergelo	r21,r3,r21
+	lwz	r3,_UC_VREGS+(22*4)(r31)
+	evmergelo	r22,r3,r22
+	lwz	r3,_UC_VREGS+(23*4)(r31)
+	evmergelo	r23,r3,r23
+	lwz	r3,_UC_VREGS+(24*4)(r31)
+	evmergelo	r24,r3,r24
+	lwz	r3,_UC_VREGS+(25*4)(r31)
+	evmergelo	r25,r3,r25
+	lwz	r3,_UC_VREGS+(26*4)(r31)
+	evmergelo	r26,r3,r26
+	lwz	r3,_UC_VREGS+(27*4)(r31)
+	evmergelo	r27,r3,r27
+	lwz	r3,_UC_VREGS+(28*4)(r31)
+	evmergelo	r28,r3,r28
+	lwz	r3,_UC_VREGS+(29*4)(r31)
+	evmergelo	r29,r3,r29
+	lwz	r3,_UC_VREGS+(30*4)(r31)
+	evmergelo	r30,r3,r30
+	lwz	r3,_UC_VREGS+(31*4)(r31)
+	evmergelo	r31,r3,r31
+	lwz	r3,_UC_VREGS+(34*4)(r31)
+	mtspefscr	r3
+.endm
+#else
+# undef __CONTEXT_ENABLE_E500
+#endif
+
+#endif /* context-e500.h */
diff --git a/ports/sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/getcontext.S b/ports/sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/getcontext.S
index 90d9dcd..8bc3c7a 100644
--- a/ports/sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/getcontext.S
+++ b/ports/sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/getcontext.S
@@ -24,6 +24,8 @@
 #include <asm/ptrace.h>
 #include "ucontext_i.h"
 
+#include <context-e500.h>
+
 #define __CONTEXT_FUNC_NAME __getcontext
 #undef __CONTEXT_ENABLE_FPRS
 #undef __CONTEXT_ENABLE_VRS
diff --git a/ports/sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/setcontext.S b/ports/sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/setcontext.S
index 06061ea..5f8653f 100644
--- a/ports/sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/setcontext.S
+++ b/ports/sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/setcontext.S
@@ -24,6 +24,8 @@
 #include <asm/ptrace.h>
 #include "ucontext_i.h"
 
+#include <context-e500.h>
+
 #define __CONTEXT_FUNC_NAME __setcontext
 #undef __CONTEXT_ENABLE_FPRS
 #undef __CONTEXT_ENABLE_VRS
diff --git a/ports/sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/swapcontext.S b/ports/sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/swapcontext.S
index 2150eec..de6d56f 100644
--- a/ports/sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/swapcontext.S
+++ b/ports/sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/swapcontext.S
@@ -24,6 +24,8 @@
 #include <asm/ptrace.h>
 #include "ucontext_i.h"
 
+#include <context-e500.h>
+
 #define __CONTEXT_FUNC_NAME __swapcontext
 #undef __CONTEXT_ENABLE_FPRS
 #undef __CONTEXT_ENABLE_VRS
diff --git a/sysdeps/unix/sysv/linux/powerpc/powerpc32/getcontext-common.S b/sysdeps/unix/sysv/linux/powerpc/powerpc32/getcontext-common.S
index 6330780..b396b22 100644
--- a/sysdeps/unix/sysv/linux/powerpc/powerpc32/getcontext-common.S
+++ b/sysdeps/unix/sysv/linux/powerpc/powerpc32/getcontext-common.S
@@ -261,6 +261,11 @@ ENTRY(__CONTEXT_FUNC_NAME)
 2: /* L(no_vec): */
 # endif
 #endif
+
+#ifdef __CONTEXT_ENABLE_E500
+	getcontext_e500
+#endif
+
 /* We need to set up parms and call sigprocmask which will clobber
    volatile registers. So before the call we need to retrieve the
    original ucontext ptr (parm1) from stack and store the UC_REGS_PTR
diff --git a/sysdeps/unix/sysv/linux/powerpc/powerpc32/setcontext-common.S b/sysdeps/unix/sysv/linux/powerpc/powerpc32/setcontext-common.S
index bedebf0..ba25092 100644
--- a/sysdeps/unix/sysv/linux/powerpc/powerpc32/setcontext-common.S
+++ b/sysdeps/unix/sysv/linux/powerpc/powerpc32/setcontext-common.S
@@ -243,6 +243,10 @@ ENTRY(__CONTEXT_FUNC_NAME)
 	lfd	fp31,_UC_FREGS+(31*8)(r31)
 #endif /* __CONTEXT_ENABLE_FPRS */
 
+#ifdef __CONTEXT_ENABLE_E500
+	setcontext_e500
+#endif
+
 	/* Restore LR and CCR, and set CTR to the NIP value */
 	lwz	r3,_UC_GREGS+(PT_LNK*4)(r31)
 	lwz	r4,_UC_GREGS+(PT_NIP*4)(r31)
diff --git a/sysdeps/unix/sysv/linux/powerpc/powerpc32/swapcontext-common.S b/sysdeps/unix/sysv/linux/powerpc/powerpc32/swapcontext-common.S
index 21c2e1f..f3e7741 100644
--- a/sysdeps/unix/sysv/linux/powerpc/powerpc32/swapcontext-common.S
+++ b/sysdeps/unix/sysv/linux/powerpc/powerpc32/swapcontext-common.S
@@ -265,6 +265,10 @@ ENTRY(__CONTEXT_FUNC_NAME)
 # endif /* __CONTEXT_ENABLE_VRS */
 #endif /* __CONTEXT_ENABLE_FPRS */
 
+#ifdef __CONTEXT_ENABLE_E500
+	getcontext_e500
+#endif
+
 /* Restore ucontext (parm1) from stack.  */
 	lwz	r12,_FRAME_PARM_SAVE1(r1)
 	li	r4,0
@@ -468,6 +472,10 @@ ENTRY(__CONTEXT_FUNC_NAME)
 	lfd	fp31,_UC_FREGS+(31*8)(r31)
 #endif /* __CONTEXT_ENABLE_FPRS */
 
+#ifdef __CONTEXT_ENABLE_E500
+	setcontext_e500
+#endif
+
 	/* Restore LR and CCR, and set CTR to the NIP value */
 	lwz	r3,_UC_GREGS+(PT_LNK*4)(r31)
 	lwz	r4,_UC_GREGS+(PT_NIP*4)(r31)

-- 
Joseph S. Myers
joseph@codesourcery.com

^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: e500 port: getcontext / setcontext / swapcontext
  2013-09-18 17:29 e500 port: getcontext / setcontext / swapcontext Joseph S. Myers
@ 2013-09-18 18:25 ` Steven Munroe
  2013-09-18 19:33   ` Joseph S. Myers
  2013-09-19 15:52 ` Ryan S. Arnold
  1 sibling, 1 reply; 5+ messages in thread
From: Steven Munroe @ 2013-09-18 18:25 UTC (permalink / raw)
  To: Joseph S. Myers; +Cc: libc-alpha, libc-ports, Ryan S. Arnold

On Wed, 2013-09-18 at 17:29 +0000, Joseph S. Myers wrote:
> This patch adds e500 support to powerpc getcontext / setcontext /
> swapcontext.  Like setjmp/longjmp, it's compile-time conditional
> rather than using HWCAPs.  Unlike setjmp/longjmp, this is following an
> external ABI: the context layout used by the Linux kernel.  Nothing is
> done about the ACC register; as it's call-clobbered and not used for
> argument passing, only the signal context case needs to handle it and
> that does it via a syscall.  (But as is generally usual for context
> code in glibc, this code doesn't otherwise try to limit GPR saving /
> restoring to avoid call-clobbered registers for which it isn't
> actually needed.)
> 
> To avoid duplication of the code sequences between getcontext /
> setcontext and swapcontext, they are put in assembler macros defined
> in a single place, and those macros are then used as needed under the
> __CONTEXT_ENABLE_E500 conditional.
> 

Joseph I must really question the wisdom of merging the e500 into the
same source with the "classical" PPC32 code. 

The unified GPR/FPR set of the e500 impacts the ABI in many different
ways from the classic PPC32 soft-float or hard-float that I fear this
will be a constant source of misunderstanding and error in the future.

I am much more comfortable keeping the e500 specific codes in different
source files and different parts of the tree.


^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: e500 port: getcontext / setcontext / swapcontext
  2013-09-18 18:25 ` Steven Munroe
@ 2013-09-18 19:33   ` Joseph S. Myers
  0 siblings, 0 replies; 5+ messages in thread
From: Joseph S. Myers @ 2013-09-18 19:33 UTC (permalink / raw)
  To: munroesj; +Cc: libc-alpha, libc-ports, Ryan S. Arnold

On Wed, 18 Sep 2013, Steven Munroe wrote:

> Joseph I must really question the wisdom of merging the e500 into the
> same source with the "classical" PPC32 code. 
> 
> The unified GPR/FPR set of the e500 impacts the ABI in many different
> ways from the classic PPC32 soft-float or hard-float that I fear this
> will be a constant source of misunderstanding and error in the future.
> 
> I am much more comfortable keeping the e500 specific codes in different
> source files and different parts of the tree.

The design is detailed at 
<https://sourceware.org/ml/libc-alpha/2013-09/msg00485.html>.  The 
function-calling ABI is identical to soft-float (the only case in which 
register high parts are used in argument passing / return is for 
e500-specific vector types).  With this port version, the other aspects of 
the glibc ABI also match soft-float, so soft-float binaries can run with 
glibc built for e500; this avoids an unnecessary proliferation of ABI 
variants (requiring their own ABI test baselines etc.).  The present 
getcontext / setcontext / swapcontext patch follows the same approach as 
the support in those functions for other optional features, with a macro 
__CONTEXT_ENABLE_E500 conditioning the e500-specific code, and in fact the 
substance of the e500 implementation is in a separate file context-e500.h 
in this case because that was the natural way to avoid code duplication in 
the swapcontext sources.

-- 
Joseph S. Myers
joseph@codesourcery.com

^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: e500 port: getcontext / setcontext / swapcontext
  2013-09-18 17:29 e500 port: getcontext / setcontext / swapcontext Joseph S. Myers
  2013-09-18 18:25 ` Steven Munroe
@ 2013-09-19 15:52 ` Ryan S. Arnold
  2013-09-26 15:39   ` Joseph S. Myers
  1 sibling, 1 reply; 5+ messages in thread
From: Ryan S. Arnold @ 2013-09-19 15:52 UTC (permalink / raw)
  To: Joseph S. Myers; +Cc: libc-alpha, libc-ports

On Wed, Sep 18, 2013 at 12:29 PM, Joseph S. Myers
<joseph@codesourcery.com> wrote:
> This patch adds e500 support to powerpc getcontext / setcontext /
> swapcontext.  Like setjmp/longjmp, it's compile-time conditional
> rather than using HWCAPs.  Unlike setjmp/longjmp, this is following an
> external ABI: the context layout used by the Linux kernel.  Nothing is
> done about the ACC register; as it's call-clobbered and not used for
> argument passing, only the signal context case needs to handle it and
> that does it via a syscall.  (But as is generally usual for context
> code in glibc, this code doesn't otherwise try to limit GPR saving /
> restoring to avoid call-clobbered registers for which it isn't
> actually needed.)
>
> To avoid duplication of the code sequences between getcontext /
> setcontext and swapcontext, they are put in assembler macros defined
> in a single place, and those macros are then used as needed under the
> __CONTEXT_ENABLE_E500 conditional.
>
> Tested with a build for e500.
>
> 2013-09-18  Joseph Myers  <joseph@codesourcery.com>
>
>         * sysdeps/unix/sysv/linux/powerpc/powerpc32/getcontext-common.S
>         (__CONTEXT_FUNC_NAME) [__CONTEXT_ENABLE_E500]: Use
>         getcontext_e500.
>         * sysdeps/unix/sysv/linux/powerpc/powerpc32/setcontext-common.S
>         (__CONTEXT_FUNC_NAME) [__CONTEXT_ENABLE_E500]: Use
>         setcontext_e500.
>         * sysdeps/unix/sysv/linux/powerpc/powerpc32/swapcontext-common.S
>         (__CONTEXT_FUNC_NAME) [__CONTEXT_ENABLE_E500]: Use getcontext_e500
>         and setcontext_e500.
>
> ports/ChangeLog.powerpc:
> 2013-09-18  Joseph Myers  <joseph@codesourcery.com>
>
>         * sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/context-e500.h:
>         New file.
>         * sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/getcontext.S:
>         Include <context-e500.h>.
>         * sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/setcontext.S:
>         Likewise.
>         * sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/swapcontext.S:
>         Likewise.

Structurally I'm fine with this patch.  It's as non-invasive as can be expected.

Ryan

^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: e500 port: getcontext / setcontext / swapcontext
  2013-09-19 15:52 ` Ryan S. Arnold
@ 2013-09-26 15:39   ` Joseph S. Myers
  0 siblings, 0 replies; 5+ messages in thread
From: Joseph S. Myers @ 2013-09-26 15:39 UTC (permalink / raw)
  To: Ryan S. Arnold; +Cc: libc-alpha, libc-ports

On Thu, 19 Sep 2013, Ryan S. Arnold wrote:

> Structurally I'm fine with this patch.  It's as non-invasive as can be 
> expected.

What about the substance of the patch - is that OK?

-- 
Joseph S. Myers
joseph@codesourcery.com

^ permalink raw reply	[flat|nested] 5+ messages in thread

end of thread, other threads:[~2013-09-26 15:39 UTC | newest]

Thread overview: 5+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2013-09-18 17:29 e500 port: getcontext / setcontext / swapcontext Joseph S. Myers
2013-09-18 18:25 ` Steven Munroe
2013-09-18 19:33   ` Joseph S. Myers
2013-09-19 15:52 ` Ryan S. Arnold
2013-09-26 15:39   ` Joseph S. Myers

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).