Skip to contentSkip to conversation

s390: allow to build with llvm's integrated assembler

PATCH In review Heiko Carstens 7 participants 41 messages May 11, 2022 – May 16, 2022

Subsystems (2)
lore
Share & tools
lore

Heiko Carstens

author maintainer
#0 lore raw

cover s390: allow to build with llvm's integrated assembler

A couple of patches which in result make it finally possible to build the kernel for s390 with llvm's integrated assembler. Several configs build without errors or warnings, and the kernel also works as expected.

Note that patch 6 ("s390/boot: workaround llvm IAS bug") reveals a miscompile. This looks like a bug in the instruction definitions of the mvc and clc instructions(?). I'd like to ask people to look into this, since this silently generated broken code.

This patch series is based on linux-next, which contains two additional required s390 specific patches to make llvm's IAS work.

Thanks,
Heiko

Heiko Carstens (8):
s390/alternatives: provide identical sized orginal/alternative sequences s390/alternatives: remove padding generation code
s390/entry: shorten OUTSIDE macro
s390/entry: workaround llvm's IAS limitations
s390/purgatory: workaround llvm's IAS limitations
s390/boot: workaround llvm IAS bug
s390/boot: do not emit debug info for assembly with llvm's IAS scripts/min-tool-version.sh: raise minimum clang version to 14.0.0 for s390

 arch/s390/Makefile                      |  2 +
 arch/s390/boot/head.S                   | 34 +++++----
 arch/s390/include/asm/alternative-asm.h | 76 +++-----------------
 arch/s390/include/asm/alternative.h     | 93 ++++++-------------------
 arch/s390/include/asm/spinlock.h        |  2 +-
 arch/s390/kernel/alternative.c          | 61 +---------------
 arch/s390/kernel/entry.S                | 39 +++++++----
 arch/s390/lib/spinlock.c                |  4 +-
 arch/s390/purgatory/head.S              | 29 ++++++--
 scripts/min-tool-version.sh             |  3 +-
 10 files changed, 104 insertions(+), 239 deletions(-)
Signature · 1 line
2.32.0
author maintainer
#1 lore raw

PATCH 1/8 s390/alternatives: provide identical sized orginal/alternative sequences

Explicitly provide identical sized original/alternative instruction sequences. This way there is no need for the s390 specific alternatives infrastructure to generate padding sequences.
The code which generates such sequences will be removed with a follow on patch.

---
 arch/s390/include/asm/spinlock.h |  2 +-
 arch/s390/kernel/entry.S         | 20 ++++++++++----------
 arch/s390/lib/spinlock.c         |  4 ++--
 3 files changed, 13 insertions(+), 13 deletions(-)
arch/s390/include/asm/spinlock.h +1 -1
index 10a460762e94..37127cd7749e 100644
@@ -79,7 +79,7 @@ static inline void arch_spin_unlock(arch_spinlock_t *lp)
 	typecheck(int, lp->lock);
 	kcsan_release();
 	asm_inline volatile(
-		ALTERNATIVE("", ".insn rre,0xb2fa0000,7,0", 49)	/* NIAI 7 */
+		ALTERNATIVE("nop", ".insn rre,0xb2fa0000,7,0", 49) /* NIAI 7 */
 		"	sth	%1,%0\n"
 		: "=R" (((unsigned short *) &lp->lock)[1])
 		: "d" (0) : "cc", "memory");
arch/s390/kernel/entry.S +10 -10
index 685ccec02a27..a6b45eaa3450 100644
@@ -53,19 +53,19 @@ STACK_INIT = STACK_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE
 _LPP_OFFSET	= __LC_LPP
 
 	.macro STBEAR address
-	ALTERNATIVE "", ".insn	s,0xb2010000,\address", 193
+	ALTERNATIVE "nop", ".insn s,0xb2010000,\address", 193
 	.endm
 
 	.macro LBEAR address
-	ALTERNATIVE "", ".insn	s,0xb2000000,\address", 193
+	ALTERNATIVE "nop", ".insn s,0xb2000000,\address", 193
 	.endm
 
 	.macro LPSWEY address,lpswe
-	ALTERNATIVE "b \lpswe", ".insn siy,0xeb0000000071,\address,0", 193
+	ALTERNATIVE "b \lpswe; nopr", ".insn siy,0xeb0000000071,\address,0", 193
 	.endm
 
 	.macro MBEAR reg
-	ALTERNATIVE "", __stringify(mvc __PT_LAST_BREAK(8,\reg),__LC_LAST_BREAK), 193
+	ALTERNATIVE "brcl 0,0", __stringify(mvc __PT_LAST_BREAK(8,\reg),__LC_LAST_BREAK), 193
 	.endm
 
 	.macro	CHECK_STACK savearea
@@ -121,16 +121,16 @@ _LPP_OFFSET	= __LC_LPP
 	.endm
 
 	.macro BPOFF
-	ALTERNATIVE "", ".insn rrf,0xb2e80000,0,0,12,0", 82
+	ALTERNATIVE "nop", ".insn rrf,0xb2e80000,0,0,12,0", 82
 	.endm
 
 	.macro BPON
-	ALTERNATIVE "", ".insn rrf,0xb2e80000,0,0,13,0", 82
+	ALTERNATIVE "nop", ".insn rrf,0xb2e80000,0,0,13,0", 82
 	.endm
 
 	.macro BPENTER tif_ptr,tif_mask
 	ALTERNATIVE "TSTMSK \tif_ptr,\tif_mask; jz .+8; .insn rrf,0xb2e80000,0,0,13,0", \
-		    "", 82
+		    "j .+12; nop; nop", 82
 	.endm
 
 	.macro BPEXIT tif_ptr,tif_mask
@@ -226,7 +226,7 @@ ENTRY(__switch_to)
 	aghi	%r3,__TASK_pid
 	mvc	__LC_CURRENT_PID(4,%r0),0(%r3)	# store pid of next
 	lmg	%r6,%r15,__SF_GPRS(%r15)	# load gprs of next task
-	ALTERNATIVE "", "lpp _LPP_OFFSET", 40
+	ALTERNATIVE "nop", "lpp _LPP_OFFSET", 40
 	BR_EX	%r14
 ENDPROC(__switch_to)
 
@@ -610,7 +610,7 @@ ENTRY(mcck_int_handler)
 	jno	0f
 	BPEXIT	__TI_flags(%r12),_TIF_ISOLATE_BP
 	stpt	__LC_EXIT_TIMER
-0:	ALTERNATIVE "", __stringify(lghi %r12,__LC_LAST_BREAK_SAVE_AREA),193
+0:	ALTERNATIVE "nop", __stringify(lghi %r12,__LC_LAST_BREAK_SAVE_AREA),193
 	LBEAR	0(%r12)
 	lmg	%r11,%r15,__PT_R11(%r11)
 	LPSWEY	__LC_RETURN_MCCK_PSW,__LC_RETURN_MCCK_LPSWE
@@ -646,7 +646,7 @@ ENTRY(mcck_int_handler)
 ENDPROC(mcck_int_handler)
 
 ENTRY(restart_int_handler)
-	ALTERNATIVE "", "lpp _LPP_OFFSET", 40
+	ALTERNATIVE "nop", "lpp _LPP_OFFSET", 40
 	stg	%r15,__LC_SAVE_AREA_RESTART
 	TSTMSK	__LC_RESTART_FLAGS,RESTART_FLAG_CTLREGS,4
 	jz	0f
arch/s390/lib/spinlock.c +2 -2
index 5e7ea8b111e8..04d4c6cf898e 100644
@@ -75,7 +75,7 @@ static inline int arch_load_niai4(int *lock)
 	int owner;
 
 	asm_inline volatile(
-		ALTERNATIVE("", ".insn rre,0xb2fa0000,4,0", 49)	/* NIAI 4 */
+		ALTERNATIVE("nop", ".insn rre,0xb2fa0000,4,0", 49) /* NIAI 4 */
 		"	l	%0,%1\n"
 		: "=d" (owner) : "Q" (*lock) : "memory");
 	return owner;
@@ -86,7 +86,7 @@ static inline int arch_cmpxchg_niai8(int *lock, int old, int new)
 	int expected = old;
 
 	asm_inline volatile(
-		ALTERNATIVE("", ".insn rre,0xb2fa0000,8,0", 49)	/* NIAI 8 */
+		ALTERNATIVE("nop", ".insn rre,0xb2fa0000,8,0", 49) /* NIAI 8 */
 		"	cs	%0,%3,%1\n"
 		: "=d" (old), "=Q" (*lock)
 		: "0" (old), "d" (new), "Q" (*lock)
Signature · 1 line
2.32.0
maintainer Acked-by on v1 1/8
#2 lore raw

Re: [PATCH 1/8] s390/alternatives: provide identical sized orginal/alternative sequences

Quoted 6 lines · Heiko Carstens

Explicitly provide identical sized original/alternative instruction sequences. This way there is no need for the s390 specific alternatives infrastructure to generate padding sequences.
The code which generates such sequences will be removed with a follow on patch.

Quoted 6 lines
---
 arch/s390/include/asm/spinlock.h |  2 +-
 arch/s390/kernel/entry.S         | 20 ++++++++++----------
 arch/s390/lib/spinlock.c         |  4 ++--
 3 files changed, 13 insertions(+), 13 deletions(-)
author maintainer
#3 lore raw

PATCH 2/8 s390/alternatives: remove padding generation code

In reply to Heiko Carstens

clang fails to handle ".if" statements in inline assembly which are heavily used in the alternatives code.

To work around this remove this code, and enforce that users of alternatives must specify original and alternative instruction sequences which have identical sizes. Add a compile time check with two ".org" statements similar to arm64.

In result not only clang can handle this, but also quite a lot of code can be removed.

---
 arch/s390/include/asm/alternative-asm.h | 76 +++-----------------
 arch/s390/include/asm/alternative.h     | 93 ++++++-------------------
 arch/s390/kernel/alternative.c          | 61 +---------------
 3 files changed, 31 insertions(+), 199 deletions(-)
arch/s390/include/asm/alternative-asm.h +9 -67
index bb3837d7387c..7db046596b93 100644
@@ -4,19 +4,6 @@
 
 #ifdef __ASSEMBLY__
 
-/*
- * Check the length of an instruction sequence. The length may not be larger
- * than 254 bytes and it has to be divisible by 2.
- */
-.macro alt_len_check start,end
-	.if ( \end - \start ) > 254
-	.error "cpu alternatives does not support instructions blocks > 254 bytes\n"
-	.endif
-	.if ( \end - \start ) % 2
-	.error "cpu alternatives instructions length is odd\n"
-	.endif
-.endm
-
 /*
  * Issue one struct alt_instr descriptor entry (need to put it into
  * the section .altinstructions, see below). This entry contains
@@ -28,66 +15,29 @@
 	.long	\alt_start - .
 	.word	\feature
 	.byte	\orig_end - \orig_start
-	.byte	\alt_end - \alt_start
-.endm
-
-/*
- * Fill up @bytes with nops. The macro emits 6-byte nop instructions
- * for the bulk of the area, possibly followed by a 4-byte and/or
- * a 2-byte nop if the size of the area is not divisible by 6.
- */
-.macro alt_pad_fill bytes
-	.rept	( \bytes ) / 6
-	brcl	0,0
-	.endr
-	.rept	( \bytes ) % 6 / 4
-	nop
-	.endr
-	.rept	( \bytes ) % 6 % 4 / 2
-	nopr
-	.endr
-.endm
-
-/*
- * Fill up @bytes with nops. If the number of bytes is larger
- * than 6, emit a jg instruction to branch over all nops, then
- * fill an area of size (@bytes - 6) with nop instructions.
- */
-.macro alt_pad bytes
-	.if ( \bytes > 0 )
-	.if ( \bytes > 6 )
-	jg	. + \bytes
-	alt_pad_fill \bytes - 6
-	.else
-	alt_pad_fill \bytes
-	.endif
-	.endif
+	.org	. - ( \orig_end - \orig_start ) + ( \alt_end - \alt_start )
+	.org	. - ( \alt_end - \alt_start ) + ( \orig_end - \orig_start )
 .endm
 
 /*
  * Define an alternative between two instructions. If @feature is
  * present, early code in apply_alternatives() replaces @oldinstr with
- * @newinstr. ".skip" directive takes care of proper instruction padding
- * in case @newinstr is longer than @oldinstr.
+ * @newinstr.
  */
 .macro ALTERNATIVE oldinstr, newinstr, feature
 	.pushsection .altinstr_replacement,"ax"
 770:	\newinstr
 771:	.popsection
 772:	\oldinstr
-773:	alt_len_check 770b, 771b
-	alt_len_check 772b, 773b
-	alt_pad ( ( 771b - 770b ) - ( 773b - 772b ) )
-774:	.pushsection .altinstructions,"a"
-	alt_entry 772b, 774b, 770b, 771b, \feature
+773:	.pushsection .altinstructions,"a"
+	alt_entry 772b, 773b, 770b, 771b, \feature
 	.popsection
 .endm
 
 /*
  * Define an alternative between two instructions. If @feature is
  * present, early code in apply_alternatives() replaces @oldinstr with
- * @newinstr. ".skip" directive takes care of proper instruction padding
- * in case @newinstr is longer than @oldinstr.
+ * @newinstr.
  */
 .macro ALTERNATIVE_2 oldinstr, newinstr1, feature1, newinstr2, feature2
 	.pushsection .altinstr_replacement,"ax"
@@ -95,17 +45,9 @@
 771:	\newinstr2
 772:	.popsection
 773:	\oldinstr
-774:	alt_len_check 770b, 771b
-	alt_len_check 771b, 772b
-	alt_len_check 773b, 774b
-	.if ( 771b - 770b > 772b - 771b )
-	alt_pad ( ( 771b - 770b ) - ( 774b - 773b ) )
-	.else
-	alt_pad ( ( 772b - 771b ) - ( 774b - 773b ) )
-	.endif
-775:	.pushsection .altinstructions,"a"
-	alt_entry 773b, 775b, 770b, 771b,\feature1
-	alt_entry 773b, 775b, 771b, 772b,\feature2
+774:	.pushsection .altinstructions,"a"
+	alt_entry 773b, 774b, 770b, 771b,\feature1
+	alt_entry 773b, 774b, 771b, 772b,\feature2
 	.popsection
 .endm
 
arch/s390/include/asm/alternative.h +20 -73

136 lines not on this page · raw message

arch/s390/kernel/alternative.c +2 -59

95 lines not on this page · raw message

Signature · 1 line
2.32.0
author maintainer
#4 lore raw

Re: [PATCH 2/8] s390/alternatives: remove padding generation code

Quoted 2 lines · Heiko Carstens

clang fails to handle ".if" statements in inline assembly which are heavily used in the alternatives code.

FWIW, I missed to add error message(s) to the changelog:

In file included from ./include/linux/spinlock.h:93:
./arch/s390/include/asm/spinlock.h:81:3: error: expected absolute expression

                ALTERNATIVE("", ".insn rre,0xb2fa0000,7,0", 49) /* NIAI 7 */
                ^
./arch/s390/include/asm/alternative.h:118:2: note: expanded from macro 'ALTERNATIVE'
        ALTINSTR_REPLACEMENT(altinstr, 1)                               \
        ^
./arch/s390/include/asm/alternative.h:113:2: note: expanded from macro 'ALTINSTR_REPLACEMENT'
        INSTR_LEN_SANITY_CHECK(altinstr_len(num))
        ^
./arch/s390/include/asm/alternative.h:62:3: note: expanded from macro 'INSTR_LEN_SANITY_CHECK'
        ".if " len " > 254\n"                                           \
         ^
<inline asm>:5:5: note: instantiated into assembly here
.if 6651b-6641b > 254
    ^
maintainer Acked-by on v1 2/8
#5 lore raw

Re: [PATCH 2/8] s390/alternatives: remove padding generation code

In reply to Heiko Carstens
Quoted 11 lines · Heiko Carstens

clang fails to handle ".if" statements in inline assembly which are heavily used in the alternatives code.

To work around this remove this code, and enforce that users of alternatives must specify original and alternative instruction sequences which have identical sizes. Add a compile time check with two ".org" statements similar to arm64.

In result not only clang can handle this, but also quite a lot of code can be removed.

Quoted 6 lines
---
 arch/s390/include/asm/alternative-asm.h | 76 +++-----------------
 arch/s390/include/asm/alternative.h     | 93 ++++++-------------------
 arch/s390/kernel/alternative.c          | 61 +---------------
 3 files changed, 31 insertions(+), 199 deletions(-)
author maintainer
#6 lore raw

PATCH 3/8 s390/entry: shorten OUTSIDE macro

In reply to Heiko Carstens

Since the minimum architecture level has been raised to z10 a shorter instruction sequence can be used to implement the OUTSIDE macro. This also reduces the number of used registers within that macro to one.

---
 arch/s390/kernel/entry.S | 8 +++-----
 1 file changed, 3 insertions(+), 5 deletions(-)
arch/s390/kernel/entry.S +3 -5
index a6b45eaa3450..e1664b45090f 100644
@@ -169,11 +169,9 @@ _LPP_OFFSET	= __LC_LPP
 	 * @outside_label: jump here if @reg is outside of [@start..@end)
 	 */
 	.macro OUTSIDE reg,start,end,outside_label
-	lgr	%r14,\reg
-	larl	%r13,\start
-	slgr	%r14,%r13
-	lghi	%r13,\end - \start
-	clgr	%r14,%r13
+	larl	%r14,\start
+	slgrk	%r14,\reg,%r14
+	clgfi	%r14,\end - \start
 	jhe	\outside_label
 	.endm
 
Signature · 1 line
2.32.0
#7 lore raw

Re: [PATCH 3/8] s390/entry: shorten OUTSIDE macro

I love your patch! Yet something to improve:

Hi Heiko,

I love your patch! Yet something to improve:

[auto build test ERROR on s390/features]
[also build test ERROR on tip/locking/core masahiroy-kbuild/for-next linus/master v5.18-rc6] [If your patch is applied to the wrong git tree, kindly drop us a note.
And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch]

url:    https://github.com/intel-lab-lkp/linux/commits/Heiko-Carstens/s390-allow-to-build-with-llvm-s-integrated-assembler/20220511-201054
base:   https://git.kernel.org/pub/scm/linux/kernel/git/s390/linux.git features
config: s390-randconfig-p001-20220512 (https://download.01.org/0day-ci/archive/20220512/[email protected]/config)
compiler: s390-linux-gcc (GCC) 11.3.0
reproduce (this is a W=1 build):
        wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross
        chmod +x ~/bin/make.cross
        # https://github.com/intel-lab-lkp/linux/commit/ad8a0a6488ba252aaa84b813dee2c949ae59d88a
        git remote add linux-review https://github.com/intel-lab-lkp/linux
        git fetch --no-tags linux-review Heiko-Carstens/s390-allow-to-build-with-llvm-s-integrated-assembler/20220511-201054
        git checkout ad8a0a6488ba252aaa84b813dee2c949ae59d88a
        # save the config file
        mkdir build_dir && cp config build_dir/.config
        COMPILER_INSTALL_PATH=$HOME/0day COMPILER=gcc-11.3.0 make.cross W=1 O=build_dir ARCH=s390 SHELL=/bin/bash arch/s390/

If you fix the issue, kindly add following tag as appropriate

All errors (new ones prefixed by >>):

arch/s390/kernel/entry.S: Assembler messages:

>> arch/s390/kernel/entry.S:378: Error: Unrecognized opcode: `slgrk'

arch/s390/kernel/entry.S:488: Error: Unrecognized opcode: slgrk' arch/s390/kernel/entry.S:489: Error: Unrecognized opcode: slgrk' arch/s390/kernel/entry.S:554: Error: Unrecognized opcode: slgrk' arch/s390/kernel/entry.S:555: Error: Unrecognized opcode: slgrk'

vim +378 arch/s390/kernel/entry.S

26a374ae7af8d7 arch/s390/kernel/entry.S   Martin Schwidefsky 2019-01-17  359  
^1da177e4c3f41 arch/s390/kernel/entry64.S Linus Torvalds     2005-04-16  360  /*
^1da177e4c3f41 arch/s390/kernel/entry64.S Linus Torvalds     2005-04-16  361   * Program check handler routine
^1da177e4c3f41 arch/s390/kernel/entry64.S Linus Torvalds     2005-04-16  362   */
^1da177e4c3f41 arch/s390/kernel/entry64.S Linus Torvalds     2005-04-16  363  
144d634a21caff arch/s390/kernel/entry64.S Jan Glauber        2011-07-24  364  ENTRY(pgm_check_handler)
56e62a73702836 arch/s390/kernel/entry.S   Sven Schnelle      2020-11-21  365  	stpt	__LC_SYS_ENTER_TIMER
d768bd892fc8f0 arch/s390/kernel/entry.S   Martin Schwidefsky 2018-01-16  366  	BPOFF
c5328901aa1db1 arch/s390/kernel/entry64.S Martin Schwidefsky 2011-12-27  367  	stmg	%r8,%r15,__LC_SAVE_AREA_SYNC
56e62a73702836 arch/s390/kernel/entry.S   Sven Schnelle      2020-11-21  368  	lg	%r12,__LC_CURRENT
56e62a73702836 arch/s390/kernel/entry.S   Sven Schnelle      2020-11-21  369  	lghi	%r10,0
c5328901aa1db1 arch/s390/kernel/entry64.S Martin Schwidefsky 2011-12-27  370  	lmg	%r8,%r9,__LC_PGM_OLD_PSW
87d5986345219a arch/s390/kernel/entry.S   Heiko Carstens     2020-11-16  371  	tmhh	%r8,0x0001		# coming from user space?
87d5986345219a arch/s390/kernel/entry.S   Heiko Carstens     2020-11-16  372  	jno	.Lpgm_skip_asce
87d5986345219a arch/s390/kernel/entry.S   Heiko Carstens     2020-11-16  373  	lctlg	%c1,%c1,__LC_KERNEL_ASCE
56e62a73702836 arch/s390/kernel/entry.S   Sven Schnelle      2020-11-21  374  	j	3f			# -> fault in user space
87d5986345219a arch/s390/kernel/entry.S   Heiko Carstens     2020-11-16  375  .Lpgm_skip_asce:
d0fc41071a6884 arch/s390/kernel/entry.S   Martin Schwidefsky 2015-06-22  376  #if IS_ENABLED(CONFIG_KVM)
0a5e2ec2647737 arch/s390/kernel/entry.S   Martin Schwidefsky 2017-10-05  377  	# cleanup critical section for program checks in sie64a
b5415c8f975506 arch/s390/kernel/entry.S   Alexander Gordeev  2021-06-07 @378  	OUTSIDE	%r9,.Lsie_gmap,.Lsie_done,1f
fbbdfca5c5535f arch/s390/kernel/entry.S   Alexander Gordeev  2021-06-18  379  	SIEEXIT
56e62a73702836 arch/s390/kernel/entry.S   Sven Schnelle      2020-11-21  380  	lghi	%r10,_PIF_GUEST_FAULT
d0fc41071a6884 arch/s390/kernel/entry.S   Martin Schwidefsky 2015-06-22  381  #endif
0b38b5e1d0e2f3 arch/s390/kernel/entry.S   Sven Schnelle      2020-01-22  382  1:	tmhh	%r8,0x4000		# PER bit set in old PSW ?
0b38b5e1d0e2f3 arch/s390/kernel/entry.S   Sven Schnelle      2020-01-22  383  	jnz	2f			# -> enabled, can't be a double fault
c5328901aa1db1 arch/s390/kernel/entry64.S Martin Schwidefsky 2011-12-27  384  	tm	__LC_PGM_ILC+3,0x80	# check for per exception
86ed42f401cb8f arch/s390/kernel/entry64.S Martin Schwidefsky 2014-12-03  385  	jnz	.Lpgm_svcper		# -> single stepped svc
0b38b5e1d0e2f3 arch/s390/kernel/entry.S   Sven Schnelle      2020-01-22  386  2:	CHECK_STACK __LC_SAVE_AREA_SYNC
dc7ee00d4771b3 arch/s390/kernel/entry64.S Martin Schwidefsky 2013-04-24  387  	aghi	%r15,-(STACK_FRAME_OVERHEAD + __PT_SIZE)
56e62a73702836 arch/s390/kernel/entry.S   Sven Schnelle      2020-11-21  388  	# CHECK_VMAP_STACK branches to stack_overflow or 4f
56e62a73702836 arch/s390/kernel/entry.S   Sven Schnelle      2020-11-21  389  	CHECK_VMAP_STACK __LC_SAVE_AREA_SYNC,4f
56e62a73702836 arch/s390/kernel/entry.S   Sven Schnelle      2020-11-21  390  3:	BPENTER __TI_flags(%r12),_TIF_ISOLATE_BP
c5328901aa1db1 arch/s390/kernel/entry64.S Martin Schwidefsky 2011-12-27  391  	lg	%r15,__LC_KERNEL_STACK
56e62a73702836 arch/s390/kernel/entry.S   Sven Schnelle      2020-11-21  392  4:	la	%r11,STACK_FRAME_OVERHEAD(%r15)
56e62a73702836 arch/s390/kernel/entry.S   Sven Schnelle      2020-11-21  393  	stg	%r10,__PT_FLAGS(%r11)
56e62a73702836 arch/s390/kernel/entry.S   Sven Schnelle      2020-11-21  394  	xc	__SF_BACKCHAIN(8,%r15),__SF_BACKCHAIN(%r15)
c5328901aa1db1 arch/s390/kernel/entry64.S Martin Schwidefsky 2011-12-27  395  	stmg	%r0,%r7,__PT_R0(%r11)
56e62a73702836 arch/s390/kernel/entry.S   Sven Schnelle      2020-11-21  396  	mvc	__PT_R8(64,%r11),__LC_SAVE_AREA_SYNC
3b051e89da70d4 arch/s390/kernel/entry.S   Sven Schnelle      2021-04-07  397  	mvc	__PT_LAST_BREAK(8,%r11),__LC_PGM_LAST_BREAK
56e62a73702836 arch/s390/kernel/entry.S   Sven Schnelle      2020-11-21  398  	stmg	%r8,%r9,__PT_PSW(%r11)
56e62a73702836 arch/s390/kernel/entry.S   Sven Schnelle      2020-11-21  399  
7041d28115e91f arch/s390/kernel/entry.S   Martin Schwidefsky 2018-01-16  400  	# clear user controlled registers to prevent speculative use
7041d28115e91f arch/s390/kernel/entry.S   Martin Schwidefsky 2018-01-16  401  	xgr	%r0,%r0
7041d28115e91f arch/s390/kernel/entry.S   Martin Schwidefsky 2018-01-16  402  	xgr	%r1,%r1
7041d28115e91f arch/s390/kernel/entry.S   Martin Schwidefsky 2018-01-16  403  	xgr	%r3,%r3
7041d28115e91f arch/s390/kernel/entry.S   Martin Schwidefsky 2018-01-16  404  	xgr	%r4,%r4
7041d28115e91f arch/s390/kernel/entry.S   Martin Schwidefsky 2018-01-16  405  	xgr	%r5,%r5
7041d28115e91f arch/s390/kernel/entry.S   Martin Schwidefsky 2018-01-16  406  	xgr	%r6,%r6
7041d28115e91f arch/s390/kernel/entry.S   Martin Schwidefsky 2018-01-16  407  	xgr	%r7,%r7
56e62a73702836 arch/s390/kernel/entry.S   Sven Schnelle      2020-11-21  408  	lgr	%r2,%r11
56e62a73702836 arch/s390/kernel/entry.S   Sven Schnelle      2020-11-21  409  	brasl	%r14,__do_pgm_check
56e62a73702836 arch/s390/kernel/entry.S   Sven Schnelle      2020-11-21  410  	tmhh	%r8,0x0001		# returning to user space?
56e62a73702836 arch/s390/kernel/entry.S   Sven Schnelle      2020-11-21  411  	jno	.Lpgm_exit_kernel
56e62a73702836 arch/s390/kernel/entry.S   Sven Schnelle      2020-11-21  412  	lctlg	%c1,%c1,__LC_USER_ASCE
56e62a73702836 arch/s390/kernel/entry.S   Sven Schnelle      2020-11-21  413  	BPEXIT __TI_flags(%r12),_TIF_ISOLATE_BP
0cd9b7230cc57b arch/s390/kernel/entry.S   Heiko Carstens     2020-11-11  414  	stpt	__LC_EXIT_TIMER
56e62a73702836 arch/s390/kernel/entry.S   Sven Schnelle      2020-11-21  415  .Lpgm_exit_kernel:
56e62a73702836 arch/s390/kernel/entry.S   Sven Schnelle      2020-11-21  416  	mvc	__LC_RETURN_PSW(16),STACK_FRAME_OVERHEAD+__PT_PSW(%r15)
3b051e89da70d4 arch/s390/kernel/entry.S   Sven Schnelle      2021-04-07  417  	LBEAR	STACK_FRAME_OVERHEAD+__PT_LAST_BREAK(%r15)
56e62a73702836 arch/s390/kernel/entry.S   Sven Schnelle      2020-11-21  418  	lmg	%r0,%r15,STACK_FRAME_OVERHEAD+__PT_R0(%r15)
3b051e89da70d4 arch/s390/kernel/entry.S   Sven Schnelle      2021-04-07  419  	LPSWEY	__LC_RETURN_PSW,__LC_RETURN_LPSWE
^1da177e4c3f41 arch/s390/kernel/entry64.S Linus Torvalds     2005-04-16  420  
Signature · 2 lines
0-DAY CI Kernel Test Service
https://01.org/lkp
#8 lore raw

Re: [PATCH 3/8] s390/entry: shorten OUTSIDE macro

In reply to Heiko Carstens
I love your patch! Yet something to improve:

Hi Heiko,

I love your patch! Yet something to improve:

[auto build test ERROR on s390/features]
[also build test ERROR on tip/locking/core masahiroy-kbuild/for-next linus/master v5.18-rc6 next-20220512] [If your patch is applied to the wrong git tree, kindly drop us a note.
And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch]

url:    https://github.com/intel-lab-lkp/linux/commits/Heiko-Carstens/s390-allow-to-build-with-llvm-s-integrated-assembler/20220511-201054
base:   https://git.kernel.org/pub/scm/linux/kernel/git/s390/linux.git features
config: s390-randconfig-r003-20220512 (https://download.01.org/0day-ci/archive/20220513/[email protected]/config)
compiler: clang version 15.0.0 (https://github.com/llvm/llvm-project 18dd123c56754edf62c7042dcf23185c3727610f)

reproduce (this is a W=1 build):
wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross chmod +x ~/bin/make.cross
# install s390 cross compiling tool for clang build
# apt-get install binutils-s390x-linux-gnu
# https://github.com/intel-lab-lkp/linux/commit/ad8a0a6488ba252aaa84b813dee2c949ae59d88a
git remote add linux-review https://github.com/intel-lab-lkp/linux
git fetch --no-tags linux-review Heiko-Carstens/s390-allow-to-build-with-llvm-s-integrated-assembler/20220511-201054 git checkout ad8a0a6488ba252aaa84b813dee2c949ae59d88a
# save the config file
mkdir build_dir && cp config build_dir/.config
COMPILER_INSTALL_PATH=$HOME/0day COMPILER=clang make.cross W=1 O=build_dir ARCH=s390 SHELL=/bin/bash

If you fix the issue, kindly add following tag as appropriate

All errors (new ones prefixed by >>):

>> <instantiation>:2:2: error: instruction requires: distinct-ops
    slgrk %r14,%r9,%r14
    ^
   arch/s390/kernel/entry.S:378:2: note: while in macro instantiation
    OUTSIDE %r9,.Lsie_gmap,.Lsie_done,1f
    ^
   <instantiation>:3:13: error: invalid operand for instruction
    clgfi %r14,.Lsie_done - .Lsie_gmap
               ^
   arch/s390/kernel/entry.S:378:2: note: while in macro instantiation
    OUTSIDE %r9,.Lsie_gmap,.Lsie_done,1f
    ^
>> <instantiation>:2:2: error: instruction requires: distinct-ops
    slgrk %r14,%r9,%r14
    ^
   <instantiation>:12:2: note: while in macro instantiation
    OUTSIDE %r9,.Lsie_gmap,.Lsie_done,0f
    ^
   arch/s390/kernel/entry.S:488:1: note: while in macro instantiation
   INT_HANDLER ext_int_handler,304,do_ext_irq
   ^
   <instantiation>:3:13: error: invalid operand for instruction
    clgfi %r14,.Lsie_done - .Lsie_gmap
               ^
   <instantiation>:12:2: note: while in macro instantiation
    OUTSIDE %r9,.Lsie_gmap,.Lsie_done,0f
    ^
   arch/s390/kernel/entry.S:488:1: note: while in macro instantiation
   INT_HANDLER ext_int_handler,304,do_ext_irq
   ^
>> <instantiation>:2:2: error: instruction requires: distinct-ops
    slgrk %r14,%r9,%r14
    ^
   <instantiation>:12:2: note: while in macro instantiation
    OUTSIDE %r9,.Lsie_gmap,.Lsie_done,0f
    ^
   arch/s390/kernel/entry.S:489:1: note: while in macro instantiation
   INT_HANDLER io_int_handler,368,do_io_irq
   ^
   <instantiation>:3:13: error: invalid operand for instruction
    clgfi %r14,.Lsie_done - .Lsie_gmap
               ^
   <instantiation>:12:2: note: while in macro instantiation
    OUTSIDE %r9,.Lsie_gmap,.Lsie_done,0f
    ^
   arch/s390/kernel/entry.S:489:1: note: while in macro instantiation
   INT_HANDLER io_int_handler,368,do_io_irq
   ^
>> <instantiation>:2:2: error: instruction requires: distinct-ops
    slgrk %r14,%r9,%r14
    ^
   arch/s390/kernel/entry.S:554:2: note: while in macro instantiation
    OUTSIDE %r9,.Lsie_gmap,.Lsie_done,6f
    ^
   <instantiation>:3:13: error: invalid operand for instruction
    clgfi %r14,.Lsie_done - .Lsie_gmap
               ^
   arch/s390/kernel/entry.S:554:2: note: while in macro instantiation
    OUTSIDE %r9,.Lsie_gmap,.Lsie_done,6f
    ^
>> <instantiation>:2:2: error: instruction requires: distinct-ops
    slgrk %r14,%r9,%r14
    ^
   arch/s390/kernel/entry.S:555:2: note: while in macro instantiation
    OUTSIDE %r9,.Lsie_entry,.Lsie_skip,4f
    ^
   <instantiation>:3:13: error: invalid operand for instruction
    clgfi %r14,.Lsie_skip - .Lsie_entry
               ^
   arch/s390/kernel/entry.S:555:2: note: while in macro instantiation
    OUTSIDE %r9,.Lsie_entry,.Lsie_skip,4f
    ^
Signature · 2 lines
0-DAY CI Kernel Test Service
https://01.org/lkp
author maintainer
#9 lore raw

Re: [PATCH 3/8] s390/entry: shorten OUTSIDE macro

In reply to Heiko Carstens
Quoted 15 lines · Heiko Carstens

Since the minimum architecture level has been raised to z10 a shorter instruction sequence can be used to implement the OUTSIDE macro. This also reduces the number of used registers within that macro to one.

---
 arch/s390/kernel/entry.S | 8 +++-----
 1 file changed, 3 insertions(+), 5 deletions(-)
arch/s390/kernel/entry.S +0 -0
index a6b45eaa3450..e1664b45090f 100644
@@ -169,11 +169,9 @@ _LPP_OFFSET	= __LC_LPP
 	 * @outside_label: jump here if @reg is outside of [@start..@end)

arch/s390/kernel/entry.S _LPP_OFFSET = __LC_LPP · line 174

Quoted 10 lines
arch/s390/kernel/entry.S +3 -5
 	 */
 	.macro OUTSIDE reg,start,end,outside_label
-	lgr	%r14,\reg
-	larl	%r13,\start
-	slgr	%r14,%r13
-	lghi	%r13,\end - \start
-	clgr	%r14,%r13
+	larl	%r14,\start
+	slgrk	%r14,\reg,%r14
+	clgfi	%r14,\end - \start

Clever me.. slgrk was added with z196, and not z10. So dropping this patch.

#10 lore raw

Re: [PATCH 3/8] s390/entry: shorten OUTSIDE macro

Quoted 22 lines · Heiko Carstens
Quoted 20 lines · Heiko Carstens

Since the minimum architecture level has been raised to z10 a shorter instruction sequence can be used to implement the OUTSIDE macro. This also reduces the number of used registers within that macro to one.

---
 arch/s390/kernel/entry.S | 8 +++-----
 1 file changed, 3 insertions(+), 5 deletions(-)
arch/s390/kernel/entry.S +0 -3
index a6b45eaa3450..e1664b45090f 100644
@@ -169,11 +169,9 @@ _LPP_OFFSET      = __LC_LPP
       * @outside_label: jump here if @reg is outside of [@start..@end)
       */
      .macro OUTSIDE reg,start,end,outside_label
-     lgr     %r14,\reg
-     larl    %r13,\start
-     slgr    %r14,%r13
Quoted 8 lines

arch/s390/kernel/entry.S _LPP_OFFSET = __LC_LPP · line 174

Quoted 5 lines
arch/s390/kernel/entry.S +3 -2
-     lghi    %r13,\end - \start
-     clgr    %r14,%r13
+     larl    %r14,\start
+     slgrk   %r14,\reg,%r14
+     clgfi   %r14,\end - \start

Clever me.. slgrk was added with z196, and not z10. So dropping this patch.

How do the version numbers work for SystemZ? Is there a list/reference you could link me to? If it's too deep a rabbit hole, then nevermind, but I would like to learn a little more about the architecture.

Signature · 2 lines
Thanks,
~Nick Desaulniers
author maintainer
#11 lore raw

Re: [PATCH 3/8] s390/entry: shorten OUTSIDE macro

Quoted 32 lines · Nick Desaulniers
Quoted 22 lines · Heiko Carstens
Quoted 20 lines · Heiko Carstens

Since the minimum architecture level has been raised to z10 a shorter instruction sequence can be used to implement the OUTSIDE macro. This also reduces the number of used registers within that macro to one.

---
 arch/s390/kernel/entry.S | 8 +++-----
 1 file changed, 3 insertions(+), 5 deletions(-)
arch/s390/kernel/entry.S +0 -3
index a6b45eaa3450..e1664b45090f 100644
@@ -169,11 +169,9 @@ _LPP_OFFSET      = __LC_LPP
       * @outside_label: jump here if @reg is outside of [@start..@end)
       */
      .macro OUTSIDE reg,start,end,outside_label
-     lgr     %r14,\reg
-     larl    %r13,\start
-     slgr    %r14,%r13
Quoted 8 lines

arch/s390/kernel/entry.S _LPP_OFFSET = __LC_LPP · line 174

Quoted 5 lines
arch/s390/kernel/entry.S +3 -2
-     lghi    %r13,\end - \start
-     clgr    %r14,%r13
+     larl    %r14,\start
+     slgrk   %r14,\reg,%r14
+     clgfi   %r14,\end - \start

Clever me.. slgrk was added with z196, and not z10. So dropping this patch.

Quoted 3 lines

How do the version numbers work for SystemZ? Is there a list/reference you could link me to? If it's too deep a rabbit hole, then nevermind, but I would like to learn a little more about the architecture.

If it is only for the machine generations the following links might help:

https://www.ibm.com/support/pages/ibm-mainframe-life-cycle-history
(see linked pdf for list of machine names)

https://en.wikipedia.org/wiki/IBM_Z

There might be better sources, but that's all I could find right now.

#12 lore raw

Re: [PATCH 3/8] s390/entry: shorten OUTSIDE macro

Quoted 38 lines · Heiko Carstens
Quoted 32 lines · Nick Desaulniers
Quoted 22 lines · Heiko Carstens
Quoted 20 lines · Heiko Carstens

Since the minimum architecture level has been raised to z10 a shorter instruction sequence can be used to implement the OUTSIDE macro. This also reduces the number of used registers within that macro to one.

---
 arch/s390/kernel/entry.S | 8 +++-----
 1 file changed, 3 insertions(+), 5 deletions(-)
arch/s390/kernel/entry.S +0 -3
index a6b45eaa3450..e1664b45090f 100644
@@ -169,11 +169,9 @@ _LPP_OFFSET      = __LC_LPP
       * @outside_label: jump here if @reg is outside of [@start..@end)
       */
      .macro OUTSIDE reg,start,end,outside_label
-     lgr     %r14,\reg
-     larl    %r13,\start
-     slgr    %r14,%r13
Quoted 8 lines

arch/s390/kernel/entry.S _LPP_OFFSET = __LC_LPP · line 174

Quoted 5 lines
arch/s390/kernel/entry.S +3 -2
-     lghi    %r13,\end - \start
-     clgr    %r14,%r13
+     larl    %r14,\start
+     slgrk   %r14,\reg,%r14
+     clgfi   %r14,\end - \start

Clever me.. slgrk was added with z196, and not z10. So dropping this patch.

Quoted 3 lines

How do the version numbers work for SystemZ? Is there a list/reference you could link me to? If it's too deep a rabbit hole, then nevermind, but I would like to learn a little more about the architecture.

Quoted 8 lines

If it is only for the machine generations the following links might help:

https://www.ibm.com/support/pages/ibm-mainframe-life-cycle-history
(see linked pdf for list of machine names)

https://en.wikipedia.org/wiki/IBM_Z

There might be better sources, but that's all I could find right now.

Interesting! Thanks for the links.
I'm guessing in our CI that we should probably pursue testing some of the newer revisions. Wasn't defconfig updated from z10 to z12 not too long ago?
So probably
CONFIG_MARCH_Z13
CONFIG_MARCH_Z14
CONFIG_MARCH_Z15
CONFIG_MARCH_Z16

All look like they're still "supported" (and I'm guessing CONFIG_MARCH_Z10 and CONFIG_MARCH_Z196 are not too much burden to continue to maintain kernel support for), with a higher emphasis perhaps on z15+z16?

Signature · 2 lines
Thanks,
~Nick Desaulniers
author maintainer
#13 lore raw

Re: [PATCH 3/8] s390/entry: shorten OUTSIDE macro

Quoted 4 lines · Nick Desaulniers

Interesting! Thanks for the links.
I'm guessing in our CI that we should probably pursue testing some of the newer revisions. Wasn't defconfig updated from z10 to z12 not too long ago?

Yes, that was 1.5 years ago (commit ac94a2911e84 ("s390: update defconfigs")).

Quoted 10 lines

So probably
CONFIG_MARCH_Z13
CONFIG_MARCH_Z14
CONFIG_MARCH_Z15
CONFIG_MARCH_Z16

All look like they're still "supported" (and I'm guessing CONFIG_MARCH_Z10 and CONFIG_MARCH_Z196 are not too much burden to continue to maintain kernel support for), with a higher emphasis perhaps on z15+z16?

That makes sense. However for the kernel it doesn't make a difference if compiled for z15 or z16 - there were no general instructions added with z16 that would make any difference for the kernel.
All new instructions that came with z16 are either only relevant for user space, or used via .insn notation in the kernel, and only used if some hardware feature is present.

author maintainer
#14 lore raw

PATCH 4/8 s390/entry: workaround llvm's IAS limitations

In reply to Heiko Carstens

llvm's integrated assembler cannot handle immediate values which are calculated with two local labels:

<instantiation>:3:13: error: invalid operand for instruction
clgfi %r14,.Lsie_done - .Lsie_gmap

Workaround this by adding clang specific code which reads the specific value from memory. Since this code is within the hot paths of the kernel and adds an additional memory reference, keep the original code, and add ifdef'ed code.

---
 arch/s390/kernel/entry.S | 11 +++++++++++
 1 file changed, 11 insertions(+)
arch/s390/kernel/entry.S +11 -0
index e1664b45090f..ff7a75078e93 100644
@@ -171,8 +171,19 @@ _LPP_OFFSET	= __LC_LPP
 	.macro OUTSIDE reg,start,end,outside_label
 	larl	%r14,\start
 	slgrk	%r14,\reg,%r14
+#ifdef CONFIG_CC_IS_CLANG
+	clgfrl	%r14,.Lrange_size\@
+#else
 	clgfi	%r14,\end - \start
+#endif
 	jhe	\outside_label
+#ifdef CONFIG_CC_IS_CLANG
+	.section .rodata, "a"
+	.align 4
+.Lrange_size\@:
+	.long	\end - \start
+	.previous
+#endif
 	.endm
 
 	.macro SIEEXIT
Signature · 1 line
2.32.0
#15 lore raw

Re: [PATCH 4/8] s390/entry: workaround llvm's IAS limitations

Hi Heiko,

Quoted 21 lines · Heiko Carstens

llvm's integrated assembler cannot handle immediate values which are calculated with two local labels:

<instantiation>:3:13: error: invalid operand for instruction
clgfi %r14,.Lsie_done - .Lsie_gmap

Workaround this by adding clang specific code which reads the specific value from memory. Since this code is within the hot paths of the kernel and adds an additional memory reference, keep the original code, and add ifdef'ed code.

---
 arch/s390/kernel/entry.S | 11 +++++++++++
 1 file changed, 11 insertions(+)
arch/s390/kernel/entry.S +0 -0
index e1664b45090f..ff7a75078e93 100644

arch/s390/kernel/entry.S _LPP_OFFSET = __LC_LPP · line 174

Quoted 4 lines
arch/s390/kernel/entry.S +1 -0
@@ -171,8 +171,19 @@ _LPP_OFFSET	= __LC_LPP
 	.macro OUTSIDE reg,start,end,outside_label
 	larl	%r14,\start
 	slgrk	%r14,\reg,%r14
+#ifdef CONFIG_CC_IS_CLANG

I intend to put this series through my build and boot test matrix later today but one fly by comment in the meantime. Should this be CONFIG_AS_IS_LLVM if this is an integrated assembler limitation, rather than a clang one?

Quoted 19 lines
arch/s390/kernel/entry.S +10 -0
+	clgfrl	%r14,.Lrange_size\@
+#else
 	clgfi	%r14,\end - \start
+#endif
 	jhe	\outside_label
+#ifdef CONFIG_CC_IS_CLANG
+	.section .rodata, "a"
+	.align 4
+.Lrange_size\@:
+	.long	\end - \start
+	.previous
+#endif
 	.endm
 
 	.macro SIEEXIT
Signature · 1 line
2.32.0

Cheers,
Nathan

author maintainer
#16 lore raw

Re: [PATCH 4/8] s390/entry: workaround llvm's IAS limitations

Quoted 19 lines · Nathan Chancellor

Hi Heiko,

Quoted 16 lines · Heiko Carstens

llvm's integrated assembler cannot handle immediate values which are calculated with two local labels:

<instantiation>:3:13: error: invalid operand for instruction
clgfi %r14,.Lsie_done - .Lsie_gmap

Workaround this by adding clang specific code which reads the specific value from memory. Since this code is within the hot paths of the kernel and adds an additional memory reference, keep the original code, and add ifdef'ed code.

---
 arch/s390/kernel/entry.S | 11 +++++++++++
 1 file changed, 11 insertions(+)
Quoted 14 lines

arch/s390/kernel/entry.S _LPP_OFFSET = __LC_LPP · line 174

Quoted 9 lines
arch/s390/kernel/entry.S +1 -0
index e1664b45090f..ff7a75078e93 100644
@@ -171,8 +171,19 @@ _LPP_OFFSET	= __LC_LPP
 	.macro OUTSIDE reg,start,end,outside_label
 	larl	%r14,\start
 	slgrk	%r14,\reg,%r14
+#ifdef CONFIG_CC_IS_CLANG

I intend to put this series through my build and boot test matrix later today but one fly by comment in the meantime. Should this be CONFIG_AS_IS_LLVM if this is an integrated assembler limitation, rather than a clang one?

Yes, that makes a lot of sense. Considering that I will drop the previous patch within this series, the new version looks like:

From fe4fb0b014378d84ae517deaea338577b2ea6ae0 Mon Sep 17 00:00:00 2001
From: Heiko Carstens <[email protected]>
Date: Sat, 7 May 2022 15:00:40 +0200
Subject: [PATCH 3/7] s390/entry: workaround llvm's IAS limitations

llvm's integrated assembler cannot handle immediate values which are calculated with two local labels:

<instantiation>:3:13: error: invalid operand for instruction
clgfi %r14,.Lsie_done - .Lsie_gmap

Workaround this by adding clang specific code which reads the specific value from memory. Since this code is within the hot paths of the kernel and adds an additional memory reference, keep the original code, and add ifdef'ed code.

---
 arch/s390/kernel/entry.S | 14 ++++++++++++--
 1 file changed, 12 insertions(+), 2 deletions(-)
arch/s390/kernel/entry.S +12 -2
index a6b45eaa3450..f2f30bfba1e9 100644
@@ -172,9 +172,19 @@ _LPP_OFFSET	= __LC_LPP
 	lgr	%r14,\reg
 	larl	%r13,\start
 	slgr	%r14,%r13
-	lghi	%r13,\end - \start
-	clgr	%r14,%r13
+#ifdef CONFIG_AS_IS_LLVM
+	clgfrl	%r14,.Lrange_size\@
+#else
+	clgfi	%r14,\end - \start
+#endif
 	jhe	\outside_label
+#ifdef CONFIG_CC_IS_CLANG
+	.section .rodata, "a"
+	.align 4
+.Lrange_size\@:
+	.long	\end - \start
+	.previous
+#endif
 	.endm
 
 	.macro SIEEXIT
Signature · 1 line
2.32.0
#17 lore raw

Re: [PATCH 4/8] s390/entry: workaround llvm's IAS limitations

Quoted 66 lines · Heiko Carstens
Quoted 19 lines · Nathan Chancellor

Hi Heiko,

Quoted 16 lines · Heiko Carstens

llvm's integrated assembler cannot handle immediate values which are calculated with two local labels:

<instantiation>:3:13: error: invalid operand for instruction
clgfi %r14,.Lsie_done - .Lsie_gmap

Workaround this by adding clang specific code which reads the specific value from memory. Since this code is within the hot paths of the kernel and adds an additional memory reference, keep the original code, and add ifdef'ed code.

---
 arch/s390/kernel/entry.S | 11 +++++++++++
 1 file changed, 11 insertions(+)
Quoted 14 lines

arch/s390/kernel/entry.S _LPP_OFFSET = __LC_LPP · line 174

Quoted 9 lines
arch/s390/kernel/entry.S +1 -0
index e1664b45090f..ff7a75078e93 100644
@@ -171,8 +171,19 @@ _LPP_OFFSET	= __LC_LPP
 	.macro OUTSIDE reg,start,end,outside_label
 	larl	%r14,\start
 	slgrk	%r14,\reg,%r14
+#ifdef CONFIG_CC_IS_CLANG

I intend to put this series through my build and boot test matrix later today but one fly by comment in the meantime. Should this be CONFIG_AS_IS_LLVM if this is an integrated assembler limitation, rather than a clang one?

Yes, that makes a lot of sense. Considering that I will drop the previous patch within this series, the new version looks like:

From fe4fb0b014378d84ae517deaea338577b2ea6ae0 Mon Sep 17 00:00:00 2001
From: Heiko Carstens <[email protected]>
Date: Sat, 7 May 2022 15:00:40 +0200
Subject: [PATCH 3/7] s390/entry: workaround llvm's IAS limitations

llvm's integrated assembler cannot handle immediate values which are calculated with two local labels:

<instantiation>:3:13: error: invalid operand for instruction
clgfi %r14,.Lsie_done - .Lsie_gmap

Workaround this by adding clang specific code which reads the specific value from memory. Since this code is within the hot paths of the kernel and adds an additional memory reference, keep the original code, and add ifdef'ed code.

---
 arch/s390/kernel/entry.S | 14 ++++++++++++--
 1 file changed, 12 insertions(+), 2 deletions(-)
arch/s390/kernel/entry.S +0 -0
index a6b45eaa3450..f2f30bfba1e9 100644
@@ -172,9 +172,19 @@ _LPP_OFFSET	= __LC_LPP
 	lgr	%r14,\reg
 	larl	%r13,\start

arch/s390/kernel/entry.S _LPP_OFFSET = __LC_LPP · line 181

Quoted 10 lines
arch/s390/kernel/entry.S +6 -2
 	slgr	%r14,%r13
-	lghi	%r13,\end - \start
-	clgr	%r14,%r13
+#ifdef CONFIG_AS_IS_LLVM
+	clgfrl	%r14,.Lrange_size\@
+#else
+	clgfi	%r14,\end - \start
+#endif
 	jhe	\outside_label
+#ifdef CONFIG_CC_IS_CLANG

I think this one also wants to be CONFIG_AS_IS_LLVM, right?

Other than that, seems fine to me, although I have no knowledge of s390 assembly so that statement probably means next to nothing :)

Cheers,
Nathan

Quoted 11 lines
arch/s390/kernel/entry.S +6 -0
+	.section .rodata, "a"
+	.align 4
+.Lrange_size\@:
+	.long	\end - \start
+	.previous
+#endif
 	.endm
 
 	.macro SIEEXIT
Signature · 1 line
2.32.0
author maintainer
#18 lore raw

Re: [PATCH 4/8] s390/entry: workaround llvm's IAS limitations

Quoted 9 lines · Nathan Chancellor

arch/s390/kernel/entry.S _LPP_OFFSET = __LC_LPP · line 181

Quoted 7 lines
arch/s390/kernel/entry.S +6 -0
+#ifdef CONFIG_AS_IS_LLVM
+	clgfrl	%r14,.Lrange_size\@
+#else
+	clgfi	%r14,\end - \start
+#endif
 	jhe	\outside_label
+#ifdef CONFIG_CC_IS_CLANG

I think this one also wants to be CONFIG_AS_IS_LLVM, right?

Yes, of course.

Quoted 2 lines

Other than that, seems fine to me, although I have no knowledge of s390 assembly so that statement probably means next to nothing :)

Thanks for having a look! :)

#19 lore raw

Re: [PATCH 4/8] s390/entry: workaround llvm's IAS limitations

In reply to Heiko Carstens
Quoted 36 lines · Heiko Carstens

From fe4fb0b014378d84ae517deaea338577b2ea6ae0 Mon Sep 17 00:00:00 2001
From: Heiko Carstens <[email protected]>
Date: Sat, 7 May 2022 15:00:40 +0200
Subject: [PATCH 3/7] s390/entry: workaround llvm's IAS limitations

llvm's integrated assembler cannot handle immediate values which are calculated with two local labels:

<instantiation>:3:13: error: invalid operand for instruction
clgfi %r14,.Lsie_done - .Lsie_gmap

Workaround this by adding clang specific code which reads the specific value from memory. Since this code is within the hot paths of the kernel and adds an additional memory reference, keep the original code, and add ifdef'ed code.

---
 arch/s390/kernel/entry.S | 14 ++++++++++++--
 1 file changed, 12 insertions(+), 2 deletions(-)
arch/s390/kernel/entry.S +2 -2
index a6b45eaa3450..f2f30bfba1e9 100644
@@ -172,9 +172,19 @@ _LPP_OFFSET	= __LC_LPP
 	lgr	%r14,\reg
 	larl	%r13,\start
 	slgr	%r14,%r13
-	lghi	%r13,\end - \start
-	clgr	%r14,%r13
+#ifdef CONFIG_AS_IS_LLVM
+	clgfrl	%r14,.Lrange_size\@
 
 
 

arch/s390/kernel/entry.S _LPP_OFFSET = __LC_LPP

Quoted 10 lines
arch/s390/kernel/entry.S +8 -0
 
+#else
+	clgfi	%r14,\end - \start
+#endif
 	jhe	\outside_label
+#ifdef CONFIG_CC_IS_CLANG
+	.section .rodata, "a"
+	.align 4
+.Lrange_size\@:
+	.long	\end - \start

Isn't the machine check handler refers to this memory before checking unrecoverable storage errors (with CHKSTG macro) as result of this change?

Quoted 7 lines
arch/s390/kernel/entry.S +2 -0
+	.previous
+#endif
 	.endm
 
 	.macro SIEEXIT
Signature · 1 line
2.32.0
author maintainer
#20 lore raw

Re: [PATCH 4/8] s390/entry: workaround llvm's IAS limitations

Quoted 16 lines · Alexander Gordeev
Quoted 16 lines
arch/s390/kernel/entry.S +5 -2
index a6b45eaa3450..f2f30bfba1e9 100644
@@ -172,9 +172,19 @@ _LPP_OFFSET	= __LC_LPP
 	lgr	%r14,\reg
 	larl	%r13,\start
 	slgr	%r14,%r13
-	lghi	%r13,\end - \start
-	clgr	%r14,%r13
+#ifdef CONFIG_AS_IS_LLVM
+	clgfrl	%r14,.Lrange_size\@
+#else
+	clgfi	%r14,\end - \start
+#endif
 	jhe	\outside_label
Quoted 8 lines

arch/s390/kernel/entry.S _LPP_OFFSET = __LC_LPP · line 185

Quoted 5 lines
arch/s390/kernel/entry.S +5 -0
+#ifdef CONFIG_CC_IS_CLANG
+	.section .rodata, "a"
+	.align 4
+.Lrange_size\@:
+	.long	\end - \start

Isn't the machine check handler refers to this memory before checking unrecoverable storage errors (with CHKSTG macro) as result of this change?

Yes, indeed. However implementing this without another register will be quite of a challenge. So what I would prefer in any case: just assume that this minimal set of memory accesses work. Actually I'd seriously like to go a bit further, and even move the checks for storage errors back to C for two reasons:

- this would make the machine check handler entry code easier again
- it would also allow to enter the machine check handler with DAT on

After all we rely anyway on the fact that at least the local lowcore + the page(s) which contain text are still accessible. Assuming that a couple of page tables also work won't make this much worse, but the code much easier.

So I'd suggest: leave this code as is, and at some later point move "rework" the early machine check handler code.

What do you think?

#21 lore raw

Re: [PATCH 4/8] s390/entry: workaround llvm's IAS limitations

I will try to get a patch for clang ready soon... /Jonas

Quoted 42 lines · Heiko Carstens
Quoted 15 lines · Alexander Gordeev
Quoted 15 lines
arch/s390/kernel/entry.S +5 -2
index a6b45eaa3450..f2f30bfba1e9 100644
@@ -172,9 +172,19 @@ _LPP_OFFSET	= __LC_LPP
  	lgr	%r14,\reg
  	larl	%r13,\start
  	slgr	%r14,%r13
-	lghi	%r13,\end - \start
-	clgr	%r14,%r13
+#ifdef CONFIG_AS_IS_LLVM
+	clgfrl	%r14,.Lrange_size\@
+#else
+	clgfi	%r14,\end - \start
+#endif
Quoted 8 lines

arch/s390/kernel/entry.S _LPP_OFFSET = __LC_LPP · line 185

Quoted 6 lines
arch/s390/kernel/entry.S +5 -0
  	jhe	\outside_label
+#ifdef CONFIG_CC_IS_CLANG
+	.section .rodata, "a"
+	.align 4
+.Lrange_size\@:
+	.long	\end - \start

Isn't the machine check handler refers to this memory before checking unrecoverable storage errors (with CHKSTG macro) as result of this change?

Yes, indeed. However implementing this without another register will be quite of a challenge. So what I would prefer in any case: just assume that this minimal set of memory accesses work. Actually I'd seriously like to go a bit further, and even move the checks for storage errors back to C for two reasons:

- this would make the machine check handler entry code easier again
- it would also allow to enter the machine check handler with DAT on

After all we rely anyway on the fact that at least the local lowcore + the page(s) which contain text are still accessible. Assuming that a couple of page tables also work won't make this much worse, but the code much easier.

So I'd suggest: leave this code as is, and at some later point move "rework" the early machine check handler code.

What do you think?

maintainer Acked-by
#22 lore raw

Re: [PATCH 4/8] s390/entry: workaround llvm's IAS limitations

In reply to Heiko Carstens
Quoted 18 lines · Heiko Carstens
Quoted 2 lines · Alexander Gordeev

Isn't the machine check handler refers to this memory before checking unrecoverable storage errors (with CHKSTG macro) as result of this change?

Yes, indeed. However implementing this without another register will be quite of a challenge. So what I would prefer in any case: just assume that this minimal set of memory accesses work. Actually I'd seriously like to go a bit further, and even move the checks for storage errors back to C for two reasons:

- this would make the machine check handler entry code easier again
- it would also allow to enter the machine check handler with DAT on

After all we rely anyway on the fact that at least the local lowcore + the page(s) which contain text are still accessible. Assuming that a couple of page tables also work won't make this much worse, but the code much easier.

Quoted 4 lines

So I'd suggest: leave this code as is, and at some later point move "rework" the early machine check handler code.

What do you think?

Sounds very reasonable. Please, find my:

Also, how such a follow-up looks to you?

	lgr	%r14,\reg
#ifdef CONFIG_AS_IS_LLVM
	larl	%r13,\start
	slgr	%r14,%r13
	clgfrl	%r14,.Lrange_size\@
#else
	slgfi	%r14,\start
	clgfi	%r14,\end - \start

#endif

author maintainer
#23 lore raw

Re: [PATCH 4/8] s390/entry: workaround llvm's IAS limitations

Quoted 8 lines · Alexander Gordeev
Quoted 4 lines

So I'd suggest: leave this code as is, and at some later point move "rework" the early machine check handler code.

What do you think?

Sounds very reasonable. Please, find my:

Thanks!

Quoted 1 line

Also, how such a follow-up looks to you?

...

Quoted 2 lines
	slgfi	%r14,\start
	clgfi	%r14,\end - \start

I think using an address as an immediate value is a step in the wrong direction, since I'd like to have all code pc-relative. And as far as I can tell this new construct would only work as long as \start has an absolute address that is low enough so that it would work / fit with slgfi.
Of course this will likely always be the case, but I still think this is not the way we should go.

author maintainer
#24 lore raw

PATCH 5/8 s390/purgatory: workaround llvm's IAS limitations

In reply to Heiko Carstens

llvm's integrated assembler cannot handle immediate values which are calculated with two local labels:

arch/s390/purgatory/head.S:139:11: error: invalid operand for instruction
aghi %r8,-(.base_crash-purgatory_start)

Workaround this by partially rewriting the code.

---
 arch/s390/purgatory/head.S | 29 ++++++++++++++++++++++-------
 1 file changed, 22 insertions(+), 7 deletions(-)
arch/s390/purgatory/head.S +22 -7
index 3d1c31e0cf3d..ac1a27a20b66 100644
@@ -44,11 +44,14 @@
 .endm
 
 .macro MEMSWAP dst,src,buf,len
-10:	cghi	\len,bufsz
+10:	larl	%r0,purgatory_end
+	larl	%r1,stack
+	slgr	%r0,%r1
+	cgr	\len,%r0
 	jh	11f
 	lgr	%r4,\len
 	j	12f
-11:	lghi	%r4,bufsz
+11:	lgr	%r4,%r0
 
 12:	MEMCPY	\buf,\dst,%r4
 	MEMCPY	\dst,\src,%r4
@@ -135,12 +138,18 @@ ENTRY(purgatory_start)
 
 .start_crash_kernel:
 	/* Location of purgatory_start in crash memory */
+	larl	%r0,.base_crash
+	larl	%r1,purgatory_start
+	slgr	%r0,%r1
 	lgr	%r8,%r13
-	aghi	%r8,-(.base_crash-purgatory_start)
+	sgr	%r8,%r0
 
 	/* Destination for this code i.e. end of memory to be swapped. */
+	larl	%r0,purgatory_end
+	larl	%r1,purgatory_start
+	slgr	%r0,%r1
 	lg	%r9,crash_size-.base_crash(%r13)
-	aghi	%r9,-(purgatory_end-purgatory_start)
+	sgr	%r9,%r0
 
 	/* Destination in crash memory, i.e. same as r9 but in crash memory. */
 	lg	%r10,crash_start-.base_crash(%r13)
@@ -149,15 +158,19 @@ ENTRY(purgatory_start)
 	/* Buffer location (in crash memory) and size. As the purgatory is
 	 * behind the point of no return it can re-use the stack as buffer.
 	 */
-	lghi	%r11,bufsz
+	larl	%r11,purgatory_end
 	larl	%r12,stack
+	slgr	%r11,%r12
 
 	MEMCPY	%r12,%r9,%r11	/* dst	-> (crash) buf */
 	MEMCPY	%r9,%r8,%r11	/* self -> dst */
 
 	/* Jump to new location. */
 	lgr	%r7,%r9
-	aghi	%r7,.jump_to_dst-purgatory_start
+	larl	%r0,.jump_to_dst
+	larl	%r1,purgatory_start
+	slgr	%r0,%r1
+	agr	%r7,%r0
 	br	%r7
 
 .jump_to_dst:
@@ -169,7 +182,9 @@ ENTRY(purgatory_start)
 
 	/* Load new buffer location after jump */
 	larl	%r7,stack
-	aghi	%r10,stack-purgatory_start
+	larl	%r0,purgatory_start
+	slgrk	%r0,%r7,%r0
+	agr	%r10,%r0
 	MEMCPY	%r10,%r7,%r11	/* (new) buf -> (crash) buf */
 
 	/* Now the code is set up to run from its designated location. Start
Signature · 1 line
2.32.0
#25 lore raw

Re: [PATCH 5/8] s390/purgatory: workaround llvm's IAS limitations

Quoted 6 lines · Heiko Carstens

llvm's integrated assembler cannot handle immediate values which are calculated with two local labels:

arch/s390/purgatory/head.S:139:11: error: invalid operand for instruction
aghi %r8,-(.base_crash-purgatory_start)

I thought this was fixed in
https://github.com/ClangBuiltLinux/linux/issues/1420
https://reviews.llvm.org/D113341
(clang-14)
?

Quoted 86 lines

Workaround this by partially rewriting the code.

---
 arch/s390/purgatory/head.S | 29 ++++++++++++++++++++++-------
 1 file changed, 22 insertions(+), 7 deletions(-)
arch/s390/purgatory/head.S +22 -8
index 3d1c31e0cf3d..ac1a27a20b66 100644
@@ -44,11 +44,14 @@
 .endm
 
 .macro MEMSWAP dst,src,buf,len
-10:    cghi    \len,bufsz
+10:    larl    %r0,purgatory_end
+       larl    %r1,stack
+       slgr    %r0,%r1
+       cgr     \len,%r0
        jh      11f
        lgr     %r4,\len
        j       12f
-11:    lghi    %r4,bufsz
+11:    lgr     %r4,%r0
 
 12:    MEMCPY  \buf,\dst,%r4
        MEMCPY  \dst,\src,%r4
@@ -135,12 +138,18 @@ ENTRY(purgatory_start)
 
 .start_crash_kernel:
        /* Location of purgatory_start in crash memory */
+       larl    %r0,.base_crash
+       larl    %r1,purgatory_start
+       slgr    %r0,%r1
        lgr     %r8,%r13
-       aghi    %r8,-(.base_crash-purgatory_start)
+       sgr     %r8,%r0
 
        /* Destination for this code i.e. end of memory to be swapped. */
+       larl    %r0,purgatory_end
+       larl    %r1,purgatory_start
+       slgr    %r0,%r1
        lg      %r9,crash_size-.base_crash(%r13)
-       aghi    %r9,-(purgatory_end-purgatory_start)
+       sgr     %r9,%r0
 
        /* Destination in crash memory, i.e. same as r9 but in crash memory. */
        lg      %r10,crash_start-.base_crash(%r13)
@@ -149,15 +158,19 @@ ENTRY(purgatory_start)
        /* Buffer location (in crash memory) and size. As the purgatory is
         * behind the point of no return it can re-use the stack as buffer.
         */
-       lghi    %r11,bufsz
+       larl    %r11,purgatory_end
        larl    %r12,stack
+       slgr    %r11,%r12
 
        MEMCPY  %r12,%r9,%r11   /* dst  -> (crash) buf */
        MEMCPY  %r9,%r8,%r11    /* self -> dst */
 
        /* Jump to new location. */
        lgr     %r7,%r9
-       aghi    %r7,.jump_to_dst-purgatory_start
+       larl    %r0,.jump_to_dst
+       larl    %r1,purgatory_start
+       slgr    %r0,%r1
+       agr     %r7,%r0
        br      %r7
 
 .jump_to_dst:
@@ -169,7 +182,9 @@ ENTRY(purgatory_start)
 
        /* Load new buffer location after jump */
        larl    %r7,stack
-       aghi    %r10,stack-purgatory_start
+       larl    %r0,purgatory_start
+       slgrk   %r0,%r7,%r0
+       agr     %r10,%r0
        MEMCPY  %r10,%r7,%r11   /* (new) buf -> (crash) buf */
 
        /* Now the code is set up to run from its designated location. Start
--

2.32.0

Signature · 2 lines
Thanks,
~Nick Desaulniers
author maintainer
#26 lore raw

Re: [PATCH 5/8] s390/purgatory: workaround llvm's IAS limitations

Quoted 5 lines · Nick Desaulniers
Quoted 4 lines · Heiko Carstens

llvm's integrated assembler cannot handle immediate values which are calculated with two local labels:

Quoted 8 lines
Quoted 2 lines

arch/s390/purgatory/head.S:139:11: error: invalid operand for instruction
aghi %r8,-(.base_crash-purgatory_start)

I thought this was fixed in
https://github.com/ClangBuiltLinux/linux/issues/1420
https://reviews.llvm.org/D113341
(clang-14)
?

Looks like the referenced fix only works for displacements, bit not for immediates.

author maintainer
#27 lore raw

Re: [PATCH 5/8] s390/purgatory: workaround llvm's IAS limitations

In reply to Heiko Carstens
Quoted 12 lines · Heiko Carstens

llvm's integrated assembler cannot handle immediate values which are calculated with two local labels:

arch/s390/purgatory/head.S:139:11: error: invalid operand for instruction
aghi %r8,-(.base_crash-purgatory_start)

Workaround this by partially rewriting the code.

---
 arch/s390/purgatory/head.S | 29 ++++++++++++++++++++++-------
 1 file changed, 22 insertions(+), 7 deletions(-)

...

Review

Quoted 4 lines
(quoted diff) +3 -1
-	aghi	%r10,stack-purgatory_start
+	larl	%r0,purgatory_start
+	slgrk	%r0,%r7,%r0
+	agr	%r10,%r0

Same problem here: slgrk does not work for MARCH=z10. So new version looks like:

From 99e011fa48d9fe90f6d3d58cf89f738afddf394f Mon Sep 17 00:00:00 2001
From: Heiko Carstens <[email protected]>
Date: Sat, 7 May 2022 20:08:55 +0200
Subject: [PATCH 4/7] s390/purgatory: workaround llvm's IAS limitations

llvm's integrated assembler cannot handle immediate values which are calculated with two local labels:

arch/s390/purgatory/head.S:139:11: error: invalid operand for instruction
aghi %r8,-(.base_crash-purgatory_start)

Workaround this by partially rewriting the code.

---
 arch/s390/purgatory/head.S | 30 +++++++++++++++++++++++-------
 1 file changed, 23 insertions(+), 7 deletions(-)
arch/s390/purgatory/head.S +23 -7
index 3d1c31e0cf3d..6f835124ee82 100644
@@ -44,11 +44,14 @@
 .endm
 
 .macro MEMSWAP dst,src,buf,len
-10:	cghi	\len,bufsz
+10:	larl	%r0,purgatory_end
+	larl	%r1,stack
+	slgr	%r0,%r1
+	cgr	\len,%r0
 	jh	11f
 	lgr	%r4,\len
 	j	12f
-11:	lghi	%r4,bufsz
+11:	lgr	%r4,%r0
 
 12:	MEMCPY	\buf,\dst,%r4
 	MEMCPY	\dst,\src,%r4
@@ -135,12 +138,18 @@ ENTRY(purgatory_start)
 
 .start_crash_kernel:
 	/* Location of purgatory_start in crash memory */
+	larl	%r0,.base_crash
+	larl	%r1,purgatory_start
+	slgr	%r0,%r1
 	lgr	%r8,%r13
-	aghi	%r8,-(.base_crash-purgatory_start)
+	sgr	%r8,%r0
 
 	/* Destination for this code i.e. end of memory to be swapped. */
+	larl	%r0,purgatory_end
+	larl	%r1,purgatory_start
+	slgr	%r0,%r1
 	lg	%r9,crash_size-.base_crash(%r13)
-	aghi	%r9,-(purgatory_end-purgatory_start)
+	sgr	%r9,%r0
 
 	/* Destination in crash memory, i.e. same as r9 but in crash memory. */
 	lg	%r10,crash_start-.base_crash(%r13)
@@ -149,15 +158,19 @@ ENTRY(purgatory_start)
 	/* Buffer location (in crash memory) and size. As the purgatory is
 	 * behind the point of no return it can re-use the stack as buffer.
 	 */
-	lghi	%r11,bufsz
+	larl	%r11,purgatory_end
 	larl	%r12,stack
+	slgr	%r11,%r12
 
 	MEMCPY	%r12,%r9,%r11	/* dst	-> (crash) buf */
 	MEMCPY	%r9,%r8,%r11	/* self -> dst */
 
 	/* Jump to new location. */
 	lgr	%r7,%r9
-	aghi	%r7,.jump_to_dst-purgatory_start
+	larl	%r0,.jump_to_dst
+	larl	%r1,purgatory_start
+	slgr	%r0,%r1
+	agr	%r7,%r0
 	br	%r7
 
 .jump_to_dst:
@@ -169,7 +182,10 @@ ENTRY(purgatory_start)
 
 	/* Load new buffer location after jump */
 	larl	%r7,stack
-	aghi	%r10,stack-purgatory_start
+	lgr	%r0,%r7
+	larl	%r1,purgatory_start
+	slgr	%r0,%r1
+	agr	%r10,%r0
 	MEMCPY	%r10,%r7,%r11	/* (new) buf -> (crash) buf */
 
 	/* Now the code is set up to run from its designated location. Start
Signature · 1 line
2.32.0
author maintainer
#28 lore raw

PATCH 6/8 s390/boot: workaround llvm IAS bug

In reply to Heiko Carstens

For at least the mvc and clc instructions llvm's integrated assembler can generate incorrect code. In particular this happens with decompressor boot code. The reason seems to be that relocations for the second displacement of each instruction are at incorrect locations (-/+: gas vs llvm IAS):

mvc __LC_IO_NEW_PSW(16),.Lnewpsw

results in

        4:      d2 0f 01 f0 00 00       mvc     496(16,%r0),0
-                       8: R_390_12     .head.text+0x10
+                       6: R_390_12     .head.text+0x10

and
clc 0(3,%r4),.L_hdr
results in

      258:      d5 02 40 00 00 00       clc     0(3,%r4),0
-                       25c: R_390_12   .head.text+0x324
+                       25a: R_390_12   .head.text+0x324

Workaround this by writing the code in a different way.

---
 arch/s390/boot/head.S | 34 +++++++++++++++++++++-------------
 1 file changed, 21 insertions(+), 13 deletions(-)
arch/s390/boot/head.S +21 -13
index 2ced90172680..8402e1cd133b 100644
@@ -42,7 +42,8 @@ ipl_start:
 # subroutine to wait for end I/O
 #
 .Lirqwait:
-	mvc	__LC_IO_NEW_PSW(16),.Lnewpsw	# set up IO interrupt psw
+	larl	%r13,.Lnewpsw		# set up IO interrupt psw
+	mvc	__LC_IO_NEW_PSW(16),0(%r13)
 	lpsw	.Lwaitpsw
 .Lioint:
 	br	%r14
@@ -155,9 +156,11 @@ ipl_start:
 	lr	%r2,%r3
 .Lnotrunc:
 	l	%r4,.Linitrd
-	clc	0(3,%r4),.L_hdr		# if it is HDRx
+	larl	%r13,.L_hdr
+	clc	0(3,%r4),0(%r13)	# if it is HDRx
 	bz	.Lagain1		# skip dataset header
-	clc	0(3,%r4),.L_eof		# if it is EOFx
+	larl	%r13,.L_eof
+	clc	0(3,%r4),0(%r13)	# if it is EOFx
 	bz	.Lagain1		# skip dateset trailer
 
 	lr	%r5,%r2
@@ -181,9 +184,11 @@ ipl_start:
 .Lrdcont:
 	l	%r2,.Linitrd
 
-	clc	0(3,%r2),.L_hdr		# skip HDRx and EOFx
+	larl	%r13,.L_hdr		# skip HDRx and EOFx
+	clc	0(3,%r2),0(%r13)
 	bz	.Lagain2
-	clc	0(3,%r2),.L_eof
+	larl	%r13,.L_eof
+	clc	0(3,%r2),0(%r13)
 	bz	.Lagain2
 
 #
@@ -260,20 +265,23 @@ SYM_CODE_START_LOCAL(startup_normal)
 	.fill	16,4,0x0
 0:	lmh	%r0,%r15,0(%r13)	# clear high-order half of gprs
 	sam64				# switch to 64 bit addressing mode
-	basr	%r13,0			# get base
-.LPG0:
-	mvc	__LC_EXT_NEW_PSW(16),.Lext_new_psw-.LPG0(%r13)
-	mvc	__LC_PGM_NEW_PSW(16),.Lpgm_new_psw-.LPG0(%r13)
-	mvc	__LC_IO_NEW_PSW(16),.Lio_new_psw-.LPG0(%r13)
+	larl	%r13,.Lext_new_psw
+	mvc	__LC_EXT_NEW_PSW(16),0(%r13)
+	larl	%r13,.Lpgm_new_psw
+	mvc	__LC_PGM_NEW_PSW(16),0(%r13)
+	larl	%r13,.Lio_new_psw
+	mvc	__LC_IO_NEW_PSW(16),0(%r13)
 	xc	0x200(256),0x200	# partially clear lowcore
 	xc	0x300(256),0x300
 	xc	0xe00(256),0xe00
 	xc	0xf00(256),0xf00
-	lctlg	%c0,%c15,.Lctl-.LPG0(%r13)	# load control registers
+	larl	%r13,.Lctl
+	lctlg	%c0,%c15,0(%r13)	# load control registers
 	stcke	__LC_BOOT_CLOCK
 	mvc	__LC_LAST_UPDATE_CLOCK(8),__LC_BOOT_CLOCK+1
-	spt	6f-.LPG0(%r13)
-	mvc	__LC_LAST_UPDATE_TIMER(8),6f-.LPG0(%r13)
+	larl	%r13,6f
+	spt	0(%r13)
+	mvc	__LC_LAST_UPDATE_TIMER(8),0(%r13)
 	larl	%r15,_stack_end-STACK_FRAME_OVERHEAD
 	brasl	%r14,sclp_early_setup_buffer
 	brasl	%r14,verify_facilities
Signature · 1 line
2.32.0
#29 lore raw

Re: [PATCH 6/8] s390/boot: workaround llvm IAS bug

Quoted 22 lines · Heiko Carstens

For at least the mvc and clc instructions llvm's integrated assembler can generate incorrect code. In particular this happens with decompressor boot code. The reason seems to be that relocations for the second displacement of each instruction are at incorrect locations (-/+: gas vs llvm IAS):

mvc __LC_IO_NEW_PSW(16),.Lnewpsw

results in

        4:      d2 0f 01 f0 00 00       mvc     496(16,%r0),0
(quoted diff) +1 -1
-                       8: R_390_12     .head.text+0x10
+                       6: R_390_12     .head.text+0x10

and
clc 0(3,%r4),.L_hdr
results in

      258:      d5 02 40 00 00 00       clc     0(3,%r4),0
(quoted diff) +1 -1
-                       25c: R_390_12   .head.text+0x324
+                       25a: R_390_12   .head.text+0x324
Quoted 3 lines

Workaround this by writing the code in a different way.

Please link to an LLVM bugreport for this.

Quoted 81 lines
---
 arch/s390/boot/head.S | 34 +++++++++++++++++++++-------------
 1 file changed, 21 insertions(+), 13 deletions(-)
arch/s390/boot/head.S +21 -14
index 2ced90172680..8402e1cd133b 100644
@@ -42,7 +42,8 @@ ipl_start:
 # subroutine to wait for end I/O
 #
 .Lirqwait:
-       mvc     __LC_IO_NEW_PSW(16),.Lnewpsw    # set up IO interrupt psw
+       larl    %r13,.Lnewpsw           # set up IO interrupt psw
+       mvc     __LC_IO_NEW_PSW(16),0(%r13)
        lpsw    .Lwaitpsw
 .Lioint:
        br      %r14
@@ -155,9 +156,11 @@ ipl_start:
        lr      %r2,%r3
 .Lnotrunc:
        l       %r4,.Linitrd
-       clc     0(3,%r4),.L_hdr         # if it is HDRx
+       larl    %r13,.L_hdr
+       clc     0(3,%r4),0(%r13)        # if it is HDRx
        bz      .Lagain1                # skip dataset header
-       clc     0(3,%r4),.L_eof         # if it is EOFx
+       larl    %r13,.L_eof
+       clc     0(3,%r4),0(%r13)        # if it is EOFx
        bz      .Lagain1                # skip dateset trailer
 
        lr      %r5,%r2
@@ -181,9 +184,11 @@ ipl_start:
 .Lrdcont:
        l       %r2,.Linitrd
 
-       clc     0(3,%r2),.L_hdr         # skip HDRx and EOFx
+       larl    %r13,.L_hdr             # skip HDRx and EOFx
+       clc     0(3,%r2),0(%r13)
        bz      .Lagain2
-       clc     0(3,%r2),.L_eof
+       larl    %r13,.L_eof
+       clc     0(3,%r2),0(%r13)
        bz      .Lagain2
 
 #
@@ -260,20 +265,23 @@ SYM_CODE_START_LOCAL(startup_normal)
        .fill   16,4,0x0
 0:     lmh     %r0,%r15,0(%r13)        # clear high-order half of gprs
        sam64                           # switch to 64 bit addressing mode
-       basr    %r13,0                  # get base
-.LPG0:
-       mvc     __LC_EXT_NEW_PSW(16),.Lext_new_psw-.LPG0(%r13)
-       mvc     __LC_PGM_NEW_PSW(16),.Lpgm_new_psw-.LPG0(%r13)
-       mvc     __LC_IO_NEW_PSW(16),.Lio_new_psw-.LPG0(%r13)
+       larl    %r13,.Lext_new_psw
+       mvc     __LC_EXT_NEW_PSW(16),0(%r13)
+       larl    %r13,.Lpgm_new_psw
+       mvc     __LC_PGM_NEW_PSW(16),0(%r13)
+       larl    %r13,.Lio_new_psw
+       mvc     __LC_IO_NEW_PSW(16),0(%r13)
        xc      0x200(256),0x200        # partially clear lowcore
        xc      0x300(256),0x300
        xc      0xe00(256),0xe00
        xc      0xf00(256),0xf00
-       lctlg   %c0,%c15,.Lctl-.LPG0(%r13)      # load control registers
+       larl    %r13,.Lctl
+       lctlg   %c0,%c15,0(%r13)        # load control registers
        stcke   __LC_BOOT_CLOCK
        mvc     __LC_LAST_UPDATE_CLOCK(8),__LC_BOOT_CLOCK+1
-       spt     6f-.LPG0(%r13)
-       mvc     __LC_LAST_UPDATE_TIMER(8),6f-.LPG0(%r13)
+       larl    %r13,6f
+       spt     0(%r13)
+       mvc     __LC_LAST_UPDATE_TIMER(8),0(%r13)
        larl    %r15,_stack_end-STACK_FRAME_OVERHEAD
        brasl   %r14,sclp_early_setup_buffer
        brasl   %r14,verify_facilities
--

2.32.0

Signature · 2 lines
Thanks,
~Nick Desaulniers
author maintainer
#30 lore raw

PATCH 7/8 s390/boot: do not emit debug info for assembly with llvm's IAS

In reply to Heiko Carstens

Commit ee6d777d3e93 ("s390/decompressor: support extra debug flags") added extra debug flags, in particular debug info is created, depending on config options.

With llvm's IAS this causes this compile warning:

arch/s390/boot/head.S:38:1: warning: DWARF2 only supports one section per compilation unit .section ".head.text","ax"
^

This is a known problem and was addressed with a commit b8a9092330da ("Kbuild: do not emit debug info for assembly with LLVM_IAS=1"). Just do the same for s390 to get rid of this warning.

---
 arch/s390/Makefile | 2 ++
 1 file changed, 2 insertions(+)
arch/s390/Makefile +2 -0
index c59efc83f020..d73611b35164 100644
@@ -20,7 +20,9 @@ LDFLAGS_vmlinux	:= -pie
 endif
 aflags_dwarf	:= -Wa,-gdwarf-2
 KBUILD_AFLAGS_DECOMPRESSOR := $(CLANG_FLAGS) -m64 -D__ASSEMBLY__
+ifndef CONFIG_AS_IS_LLVM
 KBUILD_AFLAGS_DECOMPRESSOR += $(if $(CONFIG_DEBUG_INFO),$(aflags_dwarf))
+endif
 KBUILD_CFLAGS_DECOMPRESSOR := $(CLANG_FLAGS) -m64 -O2 -mpacked-stack
 KBUILD_CFLAGS_DECOMPRESSOR += -DDISABLE_BRANCH_PROFILING -D__NO_FORTIFY
 KBUILD_CFLAGS_DECOMPRESSOR += -fno-delete-null-pointer-checks -msoft-float -mbackchain
Signature · 1 line
2.32.0
#31 lore raw

Re: [PATCH 7/8] s390/boot: do not emit debug info for assembly with llvm's IAS

Quoted 25 lines · Heiko Carstens

Commit ee6d777d3e93 ("s390/decompressor: support extra debug flags") added extra debug flags, in particular debug info is created, depending on config options.

With llvm's IAS this causes this compile warning:

arch/s390/boot/head.S:38:1: warning: DWARF2 only supports one section per compilation unit .section ".head.text","ax"
^

This is a known problem and was addressed with a commit b8a9092330da ("Kbuild: do not emit debug info for assembly with LLVM_IAS=1"). Just do the same for s390 to get rid of this warning.

---
 arch/s390/Makefile | 2 ++
 1 file changed, 2 insertions(+)
arch/s390/Makefile +0 -0
index c59efc83f020..d73611b35164 100644

arch/s390/Makefile LDFLAGS_vmlinux := -pie · line 21

Quoted 2 lines
arch/s390/Makefile +0 -0
@@ -20,7 +20,9 @@ LDFLAGS_vmlinux       := -pie
 endif
 aflags_dwarf   := -Wa,-gdwarf-2

^ or can we use a more modern variant of dwarf, like at least dwarf-4?

Quoted 10 lines
arch/s390/Makefile +2 -1
 KBUILD_AFLAGS_DECOMPRESSOR := $(CLANG_FLAGS) -m64 -D__ASSEMBLY__
+ifndef CONFIG_AS_IS_LLVM
 KBUILD_AFLAGS_DECOMPRESSOR += $(if $(CONFIG_DEBUG_INFO),$(aflags_dwarf))
+endif
 KBUILD_CFLAGS_DECOMPRESSOR := $(CLANG_FLAGS) -m64 -O2 -mpacked-stack
 KBUILD_CFLAGS_DECOMPRESSOR += -DDISABLE_BRANCH_PROFILING -D__NO_FORTIFY
 KBUILD_CFLAGS_DECOMPRESSOR += -fno-delete-null-pointer-checks -msoft-float -mbackchain
--

2.32.0

Signature · 2 lines
Thanks,
~Nick Desaulniers
author maintainer
#32 lore raw

Re: [PATCH 7/8] s390/boot: do not emit debug info for assembly with llvm's IAS

Quoted 21 lines · Nick Desaulniers
Quoted 20 lines · Heiko Carstens

Commit ee6d777d3e93 ("s390/decompressor: support extra debug flags") added extra debug flags, in particular debug info is created, depending on config options.

With llvm's IAS this causes this compile warning:

arch/s390/boot/head.S:38:1: warning: DWARF2 only supports one section per compilation unit .section ".head.text","ax"
^

This is a known problem and was addressed with a commit b8a9092330da ("Kbuild: do not emit debug info for assembly with LLVM_IAS=1"). Just do the same for s390 to get rid of this warning.

---
 arch/s390/Makefile | 2 ++
 1 file changed, 2 insertions(+)
Quoted 9 lines

arch/s390/Makefile LDFLAGS_vmlinux := -pie · line 21

Quoted 7 lines
arch/s390/Makefile +0 -0
index c59efc83f020..d73611b35164 100644
@@ -20,7 +20,9 @@ LDFLAGS_vmlinux       := -pie
 endif
 aflags_dwarf   := -Wa,-gdwarf-2

^ or can we use a more modern variant of dwarf, like at least dwarf-4?

For now I would like that this works the same like what you introduced with commit b8a9092330da ("Kbuild: do not emit debug info for assembly with LLVM_IAS=1").

I wouldn't like to have dwarf-4 enabled only for this single file, and disabled for all other asm files (assuming that dwarf-4 works).

author maintainer
#33 lore raw

PATCH 8/8 scripts/min-tool-version.sh: raise minimum clang version to 14.0.0 for s390

In reply to Heiko Carstens

Before version 14.0.0 llvm's integrated assembler fails to handle some displacement variants:

arch/s390/purgatory/head.S:108:10: error: invalid operand for instruction
lg %r11,kernel_type-.base_crash(%r13)

Instead of working around this and given that this is already fixed raise the minimum clang version from 13.0.0 to 14.0.0.

---
 scripts/min-tool-version.sh | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)
scripts/min-tool-version.sh +1 -2
index 53fe64856015..f1e8358ec19a 100755
@@ -24,9 +24,8 @@ icc)
 	echo 16.0.3
 	;;
 llvm)
-	# https://lore.kernel.org/r/YMtib5hKVyNknZt3@osiris/
 	if [ "$SRCARCH" = s390 ]; then
-		echo 13.0.0
+		echo 14.0.0
 	else
 		echo 11.0.0
 	fi
Signature · 1 line
2.32.0
Acked-by on v1 8/8
#34 lore raw

Re: [PATCH 8/8] scripts/min-tool-version.sh: raise minimum clang version to 14.0.0 for s390

Quoted 9 lines · Heiko Carstens

Before version 14.0.0 llvm's integrated assembler fails to handle some displacement variants:

arch/s390/purgatory/head.S:108:10: error: invalid operand for instruction
lg %r11,kernel_type-.base_crash(%r13)

Instead of working around this and given that this is already fixed raise the minimum clang version from 13.0.0 to 14.0.0.

Do you have the commit in LLVM that fixed this? Might be nice to link to the particular commit in the commit message. Either way:

(Thanks for the series, will pull down and test!)

If you have a github account, let me know it if you'd like to be cc'ed when we wire this up in our CI.

Quoted 24 lines
---
 scripts/min-tool-version.sh | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)
scripts/min-tool-version.sh +1 -3
index 53fe64856015..f1e8358ec19a 100755
@@ -24,9 +24,8 @@ icc)
        echo 16.0.3
        ;;
 llvm)
-       # https://lore.kernel.org/r/YMtib5hKVyNknZt3@osiris/
        if [ "$SRCARCH" = s390 ]; then
-               echo 13.0.0
+               echo 14.0.0
        else
                echo 11.0.0
        fi
--

2.32.0

Signature · 2 lines
Thanks,
~Nick Desaulniers
#35 lore raw

Re: [PATCH 8/8] scripts/min-tool-version.sh: raise minimum clang version to 14.0.0 for s390

Quoted 6 lines · Nick Desaulniers
Quoted 4 lines · Heiko Carstens

Before version 14.0.0 llvm's integrated assembler fails to handle some displacement variants:

Quoted 7 lines
Quoted 5 lines

arch/s390/purgatory/head.S:108:10: error: invalid operand for instruction
lg %r11,kernel_type-.base_crash(%r13)

Instead of working around this and given that this is already fixed raise the minimum clang version from 13.0.0 to 14.0.0.

Do you have the commit in LLVM that fixed this? Might be nice to link

Maybe it's
https://reviews.llvm.org/D113341?

Also, these are the open issues we had for the integrated assembler.
https://github.com/ClangBuiltLinux/linux/issues?q=is%3Aopen+is%3Aissue+label%3A%22%5BARCH%5D+s390%22+label%3A%22%5BTOOL%5D+integrated-as%22

Any chance you could include relevant Link tags on your commit messages for patches that address these? It makes it easier to track when/where things land if we ever intend to backport anything to stable.

Or can any of those be closed out?

Quoted 37 lines

to the particular commit in the commit message. Either way:

(Thanks for the series, will pull down and test!)

If you have a github account, let me know it if you'd like to be cc'ed when we wire this up in our CI.

Quoted 22 lines
---
 scripts/min-tool-version.sh | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)
scripts/min-tool-version.sh +1 -3
index 53fe64856015..f1e8358ec19a 100755
@@ -24,9 +24,8 @@ icc)
        echo 16.0.3
        ;;
 llvm)
-       # https://lore.kernel.org/r/YMtib5hKVyNknZt3@osiris/
        if [ "$SRCARCH" = s390 ]; then
-               echo 13.0.0
+               echo 14.0.0
        else
                echo 11.0.0
        fi
--
Quoted 2 lines

2.32.0

--
Thanks,
~Nick Desaulniers

Signature · 2 lines
Thanks,
~Nick Desaulniers
author maintainer
#36 lore raw

Re: [PATCH 8/8] scripts/min-tool-version.sh: raise minimum clang version to 14.0.0 for s390

Quoted 11 lines · Nick Desaulniers
Quoted 9 lines · Nick Desaulniers
Quoted 7 lines · Heiko Carstens

Before version 14.0.0 llvm's integrated assembler fails to handle some displacement variants:

arch/s390/purgatory/head.S:108:10: error: invalid operand for instruction
lg %r11,kernel_type-.base_crash(%r13)

Quoted 7 lines
Quoted 4 lines
Quoted 2 lines

Instead of working around this and given that this is already fixed raise the minimum clang version from 13.0.0 to 14.0.0.

Do you have the commit in LLVM that fixed this? Might be nice to link

Maybe it's
https://reviews.llvm.org/D113341?

Yes, looks like it.

Quoted 7 lines

Also, these are the open issues we had for the integrated assembler.
https://github.com/ClangBuiltLinux/linux/issues?q=is%3Aopen+is%3Aissue+label%3A%22%5BARCH%5D+s390%22+label%3A%22%5BTOOL%5D+integrated-as%22

Any chance you could include relevant Link tags on your commit messages for patches that address these? It makes it easier to track when/where things land if we ever intend to backport anything to stable.

Sure, will do.

Tested-by on the v1 series
#37 lore raw
In reply to Heiko Carstens
Quoted 5 lines · Heiko Carstens

A couple of patches which in result make it finally possible to build the kernel for s390 with llvm's integrated assembler. Several configs build without errors or warnings, and the kernel also works as expected.

Quoted 7 lines

Note that patch 6 ("s390/boot: workaround llvm IAS bug") reveals a miscompile. This looks like a bug in the instruction definitions of the mvc and clc instructions(?). I'd like to ask people to look into this, since this silently generated broken code.

This patch series is based on linux-next, which contains two additional required s390 specific patches to make llvm's IAS work.

I did a quick test of just a defconfig via:

$ ARCH=s390 CROSS_COMPILE=s390x-linux-gnu- make CC=clang -j72 defconfig all
and this assembled then booted in qemu for me. Thanks for the work
that went into this!

Sounds like Nathan is doing additional testing as well.

Quoted 29 lines

Thanks,
Heiko

Heiko Carstens (8):
s390/alternatives: provide identical sized orginal/alternative sequences s390/alternatives: remove padding generation code
s390/entry: shorten OUTSIDE macro
s390/entry: workaround llvm's IAS limitations
s390/purgatory: workaround llvm's IAS limitations
s390/boot: workaround llvm IAS bug
s390/boot: do not emit debug info for assembly with llvm's IAS scripts/min-tool-version.sh: raise minimum clang version to 14.0.0 for s390

 arch/s390/Makefile                      |  2 +
 arch/s390/boot/head.S                   | 34 +++++----
 arch/s390/include/asm/alternative-asm.h | 76 +++-----------------
 arch/s390/include/asm/alternative.h     | 93 ++++++-------------------
 arch/s390/include/asm/spinlock.h        |  2 +-
 arch/s390/kernel/alternative.c          | 61 +---------------
 arch/s390/kernel/entry.S                | 39 +++++++----
 arch/s390/lib/spinlock.c                |  4 +-
 arch/s390/purgatory/head.S              | 29 ++++++--
 scripts/min-tool-version.sh             |  3 +-
 10 files changed, 104 insertions(+), 239 deletions(-)

--
2.32.0

Signature · 2 lines
Thanks,
~Nick Desaulniers
author maintainer
#38 lore raw
Quoted 14 lines · Nick Desaulniers
Quoted 5 lines · Heiko Carstens

A couple of patches which in result make it finally possible to build the kernel for s390 with llvm's integrated assembler. Several configs build without errors or warnings, and the kernel also works as expected.

Quoted 7 lines

Note that patch 6 ("s390/boot: workaround llvm IAS bug") reveals a miscompile. This looks like a bug in the instruction definitions of the mvc and clc instructions(?). I'd like to ask people to look into this, since this silently generated broken code.

This patch series is based on linux-next, which contains two additional required s390 specific patches to make llvm's IAS work.

Quoted 6 lines

I did a quick test of just a defconfig via:

$ ARCH=s390 CROSS_COMPILE=s390x-linux-gnu- make CC=clang -j72 defconfig all
and this assembled then booted in qemu for me. Thanks for the work
that went into this!

Will add this too. Thank you!

Tested-by on the v1 series
#39 lore raw
In reply to Heiko Carstens

Hi Heiko,

Quoted 8 lines · Heiko Carstens

A couple of patches which in result make it finally possible to build the kernel for s390 with llvm's integrated assembler. Several configs build without errors or warnings, and the kernel also works as expected.

Note that patch 6 ("s390/boot: workaround llvm IAS bug") reveals a miscompile. This looks like a bug in the instruction definitions of the mvc and clc instructions(?). I'd like to ask people to look into this, since this silently generated broken code.

I think it should be pretty simple to file a bug report for this since it occurs in a standalone assembly file? I agree with Nick that there should be a bug report filed and linked to in patch 6 so that we don't lose track of it.

Quoted 16 lines

This patch series is based on linux-next, which contains two additional required s390 specific patches to make llvm's IAS work.

Thanks,
Heiko

Heiko Carstens (8):
s390/alternatives: provide identical sized orginal/alternative sequences s390/alternatives: remove padding generation code
s390/entry: shorten OUTSIDE macro
s390/entry: workaround llvm's IAS limitations
s390/purgatory: workaround llvm's IAS limitations
s390/boot: workaround llvm IAS bug
s390/boot: do not emit debug info for assembly with llvm's IAS scripts/min-tool-version.sh: raise minimum clang version to 14.0.0 for s390

Quoted 11 lines
 arch/s390/Makefile                      |  2 +
 arch/s390/boot/head.S                   | 34 +++++----
 arch/s390/include/asm/alternative-asm.h | 76 +++-----------------
 arch/s390/include/asm/alternative.h     | 93 ++++++-------------------
 arch/s390/include/asm/spinlock.h        |  2 +-
 arch/s390/kernel/alternative.c          | 61 +---------------
 arch/s390/kernel/entry.S                | 39 +++++++----
 arch/s390/lib/spinlock.c                |  4 +-
 arch/s390/purgatory/head.S              | 29 ++++++--
 scripts/min-tool-version.sh             |  3 +-
 10 files changed, 104 insertions(+), 239 deletions(-)

I applied this series to the latest s390 for-next branch (c4fb15578802) and built a few in-tree and distribution configurations with clang-14 and clang-15 then boot tested them in QEMU with a simple buildroot userspace. I did not see any new warnings or errors. This is awesome, I am excited to get this wired up in our CI!

In case it is worthwhile:

Cheers,
Nathan

author maintainer
#40 lore raw
Quoted 12 lines · Nathan Chancellor

Hi Heiko,

Quoted 8 lines · Heiko Carstens

A couple of patches which in result make it finally possible to build the kernel for s390 with llvm's integrated assembler. Several configs build without errors or warnings, and the kernel also works as expected.

Note that patch 6 ("s390/boot: workaround llvm IAS bug") reveals a miscompile. This looks like a bug in the instruction definitions of the mvc and clc instructions(?). I'd like to ask people to look into this, since this silently generated broken code.

Quoted 4 lines

I think it should be pretty simple to file a bug report for this since it occurs in a standalone assembly file? I agree with Nick that there should be a bug report filed and linked to in patch 6 so that we don't lose track of it.

https://github.com/llvm/llvm-project/issues/55411

Quoted 9 lines

I applied this series to the latest s390 for-next branch (c4fb15578802) and built a few in-tree and distribution configurations with clang-14 and clang-15 then boot tested them in QEMU with a simple buildroot userspace. I did not see any new warnings or errors. This is awesome, I am excited to get this wired up in our CI!

In case it is worthwhile:

Yes, it is. Thanks a lot!