[PATCH] x86: Fix kprobes build with non-gawk awk

[PATCH] x86: Fix kprobes build with non-gawk awk

[Jonathan Nieder]

The instruction attribute table generator fails when run by mawk
or original-awk:

 $ mawk -f arch/x86/tools/gen-insn-attr-x86.awk \
	arch/x86/lib/x86-opcode-map.txt > /dev/null
 Semantic error at 240: Second IMM error
 $ echo $?
 1

Line 240 contains "c8: ENTER Iw,Ib", which indicates that this
instruction has two immediate operands, the second of which is
one byte.  The script loops through the immediate operands using
a for loop.

Unfortunately, there is no guarantee in awk that a for (variable
in array) loop will return the indices in increasing order.
Internally, both original-awk and mawk iterate over a hash table
for this purpose, and both implementations happen to produce the
index 2 before 1.  The supposed second immediate operand is more
than one byte wide, producing the error.

So loop over the indices in increasing order instead.  As a
side-effect, with mawk this means the silly two-entry hash table
never has to be built.

Cc: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Jim Keniston <jkenisto@us.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: x86@kernel.org
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
---
 arch/x86/tools/gen-insn-attr-x86.awk |   10 +++++-----
 1 files changed, 5 insertions(+), 5 deletions(-)

diff --git a/arch/x86/tools/gen-insn-attr-x86.awk b/arch/x86/tools/gen-insn-attr-x86.awk
index e34e92a..7a68506 100644
--- a/arch/x86/tools/gen-insn-attr-x86.awk
+++ b/arch/x86/tools/gen-insn-attr-x86.awk
@@ -226,12 +226,12 @@ function add_flags(old,new) {
 }
 
 # convert operands to flags.
-function convert_operands(opnd,       i,imm,mod)
+function convert_operands(count,opnd,       i,j,imm,mod)
 {
 	imm = null
 	mod = null
-	for (i in opnd) {
-		i  = opnd[i]
+	for (j = 1; j <= count; j++) {
+		i = opnd[j]
 		if (match(i, imm_expr) == 1) {
 			if (!imm_flag[i])
 				semantic_error("Unknown imm opnd: " i)
@@ -282,8 +282,8 @@ function convert_operands(opnd,       i,imm,mod)
 		# parse one opcode
 		if (match($i, opnd_expr)) {
 			opnd = $i
-			split($(i++), opnds, ",")
-			flags = convert_operands(opnds)
+			count = split($(i++), opnds, ",")
+			flags = convert_operands(count, opnds)
 		}
 		if (match($i, ext_expr))
 			ext = $(i++)
-- 
1.6.5.4

Re: [PATCH] x86: Fix kprobes build with non-gawk awk

[Masami Hiramatsu]

Hi Jonathan,

Jonathan Nieder wrote:
> The instruction attribute table generator fails when run by mawk
> or original-awk:
> 
>  $ mawk -f arch/x86/tools/gen-insn-attr-x86.awk \
> 	arch/x86/lib/x86-opcode-map.txt > /dev/null
>  Semantic error at 240: Second IMM error
>  $ echo $?
>  1
> 
> Line 240 contains "c8: ENTER Iw,Ib", which indicates that this
> instruction has two immediate operands, the second of which is
> one byte.  The script loops through the immediate operands using
> a for loop.
> 
> Unfortunately, there is no guarantee in awk that a for (variable
> in array) loop will return the indices in increasing order.
> Internally, both original-awk and mawk iterate over a hash table
> for this purpose, and both implementations happen to produce the
> index 2 before 1.  The supposed second immediate operand is more
> than one byte wide, producing the error.

Oh, I see.

> 
> So loop over the indices in increasing order instead.  As a
> side-effect, with mawk this means the silly two-entry hash table
> never has to be built.

Thank you for fixing it!

Acked-by: Masami Hiramatsu <mhiramat@redhat.com>

> 
> Cc: Masami Hiramatsu <mhiramat@redhat.com>
> Cc: Jim Keniston <jkenisto@us.ibm.com>
> Cc: Frederic Weisbecker <fweisbec@gmail.com>
> Cc: x86@kernel.org
> Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
> ---
>  arch/x86/tools/gen-insn-attr-x86.awk |   10 +++++-----
>  1 files changed, 5 insertions(+), 5 deletions(-)
> 
> diff --git a/arch/x86/tools/gen-insn-attr-x86.awk b/arch/x86/tools/gen-insn-attr-x86.awk
> index e34e92a..7a68506 100644
> --- a/arch/x86/tools/gen-insn-attr-x86.awk
> +++ b/arch/x86/tools/gen-insn-attr-x86.awk
> @@ -226,12 +226,12 @@ function add_flags(old,new) {
>  }
>  
>  # convert operands to flags.
> -function convert_operands(opnd,       i,imm,mod)
> +function convert_operands(count,opnd,       i,j,imm,mod)
>  {
>  	imm = null
>  	mod = null
> -	for (i in opnd) {
> -		i  = opnd[i]
> +	for (j = 1; j <= count; j++) {
> +		i = opnd[j]
>  		if (match(i, imm_expr) == 1) {
>  			if (!imm_flag[i])
>  				semantic_error("Unknown imm opnd: " i)
> @@ -282,8 +282,8 @@ function convert_operands(opnd,       i,imm,mod)
>  		# parse one opcode
>  		if (match($i, opnd_expr)) {
>  			opnd = $i
> -			split($(i++), opnds, ",")
> -			flags = convert_operands(opnds)
> +			count = split($(i++), opnds, ",")
> +			flags = convert_operands(count, opnds)
>  		}
>  		if (match($i, ext_expr))
>  			ext = $(i++)

-- 
Masami Hiramatsu

Software Engineer
Hitachi Computer Products (America), Inc.
Software Solutions Division

e-mail: mhiramat@redhat.com

Re: [PATCH] x86: Fix kprobes build with non-gawk awk

[Jonathan Nieder]

Masami Hiramatsu wrote:

> Thank you for fixing it!

No problem.  Thanks for looking it over.

Regards,
Jonathan

[tip:x86/urgent] x86: Fix kprobes build with non-gawk awk

[tip-bot for Jonathan Nieder]

Commit-ID:  23637568ad0c9b5ab0ad27d2f2f26d1e9282c527
Gitweb:     http://git.kernel.org/tip/23637568ad0c9b5ab0ad27d2f2f26d1e9282c527
Author:     Jonathan Nieder <jrnieder@gmail.com>
AuthorDate: Sun, 13 Dec 2009 16:04:38 -0600
Committer:  Ingo Molnar <mingo@elte.hu>
CommitDate: Tue, 15 Dec 2009 20:35:49 +0100

x86: Fix kprobes build with non-gawk awk

The instruction attribute table generator fails when run by mawk
or original-awk:

 $ mawk -f arch/x86/tools/gen-insn-attr-x86.awk \
	arch/x86/lib/x86-opcode-map.txt > /dev/null
 Semantic error at 240: Second IMM error
 $ echo $?
 1

Line 240 contains "c8: ENTER Iw,Ib", which indicates that this
instruction has two immediate operands, the second of which is
one byte.  The script loops through the immediate operands using
a for loop.

Unfortunately, there is no guarantee in awk that a for (variable
in array) loop will return the indices in increasing order.
Internally, both original-awk and mawk iterate over a hash table
for this purpose, and both implementations happen to produce the
index 2 before 1.  The supposed second immediate operand is more
than one byte wide, producing the error.

So loop over the indices in increasing order instead.  As a
side-effect, with mawk this means the silly two-entry hash table
never has to be built.

Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Acked-by Masami Hiramatsu <mhiramat@redhat.com>
Cc: Jim Keniston <jkenisto@us.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <20091213220437.GA27718@progeny.tock>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
---
 arch/x86/tools/gen-insn-attr-x86.awk |   10 +++++-----
 1 files changed, 5 insertions(+), 5 deletions(-)

diff --git a/arch/x86/tools/gen-insn-attr-x86.awk b/arch/x86/tools/gen-insn-attr-x86.awk
index e34e92a..7a68506 100644
--- a/arch/x86/tools/gen-insn-attr-x86.awk
+++ b/arch/x86/tools/gen-insn-attr-x86.awk
@@ -226,12 +226,12 @@ function add_flags(old,new) {
 }
 
 # convert operands to flags.
-function convert_operands(opnd,       i,imm,mod)
+function convert_operands(count,opnd,       i,j,imm,mod)
 {
 	imm = null
 	mod = null
-	for (i in opnd) {
-		i  = opnd[i]
+	for (j = 1; j <= count; j++) {
+		i = opnd[j]
 		if (match(i, imm_expr) == 1) {
 			if (!imm_flag[i])
 				semantic_error("Unknown imm opnd: " i)
@@ -282,8 +282,8 @@ function convert_operands(opnd,       i,imm,mod)
 		# parse one opcode
 		if (match($i, opnd_expr)) {
 			opnd = $i
-			split($(i++), opnds, ",")
-			flags = convert_operands(opnds)
+			count = split($(i++), opnds, ",")
+			flags = convert_operands(count, opnds)
 		}
 		if (match($i, ext_expr))
 			ext = $(i++)