net #9
PATCH 1/1 net #9
The following patch removes some zero initializers from statics
Andrzej
**************************** PATCH 9 *****************************
drivers/net/dmfe.c +9 -9
@@ -261,20 +261,20 @@static char version[]__devinitdata =KERN_INFO "Davicom DM9xxx net driver, version " DMFE_VERSION "\n" ; -static int dmfe_debug =0 ; +static int dmfe_debug;static unsigned char dmfe_media_mode =DMFE_AUTO ; -static u32 dmfe_cr6_user_set =0 ; +static u32 dmfe_cr6_user_set;/* For module input parameter */ -static int debug =0 ; -static u32 cr6set =0 ; +static int debug; +static u32 cr6set;static unsigned char mode =8 ;static u8 chkmode =1 ; -static u8 HPNA_mode =0 ;/* Default: Low Power/High Speed */ -static u8 HPNA_rx_cmd =0 ;/* Default: Disable Rx remote command */ -static u8 HPNA_tx_cmd =0 ;/* Default: Don't issue remote command */ -static u8 HPNA_NoiseFloor =0 ;/* Default: HPNA NoiseFloor */ -static u8 SF_mode =0 ;/* Special Function: 1:VLAN, 2:RX Flow Control +static u8 HPNA_mode;/* Default: Low Power/High Speed */ +static u8 HPNA_rx_cmd;/* Default: Disable Rx remote command */ +static u8 HPNA_tx_cmd;/* Default: Don't issue remote command */ +static u8 HPNA_NoiseFloor;/* Default: HPNA NoiseFloor */ +static u8 SF_mode;/* Special Function: 1:VLAN, 2:RX Flow Control 4: TX pause packet */ unsigned long CrcTable[256 ] = {
drivers/net/wavelan.p.h +3 -3
@@ -699,9 +699,9 @@#ifdef MODULE /* Parameters set by insmod */ -static int io[4 ] = {0 ,0 ,0 ,0 }; -static int irq[4 ] = {0 ,0 ,0 ,0 }; -static char name[4 ][IFNAMSIZ ] = {"" ,"" ,"" ,"" }; +static int io[4 ]; +static int irq[4 ]; +static char name[4 ][IFNAMSIZ ];MODULE_PARM (io,"1-4i" );MODULE_PARM (irq,"1-4i" );MODULE_PARM (name,"1-4c" __MODULE_STRING(IFNAMSIZ ));
Signature · 4 lines
======================================================================= Andrzej M. Krzysztofowicz [email protected] phone (48)(58) 347 14 61 Faculty of Applied Phys. & Math., Technical University of Gdansk
Quoted 4 lines · Andrzej Krzysztofowicz
The following patch removes some zero initializers from statics
Andrzej
If I were you, I would fix gcc rather than making my code
unreadable.
I write source code in C rather than coding ASM in hex because
of the semantic attached to what I write, which make the code readable by me and by other, and make my code portable to other environments (for example user space). Initialisating a variable to zero as opposed to leaving it undefined has plenty of semantic attached to it.
It's the job of the compiler to make sure that all this kind
of stupid optimisation are done and the code produced is as efficient as possible and adapted to the exact characteristics of the operating envirtonment. Especially that it's probably 10 lines to add the proper option to gcc command line.
Therefore, Alan, please do not apply those kind of patches to
my drivers.
Thanks... Jean
Quoted 9 lines · Jean Tourrilhes
Quoted 4 lines · Andrzej Krzysztofowicz
The following patch removes some zero initializers from statics
Andrzej
If I were you, I would fix gcc rather than making my code
unreadable.
[...]
Quoted 2 lines
Therefore, Alan, please do not apply those kind of patches to
my drivers.
This is standard kernel cleanup that makes the resulting image smaller. These patches have been going into all areas of the kernel for quite some time.
Signature · 3 lines
Jeff Garzik | Disbelief, that's why you fail. Building 1024 | MandrakeSoft |
Quoted 4 lines · Jeff Garzik
This is standard kernel cleanup that makes the resulting image smaller. These patches have been going into all areas of the kernel for quite some time.
This doesn't make it right.
Ok, while we are on the topic : could somebody explain me why
we can't get gcc to do that for us ? What is preventing adding a gcc command line flag to do exactly that ? It's not like rocket science (simple test) and would avoid to have tediously to go through all source code, past, present and *future* to make those changes.
Bah, maybe it's too straightforward...
Jean
Quoted 7 lines · Jean Tourrilhes
Quoted 4 lines · Jeff Garzik
This is standard kernel cleanup that makes the resulting image smaller. These patches have been going into all areas of the kernel for quite some time.
Quoted 8 lines
This doesn't make it right.
Ok, while we are on the topic : could somebody explain me why
we can't get gcc to do that for us ? What is preventing adding a gcc command line flag to do exactly that ? It's not like rocket science (simple test) and would avoid to have tediously to go through all source code, past, present and *future* to make those changes.
Bah, maybe it's too straightforward...
This is ANSI C standard stuff. If a static object with a scalar type is not explicitly initialized, it is initialized to zero by default.
Sure we can get gcc to recognize that case, but why use gcc to work around code that avoids an ANSI feature?
Signature · 3 lines
Jeff Garzik | Disbelief, that's why you fail. Building 1024 | MandrakeSoft |
Quoted 17 lines · Jeff Garzik
Quoted 7 lines · Jean Tourrilhes
Quoted 4 lines · Jeff Garzik
This is standard kernel cleanup that makes the resulting image smaller. These patches have been going into all areas of the kernel for quite some time.
Quoted 8 lines
This doesn't make it right.
Ok, while we are on the topic : could somebody explain me why
we can't get gcc to do that for us ? What is preventing adding a gcc command line flag to do exactly that ? It's not like rocket science (simple test) and would avoid to have tediously to go through all source code, past, present and *future* to make those changes.
Bah, maybe it's too straightforward...
Quoted 5 lines
This is ANSI C standard stuff. If a static object with a scalar type is not explicitly initialized, it is initialized to zero by default.
Sure we can get gcc to recognize that case, but why use gcc to work around code that avoids an ANSI feature?
Good standard don't mandate the implementation. And as
somebody doing some other language said, there is more than one way to do it.
ANSI C doesn't prohibit to initialise variable to zero
(AFAIK). ANSI C doesn't prevent the compiler to optimise for this case. If I initialise the varaible 3 time to the same value in a row, gcc will remove the other two.
Let's put it another way... What's different between : for(i = 0; i < n; i++) xxx; And : i = 0; while(i++ < n) xxx; I would expect the compiler to produce code as efficient in
both cases, and leave the choice of coding style to the user.
Jean
*shrug* Well, if you want to go against the kernel standard that's fine with me. I won't put Andrzej's changes to your drivers upstream. You are going to continually see patches to clean that up, though, because it makes the end user's kernel smaller. Please consider noting this special case in a comment in each of your drivers "do not clean up static initializers" or similar.
It's really a pain in the ass to remember special cases like this, so please reconsider. Being not-like-the-others is detrimental to the long term maintainability of the overall kernel.
Regards,
Jeff
Signature · 3 lines
Jeff Garzik | Disbelief, that's why you fail. Building 1024 | MandrakeSoft |
Quoted 6 lines · Jeff Garzik
This is ANSI C standard stuff. If a static object with a scalar type is not explicitly initialized, it is initialized to zero by default.
Sure we can get gcc to recognize that case, but why use gcc to work around code that avoids an ANSI feature?
Apart from this stupid flame that I'm making, I've got my
Intel/Symbol card to work properly with the Orinoco driver. This mean that we are not far away to have the 4 main flavor of 802.11b working in 2.4.X (i.e. Lucent/Symbol/PrismII/Aironet).
See :
http://www.hpl.hp.com/personal/Jean_Tourrilhes/Linux/Orinoco.html#patches
Just to make sure we end on a positive note ;-) Now, if I
could get the card of Alan to work...
Have fun, don't take it seriously...
Jean
Quoted 9 lines · Jean Tourrilhes
Apart from this stupid flame that I'm making, I've got my
Intel/Symbol card to work properly with the Orinoco driver. This mean that we are not far away to have the 4 main flavor of 802.11b working in 2.4.X (i.e. Lucent/Symbol/PrismII/Aironet).
See :
http://www.hpl.hp.com/personal/Jean_Tourrilhes/Linux/Orinoco.html#patchesJust to make sure we end on a positive note ;-) Now, if I
could get the card of Alan to work...
Cool! Though I do have a comment on this too. :)
Are you sure priv->hw_ready is not racy? You should probably use an atomic operation like test_bit and set_bit.
Signature · 3 lines
Jeff Garzik | Disbelief, that's why you fail. Building 1024 | MandrakeSoft |
Quoted 7 lines · Jeff Garzik
*shrug* Well, if you want to go against the kernel standard that's fine with me. I won't put Andrzej's changes to your drivers upstream. You are going to continually see patches to clean that up, though, because it makes the end user's kernel smaller. Please consider noting this special case in a comment in each of your drivers "do not clean up static initializers" or similar.
Quoted 7 lines
It's really a pain in the ass to remember special cases like this, so please reconsider. Being not-like-the-others is detrimental to the long term maintainability of the overall kernel.
Regards,
Jeff
I agree with you on the special case. I don't like it
either. Anyway, most patch to my drivers are applied wether I like it or not, so I guess that I should be happy that I was notified and I should sut up my big mouth because it won't make a difference.
If I reject the patch now, I will be applied behind my
back. Been there, done that.
In other words : yes, please apply the patch.
Jean
drivers/net/wavelan.p.h line 704
Quoted 3 lines · Andrzej Krzysztofowicz
drivers/net/wavelan.p.h +1
-1-static char name[4 ][IFNAMSIZ ] = {"" ,"" ,"" ,"" }; +static char name[4 ][IFNAMSIZ ];
Ugh. Sure about that one? the variables have been pointers to zero, now they're zero...
Regards
Henning
Signature · 5 lines
Dipl.-Inf. (Univ.) Henning P. Schmiedehausen -- Geschaeftsfuehrer INTERMETA - Gesellschaft fuer Mehrwertdienste mbH [email protected] Am Schwabachgrund 22 Fon.: 09131 / 50654-0 [email protected] D-91054 Buckenhof Fax.: 09131 / 50654-20
"Henning P. Schmiedehausen wrote:"
Quoted 8 lines
drivers/net/wavelan.p.h line 704
Quoted 3 lines · Andrzej Krzysztofowicz
drivers/net/wavelan.p.h +1
-1-static char name[4 ][IFNAMSIZ ] = {"" ,"" ,"" ,"" }; +static char name[4 ][IFNAMSIZ ];Ugh. Sure about that one? the variables have been pointers to zero, now they're zero...
I do not agree. As I understand C "name" is a table, i.e. a pointer to a prealocated area of size of 4*IFNAMSIZ*sizeof(char) bytes. There is no pointers to the strings stored seprately.
[ The strings are copied into the apropriate locations in the array during
initialization ]
After applying the mentioned patch just the whole area of the array will be zeroed. Not only the first characters of name[i] i=0,1,2,3 ...
Andrzej
Signature · 4 lines
======================================================================= Andrzej M. Krzysztofowicz [email protected] phone (48)(58) 347 14 61 Faculty of Applied Phys. & Math., Technical University of Gdansk
Quoted 8 lines · Henning P. Schmiedehausen
drivers/net/wavelan.p.h line 704
Quoted 3 lines · Andrzej Krzysztofowicz
drivers/net/wavelan.p.h +1
-1-static char name[4 ][IFNAMSIZ ] = {"" ,"" ,"" ,"" }; +static char name[4 ][IFNAMSIZ ];Ugh. Sure about that one? the variables have been pointers to zero, now they're zero...
Bzzt. Arrays and pointers are not always equivalent. This code defines an area of 4*IFNAMSIZ chars. The old code then initialised each IFNAMSIZ chars to the empty string (an array starting with '\0'), not a pointer to an empty string. Test file x.c.
#define IFNAMSIZ 7
static char name[4][IFNAMSIZ];
int main(void) { return(0); }
# gcc -g x.c -o x # gdb x (gdb) ptype name type = char [4][7] (gdb) x name[0] 0x80494f4 <name>: 0x00000000 (gdb) x name[1] 0x80494fb <name+7>: 0x00000000 (gdb) x name[2] 0x8049502 <name+14>: 0x00000000 (gdb) x name[3] 0x8049509 <name+21>: 0x00000000
Zero initialisation via bss works fine. Try it with
static char name[4][IFNAMSIZ] = {"", "", "", ""};
and you get exactly the same results, at the expense of more disk space and time to load. Not much extra I know, but it all adds up.
Quoted 4 lines · Keith Owens
Quoted 2 lines · Henning P. Schmiedehausen
Andrzej Krzysztofowicz <[email protected]> writes:
Quoted 8 lines
Quoted 6 lines
drivers/net/wavelan.p.h line 704
Quoted 3 lines
drivers/net/wavelan.p.h +1
-1-static char name[4 ][IFNAMSIZ ] = {"" ,"" ,"" ,"" }; +static char name[4 ][IFNAMSIZ ];Ugh. Sure about that one? the variables have been pointers to zero, now they're zero...
Bzzt. Arrays and pointers are not always equivalent. This code
yeah, you're right. I got fooled by the fact that these are already arrays, not initialized pointers.
There _is_ a point for Java in this. ;-)
Regards
Henning
Signature · 5 lines
Dipl.-Inf. (Univ.) Henning P. Schmiedehausen -- Geschaeftsfuehrer INTERMETA - Gesellschaft fuer Mehrwertdienste mbH [email protected] Am Schwabachgrund 22 Fon.: 09131 / 50654-0 [email protected] D-91054 Buckenhof Fax.: 09131 / 50654-20
Quoted 2 lines · Jeff Garzik
This is ANSI C standard stuff. If a static object with a scalar type is not explicitly initialized, it is initialized to zero by default.
"The kernel is not in ANSI C". Who wrote this again?
Regards
Henning
Signature · 5 lines
Dipl.-Inf. (Univ.) Henning P. Schmiedehausen -- Geschaeftsfuehrer INTERMETA - Gesellschaft fuer Mehrwertdienste mbH [email protected] Am Schwabachgrund 22 Fon.: 09131 / 50654-0 [email protected] D-91054 Buckenhof Fax.: 09131 / 50654-20
Quoted 9 lines · Henning P. Schmiedehausen
drivers/net/wavelan.p.h line 704
Quoted 3 lines · Andrzej Krzysztofowicz
drivers/net/wavelan.p.h +1
-1-static char name[4 ][IFNAMSIZ ] = {"" ,"" ,"" ,"" }; +static char name[4 ][IFNAMSIZ ];Ugh. Sure about that one? the variables have been pointers to zero, now they're zero...
No, the variables were and always have been arrays, not pointers.
The previous incarnation, {"","","",""}, was actually doubly lame, because not only was it an unnecessary zeroing of the var, but using "" causes an extra string to be generated in the output asm.
Signature · 3 lines
Jeff Garzik | Disbelief, that's why you fail. Building 1024 | MandrakeSoft |
Quoted 9 lines · Jean Tourrilhes
Quoted 5 lines
This is ANSI C standard stuff. If a static object with a scalar type is not explicitly initialized, it is initialized to zero by default.
Sure we can get gcc to recognize that case, but why use gcc to work around code that avoids an ANSI feature?
Good standard don't mandate the implementation. And as
somebody doing some other language said, there is more than one way to do it.
Hmmm, I understand both sides perfectly, but what about that one:
int n; /* n=0 */
Would that be a compromise?
Regards,
Hermann
Signature · 4 lines
,_, (O,O) "There is more to life than increasing its speed." ( ) -- Gandhi -"-"--------------------------------------------------------------
Hi!
Quoted 5 lines
Quoted 3 lines
This is standard kernel cleanup that makes the resulting image smaller. These patches have been going into all areas of the kernel for quite some time.
This doesn't make it right.
Well... It does. Code should be uniform. You may like
if (
x == y
) {
printf(
"ahoj" ); }
style of indentation, but it is not okay in kernel.
Quoted 2 lines
Ok, while we are on the topic : could somebody explain me why
we can't get gcc to do that for us ? What is preventing adding a gcc
int xyzzy;
goes to BSS, while
int xyzzy = 0
goes to DATA. That is important for some folks.
Quoted 2 lines
command line flag to do exactly that ? It's not like rocket science (simple test) and would avoid to have tediously to go through all
It is actually not *so* easy to do it in gcc; there's patch, however.
Pavel
Signature · 2 lines
I'm [email protected]. "In my country we have almost anarchy and I don't care." Panos Katsaloulis describing me w.r.t. patents at [email protected]