Skip to contentSkip to conversation

Coccinelle rules for module_pci_driver

Cocci Michael Opdenacker 2 participants 5 messages Oct 16, 2014

lore
Share & tools
lore
#0 lore raw

Hi Julia,

Thank you very much for your very useful tutorial at ELCE/Linux Con!

I continued to work on my rules to detect opportunities to use the module_pci_driver macro in the Linux kernel.

I created multiple rules as you suggested, and here's where I stand:

Quoted 1 line
Quoted 1 line
Quoted 1 line

// Find opportunities to use module_pci_driver(a)

@ rule1 @
identifier f1, ret;
struct pci_driver a;
@@

- int f1(void) {
- int ret;
- ret = pci_register_driver(&a);
- return ret;
- }

@ rule2 @
identifier f2;
identifier rule1.f1;
struct pci_driver rule1.a;
@@

- void f2(void) {
- pci_unregister_driver(&a);
- }

@@
declarer name module_init;
declarer name module_exit;
declarer name module_pci_driver; // this is top-level: not instantiated in a function identifier rule1.f1;
identifier rule2.f2;
struct pci_driver rule1.a;
@@

- module_init(f1);

- module_exit(f2);

+ module_pci_driver(&a);

Quoted 1 line
Quoted 1 line
Quoted 1 line

My rules work at least partially (tests on the latest mainline kernel):

Quoted 1 line
Quoted 1 line
Quoted 1 line

spatch --sp-file exmodule.cocci --dir ../linux/drivers/scsi/arcmsr/arcmsr_hba.c
init_defs_builtins: /usr/share/coccinelle/standard.h
warning: rule2: inherited metavariable f1 not used in the -, +, or context code
HANDLING: ../linux/drivers/scsi/arcmsr/arcmsr_hba.c
(ONCE) already tagged but only removed, so safe

tmp/cocci-output-7658-d2f178-arcmsr_hba.c renamed from linux/drivers/scsi/arcmsr/arcmsr_hba.c +0 -12
diff = 
@@ -1280,18 +1280,6 @@ static void arcmsr_shutdown(struct pci_d
 	arcmsr_stop_adapter_bgrb(acb);
 	arcmsr_flush_adapter_cache(acb);
 }
-
-static int arcmsr_module_init(void)
-{
-	int error = 0;
-	error = pci_register_driver(&arcmsr_pci_driver);
-	return error;
-}
-
-static void arcmsr_module_exit(void)
-{
-	pci_unregister_driver(&arcmsr_pci_driver);
-}
 module_init(arcmsr_module_init);
 module_exit(arcmsr_module_exit);
 
Quoted 1 line
Quoted 1 line
Quoted 1 line

While the first 2 rules found what they were supposed to, the third rule doesn't produce the intended effect, i.e. replacing module_init|exit by module_pci_driver.

I know I'm not caching all possible cases (multiple ways of testing the return value of pci_register_driver), but I'd least I'd be happy if I could make this first case work :)

Once everything works, I will create rules for the various bus types in the Linux kernel (usb, i2c, platform...).

Thanks again,

Cheers,

Michael.

Signature · 4 lines
Michael Opdenacker, CEO, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com
+33 484 258 098
#1 lore raw

You can't remove two top-evel things at once.

I'll propose something.

julia

#2 lore raw
In reply to Michael Opdenacker

I suggest the following semantic patch. It is in two parts. The first part matches all of the functions/declarations to ensure that they are all present. The second part then makes the transformation.

I get a change in only one file.

julia

// Find opportunities to use module_pci_driver(a)

@r1@
identifier f1, ret;
struct pci_driver a;
@@

int f1(void) {
int ret;
ret = pci_register_driver(&a);
return ret;
}

@r2 depends on r1@
identifier f2;
struct pci_driver r1.a;
@@

void f2(void) {
pci_unregister_driver(&a);
}

@r3 depends on r2@
identifier r1.f1;
declarer name module_init;
@@

module_init(f1);

@r4 depends on r3@
identifier r2.f2;
declarer name module_exit;
@@

module_exit(f2);

// Clean up

@depends on r4@
identifier f1, ret;
struct pci_driver a;
@@

- int f1(void) {

-   int ret;
-   ret = pci_register_driver(&a);
-   return ret;

- }

@depends on r4@
identifier f2;
struct pci_driver r1.a;
@@

- void f2(void) {
- pci_unregister_driver(&a);
- }

@depends on r4@
identifier r1.f1;
declarer name module_pci_driver;
@@

- module_init(f1);
+ module_pci_driver(&a);

@depends on r4@
identifier r2.f2;
@@

- module_exit(f2);

#3 lore raw

Hi Julia,

Quoted 5 lines · Julia Lawall

I suggest the following semantic patch. It is in two parts. The first part matches all of the functions/declarations to ensure that they are all present. The second part then makes the transformation.

I get a change in only one file.

Many thanks for this new version (interesting approach that has the advantage of being progressive and repeatable) and for your explanations.

I confirm that it works for me and that I will work on making it suitable for "make coccicheck" in the Linux kernel.

Cheers,

Michael.

Signature · 4 lines
Michael Opdenacker, CEO, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com
+33 484 258 098
#4 lore raw
Quoted 7 lines · Michael Opdenacker

Hi Julia,

Quoted 4 lines · Julia Lawall

I suggest the following semantic patch. It is in two parts. The first part matches all of the functions/declarations to ensure that they are all present. The second part then makes the transformation.

Quoted 7 lines
Quoted 1 line

I get a change in only one file.

Many thanks for this new version (interesting approach that has the advantage of being progressive and repeatable) and for your explanations.

I confirm that it works for me and that I will work on making it suitable for "make coccicheck" in the Linux kernel.

Just run coccinelle/tools/sgen/sgen on it and see what happens.

julia