I made a patch and would like feedback/testers (drivers/cdrom/aztcd.c)
Hello!
I made a patch and I would like feedback.
If you can test the patch.. please let me know.
Background:
There are busy loops that time out after 8 million repetitions. I made these improvements:
* Sleep during the busy loops
* Time out is based on elapsed time
If this patch is accepted, AZT_TIMEOUT can be removed.
Best regards,
Daniel Marjamäki
diff -up a/drivers/cdrom/aztcd.c b/drivers/cdrom/aztcd.c
drivers/cdrom/aztcd.c +14 -13
@@ -179,6 +179,7 @@#include <linux/ioport.h> #include <linux/string.h> #include <linux/major.h> +#include <linux/jiffies.h> #include <linux/init.h>
@@ -308,7 +309,7 @@ static char aztDiskChanged = 1;static char aztTocUpToDate =0 ;static unsigned char aztIndatum; -static unsigned long aztTimeOutCount; +static unsigned long aztTimeOutCount, aztTimeOut;static int aztCmd =0 ;static DEFINE_SPINLOCK (aztSpin);
@@ -361,14 +362,14 @@ static int azt_bcd2bin(unsigned char bcd# define OP_OK op_ok()static void op_ok (void ) { - aztTimeOutCount =0 ; + aztTimeOut = jiffies +2 ;do { aztIndatum =inb (DATA_PORT ); - aztTimeOutCount++; -if (aztTimeOutCount >=AZT_TIMEOUT ) { +if (time_after (jiffies, aztTimeOut)) {printk ("aztcd: Error Wait OP_OK\n" );break ; } +schedule_timeout_interruptible (1 ); }while (aztIndatum !=AFL_OP_OK ); }
@@ -377,14 +378,14 @@ static void op_ok(void)# define PA_OK pa_ok()static void pa_ok (void ) { - aztTimeOutCount =0 ; + aztTimeOut = jiffies +2 ;do { aztIndatum =inb (DATA_PORT ); - aztTimeOutCount++; -if (aztTimeOutCount >=AZT_TIMEOUT ) { +if (time_after (jiffies, aztTimeOut)) {printk ("aztcd: Error Wait PA_OK\n" );break ; } +schedule_timeout_interruptible (1 ); }while (aztIndatum !=AFL_PA_OK ); }#endif
@@ -393,17 +394,17 @@ static void pa_ok(void)# define STEN_LOW sten_low()static void sten_low (void ) { - aztTimeOutCount =0 ; + aztTimeOut = jiffies +2 ;do { aztIndatum =inb (STATUS_PORT ); - aztTimeOutCount++; -if (aztTimeOutCount >=AZT_TIMEOUT ) { +if (time_after (jiffies, aztTimeOut)) {if (azt_init_end)printk ("aztcd: Error Wait STEN_LOW commands:%x\n" , aztCmd);break ; } +schedule_timeout_interruptible (1 ); }while (aztIndatum &AFL_STATUS ); }
@@ -411,14 +412,14 @@ static void sten_low(void)# define DTEN_LOW dten_low()static void dten_low (void ) { - aztTimeOutCount =0 ; + aztTimeOut = jiffies +2 ;do { aztIndatum =inb (STATUS_PORT ); - aztTimeOutCount++; -if (aztTimeOutCount >=AZT_TIMEOUT ) { +if (time_after (jiffies, aztTimeOut)) {printk ("aztcd: Error Wait DTEN_OK\n" );break ; } +schedule_timeout_interruptible (1 ); }while (aztIndatum &AFL_DATA ); }
drivers/cdrom/aztcd.c static int azt_bcd2bin(unsigned char bcd · line 372
Quoted 13 lines
static void op_ok(void) {drivers/cdrom/aztcd.c +3
-3- aztTimeOutCount =0 ; + aztTimeOut = jiffies +2 ; do { aztIndatum = inb(DATA_PORT); - aztTimeOutCount++; - if (aztTimeOutCount >= AZT_TIMEOUT) { + if (time_after(jiffies, aztTimeOut)) { printk("aztcd: Error Wait OP_OK\n"); break; } + schedule_timeout_interruptible(1);
this I think is not quite right; schedule_timeout_*() doesn't do anything unless you set current->state to something. And at that point you might as well start using msleep()!
but what you're doing is generally a good idea; busy waits as the original code did is quite wrong...
On Sun, 2005-11-20 at 16:38 +0100, Arjan van de Ven wrote: > }
Quoted 4 lines
Review
Quoted 1 line
(quoted diff) +1
-0+ schedule_timeout_interruptible(1);this I think is not quite right; schedule_timeout_*() doesn't do anything unless you set
current->stateto something.
Err, schedule_timeout_(un)interruptible sets current->state.
tglx
Quoted 14 lines · Arjan van de Ven
drivers/cdrom/aztcd.c static int azt_bcd2bin(unsigned char bcd · line 372
Quoted 13 lines
static void op_ok(void) {drivers/cdrom/aztcd.c +3
-3- aztTimeOutCount =0 ; + aztTimeOut = jiffies +2 ; do { aztIndatum = inb(DATA_PORT); - aztTimeOutCount++; - if (aztTimeOutCount >= AZT_TIMEOUT) { + if (time_after(jiffies, aztTimeOut)) { printk("aztcd: Error Wait OP_OK\n"); break; } + schedule_timeout_interruptible(1);
Quoted 3 lines
this I think is not quite right; schedule_timeout_*() doesn't do anything unless you set
current->stateto something. And at that point you might as well start usingmsleep()!
Not true, as Thomas points out. You are right for schedule_timeout(), but that's why we introduced the _interruptible() and _uninterruptible(). And there are reasons to use schedule_timeout_*() instead of msleep() [not necessarily in this case, but in general], specifically the presence of wait-queues.
Quoted 2 lines
but what you're doing is generally a good idea; busy waits as the original code did is quite wrong...
I agree, and I recommend Daniel post to LKML to get some testing / see if anyone actually uses this driver :)
Thanks,
Nish