patch-1.3.87 linux/drivers/cdrom/mcdx.c
Next file: linux/drivers/cdrom/sjcd.c
Previous file: linux/drivers/cdrom/mcd.c
Back to the patch index
Back to the overall index
- Lines: 138
- Date:
Fri Apr 12 09:49:33 1996
- Orig file:
v1.3.86/linux/drivers/cdrom/mcdx.c
- Orig date:
Fri Mar 22 09:42:18 1996
diff -u --recursive --new-file v1.3.86/linux/drivers/cdrom/mcdx.c linux/drivers/cdrom/mcdx.c
@@ -35,7 +35,7 @@
* Jon Tombs, Bjorn Ekwall (module support)
* Daniel v. Mosnenck (he sent me the Technical and Programming Reference)
* Gerd Knorr (he lent me his PhotoCD)
- * Nils Faerber and Roger E. Wolff (extensivly tested the LU portion)
+ * Nils Faerber and Roger E. Wolff (extensively tested the LU portion)
* Andreas Kies (testing the mysterious hang up's)
* ... somebody forgotten?
*
@@ -130,7 +130,7 @@
/* Per drive/controller stuff **************************************/
struct s_drive_stuff {
- /* waitquenes */
+ /* waitqueues */
struct wait_queue *busyq;
struct wait_queue *lockq;
struct wait_queue *sleepq;
@@ -145,7 +145,7 @@
/* cd infos */
struct s_diskinfo di;
struct s_multi multi;
- struct s_subqcode* toc; /* first enty of the toc array */
+ struct s_subqcode* toc; /* first entry of the toc array */
struct s_subqcode start;
struct s_subqcode stop;
int xa; /* 1 if xa disk */
@@ -592,7 +592,7 @@
CURRENT->sector,
CURRENT->nr_sectors))) {
/*INFO(("do_request() read error\n"));*/
- xwarn("do_requst() read error\n");
+ xwarn("do_request() read error\n");
if (stuffp->status & MCDX_ST_EOM) {
CURRENT->sector += CURRENT->nr_sectors;
CURRENT->nr_sectors = 0;
@@ -636,7 +636,7 @@
from the hardware status register).
If the last eject is too recent, an autoclose wouldn't probably
be what we want ..., if we can't read the CD after an autoclose
- no further autclose's will be tried */
+ no further autocloses will be tried */
if (inb((unsigned int) stuffp->rreg_status) & MCDX_RBIT_DOOR) {
if (jiffies - stuffp->ejected < ACLOSE_INHIBIT) return -EIO;
if (stuffp->autoclose) mcdx_closedoor(stuffp, 1);
@@ -847,7 +847,7 @@
static void mcdx_delay(struct s_drive_stuff *stuff, long jifs)
/* This routine is used for sleeping.
May be we could use a simple count loop w/ jumps to itself, but
- I wanna make this independend of cpu speed. [1 jiffie is 1/HZ] sec */
+ I wanna make this independent of cpu speed. [1 jiffy is 1/HZ] sec */
{
unsigned long tout = jiffies + jifs;
if (jifs < 0) return;
@@ -899,7 +899,7 @@
xinfo( "intr() irq %d status 0x%02x\n",
irq, inb((unsigned int) stuffp->rreg_data));
} else {
- xinfo( "intr() irq %d ambigous hw status\n", irq);
+ xinfo( "intr() irq %d ambiguous hw status\n", irq);
}
} else {
TRACE((IRQ, "irq() irq %d ok, status %02x\n", irq, b));
@@ -925,9 +925,9 @@
{
int st;
char c;
- int disgard;
+ int discard;
- if ((disgard = (buffer == NULL))) buffer = &c;
+ if ((discard = (buffer == NULL))) buffer = &c;
while (stuffp->lock) {
interruptible_sleep_on(&stuffp->lockq);
@@ -966,7 +966,7 @@
}
st = *bp;
sz--;
- if (!disgard) bp++;
+ if (!discard) bp++;
TRACE((TALK, "talk() got status 0x%02x\n", st));
@@ -999,7 +999,7 @@
cmd[0], tries - 1, tries == 2 ? "y" : "ies"));
st = -1; break;
}
- if (!disgard) bp++;
+ if (!discard) bp++;
TRACE((TALK, "talk() got 0x%02x\n", *(bp - 1)));
}
}
@@ -1080,7 +1080,7 @@
if (level < 1) return;
va_start(args, fmt);
if (sizeof(s) < vsprintf(s, fmt, args))
- printk(MCDX ":: dprintf exeeds limit!!\n");
+ printk(MCDX ":: dprintf exceeds limit!!\n");
else printk(MCDX ":: %s", s);
va_end(args);
}
@@ -1092,7 +1092,7 @@
va_list args;
va_start(args, fmt);
if (sizeof(s) < vsprintf(s, fmt, args))
- printk(MCDX ":: dprintf exeeds limit!!\n");
+ printk(MCDX ":: dprintf exceeds limit!!\n");
else printk(MCDX ": %s", s);
va_end(args);
}
@@ -1265,7 +1265,7 @@
mcdx_transfer(struct s_drive_stuff *stuffp,
char *p, int sector, int nr_sectors)
/* This seems to do the actually transfer. But it does more. It
- keeps track of errors ocurred and will (if possible) fall back
+ keeps track of errors occurred and will (if possible) fall back
to single speed on error.
Return: -1 on timeout or other error
else status byte (as in stuff->st) */
@@ -1279,7 +1279,7 @@
else return ans;
if (stuffp->readerrs && stuffp->readcmd == READ1X) {
- WARN(("XXX Alrady reading 1x -- no chance\n"));
+ WARN(("XXX Already reading 1x -- no chance\n"));
return -1;
}
@@ -1327,7 +1327,7 @@
do {
/* wait for the drive become idle, but first
- check for possible occured errors --- the drive
+ check for possible occurred errors --- the drive
seems to report them asynchronously */
current->timeout = jiffies + 5 * HZ;
FUNET's LINUX-ADM group, linux-adm@nic.funet.fi
TCL-scripts by Sam Shen, slshen@lbl.gov
with Sam's (original) version of this