2.6.10 include/scsi/scsi.h and drivers/scsi/constants.h

PATCH In review Joe Perches 2 participants 2 messages Jan 31, 2005 – Feb 1, 2005 SCSI

lore

Share & tools

lore

Joe Perches

author

2005-01-31 17:43 UTC #0 lore raw

PATCH 1/1 2.6.10 include/scsi/scsi.h and drivers/scsi/constants.h

Remove unnecessary ";" from #define and use sense_ macros

scsi/scsi.h renamed from scsi/scsi.h~ +1 -1

@@ -356,7 +356,7 @@
 
 #define sense_class(sense)  (((sense) >> 4) & 0x7)
 #define sense_error(sense)  ((sense) & 0xf)
-#define sense_valid(sense)  ((sense) & 0x80);
+#define sense_valid(sense)  ((sense) & 0x80)
 
 
 #define IDENTIFY_BASE       0x80

scsi/constants.c renamed from scsi/constants.c~ +6 -6

@@ -924,17 +924,17 @@
 		     const unsigned char *sense_buffer,
 		     struct request *req)
 {
-	int s, sense_class, valid, code, info;
+	int s, class, valid, code, info;
 	const char *error = NULL;
 	unsigned char asc, ascq;
 	const char *sense_txt;
 	const char *name = req->rq_disk ? req->rq_disk->disk_name : devclass;
     
-	sense_class = (sense_buffer[0] >> 4) & 0x07;
-	code = sense_buffer[0] & 0xf;
-	valid = sense_buffer[0] & 0x80;
+	class = sense_class(sense_buffer[0]);
+	code = sense_error(sense_buffer[0]);
+	valid = sense_valid(sense_buffer[0]);
     
-	if (sense_class == 7) {	/* extended sense data */
+	if (class == 7) {	/* extended sense data */
 		s = sense_buffer[7] + 8;
 		if (s > SCSI_SENSE_BUFFERSIZE)
 			s = SCSI_SENSE_BUFFERSIZE;

@@ -1002,7 +1002,7 @@
 			       sense_buffer[0], sense_buffer[2]);
 
 		printk("Non-extended sense class %d code 0x%0x\n",
-		       sense_class, code);
+		       class, code);
 		s = 4;
 	}

When things settle down a bit it might be good to get rid of those macros:
- their names are too generic (e.g. no leading "sam_")
- the sense "class" concept has long been dropped (10 years ago?)
- sense_valid() is a confusing name, that bit means the info field within the sense data is valid, not the sense data itself as the macro name implies
- sense_error() yields what is now called the "response code"
- sense_valid() assumes fixed sense data format (i.e. it yields the wrong result for descriptor format)

We can help ourselves and others to understand the SCSI code, especially the error handling, if we clean up some of this stuff. The definitive modern description of sense data can be found in SPC-3 (rev 21c) section 4.5 at www.t10.org .

Thanks for bringing those macros to my attention.

Doug Gilbert

P.S. On a more subjective note, the C++ programmer in me cringes when I see variables (re-)named "class".

Quoted 38 lines

scsi/scsi.h renamed from scsi/scsi.h~ +0 -0

 #define IDENTIFY_BASE       0x80

scsi/constants.c renamed from scsi/constants.c~ +6 -6

@@ -924,17 +924,17 @@
 		     const unsigned char *sense_buffer,
 		     struct request *req)
 {
-	int s, sense_class, valid, code, info;
+	int s, class, valid, code, info;
 	const char *error = NULL;
 	unsigned char asc, ascq;
 	const char *sense_txt;
 	const char *name = req->rq_disk ? req->rq_disk->disk_name : devclass;
 
-	sense_class = (sense_buffer[0] >> 4) & 0x07;
-	code = sense_buffer[0] & 0xf;
-	valid = sense_buffer[0] & 0x80;
+	class = sense_class(sense_buffer[0]);
+	code = sense_error(sense_buffer[0]);
+	valid = sense_valid(sense_buffer[0]);
 
-	if (sense_class == 7) {	/* extended sense data */
+	if (class == 7) {	/* extended sense data */
 		s = sense_buffer[7] + 8;
 		if (s > SCSI_SENSE_BUFFERSIZE)
 			s = SCSI_SENSE_BUFFERSIZE;

@@ -1002,7 +1002,7 @@
 			       sense_buffer[0], sense_buffer[2]);
 
 		printk("Non-extended sense class %d code 0x%0x\n",
-		       sense_class, code);
+		       class, code);
 		s = 4;
 	}