sg in lk 2.5.3-pre5

In pre5 the sg driver tries to set up both address and page+offset in scatterlist entries. Unfortunately asm-i386/pci.h's pci_map_sg() contains this check:

                if (sg[i].address && sg[i].page)
                        BUG();

which kills that strategy with the sym53c8xx driver. The sooner we get rid of address in struct scatterlist the better ... the address,page+offset duality is a mess. Strange that Andi Kleen had problems compiling ide-scsi, it compiled and worked on my UP and SMP machines (AMD and dual Celeron respectively).

Sg patch (which has been already sent to Jens) follows.

Doug Gilbert

@@ -19,7 +19,7 @@
  */
 #include <linux/config.h>
 #ifdef CONFIG_PROC_FS
- static char sg_version_str[] = "Version: 3.5.23 (20020103)";
+ static char sg_version_str[] = "Version: 3.5.23 (20020124)";
 #endif
  static int sg_version_num = 30523; /* 2 digits for each component */
 /*

@@ -76,7 +76,7 @@
 #include <linux/version.h>
 #endif /* LINUX_VERSION_CODE */
 
-#define SG_STILL_HAVE_ADDRESS_IN_SCATTERLIST
+/* #define SG_STILL_HAVE_ADDRESS_IN_SCATTERLIST */
 
 #define SG_ALLOW_DIO_DEF 0
 #define SG_ALLOW_DIO_CODE      /* compile out be commenting this define */

I must apologize. It seems the removal of scatterlist->address was a vger tree only change (I tested with vger). Mainline 2.5 still seems to have it. Nevertheless the patch probably won't hurt to apply.

-Andi

Ah makes sense. Yes the ->address change is pending still. I still don't think it's worth it to complicate ide-scsi with kmap, just have it bounced. ide-scsi will not live too much longer anyways.

Jens Axboe