Hi All, <This is a followup of the similarly named post (yesterday) - that is why removing that text, and top-posting>. Question: Is there a restriction on the bi_size field of a bio submission for submitting an I/O for being aligned in case of direct write? Observation: Moving ahead with the problem previously posted (not copying the text here else it would make this post huge - please visit http://permalink.gmane.org/gmane.linux.kernel.kernelnewbies/23158) I observed that when I create a bio using bio_map_user (and passing aligned user space address to it) and have bi_size=(PAGE_SIZE aligned units), IO is being successfully performed and error being reported in case any failure at SCSI level takes place. Whereas, as soon as I change the bi_size=(something unaligned) (and altering the first and last bio vector's offset accordingly) - IO is performed successfully in case NO failure on SCSI side. But as soon as an error (like disconnection of device from switch) takes place, bio_end bio is being called with bi_size as positive and /var/log/messages is being continuously flooded with SCSI error messages like: ---- Oct 17 21:56:07 mgs02 kernel: end_request: I/O error, dev sdl, sector 33639 Oct 17 21:56:07 mgs02 kernel: sd 2:0:0:8: SCSI error: return code = 0x10000 Oct 17 21:56:07 mgs02 kernel: end_request: I/O error, dev sdl, sector 33639 Oct 17 21:56:07 mgs02 kernel: sd 2:0:0:8: SCSI error: return code = 0x10000 Oct 17 21:56:07 mgs02 kernel: end_request: I/O error, dev sdl, sector 33639 Oct 17 21:56:07 mgs02 kernel: sd 2:0:0:8: SCSI error: return code = 0x10000 Oct 17 21:56:07 mgs02 kernel: end_request: I/O error, dev sdl, sector 33639 Oct 17 21:56:07 mgs02 kernel: sd 2:0:0:8: SCSI error: return code = 0x10000 Oct 17 21:56:07 mgs02 kernel: end_request: I/O error, dev sdl, sector 33639 Oct 17 21:56:07 mgs02 kernel: sd 2:0:0:8: SCSI error: return code = 0x10000 Oct 17 21:56:07 mgs02 kernel: end_request: I/O error, dev sdl, sector 33639 Oct 17 21:56:07 mgs02 kernel: 2:0:0:8: SCSI error: return code = 0x10000 ... ... ---- <kernel version: 2.6.16.27, SLES 10> Where sector is actually the first requested sector for which bio was submitted. These errors do not cease until the device is plugged in - and after which the user space happily moves forward (completes the IO). Also, I have placed a code like if(bio->bi_size > 0) return 1; in the end_bio function call (first lines) - to discard sub-completion calls made to the routine. the code for creating the bio is: ----- <snip> size = end_addr - addr; map_e_addr = (char *)(((__u64)end_addr + PAGE_SIZE - 1) & \ (~(PAGE_SIZE - 1))); size = (map_e_addr - map_addr); while ((map_addr)<map_e_addr) { bio = bio_map_user(bdev_get_queue(bdev), bdev,(unsigned long) \ map_addr,(unsigned long) (size), rw==READ); if(!bio){ dump(" bio_map_user failed\n"); retval = -EFAULT; return retval; } bio->bi_sector = sector; bio->bi_bdev = bdev; bio->bi_rw = rw; bio->bi_private = greq; bio->bi_end_io = mf_end_bio; bio->bi_io_vec[0].bv_offset = bsize; bio->bi_io_vec[0].bv_len = (PAGE_SIZE - bsize); if( (__u64)end_addr % PAGE_SIZE){ bsize = (unsigned long)((__u64)end_addr - ((__u64)end_addr & \ ~(PAGE_SIZE - 1))); bio->bi_io_vec[bio->bi_vcnt-1].bv_len = bsize; } bsize = 0; bio->bi_size = (unsigned long)(end_addr - start_addr); bio->bi_next = top_bio; blk_queue_bounce(bdev_get_queue(bdev), &bio); top_bio=bio; bio = NULL; total_bios++; map_addr += size; bsize_total += (end_addr - start_addr); size = (unsigned long) (map_e_addr - map_addr); if(map_addr>=map_e_addr) break; <snip> ----- Where: - start_addr and end_addr are the starting and ending addresses of the buffer allocated by the user space process requesting IO - map_addr and map_e_addr are starting and ending address of aligned user space address - map_addr aligned to PAGE_SIZE less than start_addr and map_e_addr aligned to PAGE_SIZE more than end_addr - bsize is the offset of the first page of the bio vector - (start_addr - map_addr). - It is assuring that at max 3 pages are being requested (test setup) so that request does not overflow max allowable io vectors in bio Any help/pointer/reference would be really appreciated. Also, this is a test code and hence does not confirm to good coding styles - please ignore that. Regards Shreyansh Shreyansh Jain <shreyansh.jain <at> gmail.com> writes: [removed the previous text to curtail the length of this post] -- To unsubscribe from this list: send an email with "unsubscribe kernelnewbies" to ecartis@xxxxxxxxxxxx Please read the FAQ at http://kernelnewbies.org/FAQ
