/* * Copyright 1996 The Australian National University. * Copyright 1996 Fujitsu Laboratories Limited * * This software may be distributed under the terms of the Gnu * Public License version 2 or later */ /* * ddv.c - Single AP1000 block driver. * * This block driver performs io operations to the ddv option * board. (Hopefully:) * */ #include #include #include #include #include #include #include #include #define __KERNEL_SYSCALLS__ #include #include #include #include #include #include #include #define MAJOR_NR DDV_MAJOR #include #include #include #define DDV_DEBUG 0 #define AIR_DISK 1 #define SECTOR_SIZE 512 /* we can have lots of partitions */ #define PARTN_BITS 6 #define NUM_DDVDEVS (1<> 9) - 1) /* try to read ahead a bit */ #define DDV_READ_AHEAD 64 static int have_ddv_board = 1; static unsigned num_options = 0; static unsigned this_option = 0; extern int ddv_get_mlist(unsigned mptr[],int bnum); extern int ddv_set_request(struct request *req, int request_type,int bnum,int mlist,int len,int offset); extern void ddv_load_kernel(char *opcodep); extern int ddv_restart_cpu(void); extern int ddv_mlist_available(void); static int ddv_revalidate(kdev_t dev, struct gendisk *gdev); static void ddv_geninit(struct gendisk *ignored); static void ddv_release(struct inode * inode, struct file * filp); static void ddv_request1(void); static char *ddv_opcodep = NULL; static struct request *next_request = NULL; static DECLARE_WAIT_QUEUE_HEAD(busy_wait); static int ddv_blocksizes[NUM_DDVDEVS]; /* in bytes */ int ddv_sect_length[NUM_DDVDEVS]; /* in sectors */ int ddv_blk_length[NUM_DDVDEVS]; /* in blocks */ /* these are used by the ddv_daemon, which services remote disk requests */ static struct remote_request *rem_queue = NULL; static struct remote_request *rem_queue_end; static DECLARE_WAIT_QUEUE_HEAD(ddv_daemon_wait); static int opiu_kernel_loaded = 0; static struct { unsigned reads, writes, blocks, rq_started, rq_finished, errors; unsigned sectors_read, sectors_written; } ddv_stats; static struct hd_struct partition_tables[NUM_DDVDEVS]; static struct gendisk ddv_gendisk = { MAJOR_NR, /* Major number */ DEVICE_NAME, /* Major name */ PARTN_BITS, /* Bits to shift to get real from partition */ 1 << PARTN_BITS, /* Number of partitions per real */ 1, /* maximum number of real */ #ifdef MODULE NULL, /* called from init_module */ #else ddv_geninit, /* init function */ #endif partition_tables,/* hd struct */ ddv_blk_length, /* block sizes */ 1, /* number */ (void *) NULL, /* internal */ NULL /* next */ }; struct ddv_geometry { unsigned char heads; unsigned char sectors; unsigned short cylinders; unsigned long start; }; static struct ddv_geometry ddv_geometry; struct remote_request { union { struct remote_request *next; void (*fn)(void); } u; unsigned bnum; /* how many blocks does this contain */ struct request *reqp; /* pointer to the request on the original cell */ unsigned cell; /* what cell is the request from */ struct request req; /* details of the request */ }; static void ddv_set_optadr(void) { unsigned addr = 0x11000000; OPT_IO(OBASE) = addr; MSC_IO(MSC_OPTADR) = ((addr & 0xff000000)>>16) | ((OPTION_BASE & 0xf0000000)>>24) | ((OPTION_BASE + 0x10000000)>>28); OPT_IO(PRST) = 0; } extern struct RequestTable *RTable; extern struct OPrintBufArray *PrintBufs; extern struct OAlignBufArray *AlignBufs; extern struct DiskInfo *DiskInfo; static void ddv_release(struct inode * inode, struct file * filp) { #if DEBUG printk("ddv_release started\n"); #endif sync_dev(inode->i_rdev); #if DEBUG printk("ddv_release done\n"); #endif } static unsigned in_request = 0; static unsigned req_queued = 0; static void ddv_end_request(int uptodate,struct request *req) { struct buffer_head * bh; ddv_stats.rq_finished++; /* printk("ddv_end_request(%d,%p)\n",uptodate,req); */ req->errors = 0; if (!uptodate) { printk("end_request: I/O error, dev %s, sector %lu\n", kdevname(req->rq_dev), req->sector); req->nr_sectors--; req->nr_sectors &= ~SECTOR_MASK; req->sector += (BLOCK_SIZE / SECTOR_SIZE); req->sector &= ~SECTOR_MASK; ddv_stats.errors++; } if ((bh = req->bh) != NULL) { req->bh = bh->b_reqnext; bh->b_reqnext = NULL; mark_buffer_uptodate(bh, uptodate); unlock_buffer(bh); if ((bh = req->bh) != NULL) { req->current_nr_sectors = bh->b_size >> 9; if (req->nr_sectors < req->current_nr_sectors) { req->nr_sectors = req->current_nr_sectors; printk("end_request: buffer-list destroyed\n"); } req->buffer = bh->b_data; printk("WARNING: ddv: more sectors!\n"); ddv_stats.errors++; return; } } if (req->sem != NULL) up(req->sem); req->rq_status = RQ_INACTIVE; wake_up(&wait_for_request); } /* check that a request is all OK to process */ static int request_ok(struct request *req) { int minor; if (!req) return 0; if (MAJOR(req->rq_dev) != MAJOR_NR) panic(DEVICE_NAME ": bad major number\n"); if (!buffer_locked(req->bh)) panic(DEVICE_NAME ": block not locked"); minor = MINOR(req->rq_dev); if (minor >= NUM_DDVDEVS) { printk("ddv_request: Invalid minor (%d)\n", minor); return 0; } if ((req->sector + req->current_nr_sectors) > ddv_sect_length[minor]) { printk("ddv: out of range minor=%d offset=%d len=%d sect_length=%d\n", minor,(int)req->sector,(int)req->current_nr_sectors, ddv_sect_length[minor]); return 0; } if (req->cmd != READ && req->cmd != WRITE) { printk("unknown request type %d\n",req->cmd); return 0; } /* it seems to be OK */ return 1; } static void complete_request(struct request *req,int bnum) { while (bnum--) { ddv_end_request(1,req); req = req->next; } } static int completion_pointer = 0; static void check_completion(void) { int i,bnum; struct request *req; if (!RTable) return; for (; (i=completion_pointer) != RTable->ddv_pointer && RTable->async_info[i].status == DDV_REQ_FREE; completion_pointer = INC_T(completion_pointer)) { req = (struct request *)RTable->async_info[i].argv[7]; bnum = RTable->async_info[i].bnum; if (!req || !bnum) { printk("%s(%d)\n",__FILE__,__LINE__); ddv_stats.errors++; continue; } RTable->async_info[i].status = 0; RTable->async_info[i].argv[7] = 0; complete_request(req,bnum); in_request--; } } static struct request *get_request_queue(struct request *oldq) { struct request *req,*req2; /* skip any non-active or bad requests */ skip1: if (!(req = CURRENT)) return oldq; if (req->rq_status != RQ_ACTIVE) { CURRENT = req->next; goto skip1; } if (!request_ok(req)) { ddv_end_request(0,req); CURRENT = req->next; goto skip1; } /* now grab as many as we can */ req_queued++; for (req2 = req; req2->next && req2->next->rq_status == RQ_ACTIVE && request_ok(req2->next); req2 = req2->next) req_queued++; /* leave CURRENT pointing at the bad ones */ CURRENT = req2->next; /* chop our list at that point */ req2->next = NULL; if (!oldq) return req; for (req2=oldq;req2->next;req2=req2->next) ; req2->next = req; return oldq; } static void ddv_rem_complete(struct remote_request *rem) { unsigned flags; int bnum = rem->bnum; struct request *req = rem->reqp; complete_request(req,bnum); in_request--; save_flags(flags); cli(); ddv_request1(); restore_flags(flags); } /* * The background ddv daemon. This receives remote disk requests * and processes them via the normal block operations */ static int ddv_daemon(void *unused) { current->session = 1; current->pgrp = 1; sprintf(current->comm, "ddv_daemon"); spin_lock_irq(¤t->sigmask_lock); sigfillset(¤t->blocked); /* block all signals */ recalc_sigpending(current); spin_unlock_irq(¤t->sigmask_lock); /* Give it a realtime priority. */ current->policy = SCHED_FIFO; current->priority = 32; /* Fixme --- we need to standardise our namings for POSIX.4 realtime scheduling priorities. */ printk("Started ddv_daemon\n"); while (1) { struct remote_request *rem; unsigned flags; struct buffer_head *bhlist[MAX_BNUM*4]; int i,j,minor,len,shift,offset; save_flags(flags); cli(); while (!rem_queue) { spin_lock_irq(¤t->sigmask_lock); flush_signals(current); spin_unlock_irq(¤t->sigmask_lock); interruptible_sleep_on(&ddv_daemon_wait); } rem = rem_queue; rem_queue = rem->u.next; restore_flags(flags); minor = MINOR(rem->req.rq_dev); len = rem->req.current_nr_sectors; offset = rem->req.sector; /* work out the conversion to the local block size from sectors */ for (shift=0; (SECTOR_SIZE<req.rq_dev, offset >> shift, ddv_blocksizes[minor]); if (!buffer_uptodate(bhlist[i])) ll_rw_block(READ,1,&bhlist[i]); offset += 1<u.fn = ddv_rem_complete; tnet_rpc(rem->cell,rem,sizeof(int)*3,1); } } /* receive a remote disk request */ static void ddv_rem_queue(char *data,unsigned size) { unsigned flags; struct remote_request *rem = (struct remote_request *) kmalloc(size,GFP_ATOMIC); if (!rem) { /* oh bugger! */ ddv_stats.errors++; return; } memcpy(rem,data,size); rem->u.next = NULL; save_flags(flags); cli(); /* add it to our remote request queue */ if (!rem_queue) rem_queue = rem; else rem_queue_end->u.next = rem; rem_queue_end = rem; restore_flags(flags); wake_up(&ddv_daemon_wait); } /* which disk should this request go to */ static inline unsigned pardisk_num(struct request *req) { int minor = MINOR(req->rq_dev); unsigned stripe; unsigned cell; if (minor < PARDISK_BASE) return this_option; stripe = req->sector >> STRIPE_SHIFT; cell = stripe % num_options; return cell; } /* check if a 2nd request can be tacked onto the first */ static inline int contiguous(struct request *req1,struct request *req2) { if (req2->cmd != req1->cmd || req2->rq_dev != req1->rq_dev || req2->sector != req1->sector + req1->current_nr_sectors || req2->current_nr_sectors != req1->current_nr_sectors) return 0; if (pardisk_num(req1) != pardisk_num(req2)) return 0; return 1; } static void ddv_request1(void) { struct request *req,*req1,*req2; unsigned offset,len,req_num,mlist,bnum,available=0; static unsigned mptrs[MAX_BNUM]; unsigned cell; if (in_request > REQUEST_HIGH) return; next_request = get_request_queue(next_request); while ((req = next_request)) { int minor; if (in_request >= MAX_REQUEST) return; if (in_request>1 && req_queued MAX_BNUM) available = MAX_BNUM; offset = req->sector; len = req->current_nr_sectors; minor = MINOR(req->rq_dev); mptrs[0] = (int)req->buffer; for (bnum=1,req1=req,req2=req->next; req2 && bnumnext) { mptrs[bnum++] = (int)req2->buffer; } next_request = req2; req_queued -= bnum; ddv_stats.blocks += bnum; ddv_stats.rq_started += bnum; if (req->cmd == READ) { ddv_stats.reads++; ddv_stats.sectors_read += len*bnum; } else { ddv_stats.writes++; ddv_stats.sectors_written += len*bnum; } if (minor >= PARDISK_BASE) { /* translate the request to the normal partition */ unsigned stripe; minor -= PARDISK_BASE; stripe = offset >> STRIPE_SHIFT; stripe /= num_options; offset = (stripe << STRIPE_SHIFT) + (offset & ((1<u.fn = ddv_rem_queue; rem->cell = this_option; rem->bnum = bnum; rem->req = *req; rem->reqp = req; rem->req.rq_dev = MKDEV(MAJOR_NR,minor); rem->req.sector = offset; memcpy(remlist,mptrs,sizeof(mptrs[0])*bnum); if (tnet_rpc(cell,rem,size,1) != 0) { kfree_s(rem,size); return; } } else { /* its a local request */ if ((mlist = ddv_get_mlist(mptrs,bnum)) == -1) { ddv_stats.errors++; panic("ddv: mlist corrupted"); } req_num = RTable->cell_pointer; RTable->async_info[req_num].status = req->cmd==READ?DDV_RAWREAD_REQ:DDV_RAWWRITE_REQ; RTable->async_info[req_num].bnum = bnum; RTable->async_info[req_num].argv[0] = mlist; RTable->async_info[req_num].argv[1] = len; RTable->async_info[req_num].argv[2] = offset + partition_tables[minor].start_sect; RTable->async_info[req_num].argv[3] = bnum; RTable->async_info[req_num].argv[7] = (unsigned)req; RTable->cell_pointer = INC_T(RTable->cell_pointer); } in_request++; } } static void ddv_request(void) { cli(); ddv_request1(); sti(); } static void check_printbufs(void) { int i; if (!PrintBufs) return; while (PrintBufs->option_counter != PrintBufs->cell_counter) { i = PrintBufs->cell_counter; printk("opiu (%d): ",i); if (((unsigned)PrintBufs->bufs[i].fmt) > 0x100000) printk("Error: bad format in printk at %p\n", PrintBufs->bufs[i].fmt); else printk(PrintBufs->bufs[i].fmt + OPIBUS_BASE, PrintBufs->bufs[i].args[0], PrintBufs->bufs[i].args[1], PrintBufs->bufs[i].args[2], PrintBufs->bufs[i].args[3], PrintBufs->bufs[i].args[4], PrintBufs->bufs[i].args[5]); if (++PrintBufs->cell_counter == PRINT_BUFS) PrintBufs->cell_counter = 0; } } static void ddv_interrupt(int irq, void *dev_id, struct pt_regs *regs) { unsigned long flags; save_flags(flags); cli(); OPT_IO(IRC1) = 0x80000000; check_printbufs(); check_completion(); ddv_request1(); restore_flags(flags); } static int ddv_open(struct inode * inode, struct file * filp) { int minor = MINOR(inode->i_rdev); if (!have_ddv_board || minor >= NUM_DDVDEVS) return -ENODEV; if (minor >= PARDISK_BASE) { ddv_sect_length[minor] = ddv_sect_length[minor - PARDISK_BASE]; ddv_blk_length[minor] = ddv_blk_length[minor - PARDISK_BASE]; } return 0; } static void ddv_open_reply(struct cap_request *creq) { int size = creq->size - sizeof(*creq); ddv_opcodep = (char *)kmalloc(size,GFP_ATOMIC); read_bif(ddv_opcodep, size); #if DEBUG printk("received opiu kernel of size %d\n",size); #endif if (size == 0) have_ddv_board = 0; wake_up(&busy_wait); } static void ddv_load_opiu(void) { int i; struct cap_request creq; /* if the opiu kernel is already loaded then we don't do anything */ if (!have_ddv_board || opiu_kernel_loaded) return; bif_register_request(REQ_DDVOPEN,ddv_open_reply); /* send the open request to the front end */ creq.cid = mpp_cid(); creq.type = REQ_DDVOPEN; creq.header = 0; creq.size = sizeof(creq); bif_queue(&creq,0,0); ddv_set_optadr(); while (!ddv_opcodep) sleep_on(&busy_wait); if (!have_ddv_board) return; ddv_load_kernel(ddv_opcodep); kfree(ddv_opcodep); ddv_opcodep = NULL; if (ddv_restart_cpu()) return; ddv_sect_length[0] = DiskInfo->blocks; ddv_blk_length[0] = DiskInfo->blocks >> 1; ddv_blocksizes[0] = BLOCK_SIZE; ddv_geometry.cylinders = ddv_sect_length[0] / (ddv_geometry.heads*ddv_geometry.sectors); ddv_gendisk.part[0].start_sect = 0; ddv_gendisk.part[0].nr_sects = ddv_sect_length[0]; resetup_one_dev(&ddv_gendisk, 0); for (i=0;i> 1; } /* setup the parallel partitions by multiplying the normal partition by the number of options */ for (;imax_p; start = target << gdev->minor_shift; printk("ddv_revalidate dev=%d target=%d max_p=%d start=%d\n", dev,target,max_p,start); for (i=max_p - 1; i >=0 ; i--) { int minor = start + i; kdev_t devi = MKDEV(gdev->major, minor); sync_dev(devi); invalidate_inodes(devi); invalidate_buffers(devi); gdev->part[minor].start_sect = 0; gdev->part[minor].nr_sects = 0; }; ddv_sect_length[start] = DiskInfo->blocks; ddv_blk_length[start] = DiskInfo->blocks >> 1; gdev->part[start].nr_sects = ddv_sect_length[start]; resetup_one_dev(gdev, target); printk("sect_length[%d]=%d blk_length[%d]=%d\n", start,ddv_sect_length[start], start,ddv_blk_length[start]); for (i=0;ipart[start+i].nr_sects; ddv_blk_length[start+i] = gdev->part[start+i].nr_sects >> 1; if (gdev->part[start+i].nr_sects) printk("partition[%d] start=%d length=%d\n",i, (int)gdev->part[start+i].start_sect, (int)gdev->part[start+i].nr_sects); } return 0; } static int ddv_ioctl(struct inode *inode, struct file *file, unsigned int cmd, unsigned long arg) { int err; struct ddv_geometry *loc = (struct ddv_geometry *) arg; int dev; int minor = MINOR(inode->i_rdev); if ((!inode) || !(inode->i_rdev)) return -EINVAL; dev = DEVICE_NR(inode->i_rdev); #if DEBUG printk("ddv_ioctl: cmd=%x dev=%x minor=%d\n", cmd, dev, minor); #endif switch (cmd) { case HDIO_GETGEO: printk("\tHDIO_GETGEO\n"); if (!loc) return -EINVAL; if (put_user(ddv_geometry.heads, (char *) &loc->heads)) return -EFAULT; if (put_user(ddv_geometry.sectors, (char *) &loc->sectors)) return -EFAULT; if (put_user(ddv_geometry.cylinders, (short *) &loc->cylinders)) return -EFAULT; if (put_user(ddv_geometry.start, (long *) &loc->start)) return -EFAULT; return 0; case HDIO_GET_MULTCOUNT : printk("\tHDIO_GET_MULTCOUNT\n"); return -EINVAL; case HDIO_GET_IDENTITY : printk("\tHDIO_GET_IDENTITY\n"); return -EINVAL; case HDIO_GET_NOWERR : printk("\tHDIO_GET_NOWERR\n"); return -EINVAL; case HDIO_SET_NOWERR : printk("\tHDIO_SET_NOWERR\n"); return -EINVAL; case BLKRRPART: printk("\tBLKRRPART\n"); if (!capable(CAP_SYS_ADMIN)) return -EACCES; return ddv_revalidate(inode->i_rdev,&ddv_gendisk); case BLKGETSIZE: /* Return device size */ if (put_user(ddv_sect_length[minor],(long *) arg)) return -EFAULT; #if DEBUG printk("BLKGETSIZE gave %d\n",ddv_sect_length[minor]); #endif return 0; default: printk("ddv_ioctl: Invalid cmd=%d(0x%x)\n", cmd, cmd); return -EINVAL; }; } static struct file_operations ddv_fops = { NULL, /* lseek - default */ block_read, /* read */ block_write, /* write */ NULL, /* readdir - bad */ NULL, /* poll */ ddv_ioctl, /* ioctl */ NULL, /* mmap */ ddv_open, /* open */ NULL, /* flush */ ddv_release, block_fsync /* fsync */ }; static void ddv_status(void) { if (!have_ddv_board) { printk("no ddv board\n"); return; } printk(" in_request %u req_queued %u MTable: start=%u end=%u Requests: started=%u finished=%u Requests: completion_pointer=%u ddv_pointer=%u cell_pointer=%u PrintBufs: option_counter=%u cell_counter=%u ddv_stats: reads=%u writes=%u blocks=%u ddv_stats: sectors_read=%u sectors_written=%u CURRENT=%p next_request=%p errors=%u ", in_request,req_queued, RTable->start_mtable,RTable->end_mtable, ddv_stats.rq_started,ddv_stats.rq_finished, completion_pointer,RTable->ddv_pointer,RTable->cell_pointer, PrintBufs->option_counter,PrintBufs->cell_counter, ddv_stats.reads,ddv_stats.writes,ddv_stats.blocks, ddv_stats.sectors_read,ddv_stats.sectors_written, CURRENT,next_request, ddv_stats.errors); } int ddv_init(void) { int cid; cid = mpp_cid(); if (register_blkdev(MAJOR_NR,DEVICE_NAME,&ddv_fops)) { printk("ap: unable to get major %d for ap block dev\n", MAJOR_NR); return -1; } printk("ddv_init: register dev %d\n", MAJOR_NR); blk_dev[MAJOR_NR].request_fn = DEVICE_REQUEST; read_ahead[MAJOR_NR] = DDV_READ_AHEAD; bif_add_debug_key('d',ddv_status,"DDV status"); ddv_gendisk.next = gendisk_head; gendisk_head = &ddv_gendisk; num_options = mpp_num_cells(); this_option = mpp_cid(); kernel_thread(ddv_daemon, NULL, 0); return(0); } static void ddv_geninit(struct gendisk *ignored) { int i; static int done = 0; if (done) printk("ddv_geninit already done!\n"); done = 1; printk("ddv_geninit\n"); /* request interrupt line 2 */ if (request_irq(APOPT0_IRQ,ddv_interrupt,SA_INTERRUPT,"apddv",NULL)) { printk("Failed to install ddv interrupt handler\n"); } for (i=0;inext)) if (*gdp == &ddv_gendisk) break; if (*gdp) *gdp = (*gdp)->next; free_irq(APOPT0_IRQ, NULL); blk_dev[MAJOR_NR].request_fn = 0; } #endif /* MODULE */