[SCSI] atp870u: convert to accessors and !use_sg cleanup

- convert to accessors and !use_sg cleanup
 - Probably not ready for sg-chaining

Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
Acked-by: Christoph Hellwig <hch@infradead.org>
Cc: jameshsu <jameshsu@acard.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
This commit is contained in:
Boaz Harrosh 2007-09-09 21:06:23 +03:00 committed by James Bottomley
parent c73961e563
commit fe7ed98fd4

View file

@ -471,18 +471,8 @@ go_42:
/* /*
* Complete the command * Complete the command
*/ */
if (workreq->use_sg) { scsi_dma_unmap(workreq);
pci_unmap_sg(dev->pdev,
(struct scatterlist *)workreq->request_buffer,
workreq->use_sg,
workreq->sc_data_direction);
} else if (workreq->request_bufflen &&
workreq->sc_data_direction != DMA_NONE) {
pci_unmap_single(dev->pdev,
workreq->SCp.dma_handle,
workreq->request_bufflen,
workreq->sc_data_direction);
}
spin_lock_irqsave(dev->host->host_lock, flags); spin_lock_irqsave(dev->host->host_lock, flags);
(*workreq->scsi_done) (workreq); (*workreq->scsi_done) (workreq);
#ifdef ED_DBGP #ifdef ED_DBGP
@ -624,7 +614,7 @@ static int atp870u_queuecommand(struct scsi_cmnd * req_p,
c = scmd_channel(req_p); c = scmd_channel(req_p);
req_p->sense_buffer[0]=0; req_p->sense_buffer[0]=0;
req_p->resid = 0; scsi_set_resid(req_p, 0);
if (scmd_channel(req_p) > 1) { if (scmd_channel(req_p) > 1) {
req_p->result = 0x00040000; req_p->result = 0x00040000;
done(req_p); done(req_p);
@ -722,7 +712,6 @@ static void send_s870(struct atp_unit *dev,unsigned char c)
unsigned short int tmpcip, w; unsigned short int tmpcip, w;
unsigned long l, bttl = 0; unsigned long l, bttl = 0;
unsigned int workport; unsigned int workport;
struct scatterlist *sgpnt;
unsigned long sg_count; unsigned long sg_count;
if (dev->in_snd[c] != 0) { if (dev->in_snd[c] != 0) {
@ -793,6 +782,8 @@ oktosend:
} }
printk("\n"); printk("\n");
#endif #endif
l = scsi_bufflen(workreq);
if (dev->dev_id == ATP885_DEVID) { if (dev->dev_id == ATP885_DEVID) {
j = inb(dev->baseport + 0x29) & 0xfe; j = inb(dev->baseport + 0x29) & 0xfe;
outb(j, dev->baseport + 0x29); outb(j, dev->baseport + 0x29);
@ -800,12 +791,11 @@ oktosend:
} }
if (workreq->cmnd[0] == READ_CAPACITY) { if (workreq->cmnd[0] == READ_CAPACITY) {
if (workreq->request_bufflen > 8) { if (l > 8)
workreq->request_bufflen = 0x08; l = 8;
}
} }
if (workreq->cmnd[0] == 0x00) { if (workreq->cmnd[0] == 0x00) {
workreq->request_bufflen = 0; l = 0;
} }
tmport = workport + 0x1b; tmport = workport + 0x1b;
@ -852,40 +842,8 @@ oktosend:
#ifdef ED_DBGP #ifdef ED_DBGP
printk("dev->id[%d][%d].devsp = %2x\n",c,target_id,dev->id[c][target_id].devsp); printk("dev->id[%d][%d].devsp = %2x\n",c,target_id,dev->id[c][target_id].devsp);
#endif #endif
/*
* Figure out the transfer size sg_count = scsi_dma_map(workreq);
*/
if (workreq->use_sg) {
#ifdef ED_DBGP
printk("Using SGL\n");
#endif
l = 0;
sgpnt = (struct scatterlist *) workreq->request_buffer;
sg_count = pci_map_sg(dev->pdev, sgpnt, workreq->use_sg,
workreq->sc_data_direction);
for (i = 0; i < workreq->use_sg; i++) {
if (sgpnt[i].length == 0 || workreq->use_sg > ATP870U_SCATTER) {
panic("Foooooooood fight!");
}
l += sgpnt[i].length;
}
#ifdef ED_DBGP
printk( "send_s870: workreq->use_sg %d, sg_count %d l %8ld\n", workreq->use_sg, sg_count, l);
#endif
} else if(workreq->request_bufflen && workreq->sc_data_direction != PCI_DMA_NONE) {
#ifdef ED_DBGP
printk("Not using SGL\n");
#endif
workreq->SCp.dma_handle = pci_map_single(dev->pdev, workreq->request_buffer,
workreq->request_bufflen,
workreq->sc_data_direction);
l = workreq->request_bufflen;
#ifdef ED_DBGP
printk( "send_s870: workreq->use_sg %d, l %8ld\n", workreq->use_sg, l);
#endif
} else l = 0;
/* /*
* Write transfer size * Write transfer size
*/ */
@ -938,16 +896,16 @@ oktosend:
* a linear chain. * a linear chain.
*/ */
if (workreq->use_sg) { if (l) {
sgpnt = (struct scatterlist *) workreq->request_buffer; struct scatterlist *sgpnt;
i = 0; i = 0;
for (j = 0; j < workreq->use_sg; j++) { scsi_for_each_sg(workreq, sgpnt, sg_count, j) {
bttl = sg_dma_address(&sgpnt[j]); bttl = sg_dma_address(sgpnt);
l=sg_dma_len(&sgpnt[j]); l=sg_dma_len(sgpnt);
#ifdef ED_DBGP #ifdef ED_DBGP
printk("1. bttl %x, l %x\n",bttl, l); printk("1. bttl %x, l %x\n",bttl, l);
#endif #endif
while (l > 0x10000) { while (l > 0x10000) {
(((u16 *) (prd))[i + 3]) = 0x0000; (((u16 *) (prd))[i + 3]) = 0x0000;
(((u16 *) (prd))[i + 2]) = 0x0000; (((u16 *) (prd))[i + 2]) = 0x0000;
(((u32 *) (prd))[i >> 1]) = cpu_to_le32(bttl); (((u32 *) (prd))[i >> 1]) = cpu_to_le32(bttl);
@ -965,32 +923,6 @@ oktosend:
printk("prd %4x %4x %4x %4x\n",(((unsigned short int *)prd)[0]),(((unsigned short int *)prd)[1]),(((unsigned short int *)prd)[2]),(((unsigned short int *)prd)[3])); printk("prd %4x %4x %4x %4x\n",(((unsigned short int *)prd)[0]),(((unsigned short int *)prd)[1]),(((unsigned short int *)prd)[2]),(((unsigned short int *)prd)[3]));
printk("2. bttl %x, l %x\n",bttl, l); printk("2. bttl %x, l %x\n",bttl, l);
#endif #endif
} else {
/*
* For a linear request write a chain of blocks
*/
bttl = workreq->SCp.dma_handle;
l = workreq->request_bufflen;
i = 0;
#ifdef ED_DBGP
printk("3. bttl %x, l %x\n",bttl, l);
#endif
while (l > 0x10000) {
(((u16 *) (prd))[i + 3]) = 0x0000;
(((u16 *) (prd))[i + 2]) = 0x0000;
(((u32 *) (prd))[i >> 1]) = cpu_to_le32(bttl);
l -= 0x10000;
bttl += 0x10000;
i += 0x04;
}
(((u16 *) (prd))[i + 3]) = cpu_to_le16(0x8000);
(((u16 *) (prd))[i + 2]) = cpu_to_le16(l);
(((u32 *) (prd))[i >> 1]) = cpu_to_le32(bttl);
#ifdef ED_DBGP
printk("prd %4x %4x %4x %4x\n",(((unsigned short int *)prd)[0]),(((unsigned short int *)prd)[1]),(((unsigned short int *)prd)[2]),(((unsigned short int *)prd)[3]));
printk("4. bttl %x, l %x\n",bttl, l);
#endif
} }
tmpcip += 4; tmpcip += 4;
#ifdef ED_DBGP #ifdef ED_DBGP