libata: fix ATAPI DMA alignment issues

ATAPI needs to be padded to next 4 byte boundary, if misaligned.

Original work by me, many fixes from Tejun Heo.

5 files changed, 148 insertions, 40 deletions

diff --git a/drivers/scsi/ahci.c b/drivers/scsi/ahci.c
index c2c8fa828e24..6e4bb36f8d7c 100644
--- a/drivers/scsi/ahci.c
+++ b/drivers/scsi/ahci.c
@@ -314,8 +314,15 @@ static int ahci_port_start(struct ata_port *ap)
                 return -ENOMEM;
         memset(pp, 0, sizeof(*pp));
 
+        ap->pad = dma_alloc_coherent(dev, ATA_DMA_PAD_BUF_SZ, &ap->pad_dma, GFP_KERNEL);
+        if (!ap->pad) {
+                kfree(pp);
+                return -ENOMEM;
+        }
+
         mem = dma_alloc_coherent(dev, AHCI_PORT_PRIV_DMA_SZ, &mem_dma, GFP_KERNEL);
         if (!mem) {
+                dma_free_coherent(dev, ATA_DMA_PAD_BUF_SZ, ap->pad, ap->pad_dma);
                 kfree(pp);
                 return -ENOMEM;
         }
@@ -391,6 +398,7 @@ static void ahci_port_stop(struct ata_port *ap)
         ap->private_data = NULL;
         dma_free_coherent(dev, AHCI_PORT_PRIV_DMA_SZ,
                           pp->cmd_slot, pp->cmd_slot_dma);
+        dma_free_coherent(dev, ATA_DMA_PAD_BUF_SZ, ap->pad, ap->pad_dma);
         kfree(pp);
 }
 
@@ -476,23 +484,23 @@ static void ahci_tf_read(struct ata_port *ap, struct ata_taskfile *tf)
 static void ahci_fill_sg(struct ata_queued_cmd *qc)
 {
         struct ahci_port_priv *pp = qc->ap->private_data;
-        unsigned int i;
+        struct scatterlist *sg;
+        struct ahci_sg *ahci_sg;
 
         VPRINTK("ENTER\n");
 
         /*
          * Next, the S/G list.
          */
-        for (i = 0; i < qc->n_elem; i++) {
-                u32 sg_len;
-                dma_addr_t addr;
-
-                addr = sg_dma_address(&qc->sg[i]);
-                sg_len = sg_dma_len(&qc->sg[i]);
-
-                pp->cmd_tbl_sg[i].addr = cpu_to_le32(addr & 0xffffffff);
-                pp->cmd_tbl_sg[i].addr_hi = cpu_to_le32((addr >> 16) >> 16);
-                pp->cmd_tbl_sg[i].flags_size = cpu_to_le32(sg_len - 1);
+        ahci_sg = pp->cmd_tbl_sg;
+        ata_for_each_sg(sg, qc) {
+                dma_addr_t addr = sg_dma_address(sg);
+                u32 sg_len = sg_dma_len(sg);
+
+                ahci_sg->addr = cpu_to_le32(addr & 0xffffffff);
+                ahci_sg->addr_hi = cpu_to_le32((addr >> 16) >> 16);
+                ahci_sg->flags_size = cpu_to_le32(sg_len - 1);
+                ahci_sg++;
         }
 }
 
diff --git a/drivers/scsi/libata-core.c b/drivers/scsi/libata-core.c
index e5b01997117a..943b44c3c16f 100644
--- a/drivers/scsi/libata-core.c
+++ b/drivers/scsi/libata-core.c
@@ -2156,8 +2156,9 @@ static void ata_dev_set_xfermode(struct ata_port *ap, struct ata_device *dev)
 static void ata_sg_clean(struct ata_queued_cmd *qc)
 {
         struct ata_port *ap = qc->ap;
-        struct scatterlist *sg = qc->sg;
+        struct scatterlist *sg = qc->__sg;
         int dir = qc->dma_dir;
+        void *pad_buf = NULL;
 
         assert(qc->flags & ATA_QCFLAG_DMAMAP);
         assert(sg != NULL);
@@ -2167,14 +2168,35 @@ static void ata_sg_clean(struct ata_queued_cmd *qc)
 
         DPRINTK("unmapping %u sg elements\n", qc->n_elem);
 
-        if (qc->flags & ATA_QCFLAG_SG)
+        /* if we padded the buffer out to 32-bit bound, and data
+         * xfer direction is from-device, we must copy from the
+         * pad buffer back into the supplied buffer
+         */
+        if (qc->pad_len && !(qc->tf.flags & ATA_TFLAG_WRITE))
+                pad_buf = ap->pad + (qc->tag * ATA_DMA_PAD_SZ);
+
+        if (qc->flags & ATA_QCFLAG_SG) {
                 dma_unmap_sg(ap->host_set->dev, sg, qc->n_elem, dir);
-        else
+                /* restore last sg */
+                sg[qc->orig_n_elem - 1].length += qc->pad_len;
+                if (pad_buf) {
+                        struct scatterlist *psg = &qc->pad_sgent;
+                        void *addr = kmap_atomic(psg->page, KM_IRQ0);
+                        memcpy(addr + psg->offset, pad_buf, qc->pad_len);
+                        kunmap_atomic(psg->page, KM_IRQ0);
+                }
+        } else {
                 dma_unmap_single(ap->host_set->dev, sg_dma_address(&sg[0]),
                                  sg_dma_len(&sg[0]), dir);
+                /* restore sg */
+                sg->length += qc->pad_len;
+                if (pad_buf)
+                        memcpy(qc->buf_virt + sg->length - qc->pad_len,
+                               pad_buf, qc->pad_len);
+        }
 
         qc->flags &= ~ATA_QCFLAG_DMAMAP;
-        qc->sg = NULL;
+        qc->__sg = NULL;
 }
 
 /**
@@ -2190,15 +2212,15 @@ static void ata_sg_clean(struct ata_queued_cmd *qc)
  */
 static void ata_fill_sg(struct ata_queued_cmd *qc)
 {
-        struct scatterlist *sg = qc->sg;
         struct ata_port *ap = qc->ap;
-        unsigned int idx, nelem;
+        struct scatterlist *sg;
+        unsigned int idx;
 
-        assert(sg != NULL);
+        assert(qc->__sg != NULL);
         assert(qc->n_elem > 0);
 
         idx = 0;
-        for (nelem = qc->n_elem; nelem; nelem--,sg++) {
+        ata_for_each_sg(sg, qc) {
                 u32 addr, offset;
                 u32 sg_len, len;
 
@@ -2289,11 +2311,12 @@ void ata_sg_init_one(struct ata_queued_cmd *qc, void *buf, unsigned int buflen)
         qc->flags |= ATA_QCFLAG_SINGLE;
 
         memset(&qc->sgent, 0, sizeof(qc->sgent));
-        qc->sg = &qc->sgent;
+        qc->__sg = &qc->sgent;
         qc->n_elem = 1;
+        qc->orig_n_elem = 1;
         qc->buf_virt = buf;
 
-        sg = qc->sg;
+        sg = qc->__sg;
         sg->page = virt_to_page(buf);
         sg->offset = (unsigned long) buf & ~PAGE_MASK;
         sg->length = buflen;
@@ -2317,8 +2340,9 @@ void ata_sg_init(struct ata_queued_cmd *qc, struct scatterlist *sg,
                  unsigned int n_elem)
 {
         qc->flags |= ATA_QCFLAG_SG;
-        qc->sg = sg;
+        qc->__sg = sg;
         qc->n_elem = n_elem;
+        qc->orig_n_elem = n_elem;
 }
 
 /**
@@ -2338,9 +2362,32 @@ static int ata_sg_setup_one(struct ata_queued_cmd *qc)
 {
         struct ata_port *ap = qc->ap;
         int dir = qc->dma_dir;
-        struct scatterlist *sg = qc->sg;
+        struct scatterlist *sg = qc->__sg;
         dma_addr_t dma_address;
 
+        /* we must lengthen transfers to end on a 32-bit boundary */
+        qc->pad_len = sg->length & 3;
+        if (qc->pad_len) {
+                void *pad_buf = ap->pad + (qc->tag * ATA_DMA_PAD_SZ);
+                struct scatterlist *psg = &qc->pad_sgent;
+
+                assert(qc->dev->class == ATA_DEV_ATAPI);
+
+                memset(pad_buf, 0, ATA_DMA_PAD_SZ);
+
+                if (qc->tf.flags & ATA_TFLAG_WRITE)
+                        memcpy(pad_buf, qc->buf_virt + sg->length - qc->pad_len,
+                               qc->pad_len);
+
+                sg_dma_address(psg) = ap->pad_dma + (qc->tag * ATA_DMA_PAD_SZ);
+                sg_dma_len(psg) = ATA_DMA_PAD_SZ;
+                /* trim sg */
+                sg->length -= qc->pad_len;
+
+                DPRINTK("padding done, sg->length=%u pad_len=%u\n",
+                        sg->length, qc->pad_len);
+        }
+
         dma_address = dma_map_single(ap->host_set->dev, qc->buf_virt,
                                      sg->length, dir);
         if (dma_mapping_error(dma_address))
@@ -2372,12 +2419,47 @@ static int ata_sg_setup_one(struct ata_queued_cmd *qc)
 static int ata_sg_setup(struct ata_queued_cmd *qc)
 {
         struct ata_port *ap = qc->ap;
-        struct scatterlist *sg = qc->sg;
+        struct scatterlist *sg = qc->__sg;
+        struct scatterlist *lsg = &sg[qc->n_elem - 1];
         int n_elem, dir;
 
         VPRINTK("ENTER, ata%u\n", ap->id);
         assert(qc->flags & ATA_QCFLAG_SG);
 
+        /* we must lengthen transfers to end on a 32-bit boundary */
+        qc->pad_len = lsg->length & 3;
+        if (qc->pad_len) {
+                void *pad_buf = ap->pad + (qc->tag * ATA_DMA_PAD_SZ);
+                struct scatterlist *psg = &qc->pad_sgent;
+                unsigned int offset;
+
+                assert(qc->dev->class == ATA_DEV_ATAPI);
+
+                memset(pad_buf, 0, ATA_DMA_PAD_SZ);
+
+                /*
+                 * psg->page/offset are used to copy to-be-written
+                 * data in this function or read data in ata_sg_clean.
+                 */
+                offset = lsg->offset + lsg->length - qc->pad_len;
+                psg->page = nth_page(lsg->page, offset >> PAGE_SHIFT);
+                psg->offset = offset_in_page(offset);
+
+                if (qc->tf.flags & ATA_TFLAG_WRITE) {
+                        void *addr = kmap_atomic(psg->page, KM_IRQ0);
+                        memcpy(pad_buf, addr + psg->offset, qc->pad_len);
+                        kunmap_atomic(psg->page, KM_IRQ0);
+                }
+
+                sg_dma_address(psg) = ap->pad_dma + (qc->tag * ATA_DMA_PAD_SZ);
+                sg_dma_len(psg) = ATA_DMA_PAD_SZ;
+                /* trim last sg */
+                lsg->length -= qc->pad_len;
+
+                DPRINTK("padding done, sg[%d].length=%u pad_len=%u\n",
+                        qc->n_elem - 1, lsg->length, qc->pad_len);
+        }
+
         dir = qc->dma_dir;
         n_elem = dma_map_sg(ap->host_set->dev, sg, qc->n_elem, dir);
         if (n_elem < 1)
@@ -2655,7 +2737,7 @@ static void ata_data_xfer(struct ata_port *ap, unsigned char *buf,
 static void ata_pio_sector(struct ata_queued_cmd *qc)
 {
         int do_write = (qc->tf.flags & ATA_TFLAG_WRITE);
-        struct scatterlist *sg = qc->sg;
+        struct scatterlist *sg = qc->__sg;
         struct ata_port *ap = qc->ap;
         struct page *page;
         unsigned int offset;
@@ -2705,7 +2787,7 @@ static void ata_pio_sector(struct ata_queued_cmd *qc)
 static void __atapi_pio_bytes(struct ata_queued_cmd *qc, unsigned int bytes)
 {
         int do_write = (qc->tf.flags & ATA_TFLAG_WRITE);
-        struct scatterlist *sg = qc->sg;
+        struct scatterlist *sg = qc->__sg;
         struct ata_port *ap = qc->ap;
         struct page *page;
         unsigned char *buf;
@@ -2738,7 +2820,7 @@ next_sg:
                 return;
         }
 
-        sg = &qc->sg[qc->cursg];
+        sg = &qc->__sg[qc->cursg];
 
         page = sg->page;
         offset = sg->offset + qc->cursg_ofs;
@@ -3145,7 +3227,7 @@ struct ata_queued_cmd *ata_qc_new_init(struct ata_port *ap,
 
         qc = ata_qc_new(ap);
         if (qc) {
-                qc->sg = NULL;
+                qc->__sg = NULL;
                 qc->flags = 0;
                 qc->scsicmd = NULL;
                 qc->ap = ap;
@@ -3837,6 +3919,12 @@ int ata_port_start (struct ata_port *ap)
         if (!ap->prd)
                 return -ENOMEM;
 
+        ap->pad = dma_alloc_coherent(dev, ATA_DMA_PAD_BUF_SZ, &ap->pad_dma, GFP_KERNEL);
+        if (!ap->pad) {
+                dma_free_coherent(dev, ATA_PRD_TBL_SZ, ap->prd, ap->prd_dma);
+                return -ENOMEM;
+        }
+
         DPRINTK("prd alloc, virt %p, dma %llx\n", ap->prd, (unsigned long long) ap->prd_dma);
 
         return 0;
@@ -3859,6 +3947,7 @@ void ata_port_stop (struct ata_port *ap)
         struct device *dev = ap->host_set->dev;
 
         dma_free_coherent(dev, ATA_PRD_TBL_SZ, ap->prd, ap->prd_dma);
+        dma_free_coherent(dev, ATA_DMA_PAD_BUF_SZ, ap->pad, ap->pad_dma);
 }
 
 void ata_host_stop (struct ata_host_set *host_set)
diff --git a/drivers/scsi/libata-scsi.c b/drivers/scsi/libata-scsi.c
index 104fd9a63e73..ee3f1050fb5f 100644
--- a/drivers/scsi/libata-scsi.c
+++ b/drivers/scsi/libata-scsi.c
@@ -150,10 +150,10 @@ struct ata_queued_cmd *ata_scsi_qc_new(struct ata_port *ap,
                 qc->scsidone = done;
 
                 if (cmd->use_sg) {
-                        qc->sg = (struct scatterlist *) cmd->request_buffer;
+                        qc->__sg = (struct scatterlist *) cmd->request_buffer;
                         qc->n_elem = cmd->use_sg;
                 } else {
-                        qc->sg = &qc->sgent;
+                        qc->__sg = &qc->sgent;
                         qc->n_elem = 1;
                 }
         } else {
@@ -364,6 +364,16 @@ int ata_scsi_slave_config(struct scsi_device *sdev)
                          */
                         blk_queue_max_sectors(sdev->request_queue, 2048);
                 }
+
+                /*
+                 * SATA DMA transfers must be multiples of 4 byte, so
+                 * we need to pad ATAPI transfers using an extra sg.
+                 * Decrement max hw segments accordingly.
+                 */
+                if (dev->class == ATA_DEV_ATAPI) {
+                        request_queue_t *q = sdev->request_queue;
+                        blk_queue_max_hw_segments(q, q->max_hw_segments - 1);
+                }
         }
 
         return 0;       /* scsi layer doesn't check return value, sigh */
diff --git a/drivers/scsi/sata_qstor.c b/drivers/scsi/sata_qstor.c
index ffcdeb68641c..69a9b1cf6f9c 100644
--- a/drivers/scsi/sata_qstor.c
+++ b/drivers/scsi/sata_qstor.c
@@ -268,16 +268,17 @@ static void qs_scr_write (struct ata_port *ap, unsigned int sc_reg, u32 val)
 
 static void qs_fill_sg(struct ata_queued_cmd *qc)
 {
-        struct scatterlist *sg = qc->sg;
+        struct scatterlist *sg;
         struct ata_port *ap = qc->ap;
         struct qs_port_priv *pp = ap->private_data;
         unsigned int nelem;
         u8 *prd = pp->pkt + QS_CPB_BYTES;
 
-        assert(sg != NULL);
+        assert(qc->__sg != NULL);
         assert(qc->n_elem > 0);
 
-        for (nelem = 0; nelem < qc->n_elem; nelem++,sg++) {
+        nelem = 0;
+        ata_for_each_sg(sg, qc) {
                 u64 addr;
                 u32 len;
 
@@ -291,6 +292,7 @@ static void qs_fill_sg(struct ata_queued_cmd *qc)
 
                 VPRINTK("PRD[%u] = (0x%llX, 0x%X)\n", nelem,
                                         (unsigned long long)addr, len);
+                nelem++;
         }
 }
 
diff --git a/drivers/scsi/sata_sx4.c b/drivers/scsi/sata_sx4.c
index 540a85191172..79fdbbab513e 100644
--- a/drivers/scsi/sata_sx4.c
+++ b/drivers/scsi/sata_sx4.c
@@ -449,14 +449,14 @@ static inline void pdc20621_host_pkt(struct ata_taskfile *tf, u8 *buf,
 
 static void pdc20621_dma_prep(struct ata_queued_cmd *qc)
 {
-        struct scatterlist *sg = qc->sg;
+        struct scatterlist *sg;
         struct ata_port *ap = qc->ap;
         struct pdc_port_priv *pp = ap->private_data;
         void __iomem *mmio = ap->host_set->mmio_base;
         struct pdc_host_priv *hpriv = ap->host_set->private_data;
         void __iomem *dimm_mmio = hpriv->dimm_mmio;
         unsigned int portno = ap->port_no;
-        unsigned int i, last, idx, total_len = 0, sgt_len;
+        unsigned int i, idx, total_len = 0, sgt_len;
         u32 *buf = (u32 *) &pp->dimm_buf[PDC_DIMM_HEADER_SZ];
 
         assert(qc->flags & ATA_QCFLAG_DMAMAP);
@@ -469,12 +469,11 @@ static void pdc20621_dma_prep(struct ata_queued_cmd *qc)
         /*
          * Build S/G table
          */
-        last = qc->n_elem;
         idx = 0;
-        for (i = 0; i < last; i++) {
-                buf[idx++] = cpu_to_le32(sg_dma_address(&sg[i]));
-                buf[idx++] = cpu_to_le32(sg_dma_len(&sg[i]));
-                total_len += sg_dma_len(&sg[i]);
+        ata_for_each_sg(sg, qc) {
+                buf[idx++] = cpu_to_le32(sg_dma_address(sg));
+                buf[idx++] = cpu_to_le32(sg_dma_len(sg));
+                total_len += sg_dma_len(sg);
         }
         buf[idx - 1] |= cpu_to_le32(ATA_PRD_EOT);
         sgt_len = idx * 4;