Merge remote-tracking branch 'remotes/jnsnow/tags/ide-pull-request' into staging

# gpg: Signature made Sat Jul  4 07:06:08 2015 BST using RSA key ID AAFC390E
# gpg: Good signature from "John Snow (John Huston) <jsnow@redhat.com>"
# gpg: WARNING: This key is not certified with sufficiently trusted signatures!
# gpg:          It is not certain that the signature belongs to the owner.
# Primary key fingerprint: FAEB 9711 A12C F475 812F  18F2 88A9 064D 1835 61EB
#      Subkey fingerprint: F9B7 ABDB BCAC DF95 BE76  CBD0 7DEF 8106 AAFC 390E

* remotes/jnsnow/tags/ide-pull-request: (35 commits)
  ahci: fix sdb fis semantics
  qtest/ahci: halted ncq migration test
  ahci: Do not map cmd_fis to generate response
  ahci: ncq migration
  ahci: add get_cmd_header helper
  ahci: add cmd header to ncq transfer state
  qtest/ahci: halted NCQ test
  ahci: correct ncq sector count
  ahci: correct types in NCQTransferState
  ahci: add rwerror=stop support for ncq
  ahci: factor ncq_finish out of ncq_cb
  ahci: refactor process_ncq_command
  ahci: assert is_ncq for process_ncq
  ahci: stash ncq command
  ide: add limit to .prepare_buf()
  qtest/ahci: ncq migration test
  qtest/ahci: simple ncq data test
  libqos/ahci: Force all NCQ commands to be LBA48
  libqos/ahci: set the NCQ tag on command_commit
  libqos/ahci: adjust expected NCQ interrupts
  ...

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

6 files changed, 350 insertions, 176 deletions

diff --git a/hw/ide/ahci.c b/hw/ide/ahci.c
index b4b65c100a..bb6a92f7f4 100644
--- a/hw/ide/ahci.c
+++ b/hw/ide/ahci.c
@@ -45,11 +45,11 @@ do { \
 } while (0)
 
 static void check_cmd(AHCIState *s, int port);
-static int handle_cmd(AHCIState *s,int port,int slot);
+static int handle_cmd(AHCIState *s, int port, uint8_t slot);
 static void ahci_reset_port(AHCIState *s, int port);
 static void ahci_write_fis_d2h(AHCIDevice *ad, uint8_t *cmd_fis);
 static void ahci_init_d2h(AHCIDevice *ad);
-static int ahci_dma_prepare_buf(IDEDMA *dma, int is_write);
+static int ahci_dma_prepare_buf(IDEDMA *dma, int32_t limit);
 static void ahci_commit_buf(IDEDMA *dma, uint32_t tx_bytes);
 static bool ahci_map_clb_address(AHCIDevice *ad);
 static bool ahci_map_fis_address(AHCIDevice *ad);
@@ -106,8 +106,6 @@ static uint32_t  ahci_port_read(AHCIState *s, int port, int offset)
         val = pr->scr_err;
         break;
     case PORT_SCR_ACT:
-        pr->scr_act &= ~s->dev[port].finished;
-        s->dev[port].finished = 0;
         val = pr->scr_act;
         break;
     case PORT_CMD_ISSUE:
@@ -331,8 +329,7 @@ static void  ahci_port_write(AHCIState *s, int port, int offset, uint32_t val)
     }
 }
 
-static uint64_t ahci_mem_read(void *opaque, hwaddr addr,
-                              unsigned size)
+static uint64_t ahci_mem_read_32(void *opaque, hwaddr addr)
 {
     AHCIState *s = opaque;
     uint32_t val = 0;
@@ -368,6 +365,30 @@ static uint64_t ahci_mem_read(void *opaque, hwaddr addr,
 }
 
 
+/**
+ * AHCI 1.3 section 3 ("HBA Memory Registers")
+ * Support unaligned 8/16/32 bit reads, and 64 bit aligned reads.
+ * Caller is responsible for masking unwanted higher order bytes.
+ */
+static uint64_t ahci_mem_read(void *opaque, hwaddr addr, unsigned size)
+{
+    hwaddr aligned = addr & ~0x3;
+    int ofst = addr - aligned;
+    uint64_t lo = ahci_mem_read_32(opaque, aligned);
+    uint64_t hi;
+
+    /* if < 8 byte read does not cross 4 byte boundary */
+    if (ofst + size <= 4) {
+        return lo >> (ofst * 8);
+    }
+    g_assert_cmpint(size, >, 1);
+
+    /* If the 64bit read is unaligned, we will produce undefined
+     * results. AHCI does not support unaligned 64bit reads. */
+    hi = ahci_mem_read_32(opaque, aligned + 4);
+    return (hi << 32 | lo) >> (ofst * 8);
+}
+
 
 static void ahci_mem_write(void *opaque, hwaddr addr,
                            uint64_t val, unsigned size)
@@ -483,7 +504,7 @@ static void ahci_reg_init(AHCIState *s)
 static void check_cmd(AHCIState *s, int port)
 {
     AHCIPortRegs *pr = &s->dev[port].port_regs;
-    int slot;
+    uint8_t slot;
 
     if ((pr->cmd & PORT_CMD_START) && pr->cmd_issue) {
         for (slot = 0; (slot < 32) && pr->cmd_issue; slot++) {
@@ -558,6 +579,7 @@ static void ahci_reset_port(AHCIState *s, int port)
     /* reset ncq queue */
     for (i = 0; i < AHCI_MAX_CMDS; i++) {
         NCQTransferState *ncq_tfs = &s->dev[port].ncq_tfs[i];
+        ncq_tfs->halt = false;
         if (!ncq_tfs->used) {
             continue;
         }
@@ -642,14 +664,14 @@ static void ahci_unmap_clb_address(AHCIDevice *ad)
     ad->lst = NULL;
 }
 
-static void ahci_write_fis_sdb(AHCIState *s, int port, uint32_t finished)
+static void ahci_write_fis_sdb(AHCIState *s, NCQTransferState *ncq_tfs)
 {
-    AHCIDevice *ad = &s->dev[port];
+    AHCIDevice *ad = ncq_tfs->drive;
     AHCIPortRegs *pr = &ad->port_regs;
     IDEState *ide_state;
     SDBFIS *sdb_fis;
 
-    if (!s->dev[port].res_fis ||
+    if (!ad->res_fis ||
         !(pr->cmd & PORT_CMD_FIS_RX)) {
         return;
     }
@@ -659,53 +681,35 @@ static void ahci_write_fis_sdb(AHCIState *s, int port, uint32_t finished)
 
     sdb_fis->type = SATA_FIS_TYPE_SDB;
     /* Interrupt pending & Notification bit */
-    sdb_fis->flags = (ad->hba->control_regs.irqstatus ? (1 << 6) : 0);
+    sdb_fis->flags = 0x40; /* Interrupt bit, always 1 for NCQ */
     sdb_fis->status = ide_state->status & 0x77;
     sdb_fis->error = ide_state->error;
     /* update SAct field in SDB_FIS */
-    s->dev[port].finished |= finished;
     sdb_fis->payload = cpu_to_le32(ad->finished);
 
     /* Update shadow registers (except BSY 0x80 and DRQ 0x08) */
     pr->tfdata = (ad->port.ifs[0].error << 8) |
         (ad->port.ifs[0].status & 0x77) |
         (pr->tfdata & 0x88);
+    pr->scr_act &= ~ad->finished;
+    ad->finished = 0;
 
-    ahci_trigger_irq(s, ad, PORT_IRQ_SDB_FIS);
+    /* Trigger IRQ if interrupt bit is set (which currently, it always is) */
+    if (sdb_fis->flags & 0x40) {
+        ahci_trigger_irq(s, ad, PORT_IRQ_SDB_FIS);
+    }
 }
 
 static void ahci_write_fis_pio(AHCIDevice *ad, uint16_t len)
 {
     AHCIPortRegs *pr = &ad->port_regs;
-    uint8_t *pio_fis, *cmd_fis;
-    uint64_t tbl_addr;
-    dma_addr_t cmd_len = 0x80;
+    uint8_t *pio_fis;
     IDEState *s = &ad->port.ifs[0];
 
     if (!ad->res_fis || !(pr->cmd & PORT_CMD_FIS_RX)) {
         return;
     }
 
-    /* map cmd_fis */
-    tbl_addr = le64_to_cpu(ad->cur_cmd->tbl_addr);
-    cmd_fis = dma_memory_map(ad->hba->as, tbl_addr, &cmd_len,
-                             DMA_DIRECTION_TO_DEVICE);
-
-    if (cmd_fis == NULL) {
-        DPRINTF(ad->port_no, "dma_memory_map failed in ahci_write_fis_pio");
-        ahci_trigger_irq(ad->hba, ad, PORT_IRQ_HBUS_ERR);
-        return;
-    }
-
-    if (cmd_len != 0x80) {
-        DPRINTF(ad->port_no,
-                "dma_memory_map mapped too few bytes in ahci_write_fis_pio");
-        dma_memory_unmap(ad->hba->as, cmd_fis, cmd_len,
-                         DMA_DIRECTION_TO_DEVICE, cmd_len);
-        ahci_trigger_irq(ad->hba, ad, PORT_IRQ_HBUS_ERR);
-        return;
-    }
-
     pio_fis = &ad->res_fis[RES_FIS_PSFIS];
 
     pio_fis[0] = SATA_FIS_TYPE_PIO_SETUP;
@@ -721,8 +725,8 @@ static void ahci_write_fis_pio(AHCIDevice *ad, uint16_t len)
     pio_fis[9] = s->hob_lcyl;
     pio_fis[10] = s->hob_hcyl;
     pio_fis[11] = 0;
-    pio_fis[12] = cmd_fis[12];
-    pio_fis[13] = cmd_fis[13];
+    pio_fis[12] = s->nsector & 0xFF;
+    pio_fis[13] = (s->nsector >> 8) & 0xFF;
     pio_fis[14] = 0;
     pio_fis[15] = s->status;
     pio_fis[16] = len & 255;
@@ -739,9 +743,6 @@ static void ahci_write_fis_pio(AHCIDevice *ad, uint16_t len)
     }
 
     ahci_trigger_irq(ad->hba, ad, PORT_IRQ_PIOS_FIS);
-
-    dma_memory_unmap(ad->hba->as, cmd_fis, cmd_len,
-                     DMA_DIRECTION_TO_DEVICE, cmd_len);
 }
 
 static void ahci_write_fis_d2h(AHCIDevice *ad, uint8_t *cmd_fis)
@@ -749,22 +750,12 @@ static void ahci_write_fis_d2h(AHCIDevice *ad, uint8_t *cmd_fis)
     AHCIPortRegs *pr = &ad->port_regs;
     uint8_t *d2h_fis;
     int i;
-    dma_addr_t cmd_len = 0x80;
-    int cmd_mapped = 0;
     IDEState *s = &ad->port.ifs[0];
 
     if (!ad->res_fis || !(pr->cmd & PORT_CMD_FIS_RX)) {
         return;
     }
 
-    if (!cmd_fis) {
-        /* map cmd_fis */
-        uint64_t tbl_addr = le64_to_cpu(ad->cur_cmd->tbl_addr);
-        cmd_fis = dma_memory_map(ad->hba->as, tbl_addr, &cmd_len,
-                                 DMA_DIRECTION_TO_DEVICE);
-        cmd_mapped = 1;
-    }
-
     d2h_fis = &ad->res_fis[RES_FIS_RFIS];
 
     d2h_fis[0] = SATA_FIS_TYPE_REGISTER_D2H;
@@ -780,8 +771,8 @@ static void ahci_write_fis_d2h(AHCIDevice *ad, uint8_t *cmd_fis)
     d2h_fis[9] = s->hob_lcyl;
     d2h_fis[10] = s->hob_hcyl;
     d2h_fis[11] = 0;
-    d2h_fis[12] = cmd_fis[12];
-    d2h_fis[13] = cmd_fis[13];
+    d2h_fis[12] = s->nsector & 0xFF;
+    d2h_fis[13] = (s->nsector >> 8) & 0xFF;
     for (i = 14; i < 20; i++) {
         d2h_fis[i] = 0;
     }
@@ -795,26 +786,22 @@ static void ahci_write_fis_d2h(AHCIDevice *ad, uint8_t *cmd_fis)
     }
 
     ahci_trigger_irq(ad->hba, ad, PORT_IRQ_D2H_REG_FIS);
-
-    if (cmd_mapped) {
-        dma_memory_unmap(ad->hba->as, cmd_fis, cmd_len,
-                         DMA_DIRECTION_TO_DEVICE, cmd_len);
-    }
 }
 
 static int prdt_tbl_entry_size(const AHCI_SG *tbl)
 {
+    /* flags_size is zero-based */
     return (le32_to_cpu(tbl->flags_size) & AHCI_PRDT_SIZE_MASK) + 1;
 }
 
 static int ahci_populate_sglist(AHCIDevice *ad, QEMUSGList *sglist,
-                                int32_t offset)
+                                AHCICmdHdr *cmd, int64_t limit, int32_t offset)
 {
-    AHCICmdHdr *cmd = ad->cur_cmd;
-    uint32_t opts = le32_to_cpu(cmd->opts);
-    uint64_t prdt_addr = le64_to_cpu(cmd->tbl_addr) + 0x80;
-    int sglist_alloc_hint = opts >> AHCI_CMD_HDR_PRDT_LEN;
-    dma_addr_t prdt_len = (sglist_alloc_hint * sizeof(AHCI_SG));
+    uint16_t opts = le16_to_cpu(cmd->opts);
+    uint16_t prdtl = le16_to_cpu(cmd->prdtl);
+    uint64_t cfis_addr = le64_to_cpu(cmd->tbl_addr);
+    uint64_t prdt_addr = cfis_addr + 0x80;
+    dma_addr_t prdt_len = (prdtl * sizeof(AHCI_SG));
     dma_addr_t real_prdt_len = prdt_len;
     uint8_t *prdt;
     int i;
@@ -834,7 +821,7 @@ static int ahci_populate_sglist(AHCIDevice *ad, QEMUSGList *sglist,
      * request for sector sizes up to 32K.
      */
 
-    if (!sglist_alloc_hint) {
+    if (!prdtl) {
         DPRINTF(ad->port_no, "no sg list given by guest: 0x%08x\n", opts);
         return -1;
     }
@@ -853,13 +840,12 @@ static int ahci_populate_sglist(AHCIDevice *ad, QEMUSGList *sglist,
     }
 
     /* Get entries in the PRDT, init a qemu sglist accordingly */
-    if (sglist_alloc_hint > 0) {
+    if (prdtl > 0) {
         AHCI_SG *tbl = (AHCI_SG *)prdt;
         sum = 0;
-        for (i = 0; i < sglist_alloc_hint; i++) {
-            /* flags_size is zero-based */
+        for (i = 0; i < prdtl; i++) {
             tbl_entry_size = prdt_tbl_entry_size(&tbl[i]);
-            if (offset <= (sum + tbl_entry_size)) {
+            if (offset < (sum + tbl_entry_size)) {
                 off_idx = i;
                 off_pos = offset - sum;
                 break;
@@ -874,15 +860,16 @@ static int ahci_populate_sglist(AHCIDevice *ad, QEMUSGList *sglist,
             goto out;
         }
 
-        qemu_sglist_init(sglist, qbus->parent, (sglist_alloc_hint - off_idx),
+        qemu_sglist_init(sglist, qbus->parent, (prdtl - off_idx),
                          ad->hba->as);
         qemu_sglist_add(sglist, le64_to_cpu(tbl[off_idx].addr) + off_pos,
-                        prdt_tbl_entry_size(&tbl[off_idx]) - off_pos);
+                        MIN(prdt_tbl_entry_size(&tbl[off_idx]) - off_pos,
+                            limit));
 
-        for (i = off_idx + 1; i < sglist_alloc_hint; i++) {
-            /* flags_size is zero-based */
+        for (i = off_idx + 1; i < prdtl && sglist->size < limit; i++) {
             qemu_sglist_add(sglist, le64_to_cpu(tbl[i].addr),
-                            prdt_tbl_entry_size(&tbl[i]));
+                            MIN(prdt_tbl_entry_size(&tbl[i]),
+                                limit - sglist->size));
             if (sglist->size > INT32_MAX) {
                 error_report("AHCI Physical Region Descriptor Table describes "
                              "more than 2 GiB.\n");
@@ -899,28 +886,25 @@ out:
     return r;
 }
 
-static void ncq_cb(void *opaque, int ret)
+static void ncq_err(NCQTransferState *ncq_tfs)
 {
-    NCQTransferState *ncq_tfs = (NCQTransferState *)opaque;
     IDEState *ide_state = &ncq_tfs->drive->port.ifs[0];
 
-    if (ret == -ECANCELED) {
-        return;
-    }
-    /* Clear bit for this tag in SActive */
-    ncq_tfs->drive->port_regs.scr_act &= ~(1 << ncq_tfs->tag);
+    ide_state->error = ABRT_ERR;
+    ide_state->status = READY_STAT | ERR_STAT;
+    ncq_tfs->drive->port_regs.scr_err |= (1 << ncq_tfs->tag);
+}
 
-    if (ret < 0) {
-        /* error */
-        ide_state->error = ABRT_ERR;
-        ide_state->status = READY_STAT | ERR_STAT;
-        ncq_tfs->drive->port_regs.scr_err |= (1 << ncq_tfs->tag);
-    } else {
-        ide_state->status = READY_STAT | SEEK_STAT;
+static void ncq_finish(NCQTransferState *ncq_tfs)
+{
+    /* If we didn't error out, set our finished bit. Errored commands
+     * do not get a bit set for the SDB FIS ACT register, nor do they
+     * clear the outstanding bit in scr_act (PxSACT). */
+    if (!(ncq_tfs->drive->port_regs.scr_err & (1 << ncq_tfs->tag))) {
+        ncq_tfs->drive->finished |= (1 << ncq_tfs->tag);
     }
 
-    ahci_write_fis_sdb(ncq_tfs->drive->hba, ncq_tfs->drive->port_no,
-                       (1 << ncq_tfs->tag));
+    ahci_write_fis_sdb(ncq_tfs->drive->hba, ncq_tfs);
 
     DPRINTF(ncq_tfs->drive->port_no, "NCQ transfer tag %d finished\n",
             ncq_tfs->tag);
@@ -931,6 +915,35 @@ static void ncq_cb(void *opaque, int ret)
     ncq_tfs->used = 0;
 }
 
+static void ncq_cb(void *opaque, int ret)
+{
+    NCQTransferState *ncq_tfs = (NCQTransferState *)opaque;
+    IDEState *ide_state = &ncq_tfs->drive->port.ifs[0];
+
+    if (ret == -ECANCELED) {
+        return;
+    }
+
+    if (ret < 0) {
+        bool is_read = ncq_tfs->cmd == READ_FPDMA_QUEUED;
+        BlockErrorAction action = blk_get_error_action(ide_state->blk,
+                                                       is_read, -ret);
+        if (action == BLOCK_ERROR_ACTION_STOP) {
+            ncq_tfs->halt = true;
+            ide_state->bus->error_status = IDE_RETRY_HBA;
+        } else if (action == BLOCK_ERROR_ACTION_REPORT) {
+            ncq_err(ncq_tfs);
+        }
+        blk_error_action(ide_state->blk, action, is_read, -ret);
+    } else {
+        ide_state->status = READY_STAT | SEEK_STAT;
+    }
+
+    if (!ncq_tfs->halt) {
+        ncq_finish(ncq_tfs);
+    }
+}
+
 static int is_ncq(uint8_t ata_cmd)
 {
     /* Based on SATA 3.2 section 13.6.3.2 */
@@ -946,13 +959,60 @@ static int is_ncq(uint8_t ata_cmd)
     }
 }
 
+static void execute_ncq_command(NCQTransferState *ncq_tfs)
+{
+    AHCIDevice *ad = ncq_tfs->drive;
+    IDEState *ide_state = &ad->port.ifs[0];
+    int port = ad->port_no;
+
+    g_assert(is_ncq(ncq_tfs->cmd));
+    ncq_tfs->halt = false;
+
+    switch (ncq_tfs->cmd) {
+    case READ_FPDMA_QUEUED:
+        DPRINTF(port, "NCQ reading %d sectors from LBA %"PRId64", tag %d\n",
+                ncq_tfs->sector_count, ncq_tfs->lba, ncq_tfs->tag);
+
+        DPRINTF(port, "tag %d aio read %"PRId64"\n",
+                ncq_tfs->tag, ncq_tfs->lba);
+
+        dma_acct_start(ide_state->blk, &ncq_tfs->acct,
+                       &ncq_tfs->sglist, BLOCK_ACCT_READ);
+        ncq_tfs->aiocb = dma_blk_read(ide_state->blk, &ncq_tfs->sglist,
+                                      ncq_tfs->lba, ncq_cb, ncq_tfs);
+        break;
+    case WRITE_FPDMA_QUEUED:
+        DPRINTF(port, "NCQ writing %d sectors to LBA %"PRId64", tag %d\n",
+                ncq_tfs->sector_count, ncq_tfs->lba, ncq_tfs->tag);
+
+        DPRINTF(port, "tag %d aio write %"PRId64"\n",
+                ncq_tfs->tag, ncq_tfs->lba);
+
+        dma_acct_start(ide_state->blk, &ncq_tfs->acct,
+                       &ncq_tfs->sglist, BLOCK_ACCT_WRITE);
+        ncq_tfs->aiocb = dma_blk_write(ide_state->blk, &ncq_tfs->sglist,
+                                       ncq_tfs->lba, ncq_cb, ncq_tfs);
+        break;
+    default:
+        DPRINTF(port, "error: unsupported NCQ command (0x%02x) received\n",
+                ncq_tfs->cmd);
+        qemu_sglist_destroy(&ncq_tfs->sglist);
+        ncq_err(ncq_tfs);
+    }
+}
+
+
 static void process_ncq_command(AHCIState *s, int port, uint8_t *cmd_fis,
-                                int slot)
+                                uint8_t slot)
 {
+    AHCIDevice *ad = &s->dev[port];
+    IDEState *ide_state = &ad->port.ifs[0];
     NCQFrame *ncq_fis = (NCQFrame*)cmd_fis;
     uint8_t tag = ncq_fis->tag >> 3;
-    NCQTransferState *ncq_tfs = &s->dev[port].ncq_tfs[tag];
+    NCQTransferState *ncq_tfs = &ad->ncq_tfs[tag];
+    size_t size;
 
+    g_assert(is_ncq(ncq_fis->command));
     if (ncq_tfs->used) {
         /* error - already in use */
         fprintf(stderr, "%s: tag %d already used\n", __FUNCTION__, tag);
@@ -960,75 +1020,82 @@ static void process_ncq_command(AHCIState *s, int port, uint8_t *cmd_fis,
     }
 
     ncq_tfs->used = 1;
-    ncq_tfs->drive = &s->dev[port];
+    ncq_tfs->drive = ad;
     ncq_tfs->slot = slot;
+    ncq_tfs->cmdh = &((AHCICmdHdr *)ad->lst)[slot];
+    ncq_tfs->cmd = ncq_fis->command;
     ncq_tfs->lba = ((uint64_t)ncq_fis->lba5 << 40) |
                    ((uint64_t)ncq_fis->lba4 << 32) |
                    ((uint64_t)ncq_fis->lba3 << 24) |
                    ((uint64_t)ncq_fis->lba2 << 16) |
                    ((uint64_t)ncq_fis->lba1 << 8) |
                    (uint64_t)ncq_fis->lba0;
+    ncq_tfs->tag = tag;
 
-    /* Note: We calculate the sector count, but don't currently rely on it.
-     * The total size of the DMA buffer tells us the transfer size instead. */
-    ncq_tfs->sector_count = ((uint16_t)ncq_fis->sector_count_high << 8) |
-                                ncq_fis->sector_count_low;
+    /* Sanity-check the NCQ packet */
+    if (tag != slot) {
+        DPRINTF(port, "Warn: NCQ slot (%d) did not match the given tag (%d)\n",
+                slot, tag);
+    }
 
-    DPRINTF(port, "NCQ transfer LBA from %"PRId64" to %"PRId64", "
-            "drive max %"PRId64"\n",
-            ncq_tfs->lba, ncq_tfs->lba + ncq_tfs->sector_count - 2,
-            s->dev[port].port.ifs[0].nb_sectors - 1);
+    if (ncq_fis->aux0 || ncq_fis->aux1 || ncq_fis->aux2 || ncq_fis->aux3) {
+        DPRINTF(port, "Warn: Attempt to use NCQ auxiliary fields.\n");
+    }
+    if (ncq_fis->prio || ncq_fis->icc) {
+        DPRINTF(port, "Warn: Unsupported attempt to use PRIO/ICC fields\n");
+    }
+    if (ncq_fis->fua & NCQ_FIS_FUA_MASK) {
+        DPRINTF(port, "Warn: Unsupported attempt to use Force Unit Access\n");
+    }
+    if (ncq_fis->tag & NCQ_FIS_RARC_MASK) {
+        DPRINTF(port, "Warn: Unsupported attempt to use Rebuild Assist\n");
+    }
 
-    ahci_populate_sglist(&s->dev[port], &ncq_tfs->sglist, 0);
-    ncq_tfs->tag = tag;
+    ncq_tfs->sector_count = ((ncq_fis->sector_count_high << 8) |
+                             ncq_fis->sector_count_low);
+    if (!ncq_tfs->sector_count) {
+        ncq_tfs->sector_count = 0x10000;
+    }
+    size = ncq_tfs->sector_count * 512;
+    ahci_populate_sglist(ad, &ncq_tfs->sglist, ncq_tfs->cmdh, size, 0);
 
-    switch(ncq_fis->command) {
-        case READ_FPDMA_QUEUED:
-            DPRINTF(port, "NCQ reading %d sectors from LBA %"PRId64", "
-                    "tag %d\n",
-                    ncq_tfs->sector_count-1, ncq_tfs->lba, ncq_tfs->tag);
+    if (ncq_tfs->sglist.size < size) {
+        error_report("ahci: PRDT length for NCQ command (0x%zx) "
+                     "is smaller than the requested size (0x%zx)",
+                     ncq_tfs->sglist.size, size);
+        qemu_sglist_destroy(&ncq_tfs->sglist);
+        ncq_err(ncq_tfs);
+        ahci_trigger_irq(ad->hba, ad, PORT_IRQ_OVERFLOW);
+        return;
+    } else if (ncq_tfs->sglist.size != size) {
+        DPRINTF(port, "Warn: PRDTL (0x%zx)"
+                " does not match requested size (0x%zx)",
+                ncq_tfs->sglist.size, size);
+    }
 
-            DPRINTF(port, "tag %d aio read %"PRId64"\n",
-                    ncq_tfs->tag, ncq_tfs->lba);
+    DPRINTF(port, "NCQ transfer LBA from %"PRId64" to %"PRId64", "
+            "drive max %"PRId64"\n",
+            ncq_tfs->lba, ncq_tfs->lba + ncq_tfs->sector_count - 1,
+            ide_state->nb_sectors - 1);
 
-            dma_acct_start(ncq_tfs->drive->port.ifs[0].blk, &ncq_tfs->acct,
-                           &ncq_tfs->sglist, BLOCK_ACCT_READ);
-            ncq_tfs->aiocb = dma_blk_read(ncq_tfs->drive->port.ifs[0].blk,
-                                          &ncq_tfs->sglist, ncq_tfs->lba,
-                                          ncq_cb, ncq_tfs);
-            break;
-        case WRITE_FPDMA_QUEUED:
-            DPRINTF(port, "NCQ writing %d sectors to LBA %"PRId64", tag %d\n",
-                    ncq_tfs->sector_count-1, ncq_tfs->lba, ncq_tfs->tag);
-
-            DPRINTF(port, "tag %d aio write %"PRId64"\n",
-                    ncq_tfs->tag, ncq_tfs->lba);
-
-            dma_acct_start(ncq_tfs->drive->port.ifs[0].blk, &ncq_tfs->acct,
-                           &ncq_tfs->sglist, BLOCK_ACCT_WRITE);
-            ncq_tfs->aiocb = dma_blk_write(ncq_tfs->drive->port.ifs[0].blk,
-                                           &ncq_tfs->sglist, ncq_tfs->lba,
-                                           ncq_cb, ncq_tfs);
-            break;
-        default:
-            if (is_ncq(cmd_fis[2])) {
-                DPRINTF(port,
-                        "error: unsupported NCQ command (0x%02x) received\n",
-                        cmd_fis[2]);
-            } else {
-                DPRINTF(port,
-                        "error: tried to process non-NCQ command as NCQ\n");
-            }
-            qemu_sglist_destroy(&ncq_tfs->sglist);
+    execute_ncq_command(ncq_tfs);
+}
+
+static AHCICmdHdr *get_cmd_header(AHCIState *s, uint8_t port, uint8_t slot)
+{
+    if (port >= s->ports || slot >= AHCI_MAX_CMDS) {
+        return NULL;
     }
+
+    return s->dev[port].lst ? &((AHCICmdHdr *)s->dev[port].lst)[slot] : NULL;
 }
 
 static void handle_reg_h2d_fis(AHCIState *s, int port,
-                               int slot, uint8_t *cmd_fis)
+                               uint8_t slot, uint8_t *cmd_fis)
 {
     IDEState *ide_state = &s->dev[port].port.ifs[0];
-    AHCICmdHdr *cmd = s->dev[port].cur_cmd;
-    uint32_t opts = le32_to_cpu(cmd->opts);
+    AHCICmdHdr *cmd = get_cmd_header(s, port, slot);
+    uint16_t opts = le16_to_cpu(cmd->opts);
 
     if (cmd_fis[1] & 0x0F) {
         DPRINTF(port, "Port Multiplier not supported."
@@ -1108,7 +1175,7 @@ static void handle_reg_h2d_fis(AHCIState *s, int port,
     ide_exec_cmd(&s->dev[port].port, cmd_fis[2]);
 }
 
-static int handle_cmd(AHCIState *s, int port, int slot)
+static int handle_cmd(AHCIState *s, int port, uint8_t slot)
 {
     IDEState *ide_state;
     uint64_t tbl_addr;
@@ -1126,7 +1193,7 @@ static int handle_cmd(AHCIState *s, int port, int slot)
         DPRINTF(port, "error: lst not given but cmd handled");
         return -1;
     }
-    cmd = &((AHCICmdHdr *)s->dev[port].lst)[slot];
+    cmd = get_cmd_header(s, port, slot);
     /* remember current slot handle for later */
     s->dev[port].cur_cmd = cmd;
 
@@ -1185,7 +1252,7 @@ static void ahci_start_transfer(IDEDMA *dma)
     IDEState *s = &ad->port.ifs[0];
     uint32_t size = (uint32_t)(s->data_end - s->data_ptr);
     /* write == ram -> device */
-    uint32_t opts = le32_to_cpu(ad->cur_cmd->opts);
+    uint16_t opts = le16_to_cpu(ad->cur_cmd->opts);
     int is_write = opts & AHCI_CMD_WRITE;
     int is_atapi = opts & AHCI_CMD_ATAPI;
     int has_sglist = 0;
@@ -1197,7 +1264,7 @@ static void ahci_start_transfer(IDEDMA *dma)
         goto out;
     }
 
-    if (ahci_dma_prepare_buf(dma, is_write)) {
+    if (ahci_dma_prepare_buf(dma, size)) {
         has_sglist = 1;
     }
 
@@ -1243,16 +1310,34 @@ static void ahci_restart_dma(IDEDMA *dma)
 }
 
 /**
+ * IDE/PIO restarts are handled by the core layer, but NCQ commands
+ * need an extra kick from the AHCI HBA.
+ */
+static void ahci_restart(IDEDMA *dma)
+{
+    AHCIDevice *ad = DO_UPCAST(AHCIDevice, dma, dma);
+    int i;
+
+    for (i = 0; i < AHCI_MAX_CMDS; i++) {
+        NCQTransferState *ncq_tfs = &ad->ncq_tfs[i];
+        if (ncq_tfs->halt) {
+            execute_ncq_command(ncq_tfs);
+        }
+    }
+}
+
+/**
  * Called in DMA R/W chains to read the PRDT, utilizing ahci_populate_sglist.
  * Not currently invoked by PIO R/W chains,
  * which invoke ahci_populate_sglist via ahci_start_transfer.
  */
-static int32_t ahci_dma_prepare_buf(IDEDMA *dma, int is_write)
+static int32_t ahci_dma_prepare_buf(IDEDMA *dma, int32_t limit)
 {
     AHCIDevice *ad = DO_UPCAST(AHCIDevice, dma, dma);
     IDEState *s = &ad->port.ifs[0];
 
-    if (ahci_populate_sglist(ad, &s->sg, s->io_buffer_offset) == -1) {
+    if (ahci_populate_sglist(ad, &s->sg, ad->cur_cmd,
+                             limit, s->io_buffer_offset) == -1) {
         DPRINTF(ad->port_no, "ahci_dma_prepare_buf failed.\n");
         return -1;
     }
@@ -1287,7 +1372,7 @@ static int ahci_dma_rw_buf(IDEDMA *dma, int is_write)
     uint8_t *p = s->io_buffer + s->io_buffer_index;
     int l = s->io_buffer_size - s->io_buffer_index;
 
-    if (ahci_populate_sglist(ad, &s->sg, s->io_buffer_offset)) {
+    if (ahci_populate_sglist(ad, &s->sg, ad->cur_cmd, l, s->io_buffer_offset)) {
         return 0;
     }
 
@@ -1330,6 +1415,7 @@ static void ahci_irq_set(void *opaque, int n, int level)
 
 static const IDEDMAOps ahci_dma_ops = {
     .start_dma = ahci_start_dma,
+    .restart = ahci_restart,
     .restart_dma = ahci_restart_dma,
     .start_transfer = ahci_start_transfer,
     .prepare_buf = ahci_dma_prepare_buf,
@@ -1400,6 +1486,21 @@ void ahci_reset(AHCIState *s)
     }
 }
 
+static const VMStateDescription vmstate_ncq_tfs = {
+    .name = "ncq state",
+    .version_id = 1,
+    .fields = (VMStateField[]) {
+        VMSTATE_UINT32(sector_count, NCQTransferState),
+        VMSTATE_UINT64(lba, NCQTransferState),
+        VMSTATE_UINT8(tag, NCQTransferState),
+        VMSTATE_UINT8(cmd, NCQTransferState),
+        VMSTATE_UINT8(slot, NCQTransferState),
+        VMSTATE_BOOL(used, NCQTransferState),
+        VMSTATE_BOOL(halt, NCQTransferState),
+        VMSTATE_END_OF_LIST()
+    },
+};
+
 static const VMStateDescription vmstate_ahci_device = {
     .name = "ahci port",
     .version_id = 1,
@@ -1425,14 +1526,17 @@ static const VMStateDescription vmstate_ahci_device = {
         VMSTATE_BOOL(done_atapi_packet, AHCIDevice),
         VMSTATE_INT32(busy_slot, AHCIDevice),
         VMSTATE_BOOL(init_d2h_sent, AHCIDevice),
+        VMSTATE_STRUCT_ARRAY(ncq_tfs, AHCIDevice, AHCI_MAX_CMDS,
+                             1, vmstate_ncq_tfs, NCQTransferState),
         VMSTATE_END_OF_LIST()
     },
 };
 
 static int ahci_state_post_load(void *opaque, int version_id)
 {
-    int i;
+    int i, j;
     struct AHCIDevice *ad;
+    NCQTransferState *ncq_tfs;
     AHCIState *s = opaque;
 
     for (i = 0; i < s->ports; i++) {
@@ -1444,6 +1548,37 @@ static int ahci_state_post_load(void *opaque, int version_id)
             return -1;
         }
 
+        for (j = 0; j < AHCI_MAX_CMDS; j++) {
+            ncq_tfs = &ad->ncq_tfs[j];
+            ncq_tfs->drive = ad;
+
+            if (ncq_tfs->used != ncq_tfs->halt) {
+                return -1;
+            }
+            if (!ncq_tfs->halt) {
+                continue;
+            }
+            if (!is_ncq(ncq_tfs->cmd)) {
+                return -1;
+            }
+            if (ncq_tfs->slot != ncq_tfs->tag) {
+                return -1;
+            }
+            /* If ncq_tfs->halt is justly set, the engine should be engaged,
+             * and the command list buffer should be mapped. */
+            ncq_tfs->cmdh = get_cmd_header(s, i, ncq_tfs->slot);
+            if (!ncq_tfs->cmdh) {
+                return -1;
+            }
+            ahci_populate_sglist(ncq_tfs->drive, &ncq_tfs->sglist,
+                                 ncq_tfs->cmdh, ncq_tfs->sector_count * 512,
+                                 0);
+            if (ncq_tfs->sector_count != ncq_tfs->sglist.size >> 9) {
+                return -1;
+            }
+        }
+
+
         /*
          * If an error is present, ad->busy_slot will be valid and not -1.
          * In this case, an operation is waiting to resume and will re-check
@@ -1460,7 +1595,7 @@ static int ahci_state_post_load(void *opaque, int version_id)
             if (ad->busy_slot < 0 || ad->busy_slot >= AHCI_MAX_CMDS) {
                 return -1;
             }
-            ad->cur_cmd = &((AHCICmdHdr *)ad->lst)[ad->busy_slot];
+            ad->cur_cmd = get_cmd_header(s, i, ad->busy_slot);
         }
     }
 
diff --git a/hw/ide/ahci.h b/hw/ide/ahci.h
index 501c002c31..9f5b4d20b5 100644
--- a/hw/ide/ahci.h
+++ b/hw/ide/ahci.h
@@ -195,6 +195,9 @@
 #define RECEIVE_FPDMA_QUEUED               0x65
 #define SEND_FPDMA_QUEUED                  0x64
 
+#define NCQ_FIS_FUA_MASK                   0x80
+#define NCQ_FIS_RARC_MASK                  0x01
+
 #define RES_FIS_DSFIS                      0x00
 #define RES_FIS_PSFIS                      0x20
 #define RES_FIS_RFIS                       0x40
@@ -233,7 +236,8 @@ typedef struct AHCIPortRegs {
 } AHCIPortRegs;
 
 typedef struct AHCICmdHdr {
-    uint32_t    opts;
+    uint16_t    opts;
+    uint16_t    prdtl;
     uint32_t    status;
     uint64_t    tbl_addr;
     uint32_t    reserved[4];
@@ -250,13 +254,16 @@ typedef struct AHCIDevice AHCIDevice;
 typedef struct NCQTransferState {
     AHCIDevice *drive;
     BlockAIOCB *aiocb;
+    AHCICmdHdr *cmdh;
     QEMUSGList sglist;
     BlockAcctCookie acct;
-    uint16_t sector_count;
+    uint32_t sector_count;
     uint64_t lba;
     uint8_t tag;
-    int slot;
-    int used;
+    uint8_t cmd;
+    uint8_t slot;
+    bool used;
+    bool halt;
 } NCQTransferState;
 
 struct AHCIDevice {
@@ -312,27 +319,39 @@ extern const VMStateDescription vmstate_ahci;
     .offset     = vmstate_offset_value(_state, _field, AHCIState),   \
 }
 
+/**
+ * NCQFrame is the same as a Register H2D FIS (described in SATA 3.2),
+ * but some fields have been re-mapped and re-purposed, as seen in
+ * SATA 3.2 section 13.6.4.1 ("READ FPDMA QUEUED")
+ *
+ * cmd_fis[3], feature 7:0, becomes sector count 7:0.
+ * cmd_fis[7], device 7:0, uses bit 7 as the Force Unit Access bit.
+ * cmd_fis[11], feature 15:8, becomes sector count 15:8.
+ * cmd_fis[12], count 7:0, becomes the NCQ TAG (7:3) and RARC bit (0)
+ * cmd_fis[13], count 15:8, becomes the priority value (7:6)
+ * bytes 16-19 become an le32 "auxiliary" field.
+ */
 typedef struct NCQFrame {
     uint8_t fis_type;
     uint8_t c;
     uint8_t command;
-    uint8_t sector_count_low;
+    uint8_t sector_count_low;  /* (feature 7:0) */
     uint8_t lba0;
     uint8_t lba1;
     uint8_t lba2;
-    uint8_t fua;
+    uint8_t fua;               /* (device 7:0) */
     uint8_t lba3;
     uint8_t lba4;
     uint8_t lba5;
-    uint8_t sector_count_high;
-    uint8_t tag;
-    uint8_t reserved5;
-    uint8_t reserved6;
+    uint8_t sector_count_high; /* (feature 15:8) */
+    uint8_t tag;               /* (count 0:7) */
+    uint8_t prio;              /* (count 15:8) */
+    uint8_t icc;
     uint8_t control;
-    uint8_t reserved7;
-    uint8_t reserved8;
-    uint8_t reserved9;
-    uint8_t reserved10;
+    uint8_t aux0;
+    uint8_t aux1;
+    uint8_t aux2;
+    uint8_t aux3;
 } QEMU_PACKED NCQFrame;
 
 typedef struct SDBFIS {
diff --git a/hw/ide/core.c b/hw/ide/core.c
index 1efd98af63..122e955084 100644
--- a/hw/ide/core.c
+++ b/hw/ide/core.c
@@ -716,8 +716,8 @@ static void ide_dma_cb(void *opaque, int ret)
 
     sector_num = ide_get_sector(s);
     if (n > 0) {
-        assert(s->io_buffer_size == s->sg.size);
-        dma_buf_commit(s, s->io_buffer_size);
+        assert(n * 512 == s->sg.size);
+        dma_buf_commit(s, s->sg.size);
         sector_num += n;
         ide_set_sector(s, sector_num);
         s->nsector -= n;
@@ -734,7 +734,7 @@ static void ide_dma_cb(void *opaque, int ret)
     n = s->nsector;
     s->io_buffer_index = 0;
     s->io_buffer_size = n * 512;
-    if (s->bus->dma->ops->prepare_buf(s->bus->dma, ide_cmd_is_read(s)) < 512) {
+    if (s->bus->dma->ops->prepare_buf(s->bus->dma, s->io_buffer_size) < 512) {
         /* The PRDs were too short. Reset the Active bit, but don't raise an
          * interrupt. */
         s->status = READY_STAT | SEEK_STAT;
@@ -2326,7 +2326,7 @@ static void ide_nop(IDEDMA *dma)
 {
 }
 
-static int32_t ide_nop_int32(IDEDMA *dma, int x)
+static int32_t ide_nop_int32(IDEDMA *dma, int32_t l)
 {
     return 0;
 }
@@ -2371,6 +2371,13 @@ static void ide_restart_bh(void *opaque)
      * called function can set a new error status. */
     bus->error_status = 0;
 
+    /* The HBA has generically asked to be kicked on retry */
+    if (error_status & IDE_RETRY_HBA) {
+        if (s->bus->dma->ops->restart) {
+            s->bus->dma->ops->restart(s->bus->dma);
+        }
+    }
+
     if (error_status & IDE_RETRY_DMA) {
         if (error_status & IDE_RETRY_TRIM) {
             ide_restart_dma(s, IDE_DMA_TRIM);
diff --git a/hw/ide/internal.h b/hw/ide/internal.h
index 965cc55cb8..30fdcbc5fa 100644
--- a/hw/ide/internal.h
+++ b/hw/ide/internal.h
@@ -324,7 +324,7 @@ typedef void EndTransferFunc(IDEState *);
 typedef void DMAStartFunc(IDEDMA *, IDEState *, BlockCompletionFunc *);
 typedef void DMAVoidFunc(IDEDMA *);
 typedef int DMAIntFunc(IDEDMA *, int);
-typedef int32_t DMAInt32Func(IDEDMA *, int);
+typedef int32_t DMAInt32Func(IDEDMA *, int32_t len);
 typedef void DMAu32Func(IDEDMA *, uint32_t);
 typedef void DMAStopFunc(IDEDMA *, bool);
 typedef void DMARestartFunc(void *, int, RunState);
@@ -436,6 +436,7 @@ struct IDEDMAOps {
     DMAInt32Func *prepare_buf;
     DMAu32Func *commit_buf;
     DMAIntFunc *rw_buf;
+    DMAVoidFunc *restart;
     DMAVoidFunc *restart_dma;
     DMAStopFunc *set_inactive;
     DMAVoidFunc *cmd_done;
@@ -499,6 +500,7 @@ struct IDEDevice {
 #define IDE_RETRY_READ  0x20
 #define IDE_RETRY_FLUSH 0x40
 #define IDE_RETRY_TRIM 0x80
+#define IDE_RETRY_HBA  0x100
 
 static inline IDEState *idebus_active_if(IDEBus *bus)
 {
diff --git a/hw/ide/macio.c b/hw/ide/macio.c
index dd52d50732..a55a479da6 100644
--- a/hw/ide/macio.c
+++ b/hw/ide/macio.c
@@ -499,7 +499,7 @@ static int ide_nop_int(IDEDMA *dma, int x)
     return 0;
 }
 
-static int32_t ide_nop_int32(IDEDMA *dma, int x)
+static int32_t ide_nop_int32(IDEDMA *dma, int32_t l)
 {
     return 0;
 }
diff --git a/hw/ide/pci.c b/hw/ide/pci.c
index 4afd0cfe8c..d31ff885b7 100644
--- a/hw/ide/pci.c
+++ b/hw/ide/pci.c
@@ -53,10 +53,14 @@ static void bmdma_start_dma(IDEDMA *dma, IDEState *s,
 }
 
 /**
- * Return the number of bytes successfully prepared.
- * -1 on error.
+ * Prepare an sglist based on available PRDs.
+ * @limit: How many bytes to prepare total.
+ *
+ * Returns the number of bytes prepared, -1 on error.
+ * IDEState.io_buffer_size will contain the number of bytes described
+ * by the PRDs, whether or not we added them to the sglist.
  */
-static int32_t bmdma_prepare_buf(IDEDMA *dma, int is_write)
+static int32_t bmdma_prepare_buf(IDEDMA *dma, int32_t limit)
 {
     BMDMAState *bm = DO_UPCAST(BMDMAState, dma, dma);
     IDEState *s = bmdma_active_if(bm);
@@ -75,7 +79,7 @@ static int32_t bmdma_prepare_buf(IDEDMA *dma, int is_write)
             /* end of table (with a fail safe of one page) */
             if (bm->cur_prd_last ||
                 (bm->cur_addr - bm->addr) >= BMDMA_PAGE_SIZE) {
-                return s->io_buffer_size;
+                return s->sg.size;
             }
             pci_dma_read(pci_dev, bm->cur_addr, &prd, 8);
             bm->cur_addr += 8;
@@ -90,7 +94,14 @@ static int32_t bmdma_prepare_buf(IDEDMA *dma, int is_write)
         }
         l = bm->cur_prd_len;
         if (l > 0) {
-            qemu_sglist_add(&s->sg, bm->cur_prd_addr, l);
+            uint64_t sg_len;
+
+            /* Don't add extra bytes to the SGList; consume any remaining
+             * PRDs from the guest, but ignore them. */
+            sg_len = MIN(limit - s->sg.size, bm->cur_prd_len);
+            if (sg_len) {
+                qemu_sglist_add(&s->sg, bm->cur_prd_addr, sg_len);
+            }
 
             /* Note: We limit the max transfer to be 2GiB.
              * This should accommodate the largest ATA transaction