MODULE_LICENSE("Dual BSD/GPL");
MODULE_AUTHOR("Intel Corporation");
+static bool xeon_errata_workaround = true;
+module_param(xeon_errata_workaround, bool, 0644);
+MODULE_PARM_DESC(xeon_errata_workaround, "Workaround for the Xeon Errata");
+
enum {
NTB_CONN_CLASSIC = 0,
NTB_CONN_B2B,
static struct dentry *debugfs_dir;
/* Translate memory window 0,1 to BAR 2,4 */
-#define MW_TO_BAR(mw) (mw * 2 + 2)
+#define MW_TO_BAR(mw) (mw * NTB_MAX_NUM_MW + 2)
static DEFINE_PCI_DEVICE_TABLE(ntb_pci_tbl) = {
{PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_B2B_BWD)},
*/
void __iomem *ntb_get_mw_vbase(struct ntb_device *ndev, unsigned int mw)
{
- if (mw >= NTB_NUM_MW)
+ if (mw >= ntb_max_mw(ndev))
return NULL;
return ndev->mw[mw].vbase;
*/
resource_size_t ntb_get_mw_size(struct ntb_device *ndev, unsigned int mw)
{
- if (mw >= NTB_NUM_MW)
+ if (mw >= ntb_max_mw(ndev))
return 0;
return ndev->mw[mw].bar_sz;
*/
void ntb_set_mw_addr(struct ntb_device *ndev, unsigned int mw, u64 addr)
{
- if (mw >= NTB_NUM_MW)
+ if (mw >= ntb_max_mw(ndev))
return;
dev_dbg(&ndev->pdev->dev, "Writing addr %Lx to BAR %d\n", addr,
ndev->reg_ofs.spad_read = ndev->reg_base + SNB_SPAD_OFFSET;
ndev->reg_ofs.spci_cmd = ndev->reg_base + SNB_PCICMD_OFFSET;
- if (ndev->conn_type == NTB_CONN_B2B) {
- ndev->reg_ofs.sdb = ndev->reg_base + SNB_B2B_DOORBELL_OFFSET;
- ndev->reg_ofs.spad_write = ndev->reg_base + SNB_B2B_SPAD_OFFSET;
- ndev->limits.max_spads = SNB_MAX_B2B_SPADS;
+ /* There is a Xeon hardware errata related to writes to
+ * SDOORBELL or B2BDOORBELL in conjunction with inbound access
+ * to NTB MMIO Space, which may hang the system. To workaround
+ * this use the second memory window to access the interrupt and
+ * scratch pad registers on the remote system.
+ */
+ if (xeon_errata_workaround) {
+ if (!ndev->mw[1].bar_sz)
+ return -EINVAL;
+
+ ndev->limits.max_mw = SNB_ERRATA_MAX_MW;
+ ndev->reg_ofs.spad_write = ndev->mw[1].vbase +
+ SNB_SPAD_OFFSET;
+ ndev->reg_ofs.sdb = ndev->mw[1].vbase +
+ SNB_PDOORBELL_OFFSET;
+
+ /* Set the Limit register to 4k, the minimum size, to
+ * prevent an illegal access
+ */
+ writeq(ndev->mw[1].bar_sz + 0x1000, ndev->reg_base +
+ SNB_PBAR4LMT_OFFSET);
+ } else {
+ ndev->limits.max_mw = SNB_MAX_MW;
+ ndev->reg_ofs.spad_write = ndev->reg_base +
+ SNB_B2B_SPAD_OFFSET;
+ ndev->reg_ofs.sdb = ndev->reg_base +
+ SNB_B2B_DOORBELL_OFFSET;
+
+ /* Disable the Limit register, just incase it is set to
+ * something silly
+ */
+ writeq(0, ndev->reg_base + SNB_PBAR4LMT_OFFSET);
+ }
+
+ /* The Xeon errata workaround requires setting SBAR Base
+ * addresses to known values, so that the PBAR XLAT can be
+ * pointed at SBAR0 of the remote system.
+ */
+ if (ndev->dev_type == NTB_DEV_USD) {
+ writeq(SNB_MBAR23_DSD_ADDR, ndev->reg_base +
+ SNB_PBAR2XLAT_OFFSET);
+ if (xeon_errata_workaround)
+ writeq(SNB_MBAR01_DSD_ADDR, ndev->reg_base +
+ SNB_PBAR4XLAT_OFFSET);
+ else {
+ writeq(SNB_MBAR45_DSD_ADDR, ndev->reg_base +
+ SNB_PBAR4XLAT_OFFSET);
+ /* B2B_XLAT_OFFSET is a 64bit register, but can
+ * only take 32bit writes
+ */
+ writel(SNB_MBAR01_USD_ADDR & 0xffffffff,
+ ndev->reg_base + SNB_B2B_XLAT_OFFSETL);
+ writel(SNB_MBAR01_DSD_ADDR >> 32,
+ ndev->reg_base + SNB_B2B_XLAT_OFFSETU);
+ }
+
+ writeq(SNB_MBAR01_USD_ADDR, ndev->reg_base +
+ SNB_SBAR0BASE_OFFSET);
+ writeq(SNB_MBAR23_USD_ADDR, ndev->reg_base +
+ SNB_SBAR2BASE_OFFSET);
+ writeq(SNB_MBAR45_USD_ADDR, ndev->reg_base +
+ SNB_SBAR4BASE_OFFSET);
} else {
- ndev->reg_ofs.sdb = ndev->reg_base + SNB_SDOORBELL_OFFSET;
- ndev->reg_ofs.spad_write = ndev->reg_base + SNB_SPAD_OFFSET;
- ndev->limits.max_spads = SNB_MAX_COMPAT_SPADS;
+ writeq(SNB_MBAR23_USD_ADDR, ndev->reg_base +
+ SNB_PBAR2XLAT_OFFSET);
+ if (xeon_errata_workaround)
+ writeq(SNB_MBAR01_USD_ADDR, ndev->reg_base +
+ SNB_PBAR4XLAT_OFFSET);
+ else {
+ writeq(SNB_MBAR45_USD_ADDR, ndev->reg_base +
+ SNB_PBAR4XLAT_OFFSET);
+ /* B2B_XLAT_OFFSET is a 64bit register, but can
+ * only take 32bit writes
+ */
+ writel(SNB_MBAR01_USD_ADDR & 0xffffffff,
+ ndev->reg_base + SNB_B2B_XLAT_OFFSETL);
+ writel(SNB_MBAR01_USD_ADDR >> 32,
+ ndev->reg_base + SNB_B2B_XLAT_OFFSETU);
+ }
+ writeq(SNB_MBAR01_DSD_ADDR, ndev->reg_base +
+ SNB_SBAR0BASE_OFFSET);
+ writeq(SNB_MBAR23_DSD_ADDR, ndev->reg_base +
+ SNB_SBAR2BASE_OFFSET);
+ writeq(SNB_MBAR45_DSD_ADDR, ndev->reg_base +
+ SNB_SBAR4BASE_OFFSET);
}
+ ndev->limits.max_spads = SNB_MAX_B2B_SPADS;
ndev->limits.max_db_bits = SNB_MAX_DB_BITS;
ndev->limits.msix_cnt = SNB_MSIX_CNT;
ndev->bits_per_vector = SNB_DB_BITS_PER_VEC;
ndev->limits.max_spads = BWD_MAX_COMPAT_SPADS;
}
+ ndev->limits.max_mw = BWD_MAX_MW;
ndev->limits.max_db_bits = BWD_MAX_DB_BITS;
ndev->limits.msix_cnt = BWD_MSIX_CNT;
ndev->bits_per_vector = BWD_DB_BITS_PER_VEC;
goto err2;
}
- for (i = 0; i < NTB_NUM_MW; i++) {
+ for (i = 0; i < NTB_MAX_NUM_MW; i++) {
ndev->mw[i].bar_sz = pci_resource_len(pdev, MW_TO_BAR(i));
ndev->mw[i].vbase =
ioremap_wc(pci_resource_start(pdev, MW_TO_BAR(i)),
ntb_free_callbacks(ndev);
ntb_device_free(ndev);
- for (i = 0; i < NTB_NUM_MW; i++)
+ for (i = 0; i < NTB_MAX_NUM_MW; i++)
iounmap(ndev->mw[i].vbase);
iounmap(ndev->reg_base);
#define NTB_HB_TIMEOUT msecs_to_jiffies(1000)
-#define NTB_NUM_MW 2
+#define NTB_MAX_NUM_MW 2
enum ntb_hw_event {
NTB_EVENT_SW_EVENT0 = 0,
struct pci_dev *pdev;
struct msix_entry *msix_entries;
void __iomem *reg_base;
- struct ntb_mw mw[NTB_NUM_MW];
+ struct ntb_mw mw[NTB_MAX_NUM_MW];
struct {
- unsigned int max_spads;
- unsigned int max_db_bits;
- unsigned int msix_cnt;
+ unsigned char max_mw;
+ unsigned char max_spads;
+ unsigned char max_db_bits;
+ unsigned char msix_cnt;
} limits;
struct {
void __iomem *pdb;
struct dentry *debugfs_dir;
};
+/**
+ * ntb_max_cbs() - return the max callbacks
+ * @ndev: pointer to ntb_device instance
+ *
+ * Given the ntb pointer, return the maximum number of callbacks
+ *
+ * RETURNS: the maximum number of callbacks
+ */
+static inline unsigned char ntb_max_cbs(struct ntb_device *ndev)
+{
+ return ndev->max_cbs;
+}
+
+/**
+ * ntb_max_mw() - return the max number of memory windows
+ * @ndev: pointer to ntb_device instance
+ *
+ * Given the ntb pointer, return the maximum number of memory windows
+ *
+ * RETURNS: the maximum number of memory windows
+ */
+static inline unsigned char ntb_max_mw(struct ntb_device *ndev)
+{
+ return ndev->limits.max_mw;
+}
+
/**
* ntb_hw_link_status() - return the hardware link status
* @ndev: pointer to ntb_device instance
* ntb_query_pdev() - return the pci_dev pointer
* @ndev: pointer to ntb_device instance
*
- * Given the ntb pointer return the pci_dev pointerfor the NTB hardware device
+ * Given the ntb pointer, return the pci_dev pointer for the NTB hardware device
*
* RETURNS: a pointer to the ntb pci_dev
*/
/* Reserve the uppermost bit for link interrupt */
#define SNB_MAX_DB_BITS 15
#define SNB_DB_BITS_PER_VEC 5
+#define SNB_MAX_MW 2
+#define SNB_ERRATA_MAX_MW 1
#define SNB_DB_HW_LINK 0x8000
#define SNB_SBAR2XLAT_OFFSET 0x0030
#define SNB_SBAR4XLAT_OFFSET 0x0038
#define SNB_SBAR0BASE_OFFSET 0x0040
+#define SNB_SBAR0BASE_OFFSET 0x0040
+#define SNB_SBAR2BASE_OFFSET 0x0048
+#define SNB_SBAR4BASE_OFFSET 0x0050
#define SNB_SBAR2BASE_OFFSET 0x0048
#define SNB_SBAR4BASE_OFFSET 0x0050
#define SNB_NTBCNTL_OFFSET 0x0058
#define SNB_WCCNTRL_OFFSET 0x00e0
#define SNB_B2B_SPAD_OFFSET 0x0100
#define SNB_B2B_DOORBELL_OFFSET 0x0140
-#define SNB_B2B_XLAT_OFFSET 0x0144
+#define SNB_B2B_XLAT_OFFSETL 0x0144
+#define SNB_B2B_XLAT_OFFSETU 0x0148
+
+#define SNB_MBAR01_USD_ADDR 0x000000210000000CULL
+#define SNB_MBAR23_USD_ADDR 0x000000410000000CULL
+#define SNB_MBAR45_USD_ADDR 0x000000810000000CULL
+#define SNB_MBAR01_DSD_ADDR 0x000000200000000CULL
+#define SNB_MBAR23_DSD_ADDR 0x000000400000000CULL
+#define SNB_MBAR45_DSD_ADDR 0x000000800000000CULL
#define BWD_MSIX_CNT 34
#define BWD_MAX_SPADS 16
#define BWD_MAX_COMPAT_SPADS 16
#define BWD_MAX_DB_BITS 34
#define BWD_DB_BITS_PER_VEC 1
+#define BWD_MAX_MW 2
#define BWD_PCICMD_OFFSET 0xb004
#define BWD_MBAR23_OFFSET 0xb018
#define BWD_PPD_INIT_LINK 0x0008
#define BWD_PPD_CONN_TYPE 0x0300
#define BWD_PPD_DEV_TYPE 0x1000
-
-#define BWD_PBAR2XLAT_USD_ADDR 0x0000004000000000
-#define BWD_PBAR4XLAT_USD_ADDR 0x0000008000000000
-#define BWD_MBAR23_USD_ADDR 0x000000410000000C
-#define BWD_MBAR45_USD_ADDR 0x000000810000000C
-#define BWD_PBAR2XLAT_DSD_ADDR 0x0000004100000000
-#define BWD_PBAR4XLAT_DSD_ADDR 0x0000008100000000
-#define BWD_MBAR23_DSD_ADDR 0x000000400000000C
-#define BWD_MBAR45_DSD_ADDR 0x000000800000000C
module_param(transport_mtu, uint, 0644);
MODULE_PARM_DESC(transport_mtu, "Maximum size of NTB transport packets");
-static unsigned char max_num_clients = 2;
+static unsigned char max_num_clients;
module_param(max_num_clients, byte, 0644);
MODULE_PARM_DESC(max_num_clients, "Maximum number of NTB transport clients");
struct list_head client_devs;
struct ntb_device *ndev;
- struct ntb_transport_mw mw[NTB_NUM_MW];
+ struct ntb_transport_mw *mw;
struct ntb_transport_qp *qps;
unsigned int max_qps;
unsigned long qp_bitmap;
MAX_SPAD,
};
-#define QP_TO_MW(qp) ((qp) % NTB_NUM_MW)
+#define QP_TO_MW(ndev, qp) ((qp) % ntb_max_mw(ndev))
#define NTB_QP_DEF_NUM_ENTRIES 100
#define NTB_LINK_DOWN_TIMEOUT 10
{
struct ntb_transport_qp *qp = &nt->qps[qp_num];
unsigned int rx_size, num_qps_mw;
- u8 mw_num = QP_TO_MW(qp_num);
+ u8 mw_num, mw_max;
unsigned int i;
+ mw_max = ntb_max_mw(nt->ndev);
+ mw_num = QP_TO_MW(nt->ndev, qp_num);
+
WARN_ON(nt->mw[mw_num].virt_addr == NULL);
- if (nt->max_qps % NTB_NUM_MW && mw_num < nt->max_qps % NTB_NUM_MW)
- num_qps_mw = nt->max_qps / NTB_NUM_MW + 1;
+ if (nt->max_qps % mw_max && mw_num < nt->max_qps % mw_max)
+ num_qps_mw = nt->max_qps / mw_max + 1;
else
- num_qps_mw = nt->max_qps / NTB_NUM_MW;
+ num_qps_mw = nt->max_qps / mw_max;
rx_size = (unsigned int) nt->mw[mw_num].size / num_qps_mw;
qp->remote_rx_info = nt->mw[mw_num].virt_addr +
- (qp_num / NTB_NUM_MW * rx_size);
+ (qp_num / mw_max * rx_size);
rx_size -= sizeof(struct ntb_rx_info);
qp->rx_buff = qp->remote_rx_info + 1;
int rc, i;
/* send the local info, in the opposite order of the way we read it */
- for (i = 0; i < NTB_NUM_MW; i++) {
+ for (i = 0; i < ntb_max_mw(ndev); i++) {
rc = ntb_write_remote_spad(ndev, MW0_SZ_HIGH + (i * 2),
ntb_get_mw_size(ndev, i) >> 32);
if (rc) {
}
}
- rc = ntb_write_remote_spad(ndev, NUM_MWS, NTB_NUM_MW);
+ rc = ntb_write_remote_spad(ndev, NUM_MWS, ntb_max_mw(ndev));
if (rc) {
dev_err(&pdev->dev, "Error writing %x to remote spad %d\n",
- NTB_NUM_MW, NUM_MWS);
+ ntb_max_mw(ndev), NUM_MWS);
goto out;
}
goto out;
}
- if (val != NTB_NUM_MW)
+ if (val != ntb_max_mw(ndev))
goto out;
dev_dbg(&pdev->dev, "Remote number of mws = %d\n", val);
- for (i = 0; i < NTB_NUM_MW; i++) {
+ for (i = 0; i < ntb_max_mw(ndev); i++) {
u64 val64;
rc = ntb_read_remote_spad(ndev, MW0_SZ_HIGH + (i * 2), &val);
return;
out1:
- for (i = 0; i < NTB_NUM_MW; i++)
+ for (i = 0; i < ntb_max_mw(ndev); i++)
ntb_free_mw(nt, i);
out:
if (ntb_hw_link_status(ndev))
{
struct ntb_transport_qp *qp;
unsigned int num_qps_mw, tx_size;
- u8 mw_num = QP_TO_MW(qp_num);
+ u8 mw_num, mw_max;
+
+ mw_max = ntb_max_mw(nt->ndev);
+ mw_num = QP_TO_MW(nt->ndev, qp_num);
qp = &nt->qps[qp_num];
qp->qp_num = qp_num;
qp->client_ready = NTB_LINK_DOWN;
qp->event_handler = NULL;
- if (nt->max_qps % NTB_NUM_MW && mw_num < nt->max_qps % NTB_NUM_MW)
- num_qps_mw = nt->max_qps / NTB_NUM_MW + 1;
+ if (nt->max_qps % mw_max && mw_num < nt->max_qps % mw_max)
+ num_qps_mw = nt->max_qps / mw_max + 1;
else
- num_qps_mw = nt->max_qps / NTB_NUM_MW;
+ num_qps_mw = nt->max_qps / mw_max;
tx_size = (unsigned int) ntb_get_mw_size(qp->ndev, mw_num) / num_qps_mw;
qp->rx_info = ntb_get_mw_vbase(nt->ndev, mw_num) +
- (qp_num / NTB_NUM_MW * tx_size);
+ (qp_num / mw_max * tx_size);
tx_size -= sizeof(struct ntb_rx_info);
qp->tx_mw = qp->rx_info + 1;
goto err;
}
- nt->max_qps = min(nt->ndev->max_cbs, max_num_clients);
+ nt->mw = kcalloc(ntb_max_mw(nt->ndev), sizeof(struct ntb_transport_mw),
+ GFP_KERNEL);
+ if (!nt->mw) {
+ rc = -ENOMEM;
+ goto err1;
+ }
+
+ if (max_num_clients)
+ nt->max_qps = min(ntb_max_cbs(nt->ndev), max_num_clients);
+ else
+ nt->max_qps = min(ntb_max_cbs(nt->ndev), ntb_max_mw(nt->ndev));
nt->qps = kcalloc(nt->max_qps, sizeof(struct ntb_transport_qp),
GFP_KERNEL);
if (!nt->qps) {
rc = -ENOMEM;
- goto err1;
+ goto err2;
}
nt->qp_bitmap = ((u64) 1 << nt->max_qps) - 1;
rc = ntb_register_event_callback(nt->ndev,
ntb_transport_event_callback);
if (rc)
- goto err2;
+ goto err3;
INIT_LIST_HEAD(&nt->client_devs);
rc = ntb_bus_init(nt);
if (rc)
- goto err3;
+ goto err4;
if (ntb_hw_link_status(nt->ndev))
schedule_delayed_work(&nt->link_work, 0);
return 0;
-err3:
+err4:
ntb_unregister_event_callback(nt->ndev);
-err2:
+err3:
kfree(nt->qps);
+err2:
+ kfree(nt->mw);
err1:
ntb_unregister_transport(nt->ndev);
err:
void ntb_transport_free(void *transport)
{
struct ntb_transport *nt = transport;
+ struct ntb_device *ndev = nt->ndev;
struct pci_dev *pdev;
int i;
cancel_delayed_work_sync(&nt->link_work);
- ntb_unregister_event_callback(nt->ndev);
+ ntb_unregister_event_callback(ndev);
- pdev = ntb_query_pdev(nt->ndev);
+ pdev = ntb_query_pdev(ndev);
- for (i = 0; i < NTB_NUM_MW; i++)
+ for (i = 0; i < ntb_max_mw(ndev); i++)
ntb_free_mw(nt, i);
kfree(nt->qps);
- ntb_unregister_transport(nt->ndev);
+ kfree(nt->mw);
+ ntb_unregister_transport(ndev);
kfree(nt);
}