#define MPI3MR_NAME_LENGTH 32
#define IOCNAME "%s: "
+#define MPI3MR_MAX_SECTORS 2048
+
/* Definitions for internal SGL and Chain SGL buffers */
#define MPI3MR_PAGE_SIZE_4K 4096
#define MPI3MR_SG_DEPTH (MPI3MR_PAGE_SIZE_4K / sizeof(struct mpi3_sge_common))
u8 sge_mod_mask;
u8 sge_mod_value;
u8 sge_mod_shift;
+ u8 max_dev_per_tg;
+ u16 max_io_throttle_group;
+ u16 io_throttle_data_length;
+ u16 io_throttle_low;
+ u16 io_throttle_high;
+
};
/**
};
/**
+ * struct mpi3mr_throttle_group_info - Throttle group info
+ *
+ * @io_divert: Flag indicates io divert is on or off for the TG
+ * @id: Throttle Group ID.
+ * @high: High limit to turn on throttling in 512 byte blocks
+ * @low: Low limit to turn off throttling in 512 byte blocks
+ * @pend_large_data_sz: Counter to track pending large data
+ */
+struct mpi3mr_throttle_group_info {
+ u8 io_divert;
+ u16 id;
+ u32 high;
+ u32 low;
+ atomic_t pend_large_data_sz;
+};
+
+/**
* struct tgt_dev_sas_sata - SAS/SATA device specific
* information cached from firmware given data
*
};
/**
- * struct tgt_dev_volume - virtual device specific information
+ * struct tgt_dev_vd - virtual device specific information
* cached from firmware given data
*
* @state: State of the VD
+ * @tg_id: VDs throttle group ID
+ * @high: High limit to turn on throttling in 512 byte blocks
+ * @low: Low limit to turn off throttling in 512 byte blocks
+ * @tg: Pointer to throttle group info
*/
-struct tgt_dev_volume {
+struct tgt_dev_vd {
u8 state;
+ u16 tg_id;
+ u32 tg_high;
+ u32 tg_low;
+ struct mpi3mr_throttle_group_info *tg;
};
+
/**
* union _form_spec_inf - union of device specific information
*/
union _form_spec_inf {
struct tgt_dev_sas_sata sas_sata_inf;
struct tgt_dev_pcie pcie_inf;
- struct tgt_dev_volume vol_inf;
+ struct tgt_dev_vd vd_inf;
};
* @dev_type: SAS/SATA/PCIE device type
* @is_hidden: Should be exposed to upper layers or not
* @host_exposed: Already exposed to host or not
+ * @io_throttle_enabled: I/O throttling needed or not
* @q_depth: Device specific Queue Depth
* @wwid: World wide ID
* @dev_spec: Device type specific information
u8 dev_type;
u8 is_hidden;
u8 host_exposed;
+ u8 io_throttle_enabled;
u16 q_depth;
u64 wwid;
union _form_spec_inf dev_spec;
* @dev_removed: Device removed in the Firmware
* @dev_removedelay: Device is waiting to be removed in FW
* @dev_type: Device type
+ * @io_throttle_enabled: I/O throttling needed or not
+ * @io_divert: Flag indicates io divert is on or off for the dev
+ * @throttle_group: Pointer to throttle group info
* @tgt_dev: Internal target device pointer
* @pend_count: Counter to track pending I/Os during error
* handling
u8 dev_removed;
u8 dev_removedelay;
u8 dev_type;
+ u8 io_throttle_enabled;
+ u8 io_divert;
+ struct mpi3mr_throttle_group_info *throttle_group;
struct mpi3mr_tgt_dev *tgt_dev;
u32 pend_count;
};
* @logdata_buf: Circular buffer to store log data entries
* @logdata_buf_idx: Index of entry in buffer to store
* @logdata_entry_sz: log data entry size
+ * @pend_large_data_sz: Counter to track pending large data
+ * @io_throttle_data_length: I/O size to track in 512b blocks
+ * @io_throttle_high: I/O size to start throttle in 512b blocks
+ * @io_throttle_low: I/O size to stop throttle in 512b blocks
+ * @num_io_throttle_group: Maximum number of throttle groups
+ * @throttle_groups: Pointer to throttle group info structures
*/
struct mpi3mr_ioc {
struct list_head list;
u8 *logdata_buf;
u16 logdata_buf_idx;
u16 logdata_entry_sz;
+
+ atomic_t pend_large_data_sz;
+ u32 io_throttle_data_length;
+ u32 io_throttle_high;
+ u32 io_throttle_low;
+ u16 num_io_throttle_group;
+ struct mpi3mr_throttle_group_info *throttle_groups;
};
/**
mrioc->facts.shutdown_timeout =
le16_to_cpu(facts_data->shutdown_timeout);
+ mrioc->facts.max_dev_per_tg =
+ facts_data->max_devices_per_throttle_group;
+ mrioc->facts.io_throttle_data_length =
+ le16_to_cpu(facts_data->io_throttle_data_length);
+ mrioc->facts.max_io_throttle_group =
+ le16_to_cpu(facts_data->max_io_throttle_group);
+ mrioc->facts.io_throttle_low = le16_to_cpu(facts_data->io_throttle_low);
+ mrioc->facts.io_throttle_high =
+ le16_to_cpu(facts_data->io_throttle_high);
+
+ /* Store in 512b block count */
+ if (mrioc->facts.io_throttle_data_length)
+ mrioc->io_throttle_data_length =
+ (mrioc->facts.io_throttle_data_length * 2 * 4);
+ else
+ /* set the length to 1MB + 1K to disable throttle */
+ mrioc->io_throttle_data_length = MPI3MR_MAX_SECTORS + 2;
+
+ mrioc->io_throttle_high = (mrioc->facts.io_throttle_high * 2 * 1024);
+ mrioc->io_throttle_low = (mrioc->facts.io_throttle_low * 2 * 1024);
+
ioc_info(mrioc, "ioc_num(%d), maxopQ(%d), maxopRepQ(%d), maxdh(%d),",
mrioc->facts.ioc_num, mrioc->facts.max_op_req_q,
mrioc->facts.max_op_reply_q, mrioc->facts.max_devhandle);
ioc_info(mrioc, "DMA mask %d InitialPE status 0x%x\n",
mrioc->facts.dma_mask, (facts_flags &
MPI3_IOCFACTS_FLAGS_INITIAL_PORT_ENABLE_MASK));
+ ioc_info(mrioc,
+ "max_dev_per_throttle_group(%d), max_throttle_groups(%d)\n",
+ mrioc->facts.max_dev_per_tg, mrioc->facts.max_io_throttle_group);
+ ioc_info(mrioc,
+ "io_throttle_data_len(%dKiB), io_throttle_high(%dMiB), io_throttle_low(%dMiB)\n",
+ mrioc->facts.io_throttle_data_length * 4,
+ mrioc->facts.io_throttle_high, mrioc->facts.io_throttle_low);
}
/**
int retval = 0;
u8 retry = 0;
struct mpi3_ioc_facts_data facts_data;
+ u32 sz;
retry_init:
retval = mpi3mr_bring_ioc_ready(mrioc);
mrioc->max_host_ios = mrioc->facts.max_reqs - MPI3MR_INTERNAL_CMDS_RESVD;
+ mrioc->num_io_throttle_group = mrioc->facts.max_io_throttle_group;
+ atomic_set(&mrioc->pend_large_data_sz, 0);
+
if (reset_devices)
mrioc->max_host_ios = min_t(int, mrioc->max_host_ios,
MPI3MR_HOST_IOS_KDUMP);
}
}
+ if (!mrioc->throttle_groups && mrioc->num_io_throttle_group) {
+ dprint_init(mrioc, "allocating memory for throttle groups\n");
+ sz = sizeof(struct mpi3mr_throttle_group_info);
+ mrioc->throttle_groups = (struct mpi3mr_throttle_group_info *)
+ kcalloc(mrioc->num_io_throttle_group, sz, GFP_KERNEL);
+ if (!mrioc->throttle_groups)
+ goto out_failed_noretry;
+ }
+
retval = mpi3mr_enable_events(mrioc);
if (retval) {
ioc_err(mrioc, "failed to enable events %d\n",
void mpi3mr_memset_buffers(struct mpi3mr_ioc *mrioc)
{
u16 i;
+ struct mpi3mr_throttle_group_info *tg;
mrioc->change_count = 0;
mrioc->active_poll_qcount = 0;
spin_lock_init(&mrioc->req_qinfo[i].q_lock);
mpi3mr_memset_op_req_q_buffers(mrioc, i);
}
+
+ atomic_set(&mrioc->pend_large_data_sz, 0);
+ if (mrioc->throttle_groups) {
+ tg = mrioc->throttle_groups;
+ for (i = 0; i < mrioc->num_io_throttle_group; i++, tg++) {
+ tg->id = 0;
+ tg->io_divert = 0;
+ tg->high = 0;
+ tg->low = 0;
+ atomic_set(&tg->pend_large_data_sz, 0);
+ }
+ }
}
/**
ioc_err(mrioc, "Failed to issue soft reset to the ioc\n");
goto out;
}
+ if (mrioc->num_io_throttle_group !=
+ mrioc->facts.max_io_throttle_group) {
+ ioc_err(mrioc,
+ "max io throttle group doesn't match old(%d), new(%d)\n",
+ mrioc->num_io_throttle_group,
+ mrioc->facts.max_io_throttle_group);
+ return -EPERM;
+ }
mpi3mr_flush_delayed_cmd_lists(mrioc);
mpi3mr_flush_drv_cmds(mrioc);
if (tgtdev->starget && tgtdev->starget->hostdata) {
tgt_priv = tgtdev->starget->hostdata;
tgt_priv->dev_handle = MPI3MR_INVALID_DEV_HANDLE;
+ tgt_priv->io_throttle_enabled = 0;
+ tgt_priv->io_divert = 0;
+ tgt_priv->throttle_group = NULL;
}
}
}
}
/**
+ * mpi3mr_set_io_divert_for_all_vd_in_tg -set divert for TG VDs
+ * @mrioc: Adapter instance reference
+ * @tg: Throttle group information pointer
+ * @divert_value: 1 or 0
+ *
+ * Accessor to set io_divert flag for each device associated
+ * with the given throttle group with the given value.
+ *
+ * Return: None.
+ */
+static void mpi3mr_set_io_divert_for_all_vd_in_tg(struct mpi3mr_ioc *mrioc,
+ struct mpi3mr_throttle_group_info *tg, u8 divert_value)
+{
+ unsigned long flags;
+ struct mpi3mr_tgt_dev *tgtdev;
+ struct mpi3mr_stgt_priv_data *tgt_priv;
+
+ spin_lock_irqsave(&mrioc->tgtdev_lock, flags);
+ list_for_each_entry(tgtdev, &mrioc->tgtdev_list, list) {
+ if (tgtdev->starget && tgtdev->starget->hostdata) {
+ tgt_priv = tgtdev->starget->hostdata;
+ if (tgt_priv->throttle_group == tg)
+ tgt_priv->io_divert = divert_value;
+ }
+ }
+ spin_unlock_irqrestore(&mrioc->tgtdev_lock, flags);
+}
+
+/**
* mpi3mr_print_device_event_notice - print notice related to post processing of
* device event after controller reset.
*
* @mrioc: Adapter instance reference
* @tgtdev: Target device internal structure
* @dev_pg0: New device page0
+ * @is_added: Flag to indicate the device is just added
*
* Update the information from the device page0 into the driver
* cached target device structure.
* Return: Nothing.
*/
static void mpi3mr_update_tgtdev(struct mpi3mr_ioc *mrioc,
- struct mpi3mr_tgt_dev *tgtdev, struct mpi3_device_page0 *dev_pg0)
+ struct mpi3mr_tgt_dev *tgtdev, struct mpi3_device_page0 *dev_pg0,
+ bool is_added)
{
u16 flags = 0;
- struct mpi3mr_stgt_priv_data *scsi_tgt_priv_data;
+ struct mpi3mr_stgt_priv_data *scsi_tgt_priv_data = NULL;
u8 prot_mask = 0;
tgtdev->perst_id = le16_to_cpu(dev_pg0->persistent_id);
flags = le16_to_cpu(dev_pg0->flags);
tgtdev->is_hidden = (flags & MPI3_DEVICE0_FLAGS_HIDDEN);
+ if (is_added == true)
+ tgtdev->io_throttle_enabled =
+ (flags & MPI3_DEVICE0_FLAGS_IO_THROTTLING_REQUIRED) ? 1 : 0;
+
+
if (tgtdev->starget && tgtdev->starget->hostdata) {
scsi_tgt_priv_data = (struct mpi3mr_stgt_priv_data *)
tgtdev->starget->hostdata;
scsi_tgt_priv_data->perst_id = tgtdev->perst_id;
scsi_tgt_priv_data->dev_handle = tgtdev->dev_handle;
scsi_tgt_priv_data->dev_type = tgtdev->dev_type;
+ scsi_tgt_priv_data->io_throttle_enabled =
+ tgtdev->io_throttle_enabled;
}
switch (dev_pg0->access_status) {
{
struct mpi3_device0_vd_format *vdinf =
&dev_pg0->device_specific.vd_format;
+ struct mpi3mr_throttle_group_info *tg = NULL;
+ u16 vdinf_io_throttle_group =
+ le16_to_cpu(vdinf->io_throttle_group);
- tgtdev->dev_spec.vol_inf.state = vdinf->vd_state;
+ tgtdev->dev_spec.vd_inf.state = vdinf->vd_state;
if (vdinf->vd_state == MPI3_DEVICE0_VD_STATE_OFFLINE)
tgtdev->is_hidden = 1;
+ tgtdev->dev_spec.vd_inf.tg_id = vdinf_io_throttle_group;
+ tgtdev->dev_spec.vd_inf.tg_high =
+ le16_to_cpu(vdinf->io_throttle_group_high) * 2048;
+ tgtdev->dev_spec.vd_inf.tg_low =
+ le16_to_cpu(vdinf->io_throttle_group_low) * 2048;
+ if (vdinf_io_throttle_group < mrioc->num_io_throttle_group) {
+ tg = mrioc->throttle_groups + vdinf_io_throttle_group;
+ tg->id = vdinf_io_throttle_group;
+ tg->high = tgtdev->dev_spec.vd_inf.tg_high;
+ tg->low = tgtdev->dev_spec.vd_inf.tg_low;
+ }
+ tgtdev->dev_spec.vd_inf.tg = tg;
+ if (scsi_tgt_priv_data)
+ scsi_tgt_priv_data->throttle_group = tg;
break;
}
default:
tgtdev = mpi3mr_get_tgtdev_by_handle(mrioc, dev_handle);
if (!tgtdev)
goto out;
- mpi3mr_update_tgtdev(mrioc, tgtdev, dev_pg0);
+ mpi3mr_update_tgtdev(mrioc, tgtdev, dev_pg0, false);
if (!tgtdev->is_hidden && !tgtdev->host_exposed)
mpi3mr_report_tgtdev_to_host(mrioc, perst_id);
if (tgtdev->is_hidden && tgtdev->host_exposed)
perst_id = le16_to_cpu(dev_pg0->persistent_id);
tgtdev = mpi3mr_get_tgtdev_by_perst_id(mrioc, perst_id);
if (tgtdev) {
- mpi3mr_update_tgtdev(mrioc, tgtdev, dev_pg0);
+ mpi3mr_update_tgtdev(mrioc, tgtdev, dev_pg0, true);
mpi3mr_tgtdev_put(tgtdev);
} else {
tgtdev = mpi3mr_alloc_tgtdev();
if (!tgtdev)
return -ENOMEM;
- mpi3mr_update_tgtdev(mrioc, tgtdev, dev_pg0);
+ mpi3mr_update_tgtdev(mrioc, tgtdev, dev_pg0, true);
mpi3mr_tgtdev_add_to_list(mrioc, tgtdev);
}
u32 xfer_count = 0, sense_count = 0, resp_data = 0;
u16 dev_handle = 0xFFFF;
struct scsi_sense_hdr sshdr;
+ struct mpi3mr_stgt_priv_data *stgt_priv_data = NULL;
+ struct mpi3mr_sdev_priv_data *sdev_priv_data = NULL;
+ u32 ioc_pend_data_len = 0, tg_pend_data_len = 0, data_len_blks = 0;
+ struct mpi3mr_throttle_group_info *tg = NULL;
+ u8 throttle_enabled_dev = 0;
*reply_dma = 0;
reply_desc_type = le16_to_cpu(reply_desc->reply_flags) &
goto out;
}
priv = scsi_cmd_priv(scmd);
+
+ data_len_blks = scsi_bufflen(scmd) >> 9;
+ sdev_priv_data = scmd->device->hostdata;
+ if (sdev_priv_data) {
+ stgt_priv_data = sdev_priv_data->tgt_priv_data;
+ if (stgt_priv_data) {
+ tg = stgt_priv_data->throttle_group;
+ throttle_enabled_dev =
+ stgt_priv_data->io_throttle_enabled;
+ }
+ }
+ if (unlikely((data_len_blks >= mrioc->io_throttle_data_length) &&
+ throttle_enabled_dev)) {
+ ioc_pend_data_len = atomic_sub_return(data_len_blks,
+ &mrioc->pend_large_data_sz);
+ if (tg) {
+ tg_pend_data_len = atomic_sub_return(data_len_blks,
+ &tg->pend_large_data_sz);
+ if (tg->io_divert && ((ioc_pend_data_len <=
+ mrioc->io_throttle_low) &&
+ (tg_pend_data_len <= tg->low))) {
+ tg->io_divert = 0;
+ mpi3mr_set_io_divert_for_all_vd_in_tg(
+ mrioc, tg, 0);
+ }
+ } else {
+ if (ioc_pend_data_len <= mrioc->io_throttle_low)
+ stgt_priv_data->io_divert = 0;
+ }
+ } else if (unlikely((stgt_priv_data && stgt_priv_data->io_divert))) {
+ ioc_pend_data_len = atomic_read(&mrioc->pend_large_data_sz);
+ if (!tg) {
+ if (ioc_pend_data_len <= mrioc->io_throttle_low)
+ stgt_priv_data->io_divert = 0;
+
+ } else if (ioc_pend_data_len <= mrioc->io_throttle_low) {
+ tg_pend_data_len = atomic_read(&tg->pend_large_data_sz);
+ if (tg->io_divert && (tg_pend_data_len <= tg->low)) {
+ tg->io_divert = 0;
+ mpi3mr_set_io_divert_for_all_vd_in_tg(
+ mrioc, tg, 0);
+ }
+ }
+ }
+
if (success_desc) {
scmd->result = DID_OK << 16;
goto out_success;
tgt_dev->starget = starget;
atomic_set(&scsi_tgt_priv_data->block_io, 0);
retval = 0;
+ scsi_tgt_priv_data->io_throttle_enabled =
+ tgt_dev->io_throttle_enabled;
+ if (tgt_dev->dev_type == MPI3_DEVICE_DEVFORM_VD)
+ scsi_tgt_priv_data->throttle_group =
+ tgt_dev->dev_spec.vd_inf.tg;
} else
retval = -ENXIO;
spin_unlock_irqrestore(&mrioc->tgtdev_lock, flags);
int retval = 0;
u16 dev_handle;
u16 host_tag;
- u32 scsiio_flags = 0;
+ u32 scsiio_flags = 0, data_len_blks = 0;
struct request *rq = scsi_cmd_to_rq(scmd);
int iprio_class;
u8 is_pcie_dev = 0;
+ u32 tracked_io_sz = 0;
+ u32 ioc_pend_data_len = 0, tg_pend_data_len = 0;
+ struct mpi3mr_throttle_group_info *tg = NULL;
if (mrioc->unrecoverable) {
scmd->result = DID_ERROR << 16;
goto out;
}
op_req_q = &mrioc->req_qinfo[scmd_priv_data->req_q_idx];
+ data_len_blks = scsi_bufflen(scmd) >> 9;
+ if ((data_len_blks >= mrioc->io_throttle_data_length) &&
+ stgt_priv_data->io_throttle_enabled) {
+ tracked_io_sz = data_len_blks;
+ tg = stgt_priv_data->throttle_group;
+ if (tg) {
+ ioc_pend_data_len = atomic_add_return(data_len_blks,
+ &mrioc->pend_large_data_sz);
+ tg_pend_data_len = atomic_add_return(data_len_blks,
+ &tg->pend_large_data_sz);
+ if (!tg->io_divert && ((ioc_pend_data_len >=
+ mrioc->io_throttle_high) ||
+ (tg_pend_data_len >= tg->high))) {
+ tg->io_divert = 1;
+ mpi3mr_set_io_divert_for_all_vd_in_tg(mrioc,
+ tg, 1);
+ }
+ } else {
+ ioc_pend_data_len = atomic_add_return(data_len_blks,
+ &mrioc->pend_large_data_sz);
+ if (ioc_pend_data_len >= mrioc->io_throttle_high)
+ stgt_priv_data->io_divert = 1;
+ }
+ }
+
+ if (stgt_priv_data->io_divert) {
+ scsiio_req->msg_flags |=
+ MPI3_SCSIIO_MSGFLAGS_DIVERT_TO_FIRMWARE;
+ scsiio_flags |= MPI3_SCSIIO_FLAGS_DIVERT_REASON_IO_THROTTLING;
+ }
+ scsiio_req->flags = cpu_to_le32(scsiio_flags);
if (mpi3mr_op_request_post(mrioc, op_req_q,
scmd_priv_data->mpi3mr_scsiio_req)) {
mpi3mr_clear_scmd_priv(mrioc, scmd);
retval = SCSI_MLQUEUE_HOST_BUSY;
+ if (tracked_io_sz) {
+ atomic_sub(tracked_io_sz, &mrioc->pend_large_data_sz);
+ if (tg)
+ atomic_sub(tracked_io_sz,
+ &tg->pend_large_data_sz);
+ }
goto out;
}
projects / platform / kernel / linux-starfive.git / commitdiff