Re: [PATCH v2 3/9] snic:Add meta request, handling of meta requests.

[Hannes Reinecke <hare@suse.de>]

Hi Narsimhulu,

On 03/11/2015 06:01 PM, Narsimhulu Musini wrote:
> snic_io.h contains meta request structure definition
> meta request contains high level information about firmware requests.
> such as request information, size, SGLs.
> 
> snic_io.c contains interfaces to handle meta request, firmware acknowledgment,
> and high level generic queueing interface.
> 
> Signed-off-by: Narsimhulu Musini <nmusini@cisco.com>
> Signed-off-by: Sesidhar Baddela <sebaddel@cisco.com>

Please find some comments inline.

> ---
>  drivers/scsi/snic/snic_io.c | 546 ++++++++++++++++++++++++++++++++++++++++++++
>  drivers/scsi/snic/snic_io.h | 111 +++++++++
>  2 files changed, 657 insertions(+)
>  create mode 100644 drivers/scsi/snic/snic_io.c
>  create mode 100644 drivers/scsi/snic/snic_io.h
> 
> diff --git a/drivers/scsi/snic/snic_io.c b/drivers/scsi/snic/snic_io.c
> new file mode 100644
> index 0000000..fe38c0b
> --- /dev/null
> +++ b/drivers/scsi/snic/snic_io.c
> @@ -0,0 +1,546 @@
> +/*
> + * Copyright 2014 Cisco Systems, Inc.  All rights reserved.
> + *
> + * This program is free software; you may redistribute it and/or modify
> + * it under the terms of the GNU General Public License as published by
> + * the Free Software Foundation; version 2 of the License.
> + *
> + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
> + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
> + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
> + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
> + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
> + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
> + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
> + * SOFTWARE.
> + */
> +
> +#include <linux/errno.h>
> +#include <linux/pci.h>
> +#include <linux/slab.h>
> +
> +#include <linux/interrupt.h>
> +#include <linux/workqueue.h>
> +#include <linux/spinlock.h>
> +#include <linux/mempool.h>
> +#include <scsi/scsi_tcq.h>
> +
> +#include "snic_io.h"
> +#include "snic.h"
> +#include "cq_enet_desc.h"
> +#include "snic_fwint.h"
> +#include "snic_os.h"
> +
> +static void
> +snic_wq_cmpl_frame_send(struct vnic_wq *wq,
> +			    struct cq_desc *cq_desc,
> +			    struct vnic_wq_buf *buf,
> +			    void *opaque)
> +{
> +	struct snic *snic = vnic_dev_priv(wq->vdev);
> +
> +	SNIC_BUG_ON(buf->os_buf == NULL);
> +
> +	if (snic_log_level & SNIC_DESC_LOGGING)
> +		SNIC_HOST_INFO(snic->shost,
> +			       "Ack received for snic_host_req %p.\n",
> +			       buf->os_buf);
> +
> +	SNIC_TRC(snic->shost->host_no, 0, 0,
> +		 ((u64)(buf->os_buf) - sizeof(struct snic_req_info)), 0, 0, 0);
> +	pci_unmap_single(snic->pdev, buf->dma_addr, buf->len, PCI_DMA_TODEVICE);
> +	buf->os_buf = NULL;
> +}
> +
> +static int
> +snic_wq_cmpl_handler_cont(struct vnic_dev *vdev,
> +			  struct cq_desc *cq_desc,
> +			  u8 type,
> +			  u16 q_num,
> +			  u16 cmpl_idx,
> +			  void *opaque)
> +{
> +	struct snic *snic = vnic_dev_priv(vdev);
> +	unsigned long flags;
> +
> +	SNIC_BUG_ON(q_num != 0);
> +
> +	spin_lock_irqsave(&snic->wq_lock[q_num], flags);
> +	vnic_wq_service(&snic->wq[q_num],
> +			cq_desc,
> +			cmpl_idx,
> +			snic_wq_cmpl_frame_send,
> +			NULL);
> +	spin_unlock_irqrestore(&snic->wq_lock[q_num], flags);
> +
> +	return 0;
> +} /* end of snic_cmpl_handler_cont */
> +
> +int
> +snic_wq_cmpl_handler(struct snic *snic, int work_to_do)
> +{
> +	unsigned int work_done = 0;
> +	unsigned int i;
> +
> +	snic->s_stats.misc.last_ack_time = jiffies;
> +	for (i = 0; i < snic->wq_count; i++) {
> +		work_done += vnic_cq_service(&snic->cq[i],
> +						work_to_do,
> +						snic_wq_cmpl_handler_cont,
> +						NULL);
> +	}
> +
> +	return work_done;
> +} /* end of snic_wq_cmpl_handler */
> +
> +void
> +snic_free_wq_buf(struct vnic_wq *wq, struct vnic_wq_buf *buf)
> +{
> +
> +	struct snic_host_req *req = buf->os_buf;
> +	struct snic *snic = vnic_dev_priv(wq->vdev);
> +	struct snic_req_info *rqi = NULL;
> +	unsigned long flags;
> +
> +	pci_unmap_single(snic->pdev, buf->dma_addr, buf->len, PCI_DMA_TODEVICE);
> +
> +	rqi = req_to_rqi(req);
> +	spin_lock_irqsave(&snic->spl_cmd_lock, flags);
> +	if (list_empty(&rqi->list)) {
> +		spin_unlock_irqrestore(&snic->spl_cmd_lock, flags);
> +		goto end;
> +	}
> +
> +	SNIC_BUG_ON(rqi->list.next == NULL); /* if not added to spl_cmd_list */
> +	list_del_init(&rqi->list);
> +	spin_unlock_irqrestore(&snic->spl_cmd_lock, flags);
> +
> +	if (rqi->sge_va) {
> +		snic_pci_unmap_rsp_buf(snic, rqi);
> +		kfree((void *)rqi->sge_va);
> +		rqi->sge_va = 0;
> +	}
> +	snic_req_free(snic, rqi);
> +	SNIC_HOST_INFO(snic->shost, "snic_free_wq_buf .. freed.\n");
> +
> +end:
> +	return;
> +}
> +
> +/* Criteria to select work queue in multi queue mode */
> +static int
> +snic_select_wq(struct snic *snic)
> +{
> +	/* No multi queue support for now */
> +	BUILD_BUG_ON(SNIC_WQ_MAX > 1);
> +
> +	return 0;
> +}
> +
> +int
> +snic_queue_wq_desc(struct snic *snic, void *os_buf, u16 len)
> +{
> +	dma_addr_t pa = 0;
> +	unsigned long flags;
> +	struct snic_fw_stats *fwstats = &snic->s_stats.fw;
> +	long act_reqs;
> +	int q_num = 0;
> +
> +	snic_print_desc(__func__, os_buf, len);
> +
> +	/* Map request buffer */
> +	pa = pci_map_single(snic->pdev, os_buf, len, PCI_DMA_TODEVICE);
> +	if (pci_dma_mapping_error(snic->pdev, pa)) {
> +		SNIC_HOST_ERR(snic->shost, "qdesc: PCI DMA Mapping Fail.\n");
> +
> +		return -ENOMEM;
> +	}
> +
> +	q_num = snic_select_wq(snic);
> +
> +	spin_lock_irqsave(&snic->wq_lock[q_num], flags);
> +	if (!vnic_wq_desc_avail(snic->wq)) {
> +		pci_unmap_single(snic->pdev, pa, len, PCI_DMA_TODEVICE);
> +		spin_unlock_irqrestore(&snic->wq_lock[q_num], flags);
> +		atomic64_inc(&snic->s_stats.misc.wq_alloc_fail);
> +		SNIC_DBG("host = %d, WQ is Full\n", snic->shost->host_no);
> +
> +		return -ENOMEM;
> +	}
> +
> +	snic_queue_wq_eth_desc(&snic->wq[q_num], os_buf, pa, len, 0, 0, 1);
> +	spin_unlock_irqrestore(&snic->wq_lock[q_num], flags);
> +
> +	/* Update stats */
> +	act_reqs = atomic64_inc_return(&fwstats->actv_reqs);
> +	if (act_reqs > atomic64_read(&fwstats->max_actv_reqs))
> +		atomic64_set(&fwstats->max_actv_reqs, act_reqs);
> +
> +	return 0;
> +} /* end of snic_queue_wq_desc() */
> +
> +/*
> + * snic_handle_untagged_req: Adds snic specific requests to spl_cmd_list.
> + * Purpose : Used during driver unload to clean up the requests.
> + */
> +void
> +snic_handle_untagged_req(struct snic *snic, struct snic_req_info *rqi)
> +{
> +	unsigned long flags;
> +
> +	INIT_LIST_HEAD(&rqi->list);
> +
> +	spin_lock_irqsave(&snic->spl_cmd_lock, flags);
> +	list_add_tail(&rqi->list, &snic->spl_cmd_list);
> +	spin_unlock_irqrestore(&snic->spl_cmd_lock, flags);
> +}
> +
> +static struct snic_host_req *
> +snic_req_align(struct snic_host_req *req)
> +{
> +	unsigned long sgl_ptr;
> +
> +	sgl_ptr = (unsigned long) req;
> +	/* Cache sgl list allocated address before alignment */
> +	if (sgl_ptr % SNIC_SG_DESC_ALIGN) {
> +		req = (struct snic_host_req *)
> +			   ((sgl_ptr + SNIC_SG_DESC_ALIGN - 1) &
> +			    ~(SNIC_SG_DESC_ALIGN - 1));
> +
> +	}
> +
> +	return req;
> +}
> +
> +/*
> + * snic_req_init:
> + * Allocates snic_req_info + snic_host_req + sgl data, and initializes.
> + */
> +struct snic_req_info *
> +snic_req_init(struct snic *snic, int sg_cnt)
> +{
> +	u8 typ;
> +	struct snic_req_info *rqi = NULL;
> +
> +	typ = (sg_cnt <= SNIC_REQ_CACHE_DFLT_SGL) ?
> +		SNIC_REQ_CACHE_DFLT_SGL : SNIC_REQ_CACHE_MAX_SGL;
> +
> +	rqi = mempool_alloc(snic->req_pool[typ], GFP_ATOMIC);
> +	if (!rqi) {
> +		atomic64_inc(&snic->s_stats.io.alloc_fail);
> +		SNIC_HOST_ERR(snic->shost,
> +			      "Failed to allocate memory from snic req pool id = %d\n",
> +			      typ);
> +		return rqi;
> +	}
> +
> +	memset(rqi, 0, sizeof(*rqi));
> +	rqi->rq_pool_type = typ;
> +	rqi->start_time = jiffies;
> +	rqi->req = (struct snic_host_req *) (rqi + 1);
> +	rqi->req_len = sizeof(struct snic_host_req);
> +	rqi->snic = snic;
> +
> +	BUILD_BUG_ON((sizeof(struct snic_req_info) % SNIC_SG_DESC_ALIGN) != 0);
> +	BUILD_BUG_ON((sizeof(struct snic_host_req) % SNIC_SG_DESC_ALIGN) != 0);
> +
> +	rqi->req = snic_req_align(rqi->req);
> +
Any reason why you can't allocate aligned requests directly from the
mempool?

> +	if (sg_cnt == 0)
> +		goto end;
> +
> +	rqi->req_len += (sg_cnt * sizeof(struct snic_sg_desc));
> +
> +	if (sg_cnt > atomic64_read(&snic->s_stats.io.max_sgl))
> +		atomic64_set(&snic->s_stats.io.max_sgl, sg_cnt);
> +
> +	SNIC_BUG_ON(sg_cnt > SNIC_MAX_SG_DESC_CNT);
> +	atomic64_inc(&snic->s_stats.io.sgl_cnt[sg_cnt - 1]);
> +
> +end:
> +	memset(rqi->req, 0, rqi->req_len);
> +
> +	/* pre initialization of init_ctx to support req_to_rqi */
> +	rqi->req->hdr.init_ctx = (u64) rqi;
> +
> +	SNIC_SCSI_DBG(snic->shost, "Req_alloc:rqi = %p allocatd.\n", rqi);
> +
> +	return rqi;
> +} /* end of snic_req_init */
> +
> +/*
> + * snic_abort_req_init : Inits abort request.
> + */
> +struct snic_host_req *
> +snic_abort_req_init(struct snic *snic, struct snic_req_info *rqi)
> +{
> +	struct snic_host_req *req = NULL;
> +
> +	SNIC_BUG_ON(!rqi);
> +
> +	/* If abort to be issued second time, then reuse */
> +	if (rqi->abort_req)
> +		return rqi->abort_req;
> +
> +
> +	req = mempool_alloc(snic->req_pool[SNIC_REQ_TM_CACHE], GFP_ATOMIC);
> +	if (!req) {
> +		SNIC_HOST_ERR(snic->shost, "abts:Failed to alloc tm req.\n");
> +		WARN_ON_ONCE(1);
> +
> +		return NULL;
> +	}
> +
> +	rqi->abort_req_unaligned = req;
> +	rqi->abort_req = snic_req_align(req);
Same here.

> +	req = rqi->abort_req;
> +	memset(req, 0, sizeof(struct snic_host_req));
> +	/* pre initialization of init_ctx to support req_to_rqi */
> +	req->hdr.init_ctx = (u64) rqi;
> +
> +	return req;
> +} /* end of snic_abort_req_init */
> +
> +/*
> + * snic_dr_req_init : Inits device reset req
> + */
> +struct snic_host_req *
> +snic_dr_req_init(struct snic *snic, struct snic_req_info *rqi)
> +{
> +	struct snic_host_req *req = NULL;
> +
> +	SNIC_BUG_ON(!rqi);
> +
> +	req = mempool_alloc(snic->req_pool[SNIC_REQ_TM_CACHE], GFP_ATOMIC);
> +	if (!req) {
> +		SNIC_HOST_ERR(snic->shost, "dr:Failed to alloc tm req.\n");
> +		WARN_ON_ONCE(1);
> +
> +		return NULL;
> +	}
> +
> +	SNIC_BUG_ON(rqi->dr_req_unaligned != NULL);
> +	rqi->dr_req_unaligned = req;
> +	rqi->dr_req = snic_req_align(req);
And here.

Cheers,

Hannes
-- 
Dr. Hannes Reinecke		               zSeries & Storage
hare@suse.de			               +49 911 74053 688
SUSE LINUX GmbH, Maxfeldstr. 5, 90409 Nürnberg
GF: F. Imendörffer, J. Smithard, J. Guild, D. Upmanyu, G. Norton
HRB 21284 (AG Nürnberg)
--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html