From: Bart Van Assche <bvanassche@acm.org>
To: Avri Altman <avri.altman@wdc.com>,
"James E . J . Bottomley" <jejb@linux.vnet.ibm.com>,
"Martin K . Petersen" <martin.petersen@oracle.com>,
linux-scsi@vger.kernel.org, linux-kernel@vger.kernel.org
Cc: alim.akhtar@samsung.com, asutoshd@codeaurora.org,
Zang Leigang <zangleigang@hisilicon.com>,
Avi Shchislowski <avi.shchislowski@wdc.com>,
Bean Huo <beanhuo@micron.com>,
cang@codeaurora.org, stanley.chu@mediatek.com,
MOHAMMED RAFIQ KAMAL BASHA <md.rafiq@samsung.com>,
Sang-yoon Oh <sangyoon.oh@samsung.com>,
yongmyung lee <ymhungry.lee@samsung.com>,
Jinyoung CHOI <j-young.choi@samsung.com>
Subject: Re: [RFC PATCH 07/13] scsi: scsi_dh: ufshpb: Add ufshpb state machine
Date: Fri, 15 May 2020 19:44:20 -0700 [thread overview]
Message-ID: <5fd7224f-3bbd-863f-ffd4-b049aeeb5456@acm.org> (raw)
In-Reply-To: <1589538614-24048-8-git-send-email-avri.altman@wdc.com>
On 2020-05-15 03:30, Avri Altman wrote:
> @@ -17,6 +18,13 @@
>
> #define UFSHPB_NAME "ufshpb"
>
> +#define UFSHPB_WRITE_BUFFER (0xfa)
> +#define WRITE_BUFFER_TIMEOUT (3 * HZ)
> +#define WRITE_BUFFER_RETRIES (3)
> +#define UFSHPB_READ_BUFFER (0xf9)
> +#define READ_BUFFER_TIMEOUT (3 * HZ)
> +#define READ_BUFFER_RETRIES (3)
Parentheses around expressions are normal but parentheses around
constants are unusual. I think the parentheses around constants can be
left out.
> +#define to_subregion() (container_of(work, struct ufshpb_subregion, hpb_work))
Could this have been defined as an inline function?
> @@ -76,6 +118,7 @@ struct ufshpb_subregion {
> * @writes - sum over subregions @writes
> * @region - region index
> * @active_subregions - actual active subregions
> + * @evicted - to indicated if this region is currently being evicted
> */
> struct ufshpb_region {
> struct ufshpb_subregion *subregion_tbl;
> @@ -85,6 +128,7 @@ struct ufshpb_region {
> unsigned int region;
>
> atomic_t active_subregions;
> + atomic_t evicted;
> };
Declaring a state variable as atomic_t is unusual. How are changes of
the @evicted member variable serialized?
> /**
> @@ -93,6 +137,7 @@ struct ufshpb_region {
> * @lh_map_ctx - list head of mapping context
> * @map_list_lock - to protect mapping context list operations
> * @region_tbl - regions/subregions table
> + * @pinned_map - to mark pinned regions
> * @sdev - scsi device for that lun
> * @regions_per_lun
> * @subregions_per_lun - lun size is not guaranteed to be region aligned
> @@ -105,6 +150,7 @@ struct ufshpb_dh_lun {
> struct list_head lh_map_ctx;
> spinlock_t map_list_lock;
> struct ufshpb_region *region_tbl;
> + unsigned long *pinned_map;
> struct scsi_device *sdev;
>
> unsigned int regions_per_lun;
> @@ -113,6 +159,10 @@ struct ufshpb_dh_lun {
> unsigned int max_active_regions;
>
> atomic_t active_regions;
> +
> + struct mutex eviction_lock;
> +
> + struct workqueue_struct *wq;
> };
Please document what the eviction_lock protects.
> +static inline void ufshpb_set_write_buf_cmd(unsigned char *cmd,
> + unsigned int region)
> +{
> + cmd[0] = UFSHPB_WRITE_BUFFER;
> + cmd[1] = 0x01;
> + put_unaligned_be16(region, &cmd[2]);
> +}
Please follow the example of the sd driver and use the verb "setup"
instead of "set" for functions that initialize a SCSI CDB.
> +static int ufshpb_submit_write_buf_cmd(struct scsi_device *sdev,
> + unsigned int region)
> +{
> + unsigned char cmd[10] = {};
> + struct scsi_sense_hdr sshdr = {};
> + u64 flags = REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT |
> + REQ_FAILFAST_DRIVER;
> + int timeout = WRITE_BUFFER_TIMEOUT;
> + int cmd_retries = WRITE_BUFFER_RETRIES;
> + int ret = 0;
> +
> + ufshpb_set_write_buf_cmd(cmd, region);
> +
> + ret = scsi_execute(sdev, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
> + timeout, cmd_retries, flags, 0, NULL);
> +
> + /* HPB spec does not define any error handling */
> + sdev_printk(KERN_INFO, sdev, "%s: WRITE_BUFFER %s result %d\n",
> + UFSHPB_NAME, ret ? "failed" : "succeeded", ret);
> +
> + return ret;
> +}
I don't think that unconditionally printing the result of the WRITE
BUFFER command is acceptable. How about only reporting failures?
> +static void ufshpb_set_read_buf_cmd(unsigned char *cmd, unsigned int region,
> + unsigned int subregion,
> + unsigned int alloc_len)
> +{
> + cmd[0] = UFSHPB_READ_BUFFER;
> + cmd[1] = 0x01;
> + put_unaligned_be16(region, &cmd[2]);
> + put_unaligned_be16(subregion, &cmd[4]);
> +
> + cmd[6] = alloc_len >> 16;
> + cmd[7] = (alloc_len >> 8) & 0xff;
> + cmd[8] = alloc_len & 0xff;
> + cmd[9] = 0x00;
> +}
Please use put_unaligned_be24() instead of open-coding it.
> +static int ufshpb_subregion_alloc_pages(struct ufshpb_dh_lun *hpb,
> + struct ufshpb_subregion *s)
> +{
> + struct ufshpb_map_ctx *mctx;
> +
> + spin_lock(&hpb->map_list_lock);
> + mctx = list_first_entry_or_null(&hpb->lh_map_ctx,
> + struct ufshpb_map_ctx, list);
> + if (!mctx) {
> + spin_unlock(&hpb->map_list_lock);
> + return -EINVAL;
> + }
> +
> + list_del_init(&mctx->list);
> + spin_unlock(&hpb->map_list_lock);
> +
> + s->mctx = mctx;
> + mctx->pages = (char *)__get_free_pages(GFP_KERNEL, order);
> + if (!mctx->pages)
> + return -ENOMEM;
> +
> + return 0;
> +}
Relying on higher order pages is not acceptable because memory gets
fragmented easily. See also
https://elinux.org/images/a/a8/Controlling_Linux_Memory_Fragmentation_and_Higher_Order_Allocation_Failure-_Analysis%2C_Observations_and_Results.pdf.
> + hpb->pinned_map = kcalloc(BITS_TO_LONGS(hpb->regions_per_lun),
> + sizeof(unsigned long), GFP_KERNEL);
Is this perhaps an open-coded version of bitmap_alloc()? If so, please
use bitmap_alloc() instead.
> + snprintf(wq_name, ARRAY_SIZE(wq_name), "ufshpb_wq_%d", sdev->id);
> + wq = alloc_workqueue(wq_name, WQ_HIGHPRI, WQ_MAX_ACTIVE);
> + if (!wq) {
> + ret = -ENOMEM;
> + goto out_free;
> + }
What is the purpose of the ufshpb_wq_%d workqueues? Why to allocate
dedicated workqueues instead of using one of the existing system
workqueues? If the scsi_execute() calls would be changed into
asynchronous SCSI command submission, would these workqueues still be
necessary?
Thanks,
Bart.
next prev parent reply other threads:[~2020-05-16 2:44 UTC|newest]
Thread overview: 39+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-05-15 10:30 [RFC PATCH 00/13] scsi: ufs: Add HPB Support Avri Altman
2020-05-15 10:30 ` [RFC PATCH 01/13] scsi: ufs: Add HPB parameters Avri Altman
2020-05-15 10:30 ` [RFC PATCH 02/13] scsi: ufshpb: Init part I - Read HPB config Avri Altman
2020-05-15 15:33 ` Randy Dunlap
2020-05-16 1:46 ` Bart Van Assche
2020-05-16 1:57 ` Bart Van Assche
2020-05-15 10:30 ` [RFC PATCH 03/13] scsi: scsi_dh: Introduce scsi_dh_ufshpb Avri Altman
2020-05-16 1:48 ` Bart Van Assche
2020-05-15 10:30 ` [RFC PATCH 04/13] scsi: ufs: ufshpb: Init part II - Attach scsi device Avri Altman
2020-05-16 1:52 ` Bart Van Assche
2020-05-15 10:30 ` [RFC PATCH 05/13] scsi: ufs: ufshpb: Disable HPB if no HPB-enabled luns Avri Altman
2020-05-16 2:02 ` Bart Van Assche
2020-05-15 10:30 ` [RFC PATCH 06/13] scsi: scsi_dh: ufshpb: Prepare for L2P cache management Avri Altman
2020-05-16 2:13 ` Bart Van Assche
2020-05-15 10:30 ` [RFC PATCH 07/13] scsi: scsi_dh: ufshpb: Add ufshpb state machine Avri Altman
2020-05-16 2:44 ` Bart Van Assche [this message]
2020-05-15 10:30 ` [RFC PATCH 08/13] scsi: dh: ufshpb: Activate pinned regions Avri Altman
2020-05-15 10:30 ` [RFC PATCH 09/13] scsi: ufshpb: Add response API Avri Altman
2020-05-16 3:06 ` Bart Van Assche
2020-05-15 10:30 ` [RFC PATCH 10/13] scsi: dh: ufshpb: Add ufshpb_set_params Avri Altman
2020-05-15 10:30 ` [RFC PATCH 11/13] scsi: Allow device handler set their own CDB Avri Altman
2020-05-16 3:19 ` Bart Van Assche
2020-05-15 10:30 ` [RFC PATCH 12/13] scsi: dh: ufshpb: Add prep_fn handler Avri Altman
2020-05-16 3:40 ` Bart Van Assche
2020-05-15 10:30 ` [RFC PATCH 13/13] scsi: scsi_dh: ufshpb: Add "Cold" subregions timer Avri Altman
2020-05-16 3:50 ` [RFC PATCH 00/13] scsi: ufs: Add HPB Support Bart Van Assche
2020-05-16 9:14 ` Avri Altman
2020-05-16 17:14 ` Bart Van Assche
[not found] ` <CGME20200516171420epcas2p108c570904c5117c3654d71e0a2842faa@epcms2p7>
2020-05-19 22:31 ` Another approach of UFSHPB yongmyung lee
2020-05-20 17:55 ` Christoph Hellwig
2020-05-20 21:19 ` Bart Van Assche
2020-05-22 16:35 ` Bart Van Assche
2020-05-22 16:49 ` Bart Van Assche
[not found] ` <CGME20200516171420epcas2p108c570904c5117c3654d71e0a2842faa@epcms2p4>
2020-05-25 5:40 ` Daejun Park
2020-05-25 14:56 ` Bart Van Assche
2020-05-26 6:15 ` Avri Altman
2020-05-26 17:03 ` Bart Van Assche
[not found] ` <CGME20200516171420epcas2p108c570904c5117c3654d71e0a2842faa@epcms2p3>
2020-05-27 9:11 ` Daejun Park
2020-05-27 11:46 ` Bean Huo
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=5fd7224f-3bbd-863f-ffd4-b049aeeb5456@acm.org \
--to=bvanassche@acm.org \
--cc=alim.akhtar@samsung.com \
--cc=asutoshd@codeaurora.org \
--cc=avi.shchislowski@wdc.com \
--cc=avri.altman@wdc.com \
--cc=beanhuo@micron.com \
--cc=cang@codeaurora.org \
--cc=j-young.choi@samsung.com \
--cc=jejb@linux.vnet.ibm.com \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-scsi@vger.kernel.org \
--cc=martin.petersen@oracle.com \
--cc=md.rafiq@samsung.com \
--cc=sangyoon.oh@samsung.com \
--cc=stanley.chu@mediatek.com \
--cc=ymhungry.lee@samsung.com \
--cc=zangleigang@hisilicon.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).