[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-Id: <20181105122610.1555-5-hans.ml.holmberg@owltronix.com>
Date: Mon, 5 Nov 2018 13:26:07 +0100
From: Hans Holmberg <hans.ml.holmberg@...tronix.com>
To: Matias Bjorling <mb@...htnvm.io>
Cc: Javier Gonzales <javier@...xlabs.com>, linux-block@...r.kernel.org,
linux-kernel@...r.kernel.org,
Hans Holmberg <hans.holmberg@...xlabs.com>
Subject: [PATCH v2 4/7] lightnvm: pblk: set conservative threshold for user writes
From: Hans Holmberg <hans.holmberg@...xlabs.com>
In a worst-case scenario (random writes), OP% of sectors
in each line will be invalid, and we will then need
to move data out of 100/OP% lines to free a single line.
So, to prevent the possibility of running out of lines,
temporarily block user writes when there is less than
100/OP% free lines.
Also ensure that pblk creation does not produce instances
with insufficient over provisioning.
Insufficient over-provising is not a problem on real hardware,
but often an issue when running QEMU simulations (with few lines).
100 lines is enough to create a sane instance with the standard
(11%) over provisioning.
Signed-off-by: Hans Holmberg <hans.holmberg@...xlabs.com>
---
drivers/lightnvm/pblk-init.c | 43 ++++++++++++++++++++++++------------
drivers/lightnvm/pblk-rl.c | 5 ++---
drivers/lightnvm/pblk.h | 12 +++++++++-
3 files changed, 42 insertions(+), 18 deletions(-)
diff --git a/drivers/lightnvm/pblk-init.c b/drivers/lightnvm/pblk-init.c
index 13822594647c..8b89bb26b0f1 100644
--- a/drivers/lightnvm/pblk-init.c
+++ b/drivers/lightnvm/pblk-init.c
@@ -635,13 +635,13 @@ static unsigned int calc_emeta_len(struct pblk *pblk)
return (lm->emeta_len[1] + lm->emeta_len[2] + lm->emeta_len[3]);
}
-static void pblk_set_provision(struct pblk *pblk, long nr_free_blks)
+static int pblk_set_provision(struct pblk *pblk, long nr_free_chks)
{
struct nvm_tgt_dev *dev = pblk->dev;
struct pblk_line_mgmt *l_mg = &pblk->l_mg;
struct pblk_line_meta *lm = &pblk->lm;
struct nvm_geo *geo = &dev->geo;
- sector_t provisioned;
+ sector_t provisioned, minimum;
int sec_meta, blk_meta;
if (geo->op == NVM_TARGET_DEFAULT_OP)
@@ -649,17 +649,34 @@ static void pblk_set_provision(struct pblk *pblk, long nr_free_blks)
else
pblk->op = geo->op;
- provisioned = nr_free_blks;
+ minimum = pblk_get_min_chks(pblk);
+ provisioned = nr_free_chks;
provisioned *= (100 - pblk->op);
sector_div(provisioned, 100);
- pblk->op_blks = nr_free_blks - provisioned;
+ if ((nr_free_chks - provisioned) < minimum) {
+ if (geo->op != NVM_TARGET_DEFAULT_OP) {
+ pblk_err(pblk, "OP too small to create a sane instance\n");
+ return -EINTR;
+ }
+
+ /* If the user did not specify an OP value, and PBLK_DEFAULT_OP
+ * is not enough, calculate and set sane value
+ */
+
+ provisioned = nr_free_chks - minimum;
+ pblk->op = (100 * minimum) / nr_free_chks;
+ pblk_info(pblk, "Default OP insufficient, adjusting OP to %d\n",
+ pblk->op);
+ }
+
+ pblk->op_blks = nr_free_chks - provisioned;
/* Internally pblk manages all free blocks, but all calculations based
* on user capacity consider only provisioned blocks
*/
- pblk->rl.total_blocks = nr_free_blks;
- pblk->rl.nr_secs = nr_free_blks * geo->clba;
+ pblk->rl.total_blocks = nr_free_chks;
+ pblk->rl.nr_secs = nr_free_chks * geo->clba;
/* Consider sectors used for metadata */
sec_meta = (lm->smeta_sec + lm->emeta_sec[0]) * l_mg->nr_free_lines;
@@ -667,8 +684,10 @@ static void pblk_set_provision(struct pblk *pblk, long nr_free_blks)
pblk->capacity = (provisioned - blk_meta) * geo->clba;
- atomic_set(&pblk->rl.free_blocks, nr_free_blks);
- atomic_set(&pblk->rl.free_user_blocks, nr_free_blks);
+ atomic_set(&pblk->rl.free_blocks, nr_free_chks);
+ atomic_set(&pblk->rl.free_user_blocks, nr_free_chks);
+
+ return 0;
}
static int pblk_setup_line_meta_chk(struct pblk *pblk, struct pblk_line *line,
@@ -1025,13 +1044,9 @@ static int pblk_lines_init(struct pblk *pblk)
line->state);
}
- if (!nr_free_chks) {
- pblk_err(pblk, "too many bad blocks prevent for sane instance\n");
- ret = -EINTR;
+ ret = pblk_set_provision(pblk, nr_free_chks);
+ if (ret)
goto fail_free_lines;
- }
-
- pblk_set_provision(pblk, nr_free_chks);
vfree(chunk_meta);
return 0;
diff --git a/drivers/lightnvm/pblk-rl.c b/drivers/lightnvm/pblk-rl.c
index db55a1c89997..76116d5f78e4 100644
--- a/drivers/lightnvm/pblk-rl.c
+++ b/drivers/lightnvm/pblk-rl.c
@@ -214,11 +214,10 @@ void pblk_rl_init(struct pblk_rl *rl, int budget)
struct nvm_geo *geo = &dev->geo;
struct pblk_line_mgmt *l_mg = &pblk->l_mg;
struct pblk_line_meta *lm = &pblk->lm;
- int min_blocks = lm->blk_per_line * PBLK_GC_RSV_LINE;
int sec_meta, blk_meta;
-
unsigned int rb_windows;
+
/* Consider sectors used for metadata */
sec_meta = (lm->smeta_sec + lm->emeta_sec[0]) * l_mg->nr_free_lines;
blk_meta = DIV_ROUND_UP(sec_meta, geo->clba);
@@ -226,7 +225,7 @@ void pblk_rl_init(struct pblk_rl *rl, int budget)
rl->high = pblk->op_blks - blk_meta - lm->blk_per_line;
rl->high_pw = get_count_order(rl->high);
- rl->rsv_blocks = min_blocks;
+ rl->rsv_blocks = pblk_get_min_chks(pblk);
/* This will always be a power-of-2 */
rb_windows = budget / NVM_MAX_VLBA;
diff --git a/drivers/lightnvm/pblk.h b/drivers/lightnvm/pblk.h
index f415aae600c8..e5b88a25d4d6 100644
--- a/drivers/lightnvm/pblk.h
+++ b/drivers/lightnvm/pblk.h
@@ -905,7 +905,6 @@ int pblk_recov_check_emeta(struct pblk *pblk, struct line_emeta *emeta);
#define PBLK_GC_MAX_READERS 8 /* Max number of outstanding GC reader jobs */
#define PBLK_GC_RQ_QD 128 /* Queue depth for inflight GC requests */
#define PBLK_GC_L_QD 4 /* Queue depth for inflight GC lines */
-#define PBLK_GC_RSV_LINE 1 /* Reserved lines for GC */
int pblk_gc_init(struct pblk *pblk);
void pblk_gc_exit(struct pblk *pblk, bool graceful);
@@ -1370,4 +1369,15 @@ static inline char *pblk_disk_name(struct pblk *pblk)
return disk->disk_name;
}
+
+static inline unsigned int pblk_get_min_chks(struct pblk *pblk)
+{
+ struct pblk_line_meta *lm = &pblk->lm;
+ /* In a worst-case scenario every line will have OP invalid sectors.
+ * We will then need a minimum of 1/OP lines to free up a single line
+ */
+
+ return DIV_ROUND_UP(100, pblk->op) * lm->blk_per_line;
+
+}
#endif /* PBLK_H_ */
--
2.17.1
Powered by blists - more mailing lists