Module Name: src
Committed By: jdolecek
Date: Wed Jul 19 19:39:28 UTC 2017
Modified Files:
src/sys/dev/ata [jdolecek-ncq]: ata.c atareg.h atavar.h satafis_subr.c
satafisvar.h wd.c wdvar.h
Log Message:
tighen and expand error handling, mostly for NCQ use cases:
- make retry timeout callout per xfer, i.e. retry separately
- zero whole bio struct on retry to avoid more stale state
- add a REQUEUE option, which doesn't bump retry count
- add ata_read_log_ext_ncq() for NCQ recovery
- adjust logic for activating xfers - allow next command only when
it's for same drive, several concurrent are only supported when HBA
and driver support FIS-based switching
- add new ata_timeout() which handles race between callout_stop()
and the invokation, add appropriate handling on deactivate/free paths
- stop using ch_status/ch_error in non-wdc code; later it will be dropped
completely
To generate a diff of this commit:
cvs rdiff -u -r1.132.8.18 -r1.132.8.19 src/sys/dev/ata/ata.c
cvs rdiff -u -r1.43.18.2 -r1.43.18.3 src/sys/dev/ata/atareg.h
cvs rdiff -u -r1.92.8.16 -r1.92.8.17 src/sys/dev/ata/atavar.h
cvs rdiff -u -r1.7.28.2 -r1.7.28.3 src/sys/dev/ata/satafis_subr.c
cvs rdiff -u -r1.3 -r1.3.50.1 src/sys/dev/ata/satafisvar.h
cvs rdiff -u -r1.428.2.25 -r1.428.2.26 src/sys/dev/ata/wd.c
cvs rdiff -u -r1.43.4.6 -r1.43.4.7 src/sys/dev/ata/wdvar.h
Please note that diffs are not public domain; they are subject to the
copyright notices on the relevant files.
Modified files:
Index: src/sys/dev/ata/ata.c
diff -u src/sys/dev/ata/ata.c:1.132.8.18 src/sys/dev/ata/ata.c:1.132.8.19
--- src/sys/dev/ata/ata.c:1.132.8.18 Tue Jun 27 18:36:03 2017
+++ src/sys/dev/ata/ata.c Wed Jul 19 19:39:28 2017
@@ -1,4 +1,4 @@
-/* $NetBSD: ata.c,v 1.132.8.18 2017/06/27 18:36:03 jdolecek Exp $ */
+/* $NetBSD: ata.c,v 1.132.8.19 2017/07/19 19:39:28 jdolecek Exp $ */
/*
* Copyright (c) 1998, 2001 Manuel Bouyer. All rights reserved.
@@ -25,7 +25,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: ata.c,v 1.132.8.18 2017/06/27 18:36:03 jdolecek Exp $");
+__KERNEL_RCSID(0, "$NetBSD: ata.c,v 1.132.8.19 2017/07/19 19:39:28 jdolecek Exp $");
#include "opt_ata.h"
@@ -241,14 +241,34 @@ ata_queue_get_active_xfer(struct ata_cha
return xfer;
}
+struct ata_xfer *
+ata_queue_drive_active_xfer(struct ata_channel *chp, int drive)
+{
+ struct ata_xfer *xfer = NULL;
+
+ mutex_enter(&chp->ch_lock);
+
+ TAILQ_FOREACH(xfer, &chp->ch_queue->active_xfers, c_activechain) {
+ if (xfer->c_drive == drive)
+ break;
+ }
+ KASSERT(xfer != NULL);
+
+ mutex_exit(&chp->ch_lock);
+
+ return xfer;
+}
+
static void
-ata_xfer_init(struct ata_xfer *xfer, bool zero)
+ata_xfer_init(struct ata_xfer *xfer, uint8_t slot)
{
- if (zero)
- memset(xfer, 0, sizeof(*xfer));
+ memset(xfer, 0, sizeof(*xfer));
+
+ xfer->c_slot = slot;
cv_init(&xfer->c_active, "ataact");
callout_init(&xfer->c_timo_callout, 0); /* XXX MPSAFE */
+ callout_init(&xfer->c_retry_callout, 0); /* XXX MPSAFE */
}
static void
@@ -256,6 +276,8 @@ ata_xfer_destroy(struct ata_xfer *xfer)
{
callout_halt(&xfer->c_timo_callout, NULL); /* XXX MPSAFE */
callout_destroy(&xfer->c_timo_callout);
+ callout_halt(&xfer->c_retry_callout, NULL); /* XXX MPSAFE */
+ callout_destroy(&xfer->c_retry_callout);
cv_destroy(&xfer->c_active);
}
@@ -278,7 +300,7 @@ ata_queue_alloc(uint8_t openings)
cv_init(&chq->queue_drain, "atdrn");
for (uint8_t i = 0; i < openings; i++)
- ata_xfer_init(&chq->queue_xfers[i], false);
+ ata_xfer_init(&chq->queue_xfers[i], i);
return chq;
}
@@ -1009,6 +1031,88 @@ out:
return rv;
}
+int
+ata_read_log_ext_ncq(struct ata_drive_datas *drvp, uint8_t flags,
+ uint8_t *slot, uint8_t *status, uint8_t *err)
+{
+ struct ata_xfer *xfer;
+ int rv;
+ struct ata_channel *chp = drvp->chnl_softc;
+ struct atac_softc *atac = chp->ch_atac;
+ uint8_t *tb;
+
+ ATADEBUG_PRINT(("%s\n", __func__), DEBUG_FUNCS);
+
+ /* Only NCQ ATA drives support/need this */
+ if (drvp->drive_type != ATA_DRIVET_ATA ||
+ (drvp->drive_flags & ATA_DRIVE_NCQ) == 0)
+ return EOPNOTSUPP;
+
+ xfer = ata_get_xfer_ext(chp, false, 0);
+ if (xfer == NULL) {
+ ATADEBUG_PRINT(("%s: no xfer\n", __func__),
+ DEBUG_FUNCS|DEBUG_XFERS);
+ return EAGAIN;
+ }
+
+ tb = malloc(DEV_BSIZE, M_DEVBUF, M_NOWAIT);
+ if (tb == NULL) {
+ ATADEBUG_PRINT(("%s: memory allocation failed\n", __func__),
+ DEBUG_FUNCS|DEBUG_XFERS);
+ rv = EAGAIN;
+ goto out;
+ }
+ memset(tb, 0, DEV_BSIZE);
+
+ /*
+ * We could use READ LOG DMA EXT if drive supports it (i.e.
+ * when it supports Streaming feature) to avoid PIO command,
+ * and to make this a little faster. Realistically, it
+ * should not matter.
+ */
+ xfer->c_flags |= C_IMMEDIATE;
+ xfer->c_ata_c.r_command = WDCC_READ_LOG_EXT;
+ xfer->c_ata_c.r_lba = WDCC_LOG_PAGE_NCQ;
+ xfer->c_ata_c.r_st_bmask = WDCS_DRDY;
+ xfer->c_ata_c.r_st_pmask = WDCS_DRDY;
+ xfer->c_ata_c.r_count = 1;
+ xfer->c_ata_c.r_device = WDSD_LBA;
+ xfer->c_ata_c.flags = AT_READ | AT_LBA | flags;
+ xfer->c_ata_c.timeout = 1000; /* 1s */
+ xfer->c_ata_c.data = tb;
+ xfer->c_ata_c.bcount = DEV_BSIZE;
+
+ if ((*atac->atac_bustype_ata->ata_exec_command)(drvp,
+ xfer) != ATACMD_COMPLETE) {
+ rv = EAGAIN;
+ goto out2;
+ }
+ if (xfer->c_ata_c.flags & (AT_ERROR | AT_TIMEOU | AT_DF)) {
+ rv = EINVAL;
+ goto out2;
+ }
+
+ /* XXX verify checksum and refuse if not correct (QEMU) */
+
+ if (tb[0] & WDCC_LOG_NQ) {
+ /* not a NCQ command */
+ rv = EOPNOTSUPP;
+ goto out2;
+ }
+
+ *slot = tb[0] & 0x1f;
+ *status = tb[2];
+ *err = tb[3];
+
+ rv = 0;
+
+out2:
+ free(tb, DEV_BSIZE);
+out:
+ ata_free_xfer(chp, xfer);
+ return rv;
+}
+
#if NATA_DMA
void
ata_dmaerr(struct ata_drive_datas *drvp, int flags)
@@ -1067,8 +1171,13 @@ ata_exec_xfer(struct ata_channel *chp, s
mutex_enter(&chp->ch_lock);
- /* insert at the end of command list */
- TAILQ_INSERT_TAIL(&chp->ch_queue->queue_xfer, xfer, c_xferchain);
+ /* insert at the end of command list unless specially requested */
+ if (xfer->c_flags & C_IMMEDIATE)
+ TAILQ_INSERT_HEAD(&chp->ch_queue->queue_xfer, xfer,
+ c_xferchain);
+ else
+ TAILQ_INSERT_TAIL(&chp->ch_queue->queue_xfer, xfer,
+ c_xferchain);
ATADEBUG_PRINT(("atastart from ata_exec_xfer, flags 0x%x\n",
chp->ch_flags), DEBUG_XFERS);
@@ -1086,7 +1195,7 @@ ata_exec_xfer(struct ata_channel *chp, s
* Free xfer now if it there was attempt to free it
* while we were waiting.
*/
- if (xfer->c_flags & C_FREE) {
+ if ((xfer->c_flags & (C_FREE|C_WAITTIMO)) == C_FREE) {
ata_free_xfer(chp, xfer);
return;
}
@@ -1111,6 +1220,7 @@ atastart(struct ata_channel *chp)
struct atac_softc *atac = chp->ch_atac;
struct ata_queue *chq = chp->ch_queue;
struct ata_xfer *xfer, *axfer;
+ bool immediate;
#ifdef ATA_DEBUG
int spl1, spl2;
@@ -1127,12 +1237,19 @@ atastart(struct ata_channel *chp)
mutex_enter(&chp->ch_lock);
+ KASSERT(chq->queue_active <= chq->queue_openings);
if (chq->queue_active == chq->queue_openings) {
goto out; /* channel completely busy */
}
+ /* is there a xfer ? */
+ if ((xfer = TAILQ_FIRST(&chp->ch_queue->queue_xfer)) == NULL)
+ goto out;
+
+ immediate = ISSET(xfer->c_flags, C_IMMEDIATE);
+
/* is the queue frozen? */
- if (__predict_false(chq->queue_freeze > 0)) {
+ if (__predict_false(!immediate && chq->queue_freeze > 0)) {
if (chq->queue_flags & QF_IDLE_WAIT) {
chq->queue_flags &= ~QF_IDLE_WAIT;
wakeup(&chq->queue_flags);
@@ -1140,21 +1257,23 @@ atastart(struct ata_channel *chp)
goto out; /* queue frozen */
}
- /* is there a xfer ? */
- if ((xfer = TAILQ_FIRST(&chp->ch_queue->queue_xfer)) == NULL)
- goto out;
-
/* all xfers on same queue must belong to the same channel */
KASSERT(xfer->c_chp == chp);
/*
- * Can only take NCQ command if there are no current active
- * commands, or if the active commands are NCQ. Need only check
- * first xfer.
- */
- axfer = TAILQ_FIRST(&chp->ch_queue->active_xfers);
- if (axfer && (axfer->c_flags & C_NCQ) == 0)
- goto out;
+ * Can only take the command if there are no current active
+ * commands, or if the command is NCQ and the active commands are also
+ * NCQ. If PM is in use and HBA driver doesn't support/use FIS-based
+ * switching, can only send commands to single drive.
+ * Need only check first xfer.
+ * XXX FIS-based switching - revisit
+ */
+ if (!immediate && (axfer = TAILQ_FIRST(&chp->ch_queue->active_xfers))) {
+ if (!ISSET(xfer->c_flags, C_NCQ) ||
+ !ISSET(axfer->c_flags, C_NCQ) ||
+ xfer->c_drive != axfer->c_drive)
+ goto out;
+ }
struct ata_drive_datas * const drvp = &chp->ch_drive[xfer->c_drive];
@@ -1170,11 +1289,6 @@ atastart(struct ata_channel *chp)
goto out;
}
-#ifdef DIAGNOSTIC
- if ((chp->ch_flags & ATACH_IRQ_WAIT) != 0
- && chp->ch_queue->queue_openings == 1)
- panic("atastart: channel waiting for irq");
-#endif
ATADEBUG_PRINT(("atastart: xfer %p channel %d drive %d\n", xfer,
chp->ch_channel, xfer->c_drive), DEBUG_XFERS);
if (drvp->drive_flags & ATA_DRIVE_RESET) {
@@ -1242,7 +1356,6 @@ retry:
/* zero everything after the callout member */
memset(&xfer->c_startzero, 0,
sizeof(struct ata_xfer) - offsetof(struct ata_xfer, c_startzero));
- xfer->c_slot = slot;
out:
mutex_exit(&chp->ch_lock);
@@ -1259,7 +1372,7 @@ ata_free_xfer(struct ata_channel *chp, s
mutex_enter(&chp->ch_lock);
- if (xfer->c_flags & C_WAITACT) {
+ if (xfer->c_flags & (C_WAITACT|C_WAITTIMO)) {
/* Someone is waiting for this xfer, so we can't free now */
xfer->c_flags |= C_FREE;
cv_signal(&xfer->c_active);
@@ -1318,6 +1431,9 @@ ata_deactivate_xfer(struct ata_channel *
callout_stop(&xfer->c_timo_callout);
+ if (callout_invoking(&xfer->c_timo_callout))
+ xfer->c_flags |= C_WAITTIMO;
+
TAILQ_REMOVE(&chq->active_xfers, xfer, c_activechain);
chq->active_xfers_used &= ~__BIT(xfer->c_slot);
chq->queue_active--;
@@ -1359,6 +1475,76 @@ ata_waitdrain_xfer_check(struct ata_chan
}
/*
+ * Check for race of normal transfer handling vs. timeout.
+ */
+static bool
+ata_timo_xfer_check(struct ata_xfer *xfer)
+{
+ struct ata_channel *chp = xfer->c_chp;
+ struct ata_drive_datas *drvp = &chp->ch_drive[xfer->c_drive];
+
+ mutex_enter(&chp->ch_lock);
+
+ callout_ack(&xfer->c_timo_callout);
+
+ if (xfer->c_flags & C_WAITTIMO) {
+ xfer->c_flags &= ~C_WAITTIMO;
+
+ /* Handle race vs. ata_free_xfer() */
+ if (xfer->c_flags & C_FREE) {
+ xfer->c_flags &= ~C_FREE;
+ mutex_exit(&chp->ch_lock);
+
+ aprint_normal_dev(drvp->drv_softc,
+ "xfer %d freed while invoking timeout\n",
+ xfer->c_slot);
+
+ ata_free_xfer(chp, xfer);
+ return true;
+ }
+
+ /* Handle race vs. callout_stop() in ata_deactivate_xfer() */
+ if (!callout_expired(&xfer->c_timo_callout)) {
+ mutex_exit(&chp->ch_lock);
+
+ aprint_normal_dev(drvp->drv_softc,
+ "xfer %d deactivated while invoking timeout\n",
+ xfer->c_slot);
+ return true;
+ }
+ }
+
+ mutex_exit(&chp->ch_lock);
+
+ /* No race, proceed with timeout handling */
+ return false;
+}
+
+void
+ata_timeout(void *v)
+{
+ struct ata_xfer *xfer = v;
+ int s;
+
+ ATADEBUG_PRINT(("%s: slot %d\n", __func__, xfer->c_slot),
+ DEBUG_FUNCS|DEBUG_XFERS);
+
+ s = splbio(); /* XXX MPSAFE */
+
+ if (ata_timo_xfer_check(xfer)) {
+ /* Already logged */
+ goto out;
+ }
+
+ /* Mark as timed out. Do not print anything, wd(4) will. */
+ xfer->c_flags |= C_TIMEOU;
+ xfer->c_intr(xfer->c_chp, xfer, 0);
+
+out:
+ splx(s);
+}
+
+/*
* Kill off all active xfers for a ata_channel.
*
* Must be called at splbio().
@@ -1497,12 +1683,12 @@ ata_reset_channel(struct ata_channel *ch
chp->ch_drive[drive].state = 0;
chp->ch_flags &= ~ATACH_TH_RESET;
- if ((flags & AT_RST_EMERG) == 0) {
- ata_channel_thaw(chp);
- atastart(chp);
- } else {
+ if (flags & AT_RST_EMERG) {
/* make sure that we can use polled commands */
ata_queue_reset(chp->ch_queue);
+ } else {
+ ata_channel_thaw(chp);
+ atastart(chp);
}
}
Index: src/sys/dev/ata/atareg.h
diff -u src/sys/dev/ata/atareg.h:1.43.18.2 src/sys/dev/ata/atareg.h:1.43.18.3
--- src/sys/dev/ata/atareg.h:1.43.18.2 Mon Apr 24 22:20:23 2017
+++ src/sys/dev/ata/atareg.h Wed Jul 19 19:39:28 2017
@@ -1,4 +1,4 @@
-/* $NetBSD: atareg.h,v 1.43.18.2 2017/04/24 22:20:23 jdolecek Exp $ */
+/* $NetBSD: atareg.h,v 1.43.18.3 2017/07/19 19:39:28 jdolecek Exp $ */
/*
* Copyright (c) 1998, 2001 Manuel Bouyer.
@@ -94,10 +94,17 @@
#define WDCC_RECAL 0x10 /* disk restore code -- resets cntlr */
#define WDCC_READ 0x20 /* disk read code */
+
+#define WDCC_READ_LOG_EXT 0x2f
+#define WDCC_LOG_PAGE_NCQ 0x10
+#define WDCC_LOG_NQ __BIT(7)
+
#define WDCC_WRITE 0x30 /* disk write code */
#define WDCC__LONG 0x02 /* modifier -- access ecc bytes */
#define WDCC__NORETRY 0x01 /* modifier -- no retrys */
+#define WDCC_READ_LOG_DMA_EXT 0x47 /* DMA variant of READ_LOG_EXT */
+
#define WDCC_FORMAT 0x50 /* disk format code */
#define WDCC_DIAGNOSE 0x90 /* controller diagnostic */
#define WDCC_IDP 0x91 /* initialize drive parameters */
Index: src/sys/dev/ata/atavar.h
diff -u src/sys/dev/ata/atavar.h:1.92.8.16 src/sys/dev/ata/atavar.h:1.92.8.17
--- src/sys/dev/ata/atavar.h:1.92.8.16 Tue Jun 27 18:36:03 2017
+++ src/sys/dev/ata/atavar.h Wed Jul 19 19:39:28 2017
@@ -1,4 +1,4 @@
-/* $NetBSD: atavar.h,v 1.92.8.16 2017/06/27 18:36:03 jdolecek Exp $ */
+/* $NetBSD: atavar.h,v 1.92.8.17 2017/07/19 19:39:28 jdolecek Exp $ */
/*
* Copyright (c) 1998, 2001 Manuel Bouyer.
@@ -65,8 +65,8 @@ struct ata_bio {
#define TIMEOUT 4 /* device timed out */
#define ERR_NODEV 5 /* device has been gone */
#define ERR_RESET 6 /* command was terminated by channel reset */
+#define REQUEUE 7 /* different xfer failed, requeue command */
uint8_t r_error;/* copy of error register */
- int retries;/* number of xfer retry */
struct buf *bp;
};
@@ -132,13 +132,15 @@ struct scsipi_xfer;
*/
struct ata_xfer {
struct callout c_timo_callout; /* timeout callout handle */
+ struct callout c_retry_callout; /* retry callout handle */
kcondvar_t c_active; /* somebody actively waiting for xfer */
+ int8_t c_slot; /* queue slot # */
#define c_startzero c_chp
/* Channel and drive that are to process the request. */
struct ata_channel *c_chp;
- uint16_t c_drive;
- int8_t c_slot; /* queue slot # */
+ uint16_t c_drive;
+ uint16_t c_retries; /* number of xfer retry */
volatile u_int c_flags; /* command state flags */
void *c_databuf; /* pointer to data buffer */
@@ -146,6 +148,12 @@ struct ata_xfer {
int c_skip; /* bytes already transferred */
int c_dscpoll; /* counter for dsc polling (ATAPI) */
int c_lenoff; /* offset to c_bcount (ATAPI) */
+#if 0 /* for now */
+ int c_ata_status; /* copy of ATA error + status */
+#endif
+#define ATACH_ERR_ST(error, status) ((error) << 8 | (status))
+#define ATACH_ERR(val) (((val) >> 8) & 0xff)
+#define ATACH_ST(val) (((val) >> 0) & 0xff)
union {
struct ata_bio c_bio; /* ATA transfer */
@@ -159,7 +167,6 @@ struct ata_xfer {
/* Link on the command queue. */
TAILQ_ENTRY(ata_xfer) c_xferchain;
TAILQ_ENTRY(ata_xfer) c_activechain;
- STAILQ_ENTRY(ata_xfer) c_restartchain;
/* Low-level protocol handlers. */
void (*c_start)(struct ata_channel *, struct ata_xfer *);
@@ -177,11 +184,14 @@ struct ata_xfer {
#define C_FREE 0x0040 /* call ata_free_xfer() asap */
#define C_PIOBM 0x0080 /* command uses busmastering PIO */
#define C_NCQ 0x0100 /* command is queued */
+#define C_IMMEDIATE 0x0200 /* execute command without queuing */
+#define C_WAITTIMO 0x0400 /* race vs. timeout */
/* reasons for c_kill_xfer() */
#define KILL_GONE 1 /* device is gone while xfer was active */
#define KILL_RESET 2 /* xfer was reset */
#define KILL_GONE_INACTIVE 3 /* device is gone while xfer was pending */
+#define KILL_REQUEUE 4 /* xfer must be reissued to device, no err */
/*
* While hw supports up to 32 tags, in practice we must never
@@ -383,8 +393,10 @@ struct ata_channel {
#define ATACH_TH_RUN 0x100 /* the kernel thread is working */
#define ATACH_TH_RESET 0x200 /* someone ask the thread to reset */
#define ATACH_TH_RESCAN 0x400 /* rescan requested */
+#if 1 /* for now */
uint8_t ch_status; /* copy of status register */
uint8_t ch_error; /* copy of error register */
+#endif
/* for the reset callback */
int ch_reset_flags;
@@ -483,6 +495,9 @@ void atabus_free_drives(struct ata_chann
struct ataparams;
int ata_get_params(struct ata_drive_datas *, uint8_t, struct ataparams *);
int ata_set_mode(struct ata_drive_datas *, uint8_t, uint8_t);
+int ata_read_log_ext_ncq(struct ata_drive_datas *, uint8_t, uint8_t *,
+ uint8_t *, uint8_t *);
+
/* return code for these cmds */
#define CMD_OK 0
#define CMD_ERR 1
@@ -491,10 +506,10 @@ int ata_set_mode(struct ata_drive_datas
struct ata_xfer *ata_get_xfer_ext(struct ata_channel *, bool, int8_t);
#define ata_get_xfer(chp) ata_get_xfer_ext((chp), true, 0);
void ata_free_xfer(struct ata_channel *, struct ata_xfer *);
-
void ata_deactivate_xfer(struct ata_channel *, struct ata_xfer *);
-
void ata_exec_xfer(struct ata_channel *, struct ata_xfer *);
+
+void ata_timeout(void *);
void ata_kill_pending(struct ata_drive_datas *);
void ata_kill_active(struct ata_channel *, int, int);
void ata_reset_channel(struct ata_channel *, int);
@@ -521,6 +536,8 @@ struct ata_xfer *
ata_queue_hwslot_to_xfer(struct ata_channel *, int);
struct ata_xfer *
ata_queue_get_active_xfer(struct ata_channel *);
+struct ata_xfer *
+ ata_queue_drive_active_xfer(struct ata_channel *, int);
void ata_delay(int, const char *, int);
Index: src/sys/dev/ata/satafis_subr.c
diff -u src/sys/dev/ata/satafis_subr.c:1.7.28.2 src/sys/dev/ata/satafis_subr.c:1.7.28.3
--- src/sys/dev/ata/satafis_subr.c:1.7.28.2 Wed Apr 19 20:49:17 2017
+++ src/sys/dev/ata/satafis_subr.c Wed Jul 19 19:39:28 2017
@@ -1,4 +1,4 @@
-/* $NetBSD: satafis_subr.c,v 1.7.28.2 2017/04/19 20:49:17 jdolecek Exp $ */
+/* $NetBSD: satafis_subr.c,v 1.7.28.3 2017/07/19 19:39:28 jdolecek Exp $ */
/*-
* Copyright (c) 2009 Jonathan A. Kollasch.
@@ -51,7 +51,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: satafis_subr.c,v 1.7.28.2 2017/04/19 20:49:17 jdolecek Exp $");
+__KERNEL_RCSID(0, "$NetBSD: satafis_subr.c,v 1.7.28.3 2017/07/19 19:39:28 jdolecek Exp $");
#include <sys/param.h>
#include <sys/systm.h>
@@ -160,12 +160,11 @@ satafis_rhd_construct_atapi(struct ata_x
}
#endif /* NATAPIBUS */
-void
+int
satafis_rdh_parse(struct ata_channel *chp, const uint8_t *fis)
{
- chp->ch_status = fis[rdh_status];
- chp->ch_error = fis[rdh_error];
+ return ATACH_ERR_ST(fis[rdh_error], fis[rdh_status]);
}
void
Index: src/sys/dev/ata/satafisvar.h
diff -u src/sys/dev/ata/satafisvar.h:1.3 src/sys/dev/ata/satafisvar.h:1.3.50.1
--- src/sys/dev/ata/satafisvar.h:1.3 Wed Apr 7 17:51:16 2010
+++ src/sys/dev/ata/satafisvar.h Wed Jul 19 19:39:28 2017
@@ -1,4 +1,4 @@
-/* $NetBSD: satafisvar.h,v 1.3 2010/04/07 17:51:16 jakllsch Exp $ */
+/* $NetBSD: satafisvar.h,v 1.3.50.1 2017/07/19 19:39:28 jdolecek Exp $ */
/*
* Copyright (c) 2009, 2010 Jonathan A. Kollasch.
@@ -34,7 +34,7 @@ void satafis_rhd_construct_cmd(struct at
void satafis_rhd_construct_bio(struct ata_xfer *, uint8_t *);
void satafis_rhd_construct_atapi(struct ata_xfer *, uint8_t *);
-void satafis_rdh_parse(struct ata_channel *, const uint8_t *);
+int satafis_rdh_parse(struct ata_channel *, const uint8_t *);
void satafis_rdh_cmd_readreg(struct ata_command *, const uint8_t *);
#endif /* _DEV_ATA_FISVAR_H_ */
Index: src/sys/dev/ata/wd.c
diff -u src/sys/dev/ata/wd.c:1.428.2.25 src/sys/dev/ata/wd.c:1.428.2.26
--- src/sys/dev/ata/wd.c:1.428.2.25 Mon Jul 3 19:54:44 2017
+++ src/sys/dev/ata/wd.c Wed Jul 19 19:39:28 2017
@@ -1,4 +1,4 @@
-/* $NetBSD: wd.c,v 1.428.2.25 2017/07/03 19:54:44 jdolecek Exp $ */
+/* $NetBSD: wd.c,v 1.428.2.26 2017/07/19 19:39:28 jdolecek Exp $ */
/*
* Copyright (c) 1998, 2001 Manuel Bouyer. All rights reserved.
@@ -54,7 +54,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: wd.c,v 1.428.2.25 2017/07/03 19:54:44 jdolecek Exp $");
+__KERNEL_RCSID(0, "$NetBSD: wd.c,v 1.428.2.26 2017/07/19 19:39:28 jdolecek Exp $");
#include "opt_ata.h"
@@ -116,6 +116,7 @@ int wdcdebug_wd_mask = 0x0;
#endif
#ifdef WD_CHAOS_MONKEY
+int wdcdebug_wd_cnt = 200;
int wdcdebug_wd_chaos = 0;
#endif
@@ -198,7 +199,7 @@ void wdgetdefaultlabel(struct wd_softc
void wdgetdisklabel(struct wd_softc *);
void wdstart(device_t);
void wdstart1(struct wd_softc *, struct buf *, struct ata_xfer *);
-void wdrestart(void *);
+static void wdbiorestart(void *);
void wddone(device_t, struct ata_xfer *);
static void wd_params_to_properties(struct wd_softc *);
int wd_get_params(struct wd_softc *, uint8_t, struct ataparams *);
@@ -302,14 +303,11 @@ wdattach(device_t parent, device_t self,
wd->sc_dev = self;
ATADEBUG_PRINT(("wdattach\n"), DEBUG_FUNCS | DEBUG_PROBE);
- callout_init(&wd->sc_restart_ch, 0);
- callout_setfunc(&wd->sc_restart_ch, wdrestart, wd);
mutex_init(&wd->sc_lock, MUTEX_DEFAULT, IPL_BIO);
bufq_alloc(&wd->sc_q, BUFQ_DISK_DEFAULT_STRAT, BUFQ_SORT_RAWBLOCK);
#ifdef WD_SOFTBADSECT
SLIST_INIT(&wd->sc_bslist);
#endif
- STAILQ_INIT(&wd->xfer_restart);
wd->atabus = adev->adev_bustype;
wd->drvp = adev->adev_drv_data;
@@ -525,7 +523,6 @@ wddetach(device_t self, int flags)
rnd_detach_source(&sc->rnd_source);
mutex_destroy(&sc->sc_lock);
- callout_destroy(&sc->sc_restart_ch);
sc->drvp->drive_type = ATA_DRIVET_NONE; /* no drive any more here */
sc->drvp->drive_flags = 0;
@@ -674,7 +671,7 @@ wdstart(device_t self)
bp = bufq_get(wd->sc_q);
KASSERT(bp != NULL);
- xfer->c_bio.retries = 0;
+ xfer->c_retries = 0;
wdstart1(wd, bp, xfer);
}
@@ -688,17 +685,21 @@ wdstart1(struct wd_softc *wd, struct buf
/* must be locked on entry */
KASSERT(mutex_owned(&wd->sc_lock));
+ KASSERT(bp == xfer->c_bio.bp || xfer->c_bio.bp == NULL);
+ KASSERT((xfer->c_flags & (C_WAITACT|C_FREE)) == 0);
+
+ /* Reset state, so that retries don't use stale info */
+ if (__predict_false(xfer->c_retries > 0)) {
+ xfer->c_flags = 0;
+ memset(&xfer->c_bio, 0, sizeof(xfer->c_bio));
+ }
+
xfer->c_bio.blkno = bp->b_rawblkno;
xfer->c_bio.bcount = bp->b_bcount;
xfer->c_bio.databuf = bp->b_data;
xfer->c_bio.blkdone = 0;
- KASSERT(bp == xfer->c_bio.bp || xfer->c_bio.bp == NULL);
xfer->c_bio.bp = bp;
- /* Reset state flags, so that retries don't use stale info */
- KASSERT((xfer->c_flags & (C_WAITACT|C_FREE)) == 0);
- xfer->c_flags = 0;
-
#ifdef WD_CHAOS_MONKEY
/*
* Override blkno to be over device capacity to trigger error,
@@ -706,7 +707,8 @@ wdstart1(struct wd_softc *wd, struct buf
* the command be clipped, or otherwise misinterpreted, by the
* driver or controller.
*/
- if (BUF_ISREAD(bp) && (++wdcdebug_wd_chaos % WD_CHAOS_MONKEY) == 0) {
+ if (BUF_ISREAD(bp) && xfer->c_retries == 0 && wdcdebug_wd_cnt > 0 &&
+ (++wdcdebug_wd_chaos % wdcdebug_wd_cnt) == 0) {
aprint_normal_dev(wd->sc_dev, "%s: chaos xfer %d\n",
__func__, xfer->c_slot);
xfer->c_bio.blkno = 7777777 + wd->sc_capacity;
@@ -718,7 +720,7 @@ wdstart1(struct wd_softc *wd, struct buf
* the sector number of the problem, and will eventually allow the
* transfer to succeed.
*/
- if (xfer->c_bio.retries >= WDIORETRIES_SINGLE)
+ if (xfer->c_retries >= WDIORETRIES_SINGLE)
xfer->c_bio.flags = ATA_SINGLE;
else
xfer->c_bio.flags = 0;
@@ -739,7 +741,7 @@ wdstart1(struct wd_softc *wd, struct buf
* retrying with NCQ.
*/
if (wd->drvp->drive_flags & ATA_DRIVE_NCQ &&
- (xfer->c_bio.retries == 0 || (bp->b_flags & B_MEDIA_FUA))) {
+ (xfer->c_retries == 0 || (bp->b_flags & B_MEDIA_FUA))) {
xfer->c_bio.flags |= ATA_LBA48;
xfer->c_flags |= C_NCQ;
@@ -785,7 +787,8 @@ wddone(device_t self, struct ata_xfer *x
ATADEBUG_PRINT(("wddone %s\n", device_xname(wd->sc_dev)),
DEBUG_XFERS);
- if (wddoingadump) {
+
+ if (__predict_false(wddoingadump)) {
/* just drop it to the floor */
ata_free_xfer(wd->drvp->chnl_softc, xfer);
return;
@@ -805,6 +808,9 @@ wddone(device_t self, struct ata_xfer *x
case TIMEOUT:
errmsg = "device timeout";
goto retry;
+ case REQUEUE:
+ errmsg = "requeue";
+ goto retry2;
case ERR_RESET:
errmsg = "channel reset";
goto retry2;
@@ -822,25 +828,27 @@ retry2:
diskerr(bp, "wd", errmsg, LOG_PRINTF,
xfer->c_bio.blkdone, wd->sc_dk.dk_label);
- if (xfer->c_bio.retries < WDIORETRIES)
- printf(", retrying %d", xfer->c_bio.retries + 1);
+ if (xfer->c_retries < WDIORETRIES)
+ printf(", slot %d, retry %d", xfer->c_slot,
+ xfer->c_retries + 1);
printf("\n");
if (do_perror)
wdperror(wd, xfer);
- if (xfer->c_bio.retries < WDIORETRIES) {
- xfer->c_bio.retries++;
- STAILQ_INSERT_TAIL(&wd->xfer_restart, xfer,
- c_restartchain);
-
- /*
- * Only restart the timer if it's not already pending,
- * so that we wouldn't postpone processing beyond
- * original schedule.
- */
- if (!callout_pending(&wd->sc_restart_ch)) {
- callout_schedule(&wd->sc_restart_ch,
- RECOVERYTIME);
+
+ if (xfer->c_retries < WDIORETRIES) {
+ int timo;
+
+ if (xfer->c_bio.error == REQUEUE) {
+ /* rerun ASAP, and do not count as retry */
+ timo = 1;
+ } else {
+ xfer->c_retries++;
+ timo = RECOVERYTIME;
}
+
+ callout_reset(&xfer->c_retry_callout, timo,
+ wdbiorestart, xfer);
+
mutex_exit(&wd->sc_lock);
return;
}
@@ -881,7 +889,7 @@ out:
bp->b_error = EIO;
break;
case NOERROR:
-noerror: if ((xfer->c_bio.flags & ATA_CORR) || xfer->c_bio.retries > 0)
+noerror: if ((xfer->c_bio.flags & ATA_CORR) || xfer->c_retries > 0)
aprint_error_dev(wd->sc_dev,
"soft error (corrected)\n");
break;
@@ -905,27 +913,18 @@ noerror: if ((xfer->c_bio.flags & ATA_CO
ata_channel_start(wd->drvp->chnl_softc, wd->drvp->drive);
}
-void
-wdrestart(void *v)
+static void
+wdbiorestart(void *v)
{
- struct wd_softc *wd = v;
- struct ata_xfer *xfer;
+ struct ata_xfer *xfer = v;
+ struct buf *bp = xfer->c_bio.bp;
+ struct wd_softc *wd = device_lookup_private(&wd_cd, WDUNIT(bp->b_dev));
ATADEBUG_PRINT(("wdrestart %s\n", device_xname(wd->sc_dev)),
DEBUG_XFERS);
- /*
- * Resend all failed xfers out immediatelly regardless of original
- * schedule, so that we error out reasonably fast in case of massive
- * permanent errors.
- */
mutex_enter(&wd->sc_lock);
- while (!STAILQ_EMPTY(&wd->xfer_restart)) {
- xfer = STAILQ_FIRST(&wd->xfer_restart);
- STAILQ_REMOVE_HEAD(&wd->xfer_restart, c_restartchain);
-
- wdstart1(v, xfer->c_bio.bp, xfer);
- }
+ wdstart1(wd, bp, xfer);
mutex_exit(&wd->sc_lock);
}
Index: src/sys/dev/ata/wdvar.h
diff -u src/sys/dev/ata/wdvar.h:1.43.4.6 src/sys/dev/ata/wdvar.h:1.43.4.7
--- src/sys/dev/ata/wdvar.h:1.43.4.6 Fri Jun 23 20:40:51 2017
+++ src/sys/dev/ata/wdvar.h Wed Jul 19 19:39:28 2017
@@ -1,4 +1,4 @@
-/* $NetBSD: wdvar.h,v 1.43.4.6 2017/06/23 20:40:51 jdolecek Exp $ */
+/* $NetBSD: wdvar.h,v 1.43.4.7 2017/07/19 19:39:28 jdolecek Exp $ */
/*
* Copyright (c) 1998, 2001 Manuel Bouyer.
@@ -38,7 +38,6 @@ struct wd_softc {
device_t sc_dev;
struct disk sc_dk;
struct bufq_state *sc_q;
- struct callout sc_restart_ch;
kmutex_t sc_lock;
int sc_quirks; /* any quirks drive might have */
@@ -69,8 +68,6 @@ struct wd_softc {
u_int sc_bscount;
#endif
krndsource_t rnd_source;
-
- STAILQ_HEAD(, ata_xfer) xfer_restart;
};
#endif /* _DEV_ATA_WDVAR_H_ */
