ioat: Handle channel-fatal HW errors safely

Certain invalid operations trigger hardware error conditions.  Error
conditions that only halt one channel can be detected and recovered by
resetting the channel.  Error conditions that halt the whole device are
generally not recoverable.

Add a sysctl to inject channel-fatal HW errors,
'dev.ioat.<N>.force_hw_error=1'.

When a halt due to a channel error is detected, ioat(4) blocks new
operations from being queued on the channel, completes any outstanding
operations with an error status, and resets the channel before allowing
new operations to be queued again.

Update ioat.4 to document error recovery;  document blockfill introduced
in r290021 while we are here;  document ioat_put_dmaengine() added in
r289907;  document DMA_NO_WAIT added in r289982.

Sponsored by:	EMC / Isilon Storage Division

1 files changed, 71 insertions, 8 deletions

diff --git a/share/man/man4/ioat.4 b/share/man/man4/ioat.4
index 301ab4f..62d85f1 100644
--- a/share/man/man4/ioat.4
+++ b/share/man/man4/ioat.4
@@ -24,14 +24,25 @@
 .\"
 .\" $FreeBSD$
 .\"
-.Dd August 24, 2015
+.Dd October 31, 2015
 .Dt IOAT 4
 .Os
 .Sh NAME
 .Nm I/OAT
 .Nd Intel I/O Acceleration Technology
 .Sh SYNOPSIS
+To compile this driver into your kernel,
+place the following line in your kernel configuration file:
+.Bd -ragged -offset indent
 .Cd "device ioat"
+.Ed
+.Pp
+Or, to load the driver as a module at boot, place the following line in
+.Xr loader.conf 5 :
+.Bd -literal -offset indent
+ioat_load="YES"
+.Ed
+.Pp
 In
 .Xr loader.conf 5 :
 .Pp
@@ -46,11 +57,13 @@ In
 (only critical errors; maximum of 3)
 .Pp
 .Ft typedef void
-.Fn (*bus_dmaengine_callback_t) "void *arg"
+.Fn (*bus_dmaengine_callback_t) "void *arg" "int error"
 .Pp
 .Ft bus_dmaengine_t
 .Fn ioat_get_dmaengine "uint32_t channel_index"
 .Ft void
+.Fn ioat_put_dmaengine "bus_dmaengine_t dmaengine"
+.Ft void
 .Fn ioat_acquire "bus_dmaengine_t dmaengine"
 .Ft void
 .Fn ioat_release "bus_dmaengine_t dmaengine"
@@ -65,6 +78,16 @@ In
 .Fa "uint32_t flags"
 .Fc
 .Ft struct bus_dmadesc *
+.Fo ioat_blockfill
+.Fa "bus_dmaengine_t dmaengine"
+.Fa "bus_addr_t dst"
+.Fa "uint64_t fillpattern"
+.Fa "bus_size_t len"
+.Fa "bus_dmaengine_callback_t callback_fn"
+.Fa "void *callback_arg"
+.Fa "uint32_t flags"
+.Fc
+.Ft struct bus_dmadesc *
 .Fo ioat_null
 .Fa "bus_dmaengine_t dmaengine"
 .Fa "bus_dmaengine_callback_t callback_fn"
@@ -82,7 +105,9 @@ There is a number of DMA channels per CPU package.
 Each may be used independently.
 Operations on a single channel proceed sequentially.
 .Pp
-Copy operations may be used to offload memory copies to the DMA engines.
+Blockfill operations can be used to write a 64-bit pattern to memory.
+.Pp
+Copy operations can be used to offload memory copies to the DMA engines.
 .Pp
 Null operations do nothing, but may be used to test the interrupt and callback
 mechanism.
@@ -92,6 +117,26 @@ All operations can optionally trigger an interrupt at completion with the
 flag.
 For example, a user might submit multiple operations to the same channel and
 only enable an interrupt and callback for the last operation.
+.Pp
+All operations are safe to use in a non-blocking context with the
+.Ar DMA_NO_WAIT
+flag.
+(Of course, allocations may fail and operations requested with
+.Ar DMA_NO_WAIT
+may return NULL.)
+.Pp
+All operations, as well as
+.Fn ioat_get_dmaengine ,
+can return NULL in special circumstances.
+For example, if the
+.Nm
+driver is being unloaded, or the administrator has induced a hardware reset, or
+a usage error has resulted in a hardware error state that needs to be recovered
+from.
+.Pp
+It is invalid to attempt to submit new DMA operations in a
+.Fa bus_dmaengine_callback_t
+context.
 .Sh USAGE
 A typical user will lookup the DMA engine object for a given channel with
 .Fn ioat_get_dmaengine .
@@ -101,10 +146,11 @@ the
 .Ar bus_dmaengine_t
 object for exclusive access to enqueue operations on that channel.
 Then, they will submit one or more operations using
-.Fn ioat_copy
+.Fn ioat_blockfill ,
+.Fn ioat_copy ,
 or
 .Fn ioat_null .
-Finally, they will
+After queueing one or more individual DMA operations, they will
 .Fn ioat_release
 the
 .Ar bus_dmaengine_t
@@ -114,6 +160,19 @@ The routine they provided for the
 argument will be invoked with the provided
 .Fa callback_arg
 when the operation is complete.
+When they are finished with the
+.Ar bus_dmaengine_t ,
+the user should
+.Fn ioat_put_dmaengine .
+.Pp
+Users MUST NOT block between
+.Fn ioat_acquire
+and
+.Fn ioat_release .
+Users SHOULD NOT hold
+.Ar bus_dmaengine_t
+references for a very long time to enable fault recovery and kernel module
+unload.
 .Pp
 For an example of usage, see
 .Pa src/sys/dev/ioat/ioat_test.c .
@@ -135,19 +194,23 @@ The
 .Nm
 driver was developed by
 .An \&Jim Harris Aq Mt jimharris@FreeBSD.org ,
+.An \&Carl Delsey Aq Mt carl.r.delsey@intel.com ,
 and
-.An \&Carl Delsey Aq Mt carl.r.delsey@intel.com .
+.An \&Conrad Meyer Aq Mt cem@FreeBSD.org .
 This manual page was written by
 .An \&Conrad Meyer Aq Mt cem@FreeBSD.org .
 .Sh CAVEATS
 Copy operation takes bus addresses as parameters, not virtual addresses.
 .Pp
-Copies larger than max transfer size (1MB) are not supported.
+Buffers for individual copy operations must be physically contiguous.
+.Pp
+Copies larger than max transfer size (1MB, but may vary by hardware) are not
+supported.
 Future versions will likely support this by breaking up the transfer into
 smaller sizes.
 .Sh BUGS
 The
 .Nm
-driver only supports copy and null operations at this time.
+driver only supports blockfill, copy, and null operations at this time.
 The driver does not yet support advanced DMA modes, such as XOR, that some
 I/OAT devices support.