linux-kernel - Re: ioatdma: add ioat_raid

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20130802180154.GC13747@jonmason-lab>
Date:	Fri, 2 Aug 2013 11:01:55 -0700
From:	Jon Mason <jon.mason@...el.com>
To:	Dan Williams <djbw@...com>
Cc:	Brice Goglin <Brice.Goglin@...ia.fr>,
	"Jiang, Dave" <dave.jiang@...el.com>,
	"Koul, Vinod" <vinod.koul@...el.com>,
	LKML <linux-kernel@...r.kernel.org>
Subject: Re: ioatdma: add ioat_raid_enabled module parameter

On Fri, Aug 02, 2013 at 04:57:44PM +0000, Dan Williams wrote:
> 
> 
> On 8/2/13 12:34 AM, "Brice Goglin" <Brice.Goglin@...ia.fr> wrote:
> 
> >Le 01/08/2013 19:15, Jiang, Dave a écrit :
> >> On Thu, 2013-08-01 at 10:11 -0700, Jon Mason wrote:
> >>> On Wed, Jul 31, 2013 at 03:14:07PM -0700, Jiang, Dave wrote:
> >>>> I'm ok with enabling this for people that just want to use DMA and not
> >>>> RAID. 
> >>> I might be crazy, but I'd be in favor of disabling the RAID offload by
> >>> default on non-Atom platforms.
> >>>
> >> I suppose. Technically it is disabled starting with 3.10 because of the
> >> channel switch issue. I'm ok with this disabled by default for the 3.2
> >> platforms that has broken pq-val.
> >>
> >
> >Here's a patch that may do what you guys are saying.
> >
> >Brice
> >
> >
> >
> >ioatdma: disable RAID by default when buggy and add module param
> >
> >Commit f26df1a1 added a 64-byte alignment requirement for legacy
> >operations to work around a silicon errata when mixing legacy and
> >RAID descriptors.
> >
> >RAID offload is now disabled by default on buggy 3.2 platforms.
> >Passing ioat_raid_enabled=1 force-enables it on all platforms
> >(previous behavior).
> >Passing ioat_raid_enabled=0 force-disables it everywhere.
> >
> >When RAID offload is disabled, legacy operations (memcpy, etc.)
> >can work again without alignment restrictions.
> >
> >Signed-off-by: Brice Goglin <Brice.Goglin@...ia.fr>
> >---
> > drivers/dma/ioat/dma_v3.c |   16 ++++++++++++++--
> > 1 file changed, 14 insertions(+), 2 deletions(-)
> >
> >Index: b/drivers/dma/ioat/dma_v3.c
> >===================================================================
> >--- a/drivers/dma/ioat/dma_v3.c	2013-07-31 23:06:24.163810000 +0200
> >+++ b/drivers/dma/ioat/dma_v3.c	2013-08-02 09:28:51.817037742 +0200
> >@@ -67,6 +67,11 @@
> > #include "dma.h"
> > #include "dma_v2.h"
> > 
> >+static int ioat_raid_enabled = -1;
> >+module_param(ioat_raid_enabled, int, 0444);
> >+MODULE_PARM_DESC(ioat_raid_enabled,
> >+		 "control support of RAID offload (-1=enabled unless broken [default],
> >0=disabled, 1=enabled)");
> >+
> > /* ioat hardware assumes at least two sources for raid operations */
> > #define src_cnt_to_sw(x) ((x) + 2)
> > #define src_cnt_to_hw(x) ((x) - 2)
> >@@ -1775,7 +1780,7 @@ int ioat3_dma_probe(struct ioatdma_devic
> > 	dma->device_alloc_chan_resources = ioat2_alloc_chan_resources;
> > 	dma->device_free_chan_resources = ioat2_free_chan_resources;
> > 
> >-	if (is_xeon_cb32(pdev))
> >+	if (ioat_raid_enabled == 1 && is_xeon_cb32(pdev))
> > 		dma->copy_align = 6;
> 
> Actually we can delete this now because is_xeon_cb32() already effectively
> means that raid offload will not be used.
> 
> > 
> > 	dma_cap_set(DMA_INTERRUPT, dma->cap_mask);
> >@@ -1783,7 +1788,14 @@ int ioat3_dma_probe(struct ioatdma_devic
> > 
> > 	device->cap = readl(device->reg_base + IOAT_DMA_CAP_OFFSET);
> > 
> >-	if (is_bwd_noraid(pdev))
> >+	/* disable RAID if:
> >+	 *    force-disabled by module param,
> >+	 * or not force-enabled on buggy 3.2 platforms,
> >+	 * or not actually supported.
> >+	 */
> >+	if (ioat_raid_enabled == 0
> >+	    || (ioat_raid_enabled != 1 && is_xeon_cb32(pdev))
> >+	    || is_bwd_noraid(pdev))
> > 		device->cap &= ~(IOAT_CAP_XOR | IOAT_CAP_PQ | IOAT_CAP_RAID16SS);
> > 
> > 	/* dca is incompatible with raid operations */
> >
> 
> I like Jon¹s suggestion.  Just make raid disabled by default on non-atom
> platforms.  When if a non-atom platform comes along without the previous
> restrictions it can add itself to this list.

Awesome!  This greatly increases NTB performance when using DMA
engines.

Thanks,
Jon

> 
> So let¹s drop the module parameter and just cleanup the 3.2 support to
> reflect the current reality of raid being disabled.
> 
> --
> Dan
> 
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/