From: Dmitry Torokhov <dmitry.torokhov@gmail.com> To: Andrew Morton <akpm@linux-foundation.org> Cc: "Oskar Schirmer" <os@emlix.com>, "Michael Hennerich" <Michael.Hennerich@analog.com>, "Mike Frysinger" <vapier@gentoo.org>, linux-input@vger.kernel.org, linux-kernel@vger.kernel.org, "Daniel Glöckner" <dg@emlix.com>, "Oliver Schneidewind" <osw@emlix.com>, "Johannes Weiner" <jw@emlix.com> Subject: Re: [PATCH v3] ad7877: keep dma rx buffers in seperate cache lines Date: Mon, 10 May 2010 23:05:47 -0700 [thread overview] Message-ID: <20100511060547.GA9644@core.coreip.homeip.net> (raw) In-Reply-To: <20100510142225.4bf215ef.akpm@linux-foundation.org> On Mon, May 10, 2010 at 02:22:25PM -0700, Andrew Morton wrote: > On Mon, 10 May 2010 12:42:34 +0200 > "Oskar Schirmer" <os@emlix.com> wrote: > > > With dma based spi transmission, data corruption > > is observed occasionally. With dma buffers located > > right next to msg and xfer fields, cache lines > > correctly flushed in preparation for dma usage > > may be polluted again when writing to fields > > in the same cache line. > > > > Make sure cache fields used with dma do not > > share cache lines with fields changed during > > dma handling. As both fields are part of a > > struct that is allocated via kzalloc, thus > > cache aligned, moving the fields to the 1st > > position and insert padding for alignment > > does the job. > > This sounds odd. Doesn't it imply that some code somewhere is missing > some DMA synchronisation actions? > > > > > v2: add a comment to explain why alignment is needed > > > > v3: fix the typo in comment and layout (- to end of line) > > > > diff --git a/drivers/input/touchscreen/ad7877.c b/drivers/input/touchscreen/ad7877.c > > index 885354c..9ebb1b4 100644 > > --- a/drivers/input/touchscreen/ad7877.c > > +++ b/drivers/input/touchscreen/ad7877.c > > @@ -153,15 +153,29 @@ enum { > > */ > > > > struct ser_req { > > + u16 sample; > > + /* > > + * DMA (thus cache coherency maintenance) requires the > > + * transfer buffers to live in their own cache lines. > > + */ > > + char __padalign[L1_CACHE_BYTES - sizeof(u16)]; > > It would be better to use __cacheline_aligned, rather than open-coding > things in this manner. > OK, then I have the following which I will apply unless someone shouts. -- Dmitry Input: ad7877 - keep dma rx buffers in seperate cache lines From: Oskar Schirmer <os@emlix.com> With dma based spi transmission, data corruption is observed occasionally. With dma buffers located right next to msg and xfer fields, cache lines correctly flushed in preparation for dma usage may be polluted again when writing to fields in the same cache line. Make sure cache fields used with dma do not share cache lines with fields changed during dma handling. As both fields are part of a struct that is allocated via kzalloc, thus cache aligned, moving the fields to the 1st position and insert padding for alignment does the job. Signed-off-by: Oskar Schirmer <os@emlix.com> Signed-off-by: Daniel Glöckner <dg@emlix.com> Signed-off-by: Oliver Schneidewind <osw@emlix.com> Signed-off-by: Johannes Weiner <jw@emlix.com> Acked-by: Mike Frysinger <vapier@gentoo.org> [dtor@mail.ru - changed to use ___cacheline_aligned at suggestion of akpm] Signed-off-by: Dmitry Torokhov <dtor@mail.ru> --- drivers/input/touchscreen/ad7877.c | 15 ++++++++++++--- 1 files changed, 12 insertions(+), 3 deletions(-) diff --git a/drivers/input/touchscreen/ad7877.c b/drivers/input/touchscreen/ad7877.c index e019d53..0d2d7e5 100644 --- a/drivers/input/touchscreen/ad7877.c +++ b/drivers/input/touchscreen/ad7877.c @@ -156,9 +156,14 @@ struct ser_req { u16 reset; u16 ref_on; u16 command; - u16 sample; struct spi_message msg; struct spi_transfer xfer[6]; + + /* + * DMA (thus cache coherency maintenance) requires the + * transfer buffers to live in their own cache lines. + */ + u16 sample ____cacheline_aligned; }; struct ad7877 { @@ -182,8 +187,6 @@ struct ad7877 { u8 averaging; u8 pen_down_acc_interval; - u16 conversion_data[AD7877_NR_SENSE]; - struct spi_transfer xfer[AD7877_NR_SENSE + 2]; struct spi_message msg; @@ -195,6 +198,12 @@ struct ad7877 { spinlock_t lock; struct timer_list timer; /* P: lock */ unsigned pending:1; /* P: lock */ + + /* + * DMA (thus cache coherency maintenance) requires the + * transfer buffers to live in their own cache lines. + */ + u16 conversion_data[AD7877_NR_SENSE] ____cacheline_aligned; }; static int gpio3;
WARNING: multiple messages have this Message-ID (diff)
From: Dmitry Torokhov <dmitry.torokhov@gmail.com> To: Andrew Morton <akpm@linux-foundation.org> Cc: "Oskar Schirmer" <os@emlix.com>, "Michael Hennerich" <Michael.Hennerich@analog.com>, "Mike Frysinger" <vapier@gentoo.org>, linux-input@vger.kernel.org, linux-kernel@vger.kernel.org, "Daniel Glöckner" <dg@emlix.com>, "Oliver Schneidewind" <osw@emlix.com>, "Johannes Weiner" <jw@emlix.com> Subject: Re: [PATCH v3] ad7877: keep dma rx buffers in seperate cache lines Date: Mon, 10 May 2010 23:05:47 -0700 [thread overview] Message-ID: <20100511060547.GA9644@core.coreip.homeip.net> (raw) In-Reply-To: <20100510142225.4bf215ef.akpm@linux-foundation.org> On Mon, May 10, 2010 at 02:22:25PM -0700, Andrew Morton wrote: > On Mon, 10 May 2010 12:42:34 +0200 > "Oskar Schirmer" <os@emlix.com> wrote: > > > With dma based spi transmission, data corruption > > is observed occasionally. With dma buffers located > > right next to msg and xfer fields, cache lines > > correctly flushed in preparation for dma usage > > may be polluted again when writing to fields > > in the same cache line. > > > > Make sure cache fields used with dma do not > > share cache lines with fields changed during > > dma handling. As both fields are part of a > > struct that is allocated via kzalloc, thus > > cache aligned, moving the fields to the 1st > > position and insert padding for alignment > > does the job. > > This sounds odd. Doesn't it imply that some code somewhere is missing > some DMA synchronisation actions? > > > > > v2: add a comment to explain why alignment is needed > > > > v3: fix the typo in comment and layout (- to end of line) > > > > diff --git a/drivers/input/touchscreen/ad7877.c b/drivers/input/touchscreen/ad7877.c > > index 885354c..9ebb1b4 100644 > > --- a/drivers/input/touchscreen/ad7877.c > > +++ b/drivers/input/touchscreen/ad7877.c > > @@ -153,15 +153,29 @@ enum { > > */ > > > > struct ser_req { > > + u16 sample; > > + /* > > + * DMA (thus cache coherency maintenance) requires the > > + * transfer buffers to live in their own cache lines. > > + */ > > + char __padalign[L1_CACHE_BYTES - sizeof(u16)]; > > It would be better to use __cacheline_aligned, rather than open-coding > things in this manner. > OK, then I have the following which I will apply unless someone shouts. -- Dmitry Input: ad7877 - keep dma rx buffers in seperate cache lines From: Oskar Schirmer <os@emlix.com> With dma based spi transmission, data corruption is observed occasionally. With dma buffers located right next to msg and xfer fields, cache lines correctly flushed in preparation for dma usage may be polluted again when writing to fields in the same cache line. Make sure cache fields used with dma do not share cache lines with fields changed during dma handling. As both fields are part of a struct that is allocated via kzalloc, thus cache aligned, moving the fields to the 1st position and insert padding for alignment does the job. Signed-off-by: Oskar Schirmer <os@emlix.com> Signed-off-by: Daniel Glöckner <dg@emlix.com> Signed-off-by: Oliver Schneidewind <osw@emlix.com> Signed-off-by: Johannes Weiner <jw@emlix.com> Acked-by: Mike Frysinger <vapier@gentoo.org> [dtor@mail.ru - changed to use ___cacheline_aligned at suggestion of akpm] Signed-off-by: Dmitry Torokhov <dtor@mail.ru> --- drivers/input/touchscreen/ad7877.c | 15 ++++++++++++--- 1 files changed, 12 insertions(+), 3 deletions(-) diff --git a/drivers/input/touchscreen/ad7877.c b/drivers/input/touchscreen/ad7877.c index e019d53..0d2d7e5 100644 --- a/drivers/input/touchscreen/ad7877.c +++ b/drivers/input/touchscreen/ad7877.c @@ -156,9 +156,14 @@ struct ser_req { u16 reset; u16 ref_on; u16 command; - u16 sample; struct spi_message msg; struct spi_transfer xfer[6]; + + /* + * DMA (thus cache coherency maintenance) requires the + * transfer buffers to live in their own cache lines. + */ + u16 sample ____cacheline_aligned; }; struct ad7877 { @@ -182,8 +187,6 @@ struct ad7877 { u8 averaging; u8 pen_down_acc_interval; - u16 conversion_data[AD7877_NR_SENSE]; - struct spi_transfer xfer[AD7877_NR_SENSE + 2]; struct spi_message msg; @@ -195,6 +198,12 @@ struct ad7877 { spinlock_t lock; struct timer_list timer; /* P: lock */ unsigned pending:1; /* P: lock */ + + /* + * DMA (thus cache coherency maintenance) requires the + * transfer buffers to live in their own cache lines. + */ + u16 conversion_data[AD7877_NR_SENSE] ____cacheline_aligned; }; static int gpio3; -- To unsubscribe from this list: send the line "unsubscribe linux-input" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
next prev parent reply other threads:[~2010-05-11 6:06 UTC|newest] Thread overview: 124+ messages / expand[flat|nested] mbox.gz Atom feed top 2010-05-06 10:37 [PATCH] ad7877: fix spi word size to 16 bit Oskar Schirmer 2010-05-06 10:37 ` Oskar Schirmer 2010-05-06 10:37 ` [PATCH] ad7877: keep dma rx buffers in seperate cache lines Oskar Schirmer 2010-05-06 18:46 ` Mike Frysinger 2010-05-06 18:46 ` Mike Frysinger 2010-05-07 10:15 ` Oskar Schirmer 2010-05-07 10:15 ` Oskar Schirmer 2010-05-07 18:28 ` Mike Frysinger 2010-05-07 18:28 ` Mike Frysinger 2010-05-08 22:32 ` Johannes Weiner 2010-05-08 22:32 ` Johannes Weiner 2010-05-09 4:45 ` Mike Frysinger 2010-05-09 4:45 ` Mike Frysinger 2010-05-09 8:50 ` Oskar Schirmer 2010-05-09 8:50 ` Oskar Schirmer 2010-05-07 12:07 ` Johannes Weiner 2010-05-10 10:34 ` [PATCH v2] " Oskar Schirmer 2010-05-10 10:34 ` Oskar Schirmer 2010-05-10 10:42 ` [PATCH v3] " Oskar Schirmer 2010-05-10 10:42 ` Oskar Schirmer 2010-05-10 16:39 ` Mike Frysinger 2010-05-10 20:54 ` Dmitry Torokhov 2010-05-10 21:22 ` Andrew Morton 2010-05-10 21:27 ` Mike Frysinger 2010-05-11 6:05 ` Dmitry Torokhov [this message] 2010-05-11 6:05 ` Dmitry Torokhov 2010-05-11 6:11 ` Mike Frysinger 2010-05-11 6:11 ` Mike Frysinger 2010-05-11 3:20 ` Andrew Morton 2010-05-11 6:21 ` Dmitry Torokhov 2010-05-11 6:23 ` Mike Frysinger 2010-05-11 6:23 ` Mike Frysinger 2010-05-11 6:33 ` Dmitry Torokhov 2010-05-11 6:33 ` Dmitry Torokhov 2010-05-11 6:37 ` Mike Frysinger 2010-05-11 6:42 ` Pekka Enberg 2010-05-11 13:57 ` Christoph Lameter 2010-05-11 16:52 ` Dmitry Torokhov 2010-05-11 17:31 ` Mike Frysinger 2010-05-11 18:59 ` Pekka Enberg 2010-05-11 20:03 ` Mike Frysinger 2010-05-11 20:07 ` Matt Mackall 2010-05-11 20:10 ` Mike Frysinger 2010-05-11 20:10 ` Mike Frysinger 2010-05-11 20:21 ` Christoph Lameter 2010-05-11 20:34 ` Mike Frysinger 2010-05-11 20:38 ` Pekka Enberg 2010-05-11 20:43 ` Mike Frysinger 2010-05-11 20:43 ` Mike Frysinger 2010-05-11 20:47 ` Pekka Enberg 2010-05-13 6:21 ` Paul Mundt 2010-05-11 20:46 ` Christoph Lameter 2010-05-11 20:54 ` Mike Frysinger 2010-05-11 20:54 ` Mike Frysinger 2010-05-11 21:01 ` Pekka Enberg 2010-05-11 21:11 ` Mike Frysinger 2010-05-11 21:11 ` Mike Frysinger 2010-05-12 1:58 ` FUJITA Tomonori [not found] ` <20100511214836.GH1726@emlix.com> 2010-05-11 21:53 ` Dmitry Torokhov 2010-05-11 22:39 ` Mike Frysinger 2010-05-11 22:39 ` Mike Frysinger 2010-05-12 2:07 ` [LKML] " Marc Gauthier 2010-05-12 3:03 ` Nick Piggin 2010-05-12 3:23 ` FUJITA Tomonori 2010-05-12 4:35 ` Mike Frysinger 2010-05-12 4:35 ` Mike Frysinger 2010-05-12 5:28 ` FUJITA Tomonori 2010-05-12 5:28 ` FUJITA Tomonori 2010-05-12 14:37 ` Mike Frysinger 2010-05-12 14:37 ` Mike Frysinger 2010-05-12 18:11 ` Dmitry Torokhov 2010-05-12 18:28 ` Mike Frysinger 2010-05-12 18:28 ` Mike Frysinger 2010-05-12 10:36 ` Johannes Weiner 2010-05-12 12:35 ` Marc Gauthier 2010-05-12 14:36 ` Mike Frysinger 2010-05-12 14:36 ` Mike Frysinger 2010-05-19 12:48 ` David Woodhouse 2010-05-19 13:07 ` Nick Piggin 2010-05-19 13:17 ` David Woodhouse 2010-05-19 13:36 ` Nick Piggin 2010-05-19 13:44 ` Johannes Weiner 2010-05-19 13:52 ` Nick Piggin 2010-05-19 14:38 ` FUJITA Tomonori 2010-05-19 14:58 ` David Woodhouse 2010-05-20 4:42 ` FUJITA Tomonori 2010-05-19 16:37 ` Dmitry Torokhov 2010-05-19 15:00 ` Johannes Weiner 2010-05-17 6:12 ` Michal Simek 2010-05-19 13:00 ` David Woodhouse 2010-05-11 22:37 ` Alan Cox 2010-05-11 20:22 ` Pekka Enberg 2010-05-11 14:12 ` Oskar Schirmer 2010-05-11 14:12 ` Oskar Schirmer 2010-05-06 11:18 ` [PATCH] ad7877: fix spi word size to 16 bit Hennerich, Michael 2010-05-06 11:18 ` Hennerich, Michael 2010-05-06 18:26 ` Mike Frysinger 2010-05-06 18:26 ` Mike Frysinger 2010-05-07 9:41 ` Daniel Glöckner 2010-05-07 9:41 ` Daniel Glöckner 2010-05-07 18:23 ` Mike Frysinger 2010-05-07 18:23 ` Mike Frysinger 2010-05-13 7:53 ` Dmitry Torokhov 2010-05-13 7:53 ` Dmitry Torokhov 2010-05-15 18:15 ` Oskar Schirmer 2010-05-15 18:15 ` Oskar Schirmer 2010-05-16 19:25 ` Mike Frysinger 2010-05-16 19:25 ` Mike Frysinger 2010-05-17 7:29 ` Oskar Schirmer 2010-05-17 7:29 ` Oskar Schirmer 2010-05-17 8:14 ` Hennerich, Michael 2010-05-17 8:14 ` Hennerich, Michael 2010-05-17 8:41 ` Oskar Schirmer 2010-05-17 8:41 ` Oskar Schirmer 2010-05-17 23:49 ` Mike Frysinger 2010-05-17 23:49 ` Mike Frysinger 2010-05-18 0:18 ` H Hartley Sweeten 2010-05-18 0:18 ` H Hartley Sweeten 2010-05-18 8:32 ` Oskar Schirmer 2010-05-18 8:32 ` Oskar Schirmer 2010-05-18 9:37 ` Daniel Glöckner 2010-05-18 9:37 ` Daniel Glöckner 2010-05-18 8:34 ` [PATCH v2] " Oskar Schirmer 2010-05-18 8:34 ` Oskar Schirmer
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=20100511060547.GA9644@core.coreip.homeip.net \ --to=dmitry.torokhov@gmail.com \ --cc=Michael.Hennerich@analog.com \ --cc=akpm@linux-foundation.org \ --cc=dg@emlix.com \ --cc=jw@emlix.com \ --cc=linux-input@vger.kernel.org \ --cc=linux-kernel@vger.kernel.org \ --cc=os@emlix.com \ --cc=osw@emlix.com \ --cc=vapier@gentoo.org \ /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: linkBe sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes, see mirroring instructions on how to clone and mirror all data and code used by this external index.