Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1303777 > unrolled thread

[PATCH v4.4-rc8] dmaengine: ioatdma: Squelch framesize warnings

Back to article view | Back to linux.kernel

Contents

  [PATCH v4.4-rc8] dmaengine: ioatdma: Squelch framesize warnings tim.gardner@canonical.com - 2016-01-07 19:10 +0100
    Re: [PATCH v4.4-rc8] dmaengine: ioatdma: Squelch framesize warnings Dan Williams <dan.j.williams@intel.com> - 2016-01-07 20:50 +0100
    [PATCH v4.4-rc8 v2] dmaengine: ioatdma: Squelch framesize warnings tim.gardner@canonical.com - 2016-01-08 14:40 +0100
      Re: [PATCH v4.4-rc8 v2] dmaengine: ioatdma: Squelch framesize warnings Dan Williams <dan.j.williams@intel.com> - 2016-01-08 17:40 +0100
        [PATCH v4.4-rc8 v3] dmaengine: ioatdma: Squelch framesize warnings tim.gardner@canonical.com - 2016-01-08 20:30 +0100
          Re: [PATCH v4.4-rc8 v3] dmaengine: ioatdma: Squelch framesize warnings Dan Williams <dan.j.williams@intel.com> - 2016-01-08 20:50 +0100
            Re: [PATCH v4.4-rc8 v3] dmaengine: ioatdma: Squelch framesize  warnings "Jiang, Dave" <dave.jiang@intel.com> - 2016-01-08 21:00 +0100
              [PATCH v4.4-rc8 v4] dmaengine: ioatdma: Squelch framesize warnings tim.gardner@canonical.com - 2016-01-08 22:50 +0100
                Re: [PATCH v4.4-rc8 v4] dmaengine: ioatdma: Squelch framesize  warnings Vinod Koul <vinod.koul@intel.com> - 2016-01-13 14:30 +0100

#1303777 — [PATCH v4.4-rc8] dmaengine: ioatdma: Squelch framesize warnings

From: Tim Gardner <tim.gardner@canonical.com>

  CC [M]  drivers/dma/ioat/prep.o
drivers/dma/ioat/prep.c: In function 'ioat_prep_pqxor':
drivers/dma/ioat/prep.c:682:1: warning: the frame size of 1048 bytes is larger than 1024 bytes [-Wframe-larger-than=]
 }
 ^
drivers/dma/ioat/prep.c: In function 'ioat_prep_pqxor_val':
drivers/dma/ioat/prep.c:714:1: warning: the frame size of 1048 bytes is larger than 1024 bytes [-Wframe-larger-than=]
 }

gcc version 5.3.1 20151219 (Ubuntu 5.3.1-4ubuntu1)

Cc: Vinod Koul <vinod.koul@intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Jiang <dave.jiang@intel.com>
Cc: Prarit Bhargava <prarit@redhat.com>
Cc: Nicholas Mc Guire <der.herr@hofr.at>
Cc: Jarkko Nikula <jarkko.nikula@linux.intel.com>
Signed-off-by: Tim Gardner <tim.gardner@canonical.com>
---
 drivers/dma/ioat/prep.c | 30 ++++++++++++++++++++++--------
 1 file changed, 22 insertions(+), 8 deletions(-)

diff --git a/drivers/dma/ioat/prep.c b/drivers/dma/ioat/prep.c
index 6bb4a13..ff62763 100644
--- a/drivers/dma/ioat/prep.c
+++ b/drivers/dma/ioat/prep.c
@@ -655,30 +655,41 @@ ioat_prep_pq_val(struct dma_chan *chan, dma_addr_t *pq, dma_addr_t *src,
 				     flags);
 }
 
+static char *ioat_alloc_scf(unsigned int src_cnt)
+{
+	if (!src_cnt || src_cnt > MAX_SCF)
+		return NULL;
+
+	return kzalloc(src_cnt, GFP_KERNEL);
+}
+
 struct dma_async_tx_descriptor *
 ioat_prep_pqxor(struct dma_chan *chan, dma_addr_t dst, dma_addr_t *src,
 		 unsigned int src_cnt, size_t len, unsigned long flags)
 {
-	unsigned char scf[MAX_SCF];
+	unsigned char *scf;
 	dma_addr_t pq[2];
 	struct ioatdma_chan *ioat_chan = to_ioat_chan(chan);
+	struct dma_async_tx_descriptor *desc;
 
 	if (test_bit(IOAT_CHAN_DOWN, &ioat_chan->state))
 		return NULL;
 
-	if (src_cnt > MAX_SCF)
+	scf = ioat_alloc_scf(src_cnt);
+	if (!scf)
 		return NULL;
 
-	memset(scf, 0, src_cnt);
 	pq[0] = dst;
 	flags |= DMA_PREP_PQ_DISABLE_Q;
 	pq[1] = dst; /* specify valid address for disabled result */
 
-	return src_cnt_flags(src_cnt, flags) > 8 ?
+	desc = src_cnt_flags(src_cnt, flags) > 8 ?
 		__ioat_prep_pq16_lock(chan, NULL, pq, src, src_cnt, scf, len,
 				       flags) :
 		__ioat_prep_pq_lock(chan, NULL, pq, src, src_cnt, scf, len,
 				     flags);
+	kfree(scf);
+	return desc;
 }
 
 struct dma_async_tx_descriptor *
@@ -686,14 +697,16 @@ ioat_prep_pqxor_val(struct dma_chan *chan, dma_addr_t *src,
 		     unsigned int src_cnt, size_t len,
 		     enum sum_check_flags *result, unsigned long flags)
 {
-	unsigned char scf[MAX_SCF];
+	unsigned char *scf;
 	dma_addr_t pq[2];
 	struct ioatdma_chan *ioat_chan = to_ioat_chan(chan);
+	struct dma_async_tx_descriptor *desc;
 
 	if (test_bit(IOAT_CHAN_DOWN, &ioat_chan->state))
 		return NULL;
 
-	if (src_cnt > MAX_SCF)
+	scf = ioat_alloc_scf(src_cnt);
+	if (!scf)
 		return NULL;
 
 	/* the cleanup routine only sets bits on validate failure, it
@@ -701,16 +714,17 @@ ioat_prep_pqxor_val(struct dma_chan *chan, dma_addr_t *src,
 	 */
 	*result = 0;
 
-	memset(scf, 0, src_cnt);
 	pq[0] = src[0];
 	flags |= DMA_PREP_PQ_DISABLE_Q;
 	pq[1] = pq[0]; /* specify valid address for disabled result */
 
-	return src_cnt_flags(src_cnt, flags) > 8 ?
+	desc = src_cnt_flags(src_cnt, flags) > 8 ?
 		__ioat_prep_pq16_lock(chan, result, pq, &src[1], src_cnt - 1,
 				       scf, len, flags) :
 		__ioat_prep_pq_lock(chan, result, pq, &src[1], src_cnt - 1,
 				     scf, len, flags);
+	kfree(scf);
+	return desc;
 }
 
 struct dma_async_tx_descriptor *
-- 
1.9.1

[toc] | [next] | [standalone]

#1303853

On Thu, Jan 7, 2016 at 10:07 AM,  <tim.gardner@canonical.com> wrote:
> From: Tim Gardner <tim.gardner@canonical.com>
>
>   CC [M]  drivers/dma/ioat/prep.o
> drivers/dma/ioat/prep.c: In function 'ioat_prep_pqxor':
> drivers/dma/ioat/prep.c:682:1: warning: the frame size of 1048 bytes is larger than 1024 bytes [-Wframe-larger-than=]
>  }
>  ^
> drivers/dma/ioat/prep.c: In function 'ioat_prep_pqxor_val':
> drivers/dma/ioat/prep.c:714:1: warning: the frame size of 1048 bytes is larger than 1024 bytes [-Wframe-larger-than=]
>  }
>
> gcc version 5.3.1 20151219 (Ubuntu 5.3.1-4ubuntu1)
>
> Cc: Vinod Koul <vinod.koul@intel.com>
> Cc: Dan Williams <dan.j.williams@intel.com>
> Cc: Dave Jiang <dave.jiang@intel.com>
> Cc: Prarit Bhargava <prarit@redhat.com>
> Cc: Nicholas Mc Guire <der.herr@hofr.at>
> Cc: Jarkko Nikula <jarkko.nikula@linux.intel.com>
> Signed-off-by: Tim Gardner <tim.gardner@canonical.com>
> ---
>  drivers/dma/ioat/prep.c | 30 ++++++++++++++++++++++--------
>  1 file changed, 22 insertions(+), 8 deletions(-)
>
> diff --git a/drivers/dma/ioat/prep.c b/drivers/dma/ioat/prep.c
> index 6bb4a13..ff62763 100644
> --- a/drivers/dma/ioat/prep.c
> +++ b/drivers/dma/ioat/prep.c
> @@ -655,30 +655,41 @@ ioat_prep_pq_val(struct dma_chan *chan, dma_addr_t *pq, dma_addr_t *src,
>                                      flags);
>  }
>
> +static char *ioat_alloc_scf(unsigned int src_cnt)
> +{
> +       if (!src_cnt || src_cnt > MAX_SCF)
> +               return NULL;
> +
> +       return kzalloc(src_cnt, GFP_KERNEL);

We can't do GFP_KERNEL here since prep_pqxor might be called from
non-sleeping contexts, and I don't want to incur this overhead to
every dma operation.

Instead, let's replace this with a static alloc_percpu() allocation
that gets referenced under get_cpu() (preempt disable).

[toc] | [prev] | [next] | [standalone]

#1304543 — [PATCH v4.4-rc8 v2] dmaengine: ioatdma: Squelch framesize warnings

From: Tim Gardner <tim.gardner@canonical.com>

  CC [M]  drivers/dma/ioat/prep.o
drivers/dma/ioat/prep.c: In function 'ioat_prep_pqxor':
drivers/dma/ioat/prep.c:682:1: warning: the frame size of 1048 bytes is larger than 1024 bytes [-Wframe-larger-than=]
 }
 ^
drivers/dma/ioat/prep.c: In function 'ioat_prep_pqxor_val':
drivers/dma/ioat/prep.c:714:1: warning: the frame size of 1048 bytes is larger than 1024 bytes [-Wframe-larger-than=]
 }

gcc version 5.3.1 20151219 (Ubuntu 5.3.1-4ubuntu1)

Cc: Vinod Koul <vinod.koul@intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Jiang <dave.jiang@intel.com>
Cc: Prarit Bhargava <prarit@redhat.com>
Cc: Nicholas Mc Guire <der.herr@hofr.at>
Cc: Jarkko Nikula <jarkko.nikula@linux.intel.com>
Signed-off-by: Tim Gardner <tim.gardner@canonical.com>
---

v2 - use per CPU static buffers instead of dynamically allocating memory.

 drivers/dma/ioat/prep.c | 33 +++++++++++++++++++++++++++++----
 1 file changed, 29 insertions(+), 4 deletions(-)

diff --git a/drivers/dma/ioat/prep.c b/drivers/dma/ioat/prep.c
index 6bb4a13..2c0768b 100644
--- a/drivers/dma/ioat/prep.c
+++ b/drivers/dma/ioat/prep.c
@@ -21,6 +21,8 @@
 #include <linux/dmaengine.h>
 #include <linux/dma-mapping.h>
 #include <linux/prefetch.h>
+#include <linux/percpu.h>
+#include <linux/preempt.h>
 #include "../dmaengine.h"
 #include "registers.h"
 #include "hw.h"
@@ -655,13 +657,25 @@ ioat_prep_pq_val(struct dma_chan *chan, dma_addr_t *pq, dma_addr_t *src,
 				     flags);
 }
 
+/*
+ * The scf scratch buffer is too large for an automatic variable, and
+ * we don't want to take the performance hit for dynamic allocation.
+ * Therefore, define per CPU buffers and disable preemption while in use.
+ */
+static DEFINE_PER_CPU(unsigned char [MAX_SCF], ioat_scf);
+static inline unsigned char *ioat_assign_scratch_buffer(void)
+{
+	return get_cpu_var(ioat_scf);
+}
+
 struct dma_async_tx_descriptor *
 ioat_prep_pqxor(struct dma_chan *chan, dma_addr_t dst, dma_addr_t *src,
 		 unsigned int src_cnt, size_t len, unsigned long flags)
 {
-	unsigned char scf[MAX_SCF];
+	unsigned char *scf;
 	dma_addr_t pq[2];
 	struct ioatdma_chan *ioat_chan = to_ioat_chan(chan);
+	struct dma_async_tx_descriptor *desc;
 
 	if (test_bit(IOAT_CHAN_DOWN, &ioat_chan->state))
 		return NULL;
@@ -669,16 +683,21 @@ ioat_prep_pqxor(struct dma_chan *chan, dma_addr_t dst, dma_addr_t *src,
 	if (src_cnt > MAX_SCF)
 		return NULL;
 
+	preempt_disable();
+	scf = ioat_assign_scratch_buffer();
+
 	memset(scf, 0, src_cnt);
 	pq[0] = dst;
 	flags |= DMA_PREP_PQ_DISABLE_Q;
 	pq[1] = dst; /* specify valid address for disabled result */
 
-	return src_cnt_flags(src_cnt, flags) > 8 ?
+	desc = src_cnt_flags(src_cnt, flags) > 8 ?
 		__ioat_prep_pq16_lock(chan, NULL, pq, src, src_cnt, scf, len,
 				       flags) :
 		__ioat_prep_pq_lock(chan, NULL, pq, src, src_cnt, scf, len,
 				     flags);
+	preempt_enable();
+	return desc;
 }
 
 struct dma_async_tx_descriptor *
@@ -686,9 +705,10 @@ ioat_prep_pqxor_val(struct dma_chan *chan, dma_addr_t *src,
 		     unsigned int src_cnt, size_t len,
 		     enum sum_check_flags *result, unsigned long flags)
 {
-	unsigned char scf[MAX_SCF];
+	unsigned char *scf;
 	dma_addr_t pq[2];
 	struct ioatdma_chan *ioat_chan = to_ioat_chan(chan);
+	struct dma_async_tx_descriptor *desc;
 
 	if (test_bit(IOAT_CHAN_DOWN, &ioat_chan->state))
 		return NULL;
@@ -696,6 +716,9 @@ ioat_prep_pqxor_val(struct dma_chan *chan, dma_addr_t *src,
 	if (src_cnt > MAX_SCF)
 		return NULL;
 
+	preempt_disable();
+	scf = ioat_assign_scratch_buffer();
+
 	/* the cleanup routine only sets bits on validate failure, it
 	 * does not clear bits on validate success... so clear it here
 	 */
@@ -706,11 +729,13 @@ ioat_prep_pqxor_val(struct dma_chan *chan, dma_addr_t *src,
 	flags |= DMA_PREP_PQ_DISABLE_Q;
 	pq[1] = pq[0]; /* specify valid address for disabled result */
 
-	return src_cnt_flags(src_cnt, flags) > 8 ?
+	desc = src_cnt_flags(src_cnt, flags) > 8 ?
 		__ioat_prep_pq16_lock(chan, result, pq, &src[1], src_cnt - 1,
 				       scf, len, flags) :
 		__ioat_prep_pq_lock(chan, result, pq, &src[1], src_cnt - 1,
 				     scf, len, flags);
+	preempt_enable();
+	return desc;
 }
 
 struct dma_async_tx_descriptor *
-- 
1.9.1

[toc] | [prev] | [next] | [standalone]

#1304717 — Re: [PATCH v4.4-rc8 v2] dmaengine: ioatdma: Squelch framesize warnings

On Fri, Jan 8, 2016 at 5:35 AM,  <tim.gardner@canonical.com> wrote:
> From: Tim Gardner <tim.gardner@canonical.com>
>
>   CC [M]  drivers/dma/ioat/prep.o
> drivers/dma/ioat/prep.c: In function 'ioat_prep_pqxor':
> drivers/dma/ioat/prep.c:682:1: warning: the frame size of 1048 bytes is larger than 1024 bytes [-Wframe-larger-than=]
>  }
>  ^
> drivers/dma/ioat/prep.c: In function 'ioat_prep_pqxor_val':
> drivers/dma/ioat/prep.c:714:1: warning: the frame size of 1048 bytes is larger than 1024 bytes [-Wframe-larger-than=]
>  }
>
> gcc version 5.3.1 20151219 (Ubuntu 5.3.1-4ubuntu1)
>
> Cc: Vinod Koul <vinod.koul@intel.com>
> Cc: Dan Williams <dan.j.williams@intel.com>
> Cc: Dave Jiang <dave.jiang@intel.com>
> Cc: Prarit Bhargava <prarit@redhat.com>
> Cc: Nicholas Mc Guire <der.herr@hofr.at>
> Cc: Jarkko Nikula <jarkko.nikula@linux.intel.com>
> Signed-off-by: Tim Gardner <tim.gardner@canonical.com>
> ---
>
> v2 - use per CPU static buffers instead of dynamically allocating memory.
>
>  drivers/dma/ioat/prep.c | 33 +++++++++++++++++++++++++++++----
>  1 file changed, 29 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/dma/ioat/prep.c b/drivers/dma/ioat/prep.c
> index 6bb4a13..2c0768b 100644
> --- a/drivers/dma/ioat/prep.c
> +++ b/drivers/dma/ioat/prep.c
> @@ -21,6 +21,8 @@
>  #include <linux/dmaengine.h>
>  #include <linux/dma-mapping.h>
>  #include <linux/prefetch.h>
> +#include <linux/percpu.h>
> +#include <linux/preempt.h>
>  #include "../dmaengine.h"
>  #include "registers.h"
>  #include "hw.h"
> @@ -655,13 +657,25 @@ ioat_prep_pq_val(struct dma_chan *chan, dma_addr_t *pq, dma_addr_t *src,
>                                      flags);
>  }
>
> +/*
> + * The scf scratch buffer is too large for an automatic variable, and
> + * we don't want to take the performance hit for dynamic allocation.
> + * Therefore, define per CPU buffers and disable preemption while in use.
> + */
> +static DEFINE_PER_CPU(unsigned char [MAX_SCF], ioat_scf);
> +static inline unsigned char *ioat_assign_scratch_buffer(void)
> +{
> +       return get_cpu_var(ioat_scf);

This get needs to be paired with a put_cpu_var().

> +}
> +
>  struct dma_async_tx_descriptor *
>  ioat_prep_pqxor(struct dma_chan *chan, dma_addr_t dst, dma_addr_t *src,
>                  unsigned int src_cnt, size_t len, unsigned long flags)
>  {
> -       unsigned char scf[MAX_SCF];
> +       unsigned char *scf;
>         dma_addr_t pq[2];
>         struct ioatdma_chan *ioat_chan = to_ioat_chan(chan);
> +       struct dma_async_tx_descriptor *desc;
>
>         if (test_bit(IOAT_CHAN_DOWN, &ioat_chan->state))
>                 return NULL;
> @@ -669,16 +683,21 @@ ioat_prep_pqxor(struct dma_chan *chan, dma_addr_t dst, dma_addr_t *src,
>         if (src_cnt > MAX_SCF)
>                 return NULL;
>
> +       preempt_disable();

preempt_disable() is embedded in get_cpu_var(), no need to open code it.

> +       scf = ioat_assign_scratch_buffer();
> +
>         memset(scf, 0, src_cnt);
>         pq[0] = dst;
>         flags |= DMA_PREP_PQ_DISABLE_Q;
>         pq[1] = dst; /* specify valid address for disabled result */
>
> -       return src_cnt_flags(src_cnt, flags) > 8 ?
> +       desc = src_cnt_flags(src_cnt, flags) > 8 ?
>                 __ioat_prep_pq16_lock(chan, NULL, pq, src, src_cnt, scf, len,
>                                        flags) :
>                 __ioat_prep_pq_lock(chan, NULL, pq, src, src_cnt, scf, len,
>                                      flags);
> +       preempt_enable();

put_cpu_var()...

[toc] | [prev] | [next] | [standalone]

#1304943 — [PATCH v4.4-rc8 v3] dmaengine: ioatdma: Squelch framesize warnings

From: Tim Gardner <tim.gardner@canonical.com>

  CC [M]  drivers/dma/ioat/prep.o
drivers/dma/ioat/prep.c: In function 'ioat_prep_pqxor':
drivers/dma/ioat/prep.c:682:1: warning: the frame size of 1048 bytes is larger than 1024 bytes [-Wframe-larger-than=]
 }
 ^
drivers/dma/ioat/prep.c: In function 'ioat_prep_pqxor_val':
drivers/dma/ioat/prep.c:714:1: warning: the frame size of 1048 bytes is larger than 1024 bytes [-Wframe-larger-than=]
 }

gcc version 5.3.1 20151219 (Ubuntu 5.3.1-4ubuntu1)

Cc: Vinod Koul <vinod.koul@intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Jiang <dave.jiang@intel.com>
Cc: Prarit Bhargava <prarit@redhat.com>
Cc: Nicholas Mc Guire <der.herr@hofr.at>
Cc: Jarkko Nikula <jarkko.nikula@linux.intel.com>
Signed-off-by: Tim Gardner <tim.gardner@canonical.com>
---

v2 - use per CPU static buffers instead of dynamically allocating memory.
v3 - Use get_cpu_var/put_cpu_var which implicitly control preeemption. Drop
     the wrapper function that no longer serves any purpose.

 drivers/dma/ioat/prep.c | 29 +++++++++++++++++++++++++----
 1 file changed, 25 insertions(+), 4 deletions(-)

diff --git a/drivers/dma/ioat/prep.c b/drivers/dma/ioat/prep.c
index 6bb4a13..0bd05c8 100644
--- a/drivers/dma/ioat/prep.c
+++ b/drivers/dma/ioat/prep.c
@@ -21,6 +21,7 @@
 #include <linux/dmaengine.h>
 #include <linux/dma-mapping.h>
 #include <linux/prefetch.h>
+#include <linux/percpu.h>
 #include "../dmaengine.h"
 #include "registers.h"
 #include "hw.h"
@@ -655,13 +656,22 @@ ioat_prep_pq_val(struct dma_chan *chan, dma_addr_t *pq, dma_addr_t *src,
 				     flags);
 }
 
+/*
+ * The scf scratch buffer is too large for an automatic variable, and
+ * we don't want to take the performance hit for dynamic allocation.
+ * Therefore, define per CPU buffers and use get_cpu_var()/put_cpu_var()
+ * to control preemption while the buffer is in use.
+ */
+static DEFINE_PER_CPU(unsigned char [MAX_SCF], ioat_scf);
+
 struct dma_async_tx_descriptor *
 ioat_prep_pqxor(struct dma_chan *chan, dma_addr_t dst, dma_addr_t *src,
 		 unsigned int src_cnt, size_t len, unsigned long flags)
 {
-	unsigned char scf[MAX_SCF];
+	unsigned char *scf;
 	dma_addr_t pq[2];
 	struct ioatdma_chan *ioat_chan = to_ioat_chan(chan);
+	struct dma_async_tx_descriptor *desc;
 
 	if (test_bit(IOAT_CHAN_DOWN, &ioat_chan->state))
 		return NULL;
@@ -669,16 +679,21 @@ ioat_prep_pqxor(struct dma_chan *chan, dma_addr_t dst, dma_addr_t *src,
 	if (src_cnt > MAX_SCF)
 		return NULL;
 
+	scf = get_cpu_var(ioat_scf);
+
 	memset(scf, 0, src_cnt);
 	pq[0] = dst;
 	flags |= DMA_PREP_PQ_DISABLE_Q;
 	pq[1] = dst; /* specify valid address for disabled result */
 
-	return src_cnt_flags(src_cnt, flags) > 8 ?
+	desc = src_cnt_flags(src_cnt, flags) > 8 ?
 		__ioat_prep_pq16_lock(chan, NULL, pq, src, src_cnt, scf, len,
 				       flags) :
 		__ioat_prep_pq_lock(chan, NULL, pq, src, src_cnt, scf, len,
 				     flags);
+
+	put_cpu_var(ioat_scf);
+	return desc;
 }
 
 struct dma_async_tx_descriptor *
@@ -686,9 +701,10 @@ ioat_prep_pqxor_val(struct dma_chan *chan, dma_addr_t *src,
 		     unsigned int src_cnt, size_t len,
 		     enum sum_check_flags *result, unsigned long flags)
 {
-	unsigned char scf[MAX_SCF];
+	unsigned char *scf;
 	dma_addr_t pq[2];
 	struct ioatdma_chan *ioat_chan = to_ioat_chan(chan);
+	struct dma_async_tx_descriptor *desc;
 
 	if (test_bit(IOAT_CHAN_DOWN, &ioat_chan->state))
 		return NULL;
@@ -696,6 +712,8 @@ ioat_prep_pqxor_val(struct dma_chan *chan, dma_addr_t *src,
 	if (src_cnt > MAX_SCF)
 		return NULL;
 
+	scf = get_cpu_var(ioat_scf);
+
 	/* the cleanup routine only sets bits on validate failure, it
 	 * does not clear bits on validate success... so clear it here
 	 */
@@ -706,11 +724,14 @@ ioat_prep_pqxor_val(struct dma_chan *chan, dma_addr_t *src,
 	flags |= DMA_PREP_PQ_DISABLE_Q;
 	pq[1] = pq[0]; /* specify valid address for disabled result */
 
-	return src_cnt_flags(src_cnt, flags) > 8 ?
+	desc = src_cnt_flags(src_cnt, flags) > 8 ?
 		__ioat_prep_pq16_lock(chan, result, pq, &src[1], src_cnt - 1,
 				       scf, len, flags) :
 		__ioat_prep_pq_lock(chan, result, pq, &src[1], src_cnt - 1,
 				     scf, len, flags);
+
+	put_cpu_var(ioat_scf);
+	return desc;
 }
 
 struct dma_async_tx_descriptor *
-- 
1.9.1

[toc] | [prev] | [next] | [standalone]

#1304960 — Re: [PATCH v4.4-rc8 v3] dmaengine: ioatdma: Squelch framesize warnings

On Fri, Jan 8, 2016 at 11:25 AM,  <tim.gardner@canonical.com> wrote:
> From: Tim Gardner <tim.gardner@canonical.com>
>
>   CC [M]  drivers/dma/ioat/prep.o
> drivers/dma/ioat/prep.c: In function 'ioat_prep_pqxor':
> drivers/dma/ioat/prep.c:682:1: warning: the frame size of 1048 bytes is larger than 1024 bytes [-Wframe-larger-than=]
>  }
>  ^
> drivers/dma/ioat/prep.c: In function 'ioat_prep_pqxor_val':
> drivers/dma/ioat/prep.c:714:1: warning: the frame size of 1048 bytes is larger than 1024 bytes [-Wframe-larger-than=]
>  }
>
> gcc version 5.3.1 20151219 (Ubuntu 5.3.1-4ubuntu1)
>
> Cc: Vinod Koul <vinod.koul@intel.com>
> Cc: Dan Williams <dan.j.williams@intel.com>
> Cc: Dave Jiang <dave.jiang@intel.com>
> Cc: Prarit Bhargava <prarit@redhat.com>
> Cc: Nicholas Mc Guire <der.herr@hofr.at>
> Cc: Jarkko Nikula <jarkko.nikula@linux.intel.com>
> Signed-off-by: Tim Gardner <tim.gardner@canonical.com>
> ---
>
> v2 - use per CPU static buffers instead of dynamically allocating memory.
> v3 - Use get_cpu_var/put_cpu_var which implicitly control preeemption. Drop
>      the wrapper function that no longer serves any purpose.
>
>  drivers/dma/ioat/prep.c | 29 +++++++++++++++++++++++++----
>  1 file changed, 25 insertions(+), 4 deletions(-)

Looks good to me... but now that I think about it, why is MAX_SCF set
to 1024 in the first place?  Certainly it can't be bigger than the
maximum number of sources in a single operation which is 8 to 16.
Even md raid can only support up to 256 devices in an array.  So I
think that contstant is bogus.

If we set it to 16 we may not even need the percpu change.

Dave?

[toc] | [prev] | [next] | [standalone]

#1304968 — Re: [PATCH v4.4-rc8 v3] dmaengine: ioatdma: Squelch framesize warnings

On Fri, 2016-01-08 at 19:41 +0000, Williams, Dan J wrote:
> On Fri, Jan 8, 2016 at 11:25 AM,  <tim.gardner@canonical.com> wrote:
> > From: Tim Gardner <tim.gardner@canonical.com>
> > 
> >   CC [M]  drivers/dma/ioat/prep.o
> > drivers/dma/ioat/prep.c: In function 'ioat_prep_pqxor':
> > drivers/dma/ioat/prep.c:682:1: warning: the frame size of 1048
> > bytes is larger than 1024 bytes [-Wframe-larger-than=]
> >  }
> >  ^
> > drivers/dma/ioat/prep.c: In function 'ioat_prep_pqxor_val':
> > drivers/dma/ioat/prep.c:714:1: warning: the frame size of 1048
> > bytes is larger than 1024 bytes [-Wframe-larger-than=]
> >  }
> > 
> > gcc version 5.3.1 20151219 (Ubuntu 5.3.1-4ubuntu1)
> > 
> > Cc: Vinod Koul <vinod.koul@intel.com>
> > Cc: Dan Williams <dan.j.williams@intel.com>
> > Cc: Dave Jiang <dave.jiang@intel.com>
> > Cc: Prarit Bhargava <prarit@redhat.com>
> > Cc: Nicholas Mc Guire <der.herr@hofr.at>
> > Cc: Jarkko Nikula <jarkko.nikula@linux.intel.com>
> > Signed-off-by: Tim Gardner <tim.gardner@canonical.com>
> > ---
> > 
> > v2 - use per CPU static buffers instead of dynamically allocating
> > memory.
> > v3 - Use get_cpu_var/put_cpu_var which implicitly control
> > preeemption. Drop
> >      the wrapper function that no longer serves any purpose.
> > 
> >  drivers/dma/ioat/prep.c | 29 +++++++++++++++++++++++++----
> >  1 file changed, 25 insertions(+), 4 deletions(-)
> 
> Looks good to me... but now that I think about it, why is MAX_SCF set
> to 1024 in the first place?  Certainly it can't be bigger than the
> maximum number of sources in a single operation which is 8 to 16.
> Even md raid can only support up to 256 devices in an array.  So I
> think that contstant is bogus.
> 
> If we set it to 16 we may not even need the percpu change.
> 
> Dave?

You are right. It was an arbitrary number I threw in there to address
sparse warning. It can be reduced. 256 seems reasonable. 

[toc] | [prev] | [next] | [standalone]

#1305023 — [PATCH v4.4-rc8 v4] dmaengine: ioatdma: Squelch framesize warnings

From: Tim Gardner <tim.gardner@canonical.com>

  CC [M]  drivers/dma/ioat/prep.o
drivers/dma/ioat/prep.c: In function 'ioat_prep_pqxor':
drivers/dma/ioat/prep.c:682:1: warning: the frame size of 1048 bytes is larger than 1024 bytes [-Wframe-larger-than=]
 }
 ^
drivers/dma/ioat/prep.c: In function 'ioat_prep_pqxor_val':
drivers/dma/ioat/prep.c:714:1: warning: the frame size of 1048 bytes is larger than 1024 bytes [-Wframe-larger-than=]
 }

gcc version 5.3.1 20151219 (Ubuntu 5.3.1-4ubuntu1)

Cc: Vinod Koul <vinod.koul@intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Jiang <dave.jiang@intel.com>
Cc: Prarit Bhargava <prarit@redhat.com>
Cc: Nicholas Mc Guire <der.herr@hofr.at>
Cc: Jarkko Nikula <jarkko.nikula@linux.intel.com>
Signed-off-by: Tim Gardner <tim.gardner@canonical.com>
---

v2 - use per CPU static buffers instead of dynamically allocating memory.
v3 - Use get_cpu_var/put_cpu_var which implicitly control preeemption. Drop
     the wrapper function that no longer serves any purpose.
v4 - According to discussion between Dan Williams and Dave Jiang, MAX_SCF really
     only needs to be 256. That is certainly a simpler patch then previous efforts.

     Quoting from an email regarding v3:

     Dan said, "Looks good to me... but now that I think about it, why is MAX_SCF set
     to 1024 in the first place?  Certainly it can't be bigger than the
     maximum number of sources in a single operation which is 8 to 16.
     Even md raid can only support up to 256 devices in an array.  So I
     think that contstant is bogus.
     If we set it to 16 we may not even need the percpu change.
     Dave?"

     Dave responded, "You are right. It was an arbitrary number I threw in there to address
     sparse warning. It can be reduced. 256 seems reasonable. "

 drivers/dma/ioat/prep.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/dma/ioat/prep.c b/drivers/dma/ioat/prep.c
index 6bb4a13..243421a 100644
--- a/drivers/dma/ioat/prep.c
+++ b/drivers/dma/ioat/prep.c
@@ -26,7 +26,7 @@
 #include "hw.h"
 #include "dma.h"
 
-#define MAX_SCF	1024
+#define MAX_SCF	256
 
 /* provide a lookup table for setting the source address in the base or
  * extended descriptor of an xor or pq descriptor
-- 
2.6.4

[toc] | [prev] | [next] | [standalone]

#1308410 — Re: [PATCH v4.4-rc8 v4] dmaengine: ioatdma: Squelch framesize warnings

On Fri, Jan 08, 2016 at 02:48:17PM -0700, tim.gardner@canonical.com wrote:
> From: Tim Gardner <tim.gardner@canonical.com>
> 
>   CC [M]  drivers/dma/ioat/prep.o
> drivers/dma/ioat/prep.c: In function 'ioat_prep_pqxor':
> drivers/dma/ioat/prep.c:682:1: warning: the frame size of 1048 bytes is larger than 1024 bytes [-Wframe-larger-than=]
>  }
>  ^
> drivers/dma/ioat/prep.c: In function 'ioat_prep_pqxor_val':
> drivers/dma/ioat/prep.c:714:1: warning: the frame size of 1048 bytes is larger than 1024 bytes [-Wframe-larger-than=]
>  }
> 
> gcc version 5.3.1 20151219 (Ubuntu 5.3.1-4ubuntu1)

Applied, thanks


-- 
~Vinod

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web