Skip to contentSkip to conversation

spi: spi-mtk-nor: Optimize timeout for dma read

PATCH Superseded Bayi Cheng 4 participants 9 messages Nov 3, 2022 – Nov 12, 2022

Subsystems (2)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

Bayi Cheng

author
#0 lore raw

PATCH 1/1 spi: spi-mtk-nor: Optimize timeout for dma read

bayi cheng (1):
spi: spi-mtk-nor: Optimize timeout for dma read

 drivers/spi/spi-mtk-nor.c | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)
Signature · 1 line
2.25.1
author
#1 lore raw

PATCH 1/1 spi: spi-mtk-nor: Optimize timeout for dma read

From: bayi cheng <[email protected]>

The timeout value of the current dma read is unreasonable. For example, If the spi flash clock is 26Mhz, It will takes about 1.3ms to read a 4KB data in spi mode. But the actual measurement exceeds 50s when a dma read timeout is encountered.

In order to be more accurately, It is necessary to use msecs_to_jiffies, After modification, the measured timeout value is about 130ms.

---
 drivers/spi/spi-mtk-nor.c | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)
drivers/spi/spi-mtk-nor.c +4 -3
index d167699a1a96..3d989db80ee9 100644
@@ -354,7 +354,7 @@ static int mtk_nor_dma_exec(struct mtk_nor *sp, u32 from, unsigned int length,
 			    dma_addr_t dma_addr)
 {
 	int ret = 0;
-	ulong delay;
+	ulong delay, timeout;
 	u32 reg;
 
 	writel(from, sp->base + MTK_NOR_REG_DMA_FADR);
@@ -376,15 +376,16 @@ static int mtk_nor_dma_exec(struct mtk_nor *sp, u32 from, unsigned int length,
 	mtk_nor_rmw(sp, MTK_NOR_REG_DMA_CTL, MTK_NOR_DMA_START, 0);
 
 	delay = CLK_TO_US(sp, (length + 5) * BITS_PER_BYTE);
+	timeout = (delay + 1) * 100;
 
 	if (sp->has_irq) {
 		if (!wait_for_completion_timeout(&sp->op_done,
-						 (delay + 1) * 100))
+		    msecs_to_jiffies(max_t(size_t, timeout / 1000, 10))))
 			ret = -ETIMEDOUT;
 	} else {
 		ret = readl_poll_timeout(sp->base + MTK_NOR_REG_DMA_CTL, reg,
 					 !(reg & MTK_NOR_DMA_START), delay / 3,
-					 (delay + 1) * 100);
+					 timeout);
 	}
 
 	if (ret < 0)
Signature · 1 line
2.25.1
#2 lore raw

Il 03/11/22 06:28, Bayi Cheng ha scritto:

Quoted 29 lines

From: bayi cheng <[email protected]>

The timeout value of the current dma read is unreasonable. For example, If the spi flash clock is 26Mhz, It will takes about 1.3ms to read a 4KB data in spi mode. But the actual measurement exceeds 50s when a dma read timeout is encountered.

In order to be more accurately, It is necessary to use msecs_to_jiffies, After modification, the measured timeout value is about 130ms.

---
  drivers/spi/spi-mtk-nor.c | 7 ++++---
  1 file changed, 4 insertions(+), 3 deletions(-)
drivers/spi/spi-mtk-nor.c +1 -1
index d167699a1a96..3d989db80ee9 100644
@@ -354,7 +354,7 @@ static int mtk_nor_dma_exec(struct mtk_nor *sp, u32 from, unsigned int length,
  			    dma_addr_t dma_addr)
  {
  	int ret = 0;
-	ulong delay;
+	ulong delay, timeout;
  	u32 reg;
 
  	writel(from, sp->base + MTK_NOR_REG_DMA_FADR);

drivers/spi/spi-mtk-nor.c static int mtk_nor_dma_exec(struct mtk_nor *sp, u32 from, unsigned int length, · line 383

Quoted 9 lines
drivers/spi/spi-mtk-nor.c +2 -1
@@ -376,15 +376,16 @@ static int mtk_nor_dma_exec(struct mtk_nor *sp, u32 from, unsigned int length,
  	mtk_nor_rmw(sp, MTK_NOR_REG_DMA_CTL, MTK_NOR_DMA_START, 0);
 
  	delay = CLK_TO_US(sp, (length + 5) * BITS_PER_BYTE);
+	timeout = (delay + 1) * 100;
 
  	if (sp->has_irq) {
  		if (!wait_for_completion_timeout(&sp->op_done,
-						 (delay + 1) * 100))
+		    msecs_to_jiffies(max_t(size_t, timeout / 1000, 10))))

You're giving a size_t variable to msecs_to_jiffies(), but checking jiffies.h, this function takes a const unsigned int param.
Please change the type to match that.

Aside from that, your timeout variable contains a timeout in microseconds and this means that actually using msecs_to_jiffies() is suboptimal here.

Please use usecs_to_jiffies() instead.

Regards,
Angelo

#3 lore raw

From: AngeloGioacchino Del Regno

Quoted 37 lines

Sent: 03 November 2022 09:44

Il 03/11/22 06:28, Bayi Cheng ha scritto:

Quoted 34 lines

From: bayi cheng <[email protected]>

The timeout value of the current dma read is unreasonable. For example, If the spi flash clock is 26Mhz, It will takes about 1.3ms to read a 4KB data in spi mode. But the actual measurement exceeds 50s when a dma read timeout is encountered.

In order to be more accurately, It is necessary to use msecs_to_jiffies, After modification, the measured timeout value is about 130ms.

---
  drivers/spi/spi-mtk-nor.c | 7 ++++---
  1 file changed, 4 insertions(+), 3 deletions(-)
drivers/spi/spi-mtk-nor.c +2 -1
index d167699a1a96..3d989db80ee9 100644
@@ -354,7 +354,7 @@ static int mtk_nor_dma_exec(struct mtk_nor *sp, u32 from, unsigned int length,
  			    dma_addr_t dma_addr)
  {
  	int ret = 0;
-	ulong delay;
+	ulong delay, timeout;
  	u32 reg;
 
  	writel(from, sp->base + MTK_NOR_REG_DMA_FADR);
@@ -376,15 +376,16 @@ static int mtk_nor_dma_exec(struct mtk_nor *sp, u32 from, unsigned int length,
  	mtk_nor_rmw(sp, MTK_NOR_REG_DMA_CTL, MTK_NOR_DMA_START, 0);
 
  	delay = CLK_TO_US(sp, (length + 5) * BITS_PER_BYTE);
+	timeout = (delay + 1) * 100;
 
Quoted 8 lines

drivers/spi/spi-mtk-nor.c static int mtk_nor_dma_exec(struct mtk_nor *sp, u32 from, unsigned int length, · line 383

Quoted 4 lines
drivers/spi/spi-mtk-nor.c +1 -1
  	if (sp->has_irq) {
  		if (!wait_for_completion_timeout(&sp->op_done,
-						 (delay + 1) * 100))
+		    msecs_to_jiffies(max_t(size_t, timeout / 1000, 10))))

You're giving a size_t variable to msecs_to_jiffies(), but checking jiffies.h, this function takes a const unsigned int param.
Please change the type to match that.

The type shouldn't matter at all.
What matters is the domain of the value.

Quite why you need to use max_t(size_t, ...) is another matter. timeout is ulong so max(timeout/1000, 10ul) should be fine.

But is ulong even right?
The domain of the value is almost certainly the same on 32bit and 64bit. So you almost certainly need u32 or u64.

David

Quoted 8 lines

Aside from that, your timeout variable contains a timeout in microseconds and this means that actually using msecs_to_jiffies() is suboptimal here.

Please use usecs_to_jiffies() instead.

Regards,
Angelo

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK Registration No: 1397386 (Wales)

#4 lore raw
Quoted 63 lines · David Laight

From: AngeloGioacchino Del Regno

Quoted 45 lines

Sent: 03 November 2022 09:44

Il 03/11/22 06:28, Bayi Cheng ha scritto:

Quoted 42 lines

From: bayi cheng <[email protected]>

The timeout value of the current dma read is unreasonable. For example,
If the spi flash clock is 26Mhz, It will takes about 1.3ms to read a
4KB data in spi mode. But the actual measurement exceeds 50s when a
dma read timeout is encountered.

In order to be more accurately, It is necessary to use msecs_to_jiffies,
After modification, the measured timeout value is about 130ms.

---
  drivers/spi/spi-mtk-nor.c | 7 ++++---
  1 file changed, 4 insertions(+), 3 deletions(-)
drivers/spi/spi-mtk- renamed from drivers/spi/spi-mtk-nor.c +0 -0

nor.c
index d167699a1a96..3d989db80ee9 100644

drivers/spi/spi-mtk-nor.c +1 -1
@@ -354,7 +354,7 @@ static int mtk_nor_dma_exec(struct mtk_nor
 *sp, u32 from, unsigned int length,
  			    dma_addr_t dma_addr)
  {
  	int ret = 0;
-	ulong delay;
+	ulong delay, timeout;
  	u32 reg;
 
  	writel(from, sp->base + MTK_NOR_REG_DMA_FADR);
drivers/spi/spi-mtk-nor.c +1 -0
@@ -376,15 +376,16 @@ static int mtk_nor_dma_exec(struct mtk_nor
 *sp, u32 from, unsigned int length,
  	mtk_nor_rmw(sp, MTK_NOR_REG_DMA_CTL, MTK_NOR_DMA_START,
 0);
 
  	delay = CLK_TO_US(sp, (length + 5) * BITS_PER_BYTE);
+	timeout = (delay + 1) * 100;
 
Quoted 10 lines

drivers/spi/spi-mtk-nor.c static int mtk_nor_dma_exec(struct mtk_nor *sp, u32 from, unsigned int length,

Quoted 5 lines
drivers/spi/spi-mtk-nor.c +1 -1
  	if (sp->has_irq) {
  		if (!wait_for_completion_timeout(&sp->op_done,
-						 (delay + 1) * 100))
+		    msecs_to_jiffies(max_t(size_t, timeout / 1000,
 10))))

You're giving a size_t variable to msecs_to_jiffies(), but checking jiffies.h,
this function takes a const unsigned int param. Please change the type to match that.

The type shouldn't matter at all.
What matters is the domain of the value.

Quite why you need to use max_t(size_t, ...) is another matter. timeout is ulong so max(timeout/1000, 10ul) should be fine.

Quoted 7 lines

But is ulong even right?
The domain of the value is almost certainly the same on 32bit and 64bit.
So you almost certainly need u32 or u64.

David

Hi David & Angelo

Thank you for your comments!
To sum up, I think the next version will make the following two changes:
1, The timeout value will not exceed u32, so the type of timeout will be changed from ulong to u32.
2, Change msecs_to_jiffies(max_t(size_t, timeout / 1000, 10)) to be msecs_to_jiffies(max(timeout/1000, 10ul)).

If you think these changes are not enough, please let me know, Thanks!

Best Regards,
Bayi

Quoted 15 lines
Quoted 10 lines

Aside from that, your timeout variable contains a timeout in microseconds and
this means that actually using msecs_to_jiffies() is suboptimal here.

Please use usecs_to_jiffies() instead.

Regards,
Angelo

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)

#5 lore raw
Quoted 86 lines · Bayi Cheng (程八意)
Quoted 67 lines · David Laight

From: AngeloGioacchino Del Regno

Quoted 47 lines

Sent: 03 November 2022 09:44

Il 03/11/22 06:28, Bayi Cheng ha scritto:

Quoted 44 lines

From: bayi cheng <[email protected]>

The timeout value of the current dma read is unreasonable. For example,
If the spi flash clock is 26Mhz, It will takes about 1.3ms to read a
4KB data in spi mode. But the actual measurement exceeds 50s when
a
dma read timeout is encountered.

In order to be more accurately, It is necessary to use msecs_to_jiffies,
After modification, the measured timeout value is about 130ms.

---
  drivers/spi/spi-mtk-nor.c | 7 ++++---
  1 file changed, 4 insertions(+), 3 deletions(-)
drivers/spi/spi-mtk- renamed from drivers/spi/spi-mtk-nor.c +0 -0

nor.c
index d167699a1a96..3d989db80ee9 100644

drivers/spi/spi-mtk-nor.c +1 -1
@@ -354,7 +354,7 @@ static int mtk_nor_dma_exec(struct mtk_nor
 *sp, u32 from, unsigned int length,
  			    dma_addr_t dma_addr)
  {
  	int ret = 0;
-	ulong delay;
+	ulong delay, timeout;
  	u32 reg;
 
  	writel(from, sp->base + MTK_NOR_REG_DMA_FADR);
drivers/spi/spi-mtk-nor.c +0 -0
@@ -376,15 +376,16 @@ static int mtk_nor_dma_exec(struct

mtk_nor
*sp, u32 from, unsigned int length,
mtk_nor_rmw(sp, MTK_NOR_REG_DMA_CTL, MTK_NOR_DMA_START,
0);

  	delay = CLK_TO_US(sp, (length + 5) * BITS_PER_BYTE);
drivers/spi/spi-mtk-nor.c +1 -0
+	timeout = (delay + 1) * 100;
Quoted 12 lines
Quoted 7 lines
  	if (sp->has_irq) {
  		if (!wait_for_completion_timeout(&sp->op_done,
drivers/spi/spi-mtk-nor.c +0 -1
-						 (delay + 1) *

100))

drivers/spi/spi-mtk-nor.c +1 -0
+		    msecs_to_jiffies(max_t(size_t, timeout /

1000,
10))))

You're giving a size_t variable to msecs_to_jiffies(), but checking jiffies.h,
this function takes a const unsigned int param. Please change the type to match that.

The type shouldn't matter at all.
What matters is the domain of the value.

Quite why you need to use max_t(size_t, ...) is another matter. timeout is ulong so max(timeout/1000, 10ul) should be fine.

Quoted 7 lines

But is ulong even right?
The domain of the value is almost certainly the same on 32bit and 64bit.
So you almost certainly need u32 or u64.

David

Hi David & Angelo

Thank you for your comments!
To sum up, I think the next version will make the following two changes:
1, The timeout value will not exceed u32, so the type of timeout will be changed from ulong to u32.
2, Change msecs_to_jiffies(max_t(size_t, timeout / 1000, 10)) to be msecs_to_jiffies(max(timeout / 1000, 10U)).

Quoted 6 lines

If you think these changes are not enough, please let me know, Thanks!

Best Regards,
Bayi

Hi Angelo, Hi David,

Just a gentle ping on this.
Could you please review this patch and give us some suggestion?

PS: With your permission, I will make the following changes in the next version:

Change in v2:
-Change the type of "timeout" from ulong to u32. -Replace max_t with max.

Thanks.

BRs,
Bayi Cheng

Quoted 16 lines
Quoted 16 lines
Quoted 10 lines

Aside from that, your timeout variable contains a timeout in microseconds and
this means that actually using msecs_to_jiffies() is suboptimal here.

Please use usecs_to_jiffies() instead.

Regards,
Angelo

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes,
MK1 1PT, UK
Registration No: 1397386 (Wales)

#6 lore raw

Il 11/11/22 05:16, Bayi Cheng (程八意) ha scritto:

Quoted 99 lines · Bayi Cheng (程八意)
Quoted 86 lines · David Laight
Quoted 67 lines

From: AngeloGioacchino Del Regno

Quoted 47 lines

Sent: 03 November 2022 09:44

Il 03/11/22 06:28, Bayi Cheng ha scritto:

Quoted 44 lines

From: bayi cheng <[email protected]>

The timeout value of the current dma read is unreasonable. For example,
If the spi flash clock is 26Mhz, It will takes about 1.3ms to read a
4KB data in spi mode. But the actual measurement exceeds 50s when
a
dma read timeout is encountered.

In order to be more accurately, It is necessary to use msecs_to_jiffies,
After modification, the measured timeout value is about 130ms.

---
   drivers/spi/spi-mtk-nor.c | 7 ++++---
   1 file changed, 4 insertions(+), 3 deletions(-)
drivers/spi/spi-mtk- renamed from drivers/spi/spi-mtk-nor.c +0 -0

nor.c
index d167699a1a96..3d989db80ee9 100644

drivers/spi/spi-mtk-nor.c +1 -1
@@ -354,7 +354,7 @@ static int mtk_nor_dma_exec(struct mtk_nor
 *sp, u32 from, unsigned int length,
   			    dma_addr_t dma_addr)
   {
   	int ret = 0;
-	ulong delay;
+	ulong delay, timeout;
   	u32 reg;
 
   	writel(from, sp->base + MTK_NOR_REG_DMA_FADR);
drivers/spi/spi-mtk-nor.c +0 -0
@@ -376,15 +376,16 @@ static int mtk_nor_dma_exec(struct

mtk_nor
*sp, u32 from, unsigned int length,
mtk_nor_rmw(sp, MTK_NOR_REG_DMA_CTL, MTK_NOR_DMA_START,
0);

   	delay = CLK_TO_US(sp, (length + 5) * BITS_PER_BYTE);
drivers/spi/spi-mtk-nor.c +1 -0
+	timeout = (delay + 1) * 100;
Quoted 12 lines
Quoted 7 lines
   	if (sp->has_irq) {
   		if (!wait_for_completion_timeout(&sp->op_done,
drivers/spi/spi-mtk-nor.c +0 -1
-						 (delay + 1) *

100))

drivers/spi/spi-mtk-nor.c +1 -0
+		    msecs_to_jiffies(max_t(size_t, timeout /

1000,
10))))

You're giving a size_t variable to msecs_to_jiffies(), but checking jiffies.h,
this function takes a const unsigned int param. Please change the type to match that.

The type shouldn't matter at all.
What matters is the domain of the value.

Quite why you need to use max_t(size_t, ...) is another matter. timeout is ulong so max(timeout/1000, 10ul) should be fine.

Quoted 7 lines

But is ulong even right?
The domain of the value is almost certainly the same on 32bit and 64bit.
So you almost certainly need u32 or u64.

David

Hi David & Angelo

Thank you for your comments!
To sum up, I think the next version will make the following two changes:
1, The timeout value will not exceed u32, so the type of timeout will be changed from ulong to u32.
2, Change msecs_to_jiffies(max_t(size_t, timeout / 1000, 10)) to be msecs_to_jiffies(max(timeout / 1000, 10U)).

Quoted 6 lines

If you think these changes are not enough, please let me know, Thanks!

Best Regards,
Bayi

Hi Angelo, Hi David,

Just a gentle ping on this.
Could you please review this patch and give us some suggestion?

Quoted 7 lines

PS: With your permission, I will make the following changes in the next version:

Change in v2:
-Change the type of "timeout" from ulong to u32. -Replace max_t with max.

I still recommend to use usecs_to_jiffies() when appropriate, instead of converting usecs to msecs and using msecs_to_jiffies().

As for the rest in your list: yes, please.

Regards,
Angelo

Quoted 22 lines

Thanks.

BRs,
Bayi Cheng

Quoted 16 lines
Quoted 16 lines
Quoted 10 lines

Aside from that, your timeout variable contains a timeout in microseconds and
this means that actually using msecs_to_jiffies() is suboptimal here.

Please use usecs_to_jiffies() instead.

Regards,
Angelo

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes,
MK1 1PT, UK
Registration No: 1397386 (Wales)

Signature · 6 lines
AngeloGioacchino Del Regno
Software Engineer

Collabora Ltd.
Platinum Building, St John's Innovation Park, Cambridge CB4 0DS, UK
Registered in England & Wales, no. 5513718
#7 lore raw
Quoted 118 lines · AngeloGioacchino Del Regno

Il 11/11/22 05:16, Bayi Cheng (程八意) ha scritto:

Quoted 108 lines · Bayi Cheng (程八意)
Quoted 95 lines · David Laight
Quoted 73 lines

From: AngeloGioacchino Del Regno

Quoted 53 lines

Sent: 03 November 2022 09:44

Il 03/11/22 06:28, Bayi Cheng ha scritto:

Quoted 50 lines

From: bayi cheng <[email protected]>

The timeout value of the current dma read is unreasonable. For
example,
If the spi flash clock is 26Mhz, It will takes about 1.3ms to
read a
4KB data in spi mode. But the actual measurement exceeds 50s
when
a
dma read timeout is encountered.

In order to be more accurately, It is necessary to use msecs_to_jiffies,
After modification, the measured timeout value is about 130ms.

---
   drivers/spi/spi-mtk-nor.c | 7 ++++---
   1 file changed, 4 insertions(+), 3 deletions(-)
drivers/spi/spi- renamed from drivers/spi/spi-mtk-nor.c +0 -0

mtk-
nor.c
index d167699a1a96..3d989db80ee9 100644

drivers/spi/spi-mtk-nor.c +0 -0
@@ -354,7 +354,7 @@ static int mtk_nor_dma_exec(struct

mtk_nor
*sp, u32 from, unsigned int length,
dma_addr_t dma_addr)

   {
   	int ret = 0;
drivers/spi/spi-mtk-nor.c +1 -1
-	ulong delay;
+	ulong delay, timeout;
   	u32 reg;
 
   	writel(from, sp->base + MTK_NOR_REG_DMA_FADR);
@@ -376,15 +376,16 @@ static int mtk_nor_dma_exec(struct

mtk_nor
*sp, u32 from, unsigned int length,
mtk_nor_rmw(sp, MTK_NOR_REG_DMA_CTL, MTK_NOR_DMA_START,
0);

   	delay = CLK_TO_US(sp, (length + 5) * BITS_PER_BYTE);
drivers/spi/spi-mtk-nor.c +1 -0
+	timeout = (delay + 1) * 100;
Quoted 12 lines
Quoted 7 lines
   	if (sp->has_irq) {
   		if (!wait_for_completion_timeout(&sp->op_done,
drivers/spi/spi-mtk-nor.c +0 -1
-						 (delay + 1) *

100))

drivers/spi/spi-mtk-nor.c +1 -0
+		    msecs_to_jiffies(max_t(size_t, timeout /

1000,
10))))

You're giving a size_t variable to msecs_to_jiffies(), but checking jiffies.h,
this function takes a const unsigned int param. Please change the type to match that.

The type shouldn't matter at all.
What matters is the domain of the value.

Quite why you need to use max_t(size_t, ...) is another matter. timeout is ulong so max(timeout/1000, 10ul) should be fine.

Quoted 8 lines

But is ulong even right?
The domain of the value is almost certainly the same on 32bit and
64bit.
So you almost certainly need u32 or u64.

David

Hi David & Angelo

Thank you for your comments!
To sum up, I think the next version will make the following two changes:
1, The timeout value will not exceed u32, so the type of timeout will
be changed from ulong to u32.
2, Change msecs_to_jiffies(max_t(size_t, timeout / 1000, 10)) to be
msecs_to_jiffies(max(timeout / 1000, 10U)).

Quoted 6 lines

If you think these changes are not enough, please let me know, Thanks!

Best Regards,
Bayi

Hi Angelo, Hi David,

Just a gentle ping on this.
Could you please review this patch and give us some suggestion?

Quoted 8 lines

PS: With your permission, I will make the following changes in the next
version:

Change in v2:
-Change the type of "timeout" from ulong to u32. -Replace max_t with max.

Quoted 9 lines

I still recommend to use usecs_to_jiffies() when appropriate, instead of
converting usecs to msecs and using msecs_to_jiffies().

As for the rest in your list: yes, please.

Regards,
Angelo

Hi Angelo,

Thanks for your comments!
I will change it to be usecs_to_jiffies(max(timeout, 10000U)) in the next patch.

BRs,
Bayi Cheng

Quoted 26 lines
Quoted 24 lines

Thanks.

BRs,
Bayi Cheng

Quoted 18 lines
Quoted 18 lines
Quoted 8 lines

Aside from that, your timeout variable contains a timeout in
microseconds and
this means that actually using msecs_to_jiffies() is suboptimal
here.

Quoted 4 lines

Please use usecs_to_jiffies() instead.

Regards,
Angelo

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes,
MK1 1PT, UK
Registration No: 1397386 (Wales)