Skip to contentSkip to conversation

staging: vc04_services: replace g_free_fragments_mutex with spinlock

PATCH Superseded Davidlohr Bueso 4 participants 11 messages Oct 27, 2019 – Nov 2, 2019 1 message not in the archive

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 staging: vc04_services: replace g_free_fragments_mutex with spinlock

There seems no need to be using a semaphore, or a sleeping lock in the first place: critical region is extremely short, does not call into any blocking calls and furthermore lock and unlocking operations occur in the same context.

Get rid of another semaphore user by replacing it with a spinlock.

---
This is in an effort to further reduce semaphore users in the kernel.

 .../staging/vc04_services/interface/vchiq_arm/vchiq_2835_arm.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)
drivers/staging/vc04_services/interface/vchiq_arm/vchiq_2835_arm.c +5 -5
index 8dc730cfe7a6..710d21654128 100644
@@ -63,7 +63,7 @@ static char *g_free_fragments;
 static struct semaphore g_free_fragments_sema;
 static struct device *g_dev;
 
-static DEFINE_SEMAPHORE(g_free_fragments_mutex);
+static DEFINE_SPINLOCK(g_free_fragments_lock);
 
 static irqreturn_t
 vchiq_doorbell_irq(int irq, void *dev_id);
@@ -528,11 +528,11 @@ create_pagelist(char __user *buf, size_t count, unsigned short type)
 
 		WARN_ON(g_free_fragments == NULL);
 
-		down(&g_free_fragments_mutex);
+		spin_lock(&g_free_fragments_lock);
 		fragments = g_free_fragments;
 		WARN_ON(fragments == NULL);
 		g_free_fragments = *(char **) g_free_fragments;
-		up(&g_free_fragments_mutex);
+		spin_unlock(&g_free_fragments_lock);
 		pagelist->type = PAGELIST_READ_WITH_FRAGMENTS +
 			(fragments - g_fragments_base) / g_fragments_size;
 	}
@@ -591,10 +591,10 @@ free_pagelist(struct vchiq_pagelist_info *pagelistinfo,
 			kunmap(pages[num_pages - 1]);
 		}
 
-		down(&g_free_fragments_mutex);
+		spin_lock(&g_free_fragments_lock);
 		*(char **)fragments = g_free_fragments;
 		g_free_fragments = fragments;
-		up(&g_free_fragments_mutex);
+		spin_unlock(&g_free_fragments_lock);
 		up(&g_free_fragments_sema);
 	}
 
Signature · 1 line
2.16.4

A message that is not in the archive (sent off-list or not yet received): [email protected]

#2 lore raw
Quoted 8 lines · Stefan Wahren

Hi Davidlohr,

Am 27.10.19 um 23:15 schrieb Davidlohr Bueso:

Quoted 5 lines

There seems no need to be using a semaphore, or a sleeping lock in the first place: critical region is extremely short, does not call into any blocking calls and furthermore lock and unlocking operations occur in the same context.

Quoted 7 lines
Quoted 6 lines

Get rid of another semaphore user by replacing it with a spinlock.

---
This is in an effort to further reduce semaphore users in the kernel.

thanks for this. Could please also send this to [email protected]?

Ccing.

Quoted 2 lines

I only need to know, has this been tested on the Raspberry Pi?

No testing has been done, I have no hardware to test this.

Thanks,
Davidlohr

maintainer
#3 lore raw
Quoted 12 lines · Davidlohr Bueso
Quoted 10 lines · Stefan Wahren

Hi Davidlohr,

Am 27.10.19 um 23:15 schrieb Davidlohr Bueso:

Quoted 7 lines

There seems no need to be using a semaphore, or a sleeping lock in the first place: critical region is extremely short, does not call into any blocking calls and furthermore lock and unlocking operations occur in the same context.

Get rid of another semaphore user by replacing it with a spinlock.

Quoted 7 lines
Quoted 5 lines
Quoted 4 lines

---
This is in an effort to further reduce semaphore users in the kernel.

thanks for this. Could please also send this to [email protected]?

Ccing.

I don't see a patch here :(

#4 lore raw
In reply to Davidlohr Bueso

Cc [email protected]

Quoted 57 lines · Bueso

There seems no need to be using a semaphore, or a sleeping lock in the first place: critical region is extremely short, does not call into any blocking calls and furthermore lock and unlocking operations occur in the same context.

Get rid of another semaphore user by replacing it with a spinlock.

---
This is in an effort to further reduce semaphore users in the kernel.

.../staging/vc04_services/interface/vchiq_arm/vchiq_2835_arm.c | 10 +++++-----

1 file changed, 5 insertions(+), 5 deletions(-)
drivers/staging/vc04_services/interface/vchiq_arm/vchiq_2835_arm.c +1 -1
index 8dc730cfe7a6..710d21654128 100644
@@ -63,7 +63,7 @@ static char *g_free_fragments;
 static struct semaphore g_free_fragments_sema;
 static struct device *g_dev;
 
-static DEFINE_SEMAPHORE(g_free_fragments_mutex);
+static DEFINE_SPINLOCK(g_free_fragments_lock);
 

static irqreturn_t
vchiq_doorbell_irq(int irq, void *dev_id);

drivers/staging/vc04_services/interface/vchiq_arm/vchiq_2835_arm.c +4 -4
@@ -528,11 +528,11 @@ create_pagelist(char __user *buf, size_t count, unsigned short type)
 
 		WARN_ON(g_free_fragments == NULL);
 
-		down(&g_free_fragments_mutex);
+		spin_lock(&g_free_fragments_lock);
 		fragments = g_free_fragments;
 		WARN_ON(fragments == NULL);
 		g_free_fragments = *(char **) g_free_fragments;
-		up(&g_free_fragments_mutex);
+		spin_unlock(&g_free_fragments_lock);
 		pagelist->type = PAGELIST_READ_WITH_FRAGMENTS +
 			(fragments - g_fragments_base) / g_fragments_size;
 	}
@@ -591,10 +591,10 @@ free_pagelist(struct vchiq_pagelist_info *pagelistinfo,
 			kunmap(pages[num_pages - 1]);
 		}
 
-		down(&g_free_fragments_mutex);
+		spin_lock(&g_free_fragments_lock);
 		*(char **)fragments = g_free_fragments;
 		g_free_fragments = fragments;
-		up(&g_free_fragments_mutex);
+		spin_unlock(&g_free_fragments_lock);
 		up(&g_free_fragments_sema);
 	}
 
Signature · 1 line
2.16.4
maintainer
#5 lore raw
Quoted 58 lines · Davidlohr Bueso

Cc [email protected]

Quoted 54 lines · Bueso

There seems no need to be using a semaphore, or a sleeping lock in the first place: critical region is extremely short, does not call into any blocking calls and furthermore lock and unlocking operations occur in the same context.

Get rid of another semaphore user by replacing it with a spinlock.

---
This is in an effort to further reduce semaphore users in the kernel.

.../staging/vc04_services/interface/vchiq_arm/vchiq_2835_arm.c | 10 +++++-----

1 file changed, 5 insertions(+), 5 deletions(-)
drivers/staging/vc04_services/interface/vchiq_arm/vchiq_2835_arm.c +1 -1
index 8dc730cfe7a6..710d21654128 100644
@@ -63,7 +63,7 @@ static char *g_free_fragments;
 static struct semaphore g_free_fragments_sema;
 static struct device *g_dev;
 
-static DEFINE_SEMAPHORE(g_free_fragments_mutex);
+static DEFINE_SPINLOCK(g_free_fragments_lock);
 

static irqreturn_t
vchiq_doorbell_irq(int irq, void *dev_id);

drivers/staging/vc04_services/interface/vchiq_arm/vchiq_2835_arm.c +4 -4
@@ -528,11 +528,11 @@ create_pagelist(char __user *buf, size_t count, unsigned short type)
 
 		WARN_ON(g_free_fragments == NULL);
 
-		down(&g_free_fragments_mutex);
+		spin_lock(&g_free_fragments_lock);
 		fragments = g_free_fragments;
 		WARN_ON(fragments == NULL);
 		g_free_fragments = *(char **) g_free_fragments;
-		up(&g_free_fragments_mutex);
+		spin_unlock(&g_free_fragments_lock);
 		pagelist->type = PAGELIST_READ_WITH_FRAGMENTS +
 			(fragments - g_fragments_base) / g_fragments_size;
 	}
@@ -591,10 +591,10 @@ free_pagelist(struct vchiq_pagelist_info *pagelistinfo,
 			kunmap(pages[num_pages - 1]);
 		}
 
-		down(&g_free_fragments_mutex);
+		spin_lock(&g_free_fragments_lock);
 		*(char **)fragments = g_free_fragments;
 		g_free_fragments = fragments;
-		up(&g_free_fragments_mutex);
+		spin_unlock(&g_free_fragments_lock);
 		up(&g_free_fragments_sema);
 	}
 
Quoted 3 lines
Quoted 3 lines
Signature · 1 line
2.16.4

This is obviously not in a format I can apply it in :(

#6 lore raw
Quoted 1 line · Greg KH

This is obviously not in a format I can apply it in :(

What are you talking about? I sent you the original patch, then Cc'ed the drivers mailing list. So you still have a patch you can apply... this is quite a common way of doing things (Ccing for future references to someone or another ml). I don't understand why you are hairsplitting over this patch.

Thanks,
Davidlohr

maintainer
#7 lore raw
Quoted 9 lines · Davidlohr Bueso
Quoted 1 line · Greg KH

This is obviously not in a format I can apply it in :(

What are you talking about? I sent you the original patch, then Cc'ed the drivers mailing list. So you still have a patch you can apply... this is quite a common way of doing things (Ccing for future references to someone or another ml). I don't understand why you are hairsplitting over this patch.

I don't understand what is going on at all. Is this patch already applied? If not, then yes, I need it in a format I can apply it in. If it's already applied to my tree/branch, then there's no need to send it at all.

totally confused,

greg k-h

#8 lore raw

PATCH RESEND 1/1 staging: vc04_services: replace g_free_fragments_mutex with spinlock

There is no need to be using a semaphore, or a sleeping lock in the first place: critical region is extremely short, does not call into any blocking calls and furthermore lock and unlocking operations occur in the same context.

Get rid of another semaphore user by replacing it with a spinlock.

---
This is in an effort to further reduce semaphore users in the kernel.

This is a resend, which just seems simpler given the confusions.

 .../staging/vc04_services/interface/vchiq_arm/vchiq_2835_arm.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)
drivers/staging/vc04_services/interface/vchiq_arm/vchiq_2835_arm.c +5 -5
index 8dc730cfe7a6..710d21654128 100644
@@ -63,7 +63,7 @@ static char *g_free_fragments;
 static struct semaphore g_free_fragments_sema;
 static struct device *g_dev;
 
-static DEFINE_SEMAPHORE(g_free_fragments_mutex);
+static DEFINE_SPINLOCK(g_free_fragments_lock);
 
 static irqreturn_t
 vchiq_doorbell_irq(int irq, void *dev_id);
@@ -528,11 +528,11 @@ create_pagelist(char __user *buf, size_t count, unsigned short type)
 
 		WARN_ON(g_free_fragments == NULL);
 
-		down(&g_free_fragments_mutex);
+		spin_lock(&g_free_fragments_lock);
 		fragments = g_free_fragments;
 		WARN_ON(fragments == NULL);
 		g_free_fragments = *(char **) g_free_fragments;
-		up(&g_free_fragments_mutex);
+		spin_unlock(&g_free_fragments_lock);
 		pagelist->type = PAGELIST_READ_WITH_FRAGMENTS +
 			(fragments - g_fragments_base) / g_fragments_size;
 	}
@@ -591,10 +591,10 @@ free_pagelist(struct vchiq_pagelist_info *pagelistinfo,
 			kunmap(pages[num_pages - 1]);
 		}
 
-		down(&g_free_fragments_mutex);
+		spin_lock(&g_free_fragments_lock);
 		*(char **)fragments = g_free_fragments;
 		g_free_fragments = fragments;
-		up(&g_free_fragments_mutex);
+		spin_unlock(&g_free_fragments_lock);
 		up(&g_free_fragments_sema);
 	}
 
Signature · 1 line
2.16.4
#9 lore raw

Hi Davidlohr,

Am 01.11.19 um 19:29 schrieb Davidlohr Bueso:

Quoted 31 lines

There is no need to be using a semaphore, or a sleeping lock in the first place: critical region is extremely short, does not call into any blocking calls and furthermore lock and unlocking operations occur in the same context.

Get rid of another semaphore user by replacing it with a spinlock.

---
This is in an effort to further reduce semaphore users in the kernel.

This is a resend, which just seems simpler given the confusions.

 .../staging/vc04_services/interface/vchiq_arm/vchiq_2835_arm.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)
drivers/staging/vc04_services/interface/vchiq_arm/vchiq_2835_arm.c +1 -1
index 8dc730cfe7a6..710d21654128 100644
@@ -63,7 +63,7 @@ static char *g_free_fragments;
 static struct semaphore g_free_fragments_sema;
 static struct device *g_dev;
 
-static DEFINE_SEMAPHORE(g_free_fragments_mutex);
+static DEFINE_SPINLOCK(g_free_fragments_lock);
 
 static irqreturn_t
 vchiq_doorbell_irq(int irq, void *dev_id);
@@ -528,11 +528,11 @@ create_pagelist(char __user *buf, size_t count, unsigned short type)
 

drivers/staging/vc04_services/interface/vchiq_arm/vchiq_2835_arm.c create_pagelist(char __user *buf, size_t count, unsigned short type) · line 536

Quoted 10 lines
drivers/staging/vc04_services/interface/vchiq_arm/vchiq_2835_arm.c +2 -2
 		WARN_ON(g_free_fragments == NULL);
 
-		down(&g_free_fragments_mutex);
+		spin_lock(&g_free_fragments_lock);
 		fragments = g_free_fragments;
 		WARN_ON(fragments == NULL);
 		g_free_fragments = *(char **) g_free_fragments;
-		up(&g_free_fragments_mutex);
+		spin_unlock(&g_free_fragments_lock);
 

the reason why Greg cannot apply this patch is that you are using an old or the wrong git tree.

Please make sure you use the following one (or a mirror):

https://git.kernel.org/pub/scm/linux/kernel/git/gregkh/staging.git

and operate on the branch staging-next

Please fix this up and send a new version.

Regards
Stefan

maintainer
#10 lore raw
In reply to Davidlohr Bueso
Quoted 7 lines · Davidlohr Bueso

There is no need to be using a semaphore, or a sleeping lock in the first place: critical region is extremely short, does not call into any blocking calls and furthermore lock and unlocking operations occur in the same context.

Get rid of another semaphore user by replacing it with a spinlock.

Quoted 8 lines

---
This is in an effort to further reduce semaphore users in the kernel.

This is a resend, which just seems simpler given the confusions.

 .../staging/vc04_services/interface/vchiq_arm/vchiq_2835_arm.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)

This patch does not apply to my tree at all, what did you make it against?

Please fix up and resend.

thanks,

greg k-h

#11 lore raw
In reply to Davidlohr Bueso
Quoted 10 lines · Davidlohr Bueso
Quoted 1 line · Greg KH

This is obviously not in a format I can apply it in :(

What are you talking about? I sent you the original patch, then Cc'ed the drivers mailing list. So you still have a patch you can apply... this is quite a common way of doing things (Ccing for future references to someone or another ml). I don't understand why you are hairsplitting over this patch.

I don't have the original patch either. Only the corrupted one... Maybe you did it as html and it was rejected?

regards,
dan carpenter