Skip to contentSkip to conversation

ipmi: kcs_bmc: optimize the data buffers allocation

PATCH In review Haiyue Wang 3 participants 9 messages Mar 15, 2018 – Apr 13, 2018 IPMI

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ipmi: kcs_bmc: optimize the data buffers allocation

Allocate a continuous memory block for the three KCS data buffers with related index assignment.

---
 drivers/char/ipmi/kcs_bmc.c | 10 ++++++----
 1 file changed, 6 insertions(+), 4 deletions(-)
drivers/char/ipmi/kcs_bmc.c +6 -4
index fbfc05e..dc19c0d 100644
@@ -435,6 +435,7 @@ static const struct file_operations kcs_bmc_fops = {
 struct kcs_bmc *kcs_bmc_alloc(struct device *dev, int sizeof_priv, u32 channel)
 {
 	struct kcs_bmc *kcs_bmc;
+	void *buf;
 
 	kcs_bmc = devm_kzalloc(dev, sizeof(*kcs_bmc) + sizeof_priv, GFP_KERNEL);
 	if (!kcs_bmc)
@@ -448,11 +449,12 @@ struct kcs_bmc *kcs_bmc_alloc(struct device *dev, int sizeof_priv, u32 channel)
 	mutex_init(&kcs_bmc->mutex);
 	init_waitqueue_head(&kcs_bmc->queue);
 
-	kcs_bmc->data_in = devm_kmalloc(dev, KCS_MSG_BUFSIZ, GFP_KERNEL);
-	kcs_bmc->data_out = devm_kmalloc(dev, KCS_MSG_BUFSIZ, GFP_KERNEL);
-	kcs_bmc->kbuffer = devm_kmalloc(dev, KCS_MSG_BUFSIZ, GFP_KERNEL);
-	if (!kcs_bmc->data_in || !kcs_bmc->data_out || !kcs_bmc->kbuffer)
+	buf = devm_kmalloc_array(dev, 3, KCS_MSG_BUFSIZ, GFP_KERNEL);
+	if (!buf)
 		return NULL;
+	kcs_bmc->data_in  = buf;
+	kcs_bmc->data_out = buf + KCS_MSG_BUFSIZ;
+	kcs_bmc->kbuffer  = buf + KCS_MSG_BUFSIZ * 2;
 
 	kcs_bmc->miscdev.minor = MISC_DYNAMIC_MINOR;
 	kcs_bmc->miscdev.name = dev_name(dev);
Signature · 1 line
2.7.4
#1 lore raw

Hi Corey,

The 4.17 merge window is opened now, this patch is not yet in linux-next tree,

so it will be merged into 4.18 ?

Thanks & Regards,

Haiyue

Quoted 37 lines · Haiyue Wang

Allocate a continuous memory block for the three KCS data buffers with related index assignment.

---
  drivers/char/ipmi/kcs_bmc.c | 10 ++++++----
  1 file changed, 6 insertions(+), 4 deletions(-)
drivers/char/ipmi/kcs_bmc.c +6 -4
index fbfc05e..dc19c0d 100644
@@ -435,6 +435,7 @@ static const struct file_operations kcs_bmc_fops = {
  struct kcs_bmc *kcs_bmc_alloc(struct device *dev, int sizeof_priv, u32 channel)
  {
  	struct kcs_bmc *kcs_bmc;
+	void *buf;
 
  	kcs_bmc = devm_kzalloc(dev, sizeof(*kcs_bmc) + sizeof_priv, GFP_KERNEL);
  	if (!kcs_bmc)
@@ -448,11 +449,12 @@ struct kcs_bmc *kcs_bmc_alloc(struct device *dev, int sizeof_priv, u32 channel)
  	mutex_init(&kcs_bmc->mutex);
  	init_waitqueue_head(&kcs_bmc->queue);
 
-	kcs_bmc->data_in = devm_kmalloc(dev, KCS_MSG_BUFSIZ, GFP_KERNEL);
-	kcs_bmc->data_out = devm_kmalloc(dev, KCS_MSG_BUFSIZ, GFP_KERNEL);
-	kcs_bmc->kbuffer = devm_kmalloc(dev, KCS_MSG_BUFSIZ, GFP_KERNEL);
-	if (!kcs_bmc->data_in || !kcs_bmc->data_out || !kcs_bmc->kbuffer)
+	buf = devm_kmalloc_array(dev, 3, KCS_MSG_BUFSIZ, GFP_KERNEL);
+	if (!buf)
  		return NULL;
+	kcs_bmc->data_in  = buf;
+	kcs_bmc->data_out = buf + KCS_MSG_BUFSIZ;
+	kcs_bmc->kbuffer  = buf + KCS_MSG_BUFSIZ * 2;
 
  	kcs_bmc->miscdev.minor = MISC_DYNAMIC_MINOR;
  	kcs_bmc->miscdev.name = dev_name(dev);
#2 lore raw
Quoted 6 lines · Haiyue

Hi Corey,

The 4.17 merge window is opened now, this patch is not yet in linux-next tree,

so it will be merged into 4.18 ?

Yeah, this came in kind of late, and I had some other critical things I was having to focus on, so I've been kind of out of the loop.

If it's urgent, I can work on getting it into 4.17 later, but I'd rather wait on 4.18.

-corey

Quoted 41 lines

Thanks & Regards,

Haiyue

Quoted 35 lines · Haiyue Wang

Allocate a continuous memory block for the three KCS data buffers with related index assignment.

---
  drivers/char/ipmi/kcs_bmc.c | 10 ++++++----

  1 file changed, 6 insertions(+), 4 deletions(-)
drivers/char/ipmi/kcs_bmc.c +6 -3
index fbfc05e..dc19c0d 100644
@@ -435,6 +435,7 @@ static const struct file_operations kcs_bmc_fops = {
   struct kcs_bmc *kcs_bmc_alloc(struct device *dev, int sizeof_priv, u32 channel)
   {
       struct kcs_bmc *kcs_bmc;
+    void *buf;
         kcs_bmc = devm_kzalloc(dev, sizeof(*kcs_bmc) + sizeof_priv, GFP_KERNEL);
       if (!kcs_bmc)
@@ -448,11 +449,12 @@ struct kcs_bmc *kcs_bmc_alloc(struct device 
 *dev, int sizeof_priv, u32 channel)
       mutex_init(&kcs_bmc->mutex);
       init_waitqueue_head(&kcs_bmc->queue);
   -    kcs_bmc->data_in = devm_kmalloc(dev, KCS_MSG_BUFSIZ, GFP_KERNEL);
-    kcs_bmc->data_out = devm_kmalloc(dev, KCS_MSG_BUFSIZ, GFP_KERNEL);
-    kcs_bmc->kbuffer = devm_kmalloc(dev, KCS_MSG_BUFSIZ, GFP_KERNEL);
-    if (!kcs_bmc->data_in || !kcs_bmc->data_out || !kcs_bmc->kbuffer)
+    buf = devm_kmalloc_array(dev, 3, KCS_MSG_BUFSIZ, GFP_KERNEL);
+    if (!buf)
           return NULL;
+    kcs_bmc->data_in  = buf;
+    kcs_bmc->data_out = buf + KCS_MSG_BUFSIZ;
+    kcs_bmc->kbuffer  = buf + KCS_MSG_BUFSIZ * 2;
         kcs_bmc->miscdev.minor = MISC_DYNAMIC_MINOR;
       kcs_bmc->miscdev.name = dev_name(dev);
#3 lore raw

Just a small piece of cake, not so urgent. I just try to understand

the code commitment process, such as time etc. :)

Thanks!

BR,
Haiyue

Quoted 59 lines · Corey Minyard
Quoted 6 lines · Haiyue

Hi Corey,

The 4.17 merge window is opened now, this patch is not yet in linux-next tree,

so it will be merged into 4.18 ?

Yeah, this came in kind of late, and I had some other critical things I was having to focus on, so I've been kind of out of the loop.

If it's urgent, I can work on getting it into 4.17 later, but I'd rather wait on 4.18.

-corey

Quoted 42 lines

Thanks & Regards,

Haiyue

Quoted 36 lines · Haiyue Wang

Allocate a continuous memory block for the three KCS data buffers with related index assignment.

---
  drivers/char/ipmi/kcs_bmc.c | 10 ++++++----

  1 file changed, 6 insertions(+), 4 deletions(-)
drivers/char/ipmi/kcs_bmc.c +6 -3
index fbfc05e..dc19c0d 100644
@@ -435,6 +435,7 @@ static const struct file_operations kcs_bmc_fops 
 = {
   struct kcs_bmc *kcs_bmc_alloc(struct device *dev, int sizeof_priv, u32 channel)
   {
       struct kcs_bmc *kcs_bmc;
+    void *buf;
         kcs_bmc = devm_kzalloc(dev, sizeof(*kcs_bmc) + sizeof_priv, GFP_KERNEL);
       if (!kcs_bmc)
@@ -448,11 +449,12 @@ struct kcs_bmc *kcs_bmc_alloc(struct device 
 *dev, int sizeof_priv, u32 channel)
       mutex_init(&kcs_bmc->mutex);
       init_waitqueue_head(&kcs_bmc->queue);
   -    kcs_bmc->data_in = devm_kmalloc(dev, KCS_MSG_BUFSIZ, GFP_KERNEL);
-    kcs_bmc->data_out = devm_kmalloc(dev, KCS_MSG_BUFSIZ, GFP_KERNEL);
-    kcs_bmc->kbuffer = devm_kmalloc(dev, KCS_MSG_BUFSIZ, GFP_KERNEL);
-    if (!kcs_bmc->data_in || !kcs_bmc->data_out || !kcs_bmc->kbuffer)
+    buf = devm_kmalloc_array(dev, 3, KCS_MSG_BUFSIZ, GFP_KERNEL);
+    if (!buf)
           return NULL;
+    kcs_bmc->data_in  = buf;
+    kcs_bmc->data_out = buf + KCS_MSG_BUFSIZ;
+    kcs_bmc->kbuffer  = buf + KCS_MSG_BUFSIZ * 2;
         kcs_bmc->miscdev.minor = MISC_DYNAMIC_MINOR;
       kcs_bmc->miscdev.name = dev_name(dev);
#4 lore raw
In reply to Haiyue Wang
Quoted 2 lines · Haiyue Wang

Allocate a continuous memory block for the three KCS data buffers with related index assignment.

I'm finally getting to this.

Is there a reason you want to do this?  In general, it's better to not try to outsmart your base system.  Depending on the memory allocator, in this case, you might actually use more memory.  You probably won't use any less.

In the original case, you allocate three 1000 byte buffers, resulting in 3 1024 byte slab allocated.

In the changed case, you will allocate a 3000 byte buffer, resulting in a single 4096 byte slab allocation, wasting 1024 more bytes of memory.

-corey

Quoted 34 lines
---
  drivers/char/ipmi/kcs_bmc.c | 10 ++++++----
  1 file changed, 6 insertions(+), 4 deletions(-)
drivers/char/ipmi/kcs_bmc.c +6 -4
index fbfc05e..dc19c0d 100644
@@ -435,6 +435,7 @@ static const struct file_operations kcs_bmc_fops = {
  struct kcs_bmc *kcs_bmc_alloc(struct device *dev, int sizeof_priv, u32 channel)
  {
  	struct kcs_bmc *kcs_bmc;
+	void *buf;
 
  	kcs_bmc = devm_kzalloc(dev, sizeof(*kcs_bmc) + sizeof_priv, GFP_KERNEL);
  	if (!kcs_bmc)
@@ -448,11 +449,12 @@ struct kcs_bmc *kcs_bmc_alloc(struct device *dev, int sizeof_priv, u32 channel)
  	mutex_init(&kcs_bmc->mutex);
  	init_waitqueue_head(&kcs_bmc->queue);
 
-	kcs_bmc->data_in = devm_kmalloc(dev, KCS_MSG_BUFSIZ, GFP_KERNEL);
-	kcs_bmc->data_out = devm_kmalloc(dev, KCS_MSG_BUFSIZ, GFP_KERNEL);
-	kcs_bmc->kbuffer = devm_kmalloc(dev, KCS_MSG_BUFSIZ, GFP_KERNEL);
-	if (!kcs_bmc->data_in || !kcs_bmc->data_out || !kcs_bmc->kbuffer)
+	buf = devm_kmalloc_array(dev, 3, KCS_MSG_BUFSIZ, GFP_KERNEL);
+	if (!buf)
  		return NULL;
+	kcs_bmc->data_in  = buf;
+	kcs_bmc->data_out = buf + KCS_MSG_BUFSIZ;
+	kcs_bmc->kbuffer  = buf + KCS_MSG_BUFSIZ * 2;
 
  	kcs_bmc->miscdev.minor = MISC_DYNAMIC_MINOR;
  	kcs_bmc->miscdev.name = dev_name(dev);
#5 lore raw
Quoted 11 lines · Corey Minyard
Quoted 2 lines · Haiyue Wang

Allocate a continuous memory block for the three KCS data buffers with related index assignment.

I'm finally getting to this.

Is there a reason you want to do this?  In general, it's better to not try to outsmart your base system.  Depending on the memory allocator, in this case, you might actually use more memory.  You probably won't use any less.

I got this idea from another code review, but that patch allocates 30 more the same size memory block, reducing the devm_kmalloc call will be better. For KCS only have 3, may be the key point is memory waste.

Quoted 6 lines

In the original case, you allocate three 1000 byte buffers, resulting in 3 1024 byte slab allocated.

In the changed case, you will allocate a 3000 byte buffer, resulting in a single 4096 byte slab allocation, wasting 1024 more bytes of memory.

As the kcs has memory copy between in/out/kbuffer, put them in the same page will be better ? Such as the same TLB ? (Well, I just got this from book, no real experience of memory accessing performance. And also, I was told that using space to save the time. :-)).

Just my stupid thinking. I'm OK to drop this patch if it doesn't help with performance, or something else.

BR.
Haiyue

Quoted 36 lines

-corey

Quoted 32 lines

---
  drivers/char/ipmi/kcs_bmc.c | 10 ++++++----

  1 file changed, 6 insertions(+), 4 deletions(-)
drivers/char/ipmi/kcs_bmc.c +6 -3
index fbfc05e..dc19c0d 100644
@@ -435,6 +435,7 @@ static const struct file_operations kcs_bmc_fops = {
   struct kcs_bmc *kcs_bmc_alloc(struct device *dev, int sizeof_priv, u32 channel)
   {
       struct kcs_bmc *kcs_bmc;
+    void *buf;
         kcs_bmc = devm_kzalloc(dev, sizeof(*kcs_bmc) + sizeof_priv, GFP_KERNEL);
       if (!kcs_bmc)
@@ -448,11 +449,12 @@ struct kcs_bmc *kcs_bmc_alloc(struct device 
 *dev, int sizeof_priv, u32 channel)
       mutex_init(&kcs_bmc->mutex);
       init_waitqueue_head(&kcs_bmc->queue);
   -    kcs_bmc->data_in = devm_kmalloc(dev, KCS_MSG_BUFSIZ, GFP_KERNEL);
-    kcs_bmc->data_out = devm_kmalloc(dev, KCS_MSG_BUFSIZ, GFP_KERNEL);
-    kcs_bmc->kbuffer = devm_kmalloc(dev, KCS_MSG_BUFSIZ, GFP_KERNEL);
-    if (!kcs_bmc->data_in || !kcs_bmc->data_out || !kcs_bmc->kbuffer)
+    buf = devm_kmalloc_array(dev, 3, KCS_MSG_BUFSIZ, GFP_KERNEL);
+    if (!buf)
           return NULL;
+    kcs_bmc->data_in  = buf;
+    kcs_bmc->data_out = buf + KCS_MSG_BUFSIZ;
+    kcs_bmc->kbuffer  = buf + KCS_MSG_BUFSIZ * 2;
         kcs_bmc->miscdev.minor = MISC_DYNAMIC_MINOR;
       kcs_bmc->miscdev.name = dev_name(dev);
#6 lore raw

Hi Corey,

Since IPMI 2.0 just defined minimum, no maximum:

----

KCS/SMIC Input : Required: 40 bytes IPMI Message, minimum

KCS/SMIC Output : Required: 38 bytes IPMI Message, minimum

----

We can enlarge the block size for avoiding waste, and make our driver

support most worst message size case. And I think this patch make checking

simple (from 3 to 1), and the code clean, this is the biggest reason I want to

change. The TLB is just memory management study from book, no data to

support access improvement. :)

BR,

Haiyue

Quoted 73 lines · Haiyue
Quoted 11 lines · Corey Minyard
Quoted 2 lines · Haiyue Wang

Allocate a continuous memory block for the three KCS data buffers with related index assignment.

I'm finally getting to this.

Is there a reason you want to do this?  In general, it's better to not try to outsmart your base system.  Depending on the memory allocator, in this case, you might actually use more memory.  You probably won't use any less.

I got this idea from another code review, but that patch allocates 30 more the same size memory block, reducing the devm_kmalloc call will be better. For KCS only have 3, may be the key point is memory waste.

Quoted 6 lines

In the original case, you allocate three 1000 byte buffers, resulting in 3 1024 byte slab allocated.

In the changed case, you will allocate a 3000 byte buffer, resulting in a single 4096 byte slab allocation, wasting 1024 more bytes of memory.

As the kcs has memory copy between in/out/kbuffer, put them in the same page will be better ? Such as the same TLB ? (Well, I just got this from book, no real experience of memory accessing performance. And also, I was told that using space to save the time. :-)).

Just my stupid thinking. I'm OK to drop this patch if it doesn't help with performance, or something else.

BR.
Haiyue

Quoted 37 lines

-corey

Quoted 33 lines

---
  drivers/char/ipmi/kcs_bmc.c | 10 ++++++----

  1 file changed, 6 insertions(+), 4 deletions(-)
drivers/char/ipmi/kcs_bmc.c +6 -3
index fbfc05e..dc19c0d 100644
@@ -435,6 +435,7 @@ static const struct file_operations kcs_bmc_fops 
 = {
   struct kcs_bmc *kcs_bmc_alloc(struct device *dev, int sizeof_priv, u32 channel)
   {
       struct kcs_bmc *kcs_bmc;
+    void *buf;
         kcs_bmc = devm_kzalloc(dev, sizeof(*kcs_bmc) + sizeof_priv, GFP_KERNEL);
       if (!kcs_bmc)
@@ -448,11 +449,12 @@ struct kcs_bmc *kcs_bmc_alloc(struct device 
 *dev, int sizeof_priv, u32 channel)
       mutex_init(&kcs_bmc->mutex);
       init_waitqueue_head(&kcs_bmc->queue);
   -    kcs_bmc->data_in = devm_kmalloc(dev, KCS_MSG_BUFSIZ, GFP_KERNEL);
-    kcs_bmc->data_out = devm_kmalloc(dev, KCS_MSG_BUFSIZ, GFP_KERNEL);
-    kcs_bmc->kbuffer = devm_kmalloc(dev, KCS_MSG_BUFSIZ, GFP_KERNEL);
-    if (!kcs_bmc->data_in || !kcs_bmc->data_out || !kcs_bmc->kbuffer)
+    buf = devm_kmalloc_array(dev, 3, KCS_MSG_BUFSIZ, GFP_KERNEL);
+    if (!buf)
           return NULL;
+    kcs_bmc->data_in  = buf;
+    kcs_bmc->data_out = buf + KCS_MSG_BUFSIZ;
+    kcs_bmc->kbuffer  = buf + KCS_MSG_BUFSIZ * 2;
         kcs_bmc->miscdev.minor = MISC_DYNAMIC_MINOR;
       kcs_bmc->miscdev.name = dev_name(dev);
#7 lore raw
Quoted 10 lines · Haiyue

Hi Corey,

Since IPMI 2.0 just defined minimum, no maximum:

----

KCS/SMIC Input : Required: 40 bytes IPMI Message, minimum

KCS/SMIC Output : Required: 38 bytes IPMI Message, minimum

Yes, though there are practical maximums that are much smaller than 1000 bytes.

Quoted 11 lines

----

We can enlarge the block size for avoiding waste, and make our driver

support most worst message size case. And I think this patch make checking

simple (from 3 to 1), and the code clean, this is the biggest reason I want to

change. The TLB is just memory management study from book, no data to

support access improvement. :)

I would argue that the way it is now expresses the intent of the code better than one allocation split into three parts.  Expressing your intent is more important than the number of checks and a minuscule performance improvement.  For me it makes the code easier to understand.  If you had a tool that checked for out-of-bounds memory access, then a single allocation might not find an overrun between the parts.  Smaller allocations tend to result in less memory fragmentation.

My preference is to leave it as it is.  However, it's not that important, and if you really want this patch, I can include it.

Thanks,

-corey

Quoted 82 lines

BR,

Haiyue

Quoted 74 lines · Haiyue
Quoted 11 lines · Corey Minyard
Quoted 2 lines · Haiyue Wang

Allocate a continuous memory block for the three KCS data buffers with related index assignment.

I'm finally getting to this.

Is there a reason you want to do this?  In general, it's better to not try to outsmart your base system.  Depending on the memory allocator, in this case, you might actually use more memory.  You probably won't use any less.

I got this idea from another code review, but that patch allocates 30 more the same size memory block, reducing the devm_kmalloc call will be better. For KCS only have 3, may be the key point is memory waste.

Quoted 6 lines

In the original case, you allocate three 1000 byte buffers, resulting in 3 1024 byte slab allocated.

In the changed case, you will allocate a 3000 byte buffer, resulting in a single 4096 byte slab allocation, wasting 1024 more bytes of memory.

As the kcs has memory copy between in/out/kbuffer, put them in the same page will be better ? Such as the same TLB ? (Well, I just got this from book, no real experience of memory accessing performance. And also, I was told that using space to save the time. :-)).

Just my stupid thinking. I'm OK to drop this patch if it doesn't help with performance, or something else.

BR.
Haiyue

Quoted 38 lines

-corey

Quoted 34 lines

---
  drivers/char/ipmi/kcs_bmc.c | 10 ++++++----

  1 file changed, 6 insertions(+), 4 deletions(-)
drivers/char/ipmi/kcs_bmc.c +6 -3
index fbfc05e..dc19c0d 100644
@@ -435,6 +435,7 @@ static const struct file_operations 
 kcs_bmc_fops = {
   struct kcs_bmc *kcs_bmc_alloc(struct device *dev, int sizeof_priv, u32 channel)
   {
       struct kcs_bmc *kcs_bmc;
+    void *buf;
         kcs_bmc = devm_kzalloc(dev, sizeof(*kcs_bmc) + sizeof_priv, GFP_KERNEL);
       if (!kcs_bmc)
@@ -448,11 +449,12 @@ struct kcs_bmc *kcs_bmc_alloc(struct device 
 *dev, int sizeof_priv, u32 channel)
       mutex_init(&kcs_bmc->mutex);
       init_waitqueue_head(&kcs_bmc->queue);
   -    kcs_bmc->data_in = devm_kmalloc(dev, KCS_MSG_BUFSIZ, GFP_KERNEL);
-    kcs_bmc->data_out = devm_kmalloc(dev, KCS_MSG_BUFSIZ, 
 GFP_KERNEL);
-    kcs_bmc->kbuffer = devm_kmalloc(dev, KCS_MSG_BUFSIZ, GFP_KERNEL);
-    if (!kcs_bmc->data_in || !kcs_bmc->data_out || !kcs_bmc->kbuffer)
+    buf = devm_kmalloc_array(dev, 3, KCS_MSG_BUFSIZ, GFP_KERNEL);
+    if (!buf)
           return NULL;
+    kcs_bmc->data_in  = buf;
+    kcs_bmc->data_out = buf + KCS_MSG_BUFSIZ;
+    kcs_bmc->kbuffer  = buf + KCS_MSG_BUFSIZ * 2;
         kcs_bmc->miscdev.minor = MISC_DYNAMIC_MINOR;

      kcs_bmc->miscdev.name = dev_name(dev);

#8 lore raw
Quoted 27 lines · Corey Minyard
Quoted 10 lines · Haiyue

Hi Corey,

Since IPMI 2.0 just defined minimum, no maximum:

----

KCS/SMIC Input : Required: 40 bytes IPMI Message, minimum

KCS/SMIC Output : Required: 38 bytes IPMI Message, minimum

Yes, though there are practical maximums that are much smaller than 1000 bytes.

Quoted 11 lines

----

We can enlarge the block size for avoiding waste, and make our driver

support most worst message size case. And I think this patch make checking

simple (from 3 to 1), and the code clean, this is the biggest reason I want to

change. The TLB is just memory management study from book, no data to

support access improvement. :)

Quoted 8 lines

I would argue that the way it is now expresses the intent of the code better than one allocation split into three parts.  Expressing your intent is more important than the number of checks and a minuscule performance improvement.  For me it makes the code easier to understand.  If you had a tool that checked for out-of-bounds memory access, then a single allocation might not find an overrun between the parts.  Smaller allocations tend to result in less memory fragmentation.

When I wrote the commit, I felt that the message was not so professional, and the reason sounded weak. The driver development is a complex work, needs considering more things, not just one. Thanks for your patience.

Quoted 3 lines

My preference is to leave it as it is.  However, it's not that important, and if you really want this patch, I can include it.

So leave it as it is, abandon this patch. :-)

BTW, another patch about KCS BMC chip support:
https://lkml.org/lkml/2018/3/22/284
Look forward your reviewing, I've tried my best to make it better.

Quoted 89 lines

Thanks,

-corey

Quoted 84 lines

BR,

Haiyue

Quoted 76 lines · Haiyue
Quoted 11 lines · Corey Minyard
Quoted 2 lines · Haiyue Wang

Allocate a continuous memory block for the three KCS data buffers with related index assignment.

I'm finally getting to this.

Is there a reason you want to do this?  In general, it's better to not try to outsmart your base system.  Depending on the memory allocator, in this case, you might actually use more memory.  You probably won't use any less.

I got this idea from another code review, but that patch allocates 30 more the same size memory block, reducing the devm_kmalloc call will be better. For KCS only have 3, may be the key point is memory waste.

Quoted 6 lines

In the original case, you allocate three 1000 byte buffers, resulting in 3 1024 byte slab allocated.

In the changed case, you will allocate a 3000 byte buffer, resulting in a single 4096 byte slab allocation, wasting 1024 more bytes of memory.

As the kcs has memory copy between in/out/kbuffer, put them in the same page will be better ? Such as the same TLB ? (Well, I just got this from book, no real experience of memory accessing performance. And also, I was told that using space to save the time. :-)).

Just my stupid thinking. I'm OK to drop this patch if it doesn't help with performance, or something else.

BR.
Haiyue

Quoted 40 lines

-corey

Quoted 36 lines

---
  drivers/char/ipmi/kcs_bmc.c | 10 ++++++----

  1 file changed, 6 insertions(+), 4 deletions(-)
drivers/char/ipmi/kcs_bmc.c +6 -3
index fbfc05e..dc19c0d 100644
@@ -435,6 +435,7 @@ static const struct file_operations 
 kcs_bmc_fops = {
   struct kcs_bmc *kcs_bmc_alloc(struct device *dev, int sizeof_priv, u32 channel)
   {
       struct kcs_bmc *kcs_bmc;
+    void *buf;
         kcs_bmc = devm_kzalloc(dev, sizeof(*kcs_bmc) + sizeof_priv, GFP_KERNEL);
       if (!kcs_bmc)
@@ -448,11 +449,12 @@ struct kcs_bmc *kcs_bmc_alloc(struct device 
 *dev, int sizeof_priv, u32 channel)
       mutex_init(&kcs_bmc->mutex);
       init_waitqueue_head(&kcs_bmc->queue);
   -    kcs_bmc->data_in = devm_kmalloc(dev, KCS_MSG_BUFSIZ, GFP_KERNEL);
-    kcs_bmc->data_out = devm_kmalloc(dev, KCS_MSG_BUFSIZ, 
 GFP_KERNEL);
-    kcs_bmc->kbuffer = devm_kmalloc(dev, KCS_MSG_BUFSIZ, 
 GFP_KERNEL);
-    if (!kcs_bmc->data_in || !kcs_bmc->data_out || 
 !kcs_bmc->kbuffer)
+    buf = devm_kmalloc_array(dev, 3, KCS_MSG_BUFSIZ, GFP_KERNEL);
+    if (!buf)
           return NULL;
+    kcs_bmc->data_in  = buf;
+    kcs_bmc->data_out = buf + KCS_MSG_BUFSIZ;
+    kcs_bmc->kbuffer  = buf + KCS_MSG_BUFSIZ * 2;

        kcs_bmc->miscdev.minor = MISC_DYNAMIC_MINOR; kcs_bmc->miscdev.name = dev_name(dev);