change kmalloc into vmalloc for large memory allocations

change kmalloc into vmalloc for large memory allocations

[Wang, Yalin]

Hi 


I find there is some drivers use kmalloc to allocate large 
Memorys during module_init,  some can be changed to use vmalloc
To save some low mem, I add log in kernel to track ,
And list them here:

https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/drivers/usb/gadget/f_mass_storage.c?h=master#n2724

https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/sound/soc/soc-core.c?h=master#n3772

https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/net/netfilter/nf_conntrack_ftp.c?h=master#n603
https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/net/netfilter/nf_conntrack_h323_main.c?h=master#n1849
https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/net/netfilter/nf_conntrack_irc.c?h=master#n247
https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/net/netfilter/nf_conntrack_sane.c?h=master#n195

https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/drivers/input/evdev.c?h=master#n403


they allocate large memory from 10k~64K , 
this will use lots of low mem, instead if we use 
vmalloc for these drivers , it will be good for some devices
like smart phone, we often encounter some errors like kmalloc failed 
because there is not enough low mem , especially when the device has physical 
memory less than 1GB .


could this module change the memory allocation into vmalloc ?

I was thinking that if we can introduce a helper function in kmalloc.h like this :

Kmalloc(size, flags)
{
	If (size > PAGE_SIZE && flags&CAN_USE_VMALLOC_FLAG)
		return vmalloc(size);
	Else
		return real_kmalloc(size);
}

Kfree(ptr)
{
	If (is_vmalloc_addr(ptr))
		Vfree(ptr);
	Else
		Kfree(ptr);
}


But we need add some flags to ensure always use kmalloc for
Some special use (dma etc..)

How do you think of it ?

Thanks

change kmalloc into vmalloc for large memory allocations

[Huang Shijie]

? 2014?02?28? 16:15, Wang, Yalin ??:
> could this module change the memory allocation into vmalloc ?
the buffer allocated by vmalloc can not be used for the DMA.

I think that's why drivers use the kmalloc.

thanks
Huang Shijie

change kmalloc into vmalloc for large memory allocations

[Wang, Yalin]

Hi  


Yeah,  
Dma buffer must be allocated by kmalloc,

But the modules I list should can all be changed to use
vmalloc, because the buffer is only used by software,
Not by any hardware .


Thanks

-----Original Message-----
From: Huang Shijie [mailto:b32955 at freescale.com] 
Sent: Friday, February 28, 2014 4:55 PM
To: Wang, Yalin
Cc: 'linux-kernel at vger.kernel.org'; 'linux-arm-msm at vger.kernel.org'; 'linux-arm-kernel at lists.infradead.org'; 'linux-input at vger.kernel.org'; 'balbi at ti.com'; 'gregkh at linuxfoundation.org'; 'lrg at ti.com'; 'broonie at opensource.wolfsonmicro.com'; 'perex at perex.cz'; 'tiwai at suse.de'; 'pablo at netfilter.org'; 'kaber at trash.net'; 'davem at davemloft.net'; 'rostedt at goodmis.org'; 'fweisbec at gmail.com'; 'mingo at redhat.com'; 'dmitry.torokhov at gmail.com'; 'rydberg at euromail.se'; 'linux-usb at vger.kernel.org'; 'alsa-devel at alsa-project.org'; 'netfilter-devel at vger.kernel.org'; 'netfilter at vger.kernel.org'; 'coreteam at netfilter.org'; 'netdev at vger.kernel.org'
Subject: Re: change kmalloc into vmalloc for large memory allocations

? 2014?02?28? 16:15, Wang, Yalin ??:
> could this module change the memory allocation into vmalloc ?
the buffer allocated by vmalloc can not be used for the DMA.

I think that's why drivers use the kmalloc.

thanks
Huang Shijie

change kmalloc into vmalloc for large memory allocations

['gregkh@linuxfoundation.org']

On Fri, Feb 28, 2014 at 05:20:08PM +0800, Wang, Yalin wrote:
> Hi  
> 
> 
> Yeah,  
> Dma buffer must be allocated by kmalloc,
> 
> But the modules I list should can all be changed to use
> vmalloc, because the buffer is only used by software,
> Not by any hardware .

Are you sure about that?  The USB gadget driver needs DMA memory from
what I can tell, have you tried your change out on a system that does
not allow the USB controller to access non-DMA memory?

And I agree with Steve, just fix the individual drivers, don't do a
"hidden" change of where the memory is allocated from, that's not a good
idea and will cause problems later.

greg k-h

change kmalloc into vmalloc for large memory allocations

[Wang, Yalin]

Hi  greg,

I am sorry,
I make a mistake ,
You are right ,
https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/drivers/usb/gadget/f_mass_storage.c?h=master#n2724

this one should not changed to use vmalloc,
the buffer will be used by DMA?

others should be safe to change:

https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/sound/soc/soc-core.c?h=master#n3772

https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/net/netfilter/nf_conntrack_ftp.c?h=master#n603
https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/net/netfilter/nf_conntrack_h323_main.c?h=master#n1849
https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/net/netfilter/nf_conntrack_irc.c?h=master#n247
https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/net/netfilter/nf_conntrack_sane.c?h=master#n195

https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/drivers/input/evdev.c?h=master#n403


Thanks 
-----Original Message-----
From: 'gregkh@linuxfoundation.org' [mailto:gregkh at linuxfoundation.org] 
Sent: Saturday, March 01, 2014 12:33 AM
To: Wang, Yalin
Cc: 'Huang Shijie'; 'linux-kernel at vger.kernel.org'; 'linux-arm-msm at vger.kernel.org'; 'linux-arm-kernel at lists.infradead.org'; 'linux-input at vger.kernel.org'; 'balbi at ti.com'; 'lrg at ti.com'; 'broonie at opensource.wolfsonmicro.com'; 'perex at perex.cz'; 'tiwai at suse.de'; 'pablo at netfilter.org'; 'kaber at trash.net'; 'davem at davemloft.net'; 'rostedt at goodmis.org'; 'fweisbec at gmail.com'; 'mingo at redhat.com'; 'dmitry.torokhov at gmail.com'; 'rydberg at euromail.se'; 'linux-usb at vger.kernel.org'; 'alsa-devel at alsa-project.org'; 'netfilter-devel at vger.kernel.org'; 'netfilter at vger.kernel.org'; 'coreteam at netfilter.org'; 'netdev at vger.kernel.org'
Subject: Re: change kmalloc into vmalloc for large memory allocations

On Fri, Feb 28, 2014 at 05:20:08PM +0800, Wang, Yalin wrote:
> Hi
> 
> 
> Yeah,
> Dma buffer must be allocated by kmalloc,
> 
> But the modules I list should can all be changed to use vmalloc, 
> because the buffer is only used by software, Not by any hardware .

Are you sure about that?  The USB gadget driver needs DMA memory from what I can tell, have you tried your change out on a system that does not allow the USB controller to access non-DMA memory?

And I agree with Steve, just fix the individual drivers, don't do a "hidden" change of where the memory is allocated from, that's not a good idea and will cause problems later.

greg k-h

change kmalloc into vmalloc for large memory allocations

['gregkh@linuxfoundation.org']

On Mon, Mar 03, 2014 at 10:51:23AM +0800, Wang, Yalin wrote:
> Hi  greg,
> 
> I am sorry,
> I make a mistake ,
> You are right ,
> https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/drivers/usb/gadget/f_mass_storage.c?h=master#n2724
> 
> this one should not changed to use vmalloc,
> the buffer will be used by DMA?

Which is why a wrapper function will never work.

> others should be safe to change:
> 
> https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/sound/soc/soc-core.c?h=master#n3772
> 
> https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/net/netfilter/nf_conntrack_ftp.c?h=master#n603
> https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/net/netfilter/nf_conntrack_h323_main.c?h=master#n1849
> https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/net/netfilter/nf_conntrack_irc.c?h=master#n247
> https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/net/netfilter/nf_conntrack_sane.c?h=master#n195
> 
> https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/drivers/input/evdev.c?h=master#n403

Then send individual patches for these and see what happens.

greg k-h

change kmalloc into vmalloc for large memory allocations

[Wang, Yalin]

Hi  greg,

For 
https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/drivers/input/evdev.c?h=master#n403

there have been a patch for kmalloc failed ,

for 
https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/sound/soc/soc-core.c?h=master#n3772

I have not change it , need some more code to change in devm_kzalloc ..

I make a patch for netfilter part :

https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/net/netfilter/nf_conntrack_ftp.c?h=master#n603
https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/net/netfilter/nf_conntrack_h323_main.c?h=master#n1849
https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/net/netfilter/nf_conntrack_irc.c?h=master#n247
https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/net/netfilter/nf_conntrack_sane.c?h=master#n195

seems work well ,
these module will allocate 64KB large memory,
is it possible to be merged ?

Thanks 

-- >8 --
diff --git a/net/netfilter/nf_conntrack_ftp.c b/net/netfilter/nf_conntrack_ftp.c
index b8a0924..0e92b0d 100644
--- a/net/netfilter/nf_conntrack_ftp.c
+++ b/net/netfilter/nf_conntrack_ftp.c
@@ -14,7 +14,7 @@
 #include <linux/moduleparam.h>
 #include <linux/netfilter.h>
 #include <linux/ip.h>
-#include <linux/slab.h>
+#include <linux/vmalloc.h>
 #include <linux/ipv6.h>
 #include <linux/ctype.h>
 #include <linux/inet.h>
@@ -593,14 +593,14 @@ static void nf_conntrack_ftp_fini(void)
 		}
 	}
 
-	kfree(ftp_buffer);
+	vfree(ftp_buffer);
 }
 
 static int __init nf_conntrack_ftp_init(void)
 {
 	int i, j = -1, ret = 0;
 
-	ftp_buffer = kmalloc(65536, GFP_KERNEL);
+	ftp_buffer = vmalloc(65536, GFP_KERNEL);
 	if (!ftp_buffer)
 		return -ENOMEM;
 
diff --git a/net/netfilter/nf_conntrack_h323_main.c b/net/netfilter/nf_conntrack_h323_main.c
index 70866d1..49ae092 100644
--- a/net/netfilter/nf_conntrack_h323_main.c
+++ b/net/netfilter/nf_conntrack_h323_main.c
@@ -18,7 +18,7 @@
 #include <linux/inet.h>
 #include <linux/in.h>
 #include <linux/ip.h>
-#include <linux/slab.h>
+#include <linux/vmalloc.h>
 #include <linux/udp.h>
 #include <linux/tcp.h>
 #include <linux/skbuff.h>
@@ -1837,7 +1837,7 @@ static void __exit nf_conntrack_h323_fini(void)
 	nf_conntrack_helper_unregister(&nf_conntrack_helper_q931[1]);
 	nf_conntrack_helper_unregister(&nf_conntrack_helper_q931[0]);
 	nf_conntrack_helper_unregister(&nf_conntrack_helper_h245);
-	kfree(h323_buffer);
+	vfree(h323_buffer);
 	pr_debug("nf_ct_h323: fini\n");
 }
 
@@ -1846,7 +1846,7 @@ static int __init nf_conntrack_h323_init(void)
 {
 	int ret;
 
-	h323_buffer = kmalloc(65536, GFP_KERNEL);
+	h323_buffer = vmalloc(65536, GFP_KERNEL);
 	if (!h323_buffer)
 		return -ENOMEM;
 	ret = nf_conntrack_helper_register(&nf_conntrack_helper_h245);
@@ -1876,7 +1876,7 @@ err3:
 err2:
 	nf_conntrack_helper_unregister(&nf_conntrack_helper_h245);
 err1:
-	kfree(h323_buffer);
+	vfree(h323_buffer);
 	return ret;
 }
 
diff --git a/net/netfilter/nf_conntrack_irc.c b/net/netfilter/nf_conntrack_irc.c
index 0fd2976..b57df10 100644
--- a/net/netfilter/nf_conntrack_irc.c
+++ b/net/netfilter/nf_conntrack_irc.c
@@ -16,7 +16,7 @@
 #include <linux/ip.h>
 #include <linux/tcp.h>
 #include <linux/netfilter.h>
-#include <linux/slab.h>
+#include <linux/vmalloc.h>
 
 #include <net/netfilter/nf_conntrack.h>
 #include <net/netfilter/nf_conntrack_expect.h>
@@ -244,7 +244,7 @@ static int __init nf_conntrack_irc_init(void)
 	irc_exp_policy.max_expected = max_dcc_channels;
 	irc_exp_policy.timeout = dcc_timeout;
 
-	irc_buffer = kmalloc(65536, GFP_KERNEL);
+	irc_buffer = vmalloc(65536);
 	if (!irc_buffer)
 		return -ENOMEM;
 
@@ -285,7 +285,7 @@ static void nf_conntrack_irc_fini(void)
 
 	for (i = 0; i < ports_c; i++)
 		nf_conntrack_helper_unregister(&irc[i]);
-	kfree(irc_buffer);
+	vfree(irc_buffer);
 }
 
 module_init(nf_conntrack_irc_init);
diff --git a/net/netfilter/nf_conntrack_sane.c b/net/netfilter/nf_conntrack_sane.c
index 4a2134f..a4c8bf3 100644
--- a/net/netfilter/nf_conntrack_sane.c
+++ b/net/netfilter/nf_conntrack_sane.c
@@ -20,7 +20,7 @@
 #include <linux/module.h>
 #include <linux/moduleparam.h>
 #include <linux/netfilter.h>
-#include <linux/slab.h>
+#include <linux/vmalloc.h>
 #include <linux/in.h>
 #include <linux/tcp.h>
 #include <net/netfilter/nf_conntrack.h>
@@ -185,14 +185,14 @@ static void nf_conntrack_sane_fini(void)
 		}
 	}
 
-	kfree(sane_buffer);
+	vfree(sane_buffer);
 }
 
 static int __init nf_conntrack_sane_init(void)
 {
 	int i, j = -1, ret = 0;
 
-	sane_buffer = kmalloc(65536, GFP_KERNEL);
+	sane_buffer = vmalloc(65536);
 	if (!sane_buffer)
 		return -ENOMEM;



-----Original Message-----
From: 'gregkh@linuxfoundation.org' [mailto:gregkh at linuxfoundation.org] 
Sent: Monday, March 03, 2014 11:08 AM
To: Wang, Yalin
Cc: 'Huang Shijie'; 'linux-kernel at vger.kernel.org'; 'linux-arm-msm at vger.kernel.org'; 'linux-arm-kernel at lists.infradead.org'; 'linux-input at vger.kernel.org'; 'balbi at ti.com'; 'lrg at ti.com'; 'broonie at opensource.wolfsonmicro.com'; 'perex at perex.cz'; 'tiwai at suse.de'; 'pablo at netfilter.org'; 'kaber at trash.net'; 'davem at davemloft.net'; 'rostedt at goodmis.org'; 'fweisbec at gmail.com'; 'mingo at redhat.com'; 'dmitry.torokhov at gmail.com'; 'rydberg at euromail.se'; 'linux-usb at vger.kernel.org'; 'alsa-devel at alsa-project.org'; 'netfilter-devel at vger.kernel.org'; 'netfilter at vger.kernel.org'; 'coreteam at netfilter.org'; 'netdev at vger.kernel.org'
Subject: Re: change kmalloc into vmalloc for large memory allocations

On Mon, Mar 03, 2014 at 10:51:23AM +0800, Wang, Yalin wrote:
> Hi  greg,
> 
> I am sorry,
> I make a mistake ,
> You are right ,
> https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/drive
> rs/usb/gadget/f_mass_storage.c?h=master#n2724
> 
> this one should not changed to use vmalloc, the buffer will be used by 
> DMA?

Which is why a wrapper function will never work.

> others should be safe to change:
> 
> https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/sound
> /soc/soc-core.c?h=master#n3772
> 
> https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/net/n
> etfilter/nf_conntrack_ftp.c?h=master#n603
> https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/net/n
> etfilter/nf_conntrack_h323_main.c?h=master#n1849
> https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/net/n
> etfilter/nf_conntrack_irc.c?h=master#n247
> https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/net/n
> etfilter/nf_conntrack_sane.c?h=master#n195
> 
> https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/drive
> rs/input/evdev.c?h=master#n403

Then send individual patches for these and see what happens.

greg k-h

change kmalloc into vmalloc for large memory allocations

['gregkh@linuxfoundation.org']

On Mon, Mar 03, 2014 at 04:00:59PM +0800, Wang, Yalin wrote:
> Hi  greg,

Please read Documentation/SubmittingPatches for how to properly submit
kernel patches in a form that they could be accepted.

thanks,

greg k-h

[PATCH] netfilter:Change nf_conntrack modules to use vmalloc

[Wang, Yalin]

This patch change some nf_conntrack modules to
use vmalloc to allocate large memory instead of kmalloc,
to save low memorys.

Signed-off-by: Yalin Wang <yalin.wang@sonymobile.com>
---
 net/netfilter/nf_conntrack_ftp.c       | 6 +++---
 net/netfilter/nf_conntrack_h323_main.c | 8 ++++----
 net/netfilter/nf_conntrack_irc.c       | 6 +++---
 net/netfilter/nf_conntrack_sane.c      | 6 +++---
 4 files changed, 13 insertions(+), 13 deletions(-)

diff --git a/net/netfilter/nf_conntrack_ftp.c b/net/netfilter/nf_conntrack_ftp.c
index b8a0924..aad7eca 100644
--- a/net/netfilter/nf_conntrack_ftp.c
+++ b/net/netfilter/nf_conntrack_ftp.c
@@ -14,7 +14,7 @@
 #include <linux/moduleparam.h>
 #include <linux/netfilter.h>
 #include <linux/ip.h>
-#include <linux/slab.h>
+#include <linux/vmalloc.h>
 #include <linux/ipv6.h>
 #include <linux/ctype.h>
 #include <linux/inet.h>
@@ -593,14 +593,14 @@ static void nf_conntrack_ftp_fini(void)
 		}
 	}
 
-	kfree(ftp_buffer);
+	vfree(ftp_buffer);
 }
 
 static int __init nf_conntrack_ftp_init(void)
 {
 	int i, j = -1, ret = 0;
 
-	ftp_buffer = kmalloc(65536, GFP_KERNEL);
+	ftp_buffer = vmalloc(65536);
 	if (!ftp_buffer)
 		return -ENOMEM;
 
diff --git a/net/netfilter/nf_conntrack_h323_main.c b/net/netfilter/nf_conntrack_h323_main.c
index 70866d1..5baee1a 100644
--- a/net/netfilter/nf_conntrack_h323_main.c
+++ b/net/netfilter/nf_conntrack_h323_main.c
@@ -18,7 +18,7 @@
 #include <linux/inet.h>
 #include <linux/in.h>
 #include <linux/ip.h>
-#include <linux/slab.h>
+#include <linux/vmalloc.h>
 #include <linux/udp.h>
 #include <linux/tcp.h>
 #include <linux/skbuff.h>
@@ -1837,7 +1837,7 @@ static void __exit nf_conntrack_h323_fini(void)
 	nf_conntrack_helper_unregister(&nf_conntrack_helper_q931[1]);
 	nf_conntrack_helper_unregister(&nf_conntrack_helper_q931[0]);
 	nf_conntrack_helper_unregister(&nf_conntrack_helper_h245);
-	kfree(h323_buffer);
+	vfree(h323_buffer);
 	pr_debug("nf_ct_h323: fini\n");
 }
 
@@ -1846,7 +1846,7 @@ static int __init nf_conntrack_h323_init(void)
 {
 	int ret;
 
-	h323_buffer = kmalloc(65536, GFP_KERNEL);
+	h323_buffer = vmalloc(65536);
 	if (!h323_buffer)
 		return -ENOMEM;
 	ret = nf_conntrack_helper_register(&nf_conntrack_helper_h245);
@@ -1876,7 +1876,7 @@ err3:
 err2:
 	nf_conntrack_helper_unregister(&nf_conntrack_helper_h245);
 err1:
-	kfree(h323_buffer);
+	vfree(h323_buffer);
 	return ret;
 }
 
diff --git a/net/netfilter/nf_conntrack_irc.c b/net/netfilter/nf_conntrack_irc.c
index 0fd2976..b57df10 100644
--- a/net/netfilter/nf_conntrack_irc.c
+++ b/net/netfilter/nf_conntrack_irc.c
@@ -16,7 +16,7 @@
 #include <linux/ip.h>
 #include <linux/tcp.h>
 #include <linux/netfilter.h>
-#include <linux/slab.h>
+#include <linux/vmalloc.h>
 
 #include <net/netfilter/nf_conntrack.h>
 #include <net/netfilter/nf_conntrack_expect.h>
@@ -244,7 +244,7 @@ static int __init nf_conntrack_irc_init(void)
 	irc_exp_policy.max_expected = max_dcc_channels;
 	irc_exp_policy.timeout = dcc_timeout;
 
-	irc_buffer = kmalloc(65536, GFP_KERNEL);
+	irc_buffer = vmalloc(65536);
 	if (!irc_buffer)
 		return -ENOMEM;
 
@@ -285,7 +285,7 @@ static void nf_conntrack_irc_fini(void)
 
 	for (i = 0; i < ports_c; i++)
 		nf_conntrack_helper_unregister(&irc[i]);
-	kfree(irc_buffer);
+	vfree(irc_buffer);
 }
 
 module_init(nf_conntrack_irc_init);
diff --git a/net/netfilter/nf_conntrack_sane.c b/net/netfilter/nf_conntrack_sane.c
index 4a2134f..a4c8bf3 100644
--- a/net/netfilter/nf_conntrack_sane.c
+++ b/net/netfilter/nf_conntrack_sane.c
@@ -20,7 +20,7 @@
 #include <linux/module.h>
 #include <linux/moduleparam.h>
 #include <linux/netfilter.h>
-#include <linux/slab.h>
+#include <linux/vmalloc.h>
 #include <linux/in.h>
 #include <linux/tcp.h>
 #include <net/netfilter/nf_conntrack.h>
@@ -185,14 +185,14 @@ static void nf_conntrack_sane_fini(void)
 		}
 	}
 
-	kfree(sane_buffer);
+	vfree(sane_buffer);
 }
 
 static int __init nf_conntrack_sane_init(void)
 {
 	int i, j = -1, ret = 0;
 
-	sane_buffer = kmalloc(65536, GFP_KERNEL);
+	sane_buffer = vmalloc(65536);
 	if (!sane_buffer)
 		return -ENOMEM;
 
-- 
1.8.2.2

change kmalloc into vmalloc for large memory allocations

[Steven Rostedt]

On Fri, 28 Feb 2014 16:15:23 +0800
"Wang, Yalin" <Yalin.Wang@sonymobile.com> wrote:

> Kfree(ptr)
> {
> 	If (is_vmalloc_addr(ptr))
> 		Vfree(ptr);
> 	Else
> 		Kfree(ptr);
> }
> 
> 
> But we need add some flags to ensure always use kmalloc for
> Some special use (dma etc..)
> 
> How do you think of it ?

But vmalloc also takes up tlb entries. The more vmalloc space you have,
the more tlb entries that will need to be used, which will have a
performance affect on the entire system.

I would be against making kmalloc() secretly doing a vmalloc. If it is
better for a driver to use a vmalloc on large items, than change the
driver. Don't change a core function.

-- Steve

change kmalloc into vmalloc for large memory allocations

[Takashi Iwai]

At Fri, 28 Feb 2014 16:15:23 +0800,
Wang, Yalin wrote:
> 
> Hi 
> 
> 
> I find there is some drivers use kmalloc to allocate large 
> Memorys during module_init,  some can be changed to use vmalloc
> To save some low mem, I add log in kernel to track ,
> And list them here:
> 
> https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/drivers/usb/gadget/f_mass_storage.c?h=master#n2724
> 
> https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/sound/soc/soc-core.c?h=master#n3772

At least the ASoC runtime case doesn't use the allocated memory as
buffer, and they are allocated only once per device, thus it shouldn't
be the problem you stated.  If it really consumes so much memory, we
need to rethink, instead of allocating an array but allocate each
object, for example.


thanks,

Takashi


> https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/net/netfilter/nf_conntrack_ftp.c?h=master#n603
> https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/net/netfilter/nf_conntrack_h323_main.c?h=master#n1849
> https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/net/netfilter/nf_conntrack_irc.c?h=master#n247
> https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/net/netfilter/nf_conntrack_sane.c?h=master#n195
> 
> https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/drivers/input/evdev.c?h=master#n403
> 
> 
> they allocate large memory from 10k~64K , 
> this will use lots of low mem, instead if we use 
> vmalloc for these drivers , it will be good for some devices
> like smart phone, we often encounter some errors like kmalloc failed 
> because there is not enough low mem , especially when the device has physical 
> memory less than 1GB .
> 
> 
> could this module change the memory allocation into vmalloc ?
> 
> I was thinking that if we can introduce a helper function in kmalloc.h like this :
> 
> Kmalloc(size, flags)
> {
> 	If (size > PAGE_SIZE && flags&CAN_USE_VMALLOC_FLAG)
> 		return vmalloc(size);
> 	Else
> 		return real_kmalloc(size);
> }
> 
> Kfree(ptr)
> {
> 	If (is_vmalloc_addr(ptr))
> 		Vfree(ptr);
> 	Else
> 		Kfree(ptr);
> }
> 
> 
> But we need add some flags to ensure always use kmalloc for
> Some special use (dma etc..)
> 
> How do you think of it ?
> 
> Thanks
> 
> 
> 

change kmalloc into vmalloc for large memory allocations

[Wang, Yalin]

Hi  Takashi,

For this one:
https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/sound/soc/soc-core.c?h=master#n3772

in my kernel log,
the allocation size 62496,
about 62KB,
I think need to  change to vmalloc .
How do you think of it ?

Logs:
<4>[   11.469789] ------------[ cut here ]------------
<4>[   11.469803] WARNING: at /media/00DE7FE2DE7FCE82/jb-mr2-yukon/kernel/include/linux/slub_def.h:265 __kmalloc+0x38/0x29c()
<4>[   11.469811] size=62496 flags=80d0
<4>[   11.469815] Modules linked in:
<4>[   11.469835] [<c010c534>] (unwind_backtrace+0x0/0x11c) from [<c018ab04>] (warn_slowpath_common+0x4c/0x64)
<4>[   11.469850] [<c018ab04>] (warn_slowpath_common+0x4c/0x64) from [<c018ab9c>] (warn_slowpath_fmt+0x2c/0x3c)
<4>[   11.469864] [<c018ab9c>] (warn_slowpath_fmt+0x2c/0x3c) from [<c0249ab4>] (__kmalloc+0x38/0x29c)
<4>[   11.469879] [<c0249ab4>] (__kmalloc+0x38/0x29c) from [<c06a0124>] (snd_soc_register_card+0x184/0x104c)
<4>[   11.469896] [<c06a0124>] (snd_soc_register_card+0x184/0x104c) from [<c088183c>] (msm8226_asoc_machine_probe+0x298/0x7f0)
<4>[   11.469913] [<c088183c>] (msm8226_asoc_machine_probe+0x298/0x7f0) from [<c044b1fc>] (platform_drv_probe+0x14/0x18)
<4>[   11.469927] [<c044b1fc>] (platform_drv_probe+0x14/0x18) from [<c0449f2c>] (driver_probe_device+0x134/0x334)
<4>[   11.469940] [<c0449f2c>] (driver_probe_device+0x134/0x334) from [<c044a194>] (__driver_attach+0x68/0x8c)
<4>[   11.469953] [<c044a194>] (__driver_attach+0x68/0x8c) from [<c044849c>] (bus_for_each_dev+0x48/0x80)
<4>[   11.469965] [<c044849c>] (bus_for_each_dev+0x48/0x80) from [<c0449528>] (bus_add_driver+0x100/0x26c)
<4>[   11.469978] [<c0449528>] (bus_add_driver+0x100/0x26c) from [<c044a66c>] (driver_register+0x9c/0x120)
<4>[   11.469992] [<c044a66c>] (driver_register+0x9c/0x120) from [<c010052c>] (do_one_initcall+0x90/0x160)
<4>[   11.470007] [<c010052c>] (do_one_initcall+0x90/0x160) from [<c0d00b74>] (kernel_init+0xec/0x1a8)
<4>[   11.470022] [<c0d00b74>] (kernel_init+0xec/0x1a8) from [<c0106aec>] (kernel_thread_exit+0x0/0x8)
<4>[   11.470030] ---[ end trace 1b75b31a2719ed4d ]---

BRs/ Yalin
-----Original Message-----
From: Takashi Iwai [mailto:tiwai at suse.de] 
Sent: Friday, February 28, 2014 10:20 PM
To: Wang, Yalin
Cc: 'linux-kernel at vger.kernel.org'; 'linux-arm-msm at vger.kernel.org'; 'linux-arm-kernel at lists.infradead.org'; 'linux-input at vger.kernel.org'; 'balbi at ti.com'; 'gregkh at linuxfoundation.org'; 'lrg at ti.com'; 'broonie at opensource.wolfsonmicro.com'; 'perex at perex.cz'; 'pablo at netfilter.org'; 'kaber at trash.net'; 'davem at davemloft.net'; 'rostedt at goodmis.org'; 'fweisbec at gmail.com'; 'mingo at redhat.com'; 'dmitry.torokhov at gmail.com'; 'rydberg at euromail.se'; 'linux-usb at vger.kernel.org'; 'alsa-devel at alsa-project.org'; 'netfilter-devel at vger.kernel.org'; 'netfilter at vger.kernel.org'; 'coreteam at netfilter.org'; 'netdev at vger.kernel.org'
Subject: Re: change kmalloc into vmalloc for large memory allocations

At Fri, 28 Feb 2014 16:15:23 +0800,
Wang, Yalin wrote:
> 
> Hi
> 
> 
> I find there is some drivers use kmalloc to allocate large Memorys 
> during module_init,  some can be changed to use vmalloc To save some 
> low mem, I add log in kernel to track , And list them here:
> 
> https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/drive
> rs/usb/gadget/f_mass_storage.c?h=master#n2724
> 
> https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/sound
> /soc/soc-core.c?h=master#n3772

At least the ASoC runtime case doesn't use the allocated memory as buffer, and they are allocated only once per device, thus it shouldn't be the problem you stated.  If it really consumes so much memory, we need to rethink, instead of allocating an array but allocate each object, for example.


thanks,

Takashi


> https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/net/n
> etfilter/nf_conntrack_ftp.c?h=master#n603
> https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/net/n
> etfilter/nf_conntrack_h323_main.c?h=master#n1849
> https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/net/n
> etfilter/nf_conntrack_irc.c?h=master#n247
> https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/net/n
> etfilter/nf_conntrack_sane.c?h=master#n195
> 
> https://git.kernel.org/cgit/linux/kernel/git/will/linux.git/tree/drive
> rs/input/evdev.c?h=master#n403
> 
> 
> they allocate large memory from 10k~64K , this will use lots of low 
> mem, instead if we use vmalloc for these drivers , it will be good for 
> some devices like smart phone, we often encounter some errors like 
> kmalloc failed because there is not enough low mem , especially when 
> the device has physical memory less than 1GB .
> 
> 
> could this module change the memory allocation into vmalloc ?
> 
> I was thinking that if we can introduce a helper function in kmalloc.h like this :
> 
> Kmalloc(size, flags)
> {
> 	If (size > PAGE_SIZE && flags&CAN_USE_VMALLOC_FLAG)
> 		return vmalloc(size);
> 	Else
> 		return real_kmalloc(size);
> }
> 
> Kfree(ptr)
> {
> 	If (is_vmalloc_addr(ptr))
> 		Vfree(ptr);
> 	Else
> 		Kfree(ptr);
> }
> 
> 
> But we need add some flags to ensure always use kmalloc for Some 
> special use (dma etc..)
> 
> How do you think of it ?
> 
> Thanks
> 
> 
>