After receiving a number of inquiries regarding a Linux driver for the RTL8723AU
device with USB ID 0bda:1724, I asked my contacts in the USB group at Realtek if
such a driver were available. They responded that there is one, but not on their
public FTP site; however, they gave me the necessary info to get it from their
private site, and also granted permission for me to make it publicly available.
Therefore, I have created a git repository that can be copied using the command:
git clone http://github.com/lwfinger/rtl8723au.git
Once cloned, the command sequence 'make && sudo make install' in the resulting
source directory will build and install the driver, which is named 8723au. At
the moment, it does not require external firmware. The driver will build on all
kernel versions up to, and including 3.9. I have not tested on kernels older
than 3.7, but the code contains conditional statements testing against kernel
2.6.18. It should build on any kernel new enough for the base hardware.
My plan is to evolve this driver into a form that is suitable for inclusion in
the staging tree of the kernel; however, that will take some effort and time,
which is why I am making the intermediate results available now. As that work
proceeds, the git version will evolve into code that can be submitted to the
kernel with only small changes.
Please be aware that I do not have any hardware that uses this driver, thus I
can only guarantee that the code compiles; however, Joon Ro has tested the
driver and he assures me that it works. Together we have been eliminating
spurious log entries.
The 8723a chip also includes Bluetooth hardware. I have not received any
indication that it is active in the RTL8723AU, but it certainly is in the
RTL8723AE, the PCIe version. In a separate activity, I am working with Realtek
to prepare a kernel driver for this portion of the device.
I welcome your comments, bug reports, patches, suggestions for changes, and
testing. In addition, please spread the word to anyone that has been seeking
this driver.
Thanks,
Larry
On 04/01/2013 09:26 AM, Larry Finger wrote:
>
> My plan is to evolve this driver into a form that is suitable for
> inclusion in the staging tree of the kernel; however, that will take
> some effort and time, which is why I am making the intermediate results
> available now. As that work proceeds, the git version will evolve into
> code that can be submitted to the kernel with only small changes.
>
> Please be aware that I do not have any hardware that uses this driver,
> thus I can only guarantee that the code compiles; however, Joon Ro has
> tested the driver and he assures me that it works. Together we have been
> eliminating spurious log entries.
>
> The 8723a chip also includes Bluetooth hardware. I have not received any
> indication that it is active in the RTL8723AU, but it certainly is in
> the RTL8723AE, the PCIe version. In a separate activity, I am working
> with Realtek to prepare a kernel driver for this portion of the device.
>
So I can verify that this driver works against 3.8.11 (not against
current -linus because of a build failure), and yes, there is Bluetooth
in this device. In fact, it seems to work better than some in-tree
drivers I have seen in the past.
However, I'm confused as to the above and what it implies for the
purpose of the staging tree. Wasn't the whole point of the staging tree
to have a working-but-ugly driver upstream and have the work to clean it
up happen in the mainline tree instead of off somewhere.
-hpa
On Wed, May 08, 2013 at 11:06:44AM -0500, Larry Finger wrote:
> On 05/07/2013 09:32 PM, H. Peter Anvin wrote:
> >
> >So I can verify that this driver works against 3.8.11 (not against
> >current -linus because of a build failure), and yes, there is Bluetooth
> >in this device. In fact, it seems to work better than some in-tree
> >drivers I have seen in the past.
> >
> >However, I'm confused as to the above and what it implies for the
> >purpose of the staging tree. Wasn't the whole point of the staging tree
> >to have a working-but-ugly driver upstream and have the work to clean it
> >up happen in the mainline tree instead of off somewhere.
>
> Yes, but even staging has some standards.
I have standards? That's a first :)
> For example, they frown on inclusion of firmware in a massive data
> block in a file in the driver - it should be in a separate file the
> /lib/firmware/ tree.
Sure, that would be "nice", but it's not something that I should reject
a driver for, have I done that?
> In addition, the mainline drivers/net/wireless/ tree is only available
> to devices that use mac80211 for the software-based MAC layer. That
> conversion is a serious effort that may never be done, thus the
> 8723au driver may never leave staging.
That's a bigger deal, but note, some drivers have done it, it is not
impossible.
thanks,
greg k-h
On Wed, May 08, 2013 at 09:29:10AM -0700, H. Peter Anvin wrote:
> Unlikely, since the kernel source tree is clean.
>
> The error is on os_dep/os_intfs.c:
>
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c: In function
> ?rtw_proc_init_one?:
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:315:3: error: implicit
> declaration of function ?create_proc_entry?
create_proc_entry is gone (and good riddance); use proc_create()
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:322:3: error: implicit
> declaration of function ?create_proc_read_entry?
ditto.
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:370:7: error:
> dereferencing pointer to incomplete type
... and don't poke in proc_dir_entry innards, while we are at it.
On 05/08/2013 11:29 AM, H. Peter Anvin wrote:
> On 05/08/2013 09:06 AM, Larry Finger wrote:
>> On 05/07/2013 09:32 PM, H. Peter Anvin wrote:
>>>
>>> So I can verify that this driver works against 3.8.11 (not against
>>> current -linus because of a build failure), and yes, there is Bluetooth
>>> in this device. In fact, it seems to work better than some in-tree
>>> drivers I have seen in the past.
>>>
>>> However, I'm confused as to the above and what it implies for the
>>> purpose of the staging tree. Wasn't the whole point of the staging tree
>>> to have a working-but-ugly driver upstream and have the work to clean it
>>> up happen in the mainline tree instead of off somewhere.
>>
>> Yes, but even staging has some standards. For example, they frown on
>> inclusion of firmware in a massive data block in a file in the driver -
>> it should be in a separate file the /lib/firmware/ tree. In addition,
>> the mainline drivers/net/wireless/ tree is only available to devices
>> that use mac80211 for the software-based MAC layer. That conversion is a
>> serious effort that may never be done, thus the 8723au driver may never
>> leave staging.
>>
>> What are the details of the build failure on the mainline tree? It
>> builds fine here. One possibility is that your source tree has a
>> leftover copy of include/linux/version.h that contains a stale coded
>> value of the kernel version. I just pushed a Makefile fix that causes
>> the compiler to look in the new location for version.h first. If that
>> doesn't work, please post the error.
>>
>
> Unlikely, since the kernel source tree is clean.
>
> The error is on os_dep/os_intfs.c:
>
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c: In function ?rtw_proc_init_one?:
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:315:3: error: implicit declaration
> of function ?create_proc_entry? [-Werror=implicit-function-declaration]
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:315:11: warning: assignment makes
> pointer from integer without a cast [enabled by default]
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:322:3: error: implicit declaration
> of function ?create_proc_read_entry? [-Werror=implicit-function-declaration]
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:322:9: warning: assignment makes
> pointer from integer without a cast [enabled by default]
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:333:21: warning: assignment makes
> pointer from integer without a cast [enabled by default]
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:364:8: warning: assignment makes
> pointer from integer without a cast [enabled by default]
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:370:7: error: dereferencing pointer
> to incomplete type
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:372:8: warning: assignment makes
> pointer from integer without a cast [enabled by default]
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:378:7: error: dereferencing pointer
> to incomplete type
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:381:8: warning: assignment makes
> pointer from integer without a cast [enabled by default]
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:389:8: warning: assignment makes
> pointer from integer without a cast [enabled by default]
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:397:8: warning: assignment makes
> pointer from integer without a cast [enabled by default]
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:405:8: warning: assignment makes
> pointer from integer without a cast [enabled by default]
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:412:8: warning: assignment makes
> pointer from integer without a cast [enabled by default]
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:419:8: warning: assignment makes
> pointer from integer without a cast [enabled by default]
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:426:8: warning: assignment makes
> pointer from integer without a cast [enabled by default]
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:433:8: warning: assignment makes
> pointer from integer without a cast [enabled by default]
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:440:8: warning: assignment makes
> pointer from integer without a cast [enabled by default]
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:447:8: warning: assignment makes
> pointer from integer without a cast [enabled by default]
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:454:8: warning: assignment makes
> pointer from integer without a cast [enabled by default]
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:461:8: warning: assignment makes
> pointer from integer without a cast [enabled by default]
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:468:8: warning: assignment makes
> pointer from integer without a cast [enabled by default]
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:475:8: warning: assignment makes
> pointer from integer without a cast [enabled by default]
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:482:8: warning: assignment makes
> pointer from integer without a cast [enabled by default]
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:489:8: warning: assignment makes
> pointer from integer without a cast [enabled by default]
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:496:8: warning: assignment makes
> pointer from integer without a cast [enabled by default]
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:505:9: warning: assignment makes
> pointer from integer without a cast [enabled by default]
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:512:9: warning: assignment makes
> pointer from integer without a cast [enabled by default]
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:522:8: warning: assignment makes
> pointer from integer without a cast [enabled by default]
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:539:8: warning: assignment makes
> pointer from integer without a cast [enabled by default]
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:545:7: error: dereferencing pointer
> to incomplete type
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:547:8: warning: assignment makes
> pointer from integer without a cast [enabled by default]
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:553:7: error: dereferencing pointer
> to incomplete type
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:555:8: warning: assignment makes
> pointer from integer without a cast [enabled by default]
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:561:7: error: dereferencing pointer
> to incomplete type
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:563:8: warning: assignment makes
> pointer from integer without a cast [enabled by default]
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:569:7: error: dereferencing pointer
> to incomplete type
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:572:8: warning: assignment makes
> pointer from integer without a cast [enabled by default]
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:576:8: warning: assignment makes
> pointer from integer without a cast [enabled by default]
> /home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:582:7: error: dereferencing pointer
> to incomplete type
I just pushed a "fix" that will allow the driver to build on kernel 3.10-rc0
(aka 3.9-git). This one removes all the procfs stuff. It will be added back in
/sysfs later.
Larry
On 05/08/2013 11:23 AM, Greg KH wrote:
> On Wed, May 08, 2013 at 11:06:44AM -0500, Larry Finger wrote:
>> On 05/07/2013 09:32 PM, H. Peter Anvin wrote:
>>>
>>> So I can verify that this driver works against 3.8.11 (not against
>>> current -linus because of a build failure), and yes, there is Bluetooth
>>> in this device. In fact, it seems to work better than some in-tree
>>> drivers I have seen in the past.
>>>
>>> However, I'm confused as to the above and what it implies for the
>>> purpose of the staging tree. Wasn't the whole point of the staging tree
>>> to have a working-but-ugly driver upstream and have the work to clean it
>>> up happen in the mainline tree instead of off somewhere.
>>
>> Yes, but even staging has some standards.
>
> I have standards? That's a first :)
>
>> For example, they frown on inclusion of firmware in a massive data
>> block in a file in the driver - it should be in a separate file the
>> /lib/firmware/ tree.
>
> Sure, that would be "nice", but it's not something that I should reject
> a driver for, have I done that?
No. You merely "recommended" that it be the #1 item in the TODO list, which
forced me to fix it fairly quickly for r8712u.
>
>> In addition, the mainline drivers/net/wireless/ tree is only available
>> to devices that use mac80211 for the software-based MAC layer. That
>> conversion is a serious effort that may never be done, thus the
>> 8723au driver may never leave staging.
>
> That's a bigger deal, but note, some drivers have done it, it is not
> impossible.
Most of them were in the kernel before mac80211, and have been given a pass as
part of a "grandfather" clause.
Unfortunately, I forgot the main reason for development in an off-line repo. I
do not have one of these devices, and I have drawn the line at purchasing a
$1000 Lenovo tablet just to get one. Making "small" untested changes in the
off-line repo for the users to verify is a lot simpler than working through
staging, no matter how accommodating you are. :)
If someone wants to donate a device, that would speed up the process.
Larry
On 05/08/2013 09:06 AM, Larry Finger wrote:
> On 05/07/2013 09:32 PM, H. Peter Anvin wrote:
>>
>> So I can verify that this driver works against 3.8.11 (not against
>> current -linus because of a build failure), and yes, there is Bluetooth
>> in this device. In fact, it seems to work better than some in-tree
>> drivers I have seen in the past.
>>
>> However, I'm confused as to the above and what it implies for the
>> purpose of the staging tree. Wasn't the whole point of the staging tree
>> to have a working-but-ugly driver upstream and have the work to clean it
>> up happen in the mainline tree instead of off somewhere.
>
> Yes, but even staging has some standards. For example, they frown on
> inclusion of firmware in a massive data block in a file in the driver -
> it should be in a separate file the /lib/firmware/ tree. In addition,
> the mainline drivers/net/wireless/ tree is only available to devices
> that use mac80211 for the software-based MAC layer. That conversion is a
> serious effort that may never be done, thus the 8723au driver may never
> leave staging.
>
> What are the details of the build failure on the mainline tree? It
> builds fine here. One possibility is that your source tree has a
> leftover copy of include/linux/version.h that contains a stale coded
> value of the kernel version. I just pushed a Makefile fix that causes
> the compiler to look in the new location for version.h first. If that
> doesn't work, please post the error.
>
Unlikely, since the kernel source tree is clean.
The error is on os_dep/os_intfs.c:
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c: In function
?rtw_proc_init_one?:
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:315:3: error: implicit
declaration of function ?create_proc_entry?
[-Werror=implicit-function-declaration]
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:315:11: warning: assignment
makes pointer from integer without a cast [enabled by default]
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:322:3: error: implicit
declaration of function ?create_proc_read_entry?
[-Werror=implicit-function-declaration]
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:322:9: warning: assignment
makes pointer from integer without a cast [enabled by default]
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:333:21: warning: assignment
makes pointer from integer without a cast [enabled by default]
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:364:8: warning: assignment
makes pointer from integer without a cast [enabled by default]
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:370:7: error: dereferencing
pointer to incomplete type
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:372:8: warning: assignment
makes pointer from integer without a cast [enabled by default]
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:378:7: error: dereferencing
pointer to incomplete type
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:381:8: warning: assignment
makes pointer from integer without a cast [enabled by default]
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:389:8: warning: assignment
makes pointer from integer without a cast [enabled by default]
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:397:8: warning: assignment
makes pointer from integer without a cast [enabled by default]
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:405:8: warning: assignment
makes pointer from integer without a cast [enabled by default]
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:412:8: warning: assignment
makes pointer from integer without a cast [enabled by default]
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:419:8: warning: assignment
makes pointer from integer without a cast [enabled by default]
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:426:8: warning: assignment
makes pointer from integer without a cast [enabled by default]
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:433:8: warning: assignment
makes pointer from integer without a cast [enabled by default]
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:440:8: warning: assignment
makes pointer from integer without a cast [enabled by default]
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:447:8: warning: assignment
makes pointer from integer without a cast [enabled by default]
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:454:8: warning: assignment
makes pointer from integer without a cast [enabled by default]
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:461:8: warning: assignment
makes pointer from integer without a cast [enabled by default]
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:468:8: warning: assignment
makes pointer from integer without a cast [enabled by default]
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:475:8: warning: assignment
makes pointer from integer without a cast [enabled by default]
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:482:8: warning: assignment
makes pointer from integer without a cast [enabled by default]
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:489:8: warning: assignment
makes pointer from integer without a cast [enabled by default]
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:496:8: warning: assignment
makes pointer from integer without a cast [enabled by default]
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:505:9: warning: assignment
makes pointer from integer without a cast [enabled by default]
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:512:9: warning: assignment
makes pointer from integer without a cast [enabled by default]
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:522:8: warning: assignment
makes pointer from integer without a cast [enabled by default]
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:539:8: warning: assignment
makes pointer from integer without a cast [enabled by default]
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:545:7: error: dereferencing
pointer to incomplete type
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:547:8: warning: assignment
makes pointer from integer without a cast [enabled by default]
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:553:7: error: dereferencing
pointer to incomplete type
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:555:8: warning: assignment
makes pointer from integer without a cast [enabled by default]
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:561:7: error: dereferencing
pointer to incomplete type
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:563:8: warning: assignment
makes pointer from integer without a cast [enabled by default]
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:569:7: error: dereferencing
pointer to incomplete type
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:572:8: warning: assignment
makes pointer from integer without a cast [enabled by default]
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:576:8: warning: assignment
makes pointer from integer without a cast [enabled by default]
/home/hpa/kernel/rtl8723au/os_dep/os_intfs.c:582:7: error: dereferencing
pointer to incomplete type
On 05/07/2013 09:32 PM, H. Peter Anvin wrote:
>
> So I can verify that this driver works against 3.8.11 (not against
> current -linus because of a build failure), and yes, there is Bluetooth
> in this device. In fact, it seems to work better than some in-tree
> drivers I have seen in the past.
>
> However, I'm confused as to the above and what it implies for the
> purpose of the staging tree. Wasn't the whole point of the staging tree
> to have a working-but-ugly driver upstream and have the work to clean it
> up happen in the mainline tree instead of off somewhere.
Yes, but even staging has some standards. For example, they frown on inclusion
of firmware in a massive data block in a file in the driver - it should be in a
separate file the /lib/firmware/ tree. In addition, the mainline
drivers/net/wireless/ tree is only available to devices that use mac80211 for
the software-based MAC layer. That conversion is a serious effort that may never
be done, thus the 8723au driver may never leave staging.
What are the details of the build failure on the mainline tree? It builds fine
here. One possibility is that your source tree has a leftover copy of
include/linux/version.h that contains a stale coded value of the kernel version.
I just pushed a Makefile fix that causes the compiler to look in the new
location for version.h first. If that doesn't work, please post the error.
Larry