Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1400137 > unrolled thread

Re: [PATCH v5 00/11] serial: 8250: split LPSS to 8250_lpss, enable DMA on Quark

Back to article view | Back to linux.kernel

This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by below is the oldest one visible, not the original post.

Contents

  Re: [PATCH v5 00/11] serial: 8250: split LPSS to 8250_lpss, enable  DMA on Quark Andy Shevchenko <andriy.shevchenko@linux.intel.com> - 2016-05-12 17:10 +0200
    Re: [PATCH v5 00/11] serial: 8250: split LPSS to 8250_lpss, enable  DMA on Quark Vinod Koul <vinod.koul@intel.com> - 2016-05-12 18:00 +0200
    Re: [PATCH v5 00/11] serial: 8250: split LPSS to 8250_lpss, enable  DMA on Quark Bryan O'Donoghue <pure.logic@nexus-software.ie> - 2016-05-19 03:20 +0200
      Re: [PATCH v5 00/11] serial: 8250: split LPSS to 8250_lpss, enable  DMA on Quark Andy Shevchenko <andriy.shevchenko@linux.intel.com> - 2016-05-24 19:40 +0200
        Re: [PATCH v5 00/11] serial: 8250: split LPSS to 8250_lpss, enable  DMA on Quark Andy Shevchenko <andriy.shevchenko@linux.intel.com> - 2016-05-24 20:10 +0200
          Re: [PATCH v5 00/11] serial: 8250: split LPSS to 8250_lpss, enable  DMA on Quark Bryan O'Donoghue <pure.logic@nexus-software.ie> - 2016-05-26 18:40 +0200
        Re: [PATCH v5 00/11] serial: 8250: split LPSS to 8250_lpss, enable  DMA on Quark Bryan O'Donoghue <pure.logic@nexus-software.ie> - 2016-05-26 18:20 +0200

#1400137 — Re: [PATCH v5 00/11] serial: 8250: split LPSS to 8250_lpss, enable DMA on Quark

On Fri, 2016-05-06 at 18:17 +0300, Andy Shevchenko wrote:
> This is combined series of two things:
> - split out the Intel LPSS specific driver from 8250_pci into
> 8250_lpss
> - enable DMA support on Intel Quark UART
> 
> The patch has been tested on few Intel SoCs / platforms. In any case I
> would
> like to ask Bryan to do independent test.
> 
> This is targeting serial subsystem, thus it would be nice to get and
> Ack from
> Vinod first. Moreover, the series depends on [1] that is now in
> Vinod's
> slave-dma tree.
> 
> That's why I asked Vinod to create immutable tag / branch for the [1]
> and the
> dependants (at least one more, which is sata_dwc_460ex) can use it.
> 
> The series can be reached in the branch located at [2].
> 
> [1] http://www.spinics.net/lists/kernel/msg2244475.html
> [2] https://bitbucket.org/andy-shev/linux/branch/topic%2Fdw%2Fqrk
> 

Do we have a chance to get this? Requirement, i.e. topic/dw branch of
Vinod's slave-dma tree is available and has all prerequisites.

> Since v4:
> - fix an issue what kernel test bot reported (lkp)
> - add a list of supported SoC to Kconfig (Bryan)
> - rename polarity -> hs_polarity (Bryan)
> 
> Since v3:
> - rebase on top of recent tty-next branch
> - add Reviewed-by tag from Bryan
> 
> Since v2:
> - add tags
> - rebase on top of new version of [1]
> 
> Since v1:
> - address most of Peter's comments (mostly changelog to patch 8)
> - add tag to patch 5
> - drop patch 6 from the series to be separately dealt with
> 
> Andy Shevchenko (11):
>   dmaengine: dw: keep copy of custom slave config in dwc
>   dmaengine: dw: provide probe(), remove() stubs for users
>   dmaengine: dw: set polarity of handshake interface
>   dmaengine: dw: override LLP support if asked in platform data
>   serial: 8250_dma: switch to new dmaengine_terminate_* API
>   serial: 8250_dma: adjust DMA address of the UART
>   serial: 8250: enable AFE on ports where FIFO is 16 bytes
>   serial: 8250_lpss: split LPSS driver to separate module
>   serial: 8250_lpss: move Quark code from PCI driver
>   serial: 8250_lpss: enable MSI for Intel Quark
>   serial: 8250_lpss: enable DMA on Intel Quark UART
> 
>  drivers/dma/dw/core.c                |  43 +++--
>  drivers/dma/dw/regs.h                |   5 +-
>  drivers/tty/serial/8250/8250.h       |   5 +
>  drivers/tty/serial/8250/8250_dma.c   |  14 +-
>  drivers/tty/serial/8250/8250_lpss.c  | 357
> +++++++++++++++++++++++++++++++++++
>  drivers/tty/serial/8250/8250_pci.c   | 257 +------------------------
>  drivers/tty/serial/8250/8250_port.c  |   9 +-
>  drivers/tty/serial/8250/Kconfig      |  16 +-
>  drivers/tty/serial/8250/Makefile     |   1 +
>  include/linux/dma/dw.h               |   5 +
>  include/linux/platform_data/dma-dw.h |   4 +
>  11 files changed, 430 insertions(+), 286 deletions(-)
>  create mode 100644 drivers/tty/serial/8250/8250_lpss.c
> 

-- 
Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Intel Finland Oy

[toc] | [next] | [standalone]

#1400200

On Thu, May 12, 2016 at 06:06:41PM +0300, Andy Shevchenko wrote:
> On Fri, 2016-05-06 at 18:17 +0300, Andy Shevchenko wrote:
> > This is combined series of two things:
> > - split out the Intel LPSS specific driver from 8250_pci into
> > 8250_lpss
> > - enable DMA support on Intel Quark UART
> > 
> > The patch has been tested on few Intel SoCs / platforms. In any case I
> > would
> > like to ask Bryan to do independent test.
> > 
> > This is targeting serial subsystem, thus it would be nice to get and
> > Ack from
> > Vinod first. Moreover, the series depends on [1] that is now in
> > Vinod's
> > slave-dma tree.
> > 
> > That's why I asked Vinod to create immutable tag / branch for the [1]
> > and the
> > dependants (at least one more, which is sata_dwc_460ex) can use it.
> > 
> > The series can be reached in the branch located at [2].
> > 
> > [1] http://www.spinics.net/lists/kernel/msg2244475.html
> > [2] https://bitbucket.org/andy-shev/linux/branch/topic%2Fdw%2Fqrk
> > 
> 
> Do we have a chance to get this? Requirement, i.e. topic/dw branch of
> Vinod's slave-dma tree is available and has all prerequisites.

All dmaengine patches

Acked-by: Vinod Koul <vinod.koul@intel.com>

Thanks
-- 
~Vinod

> 
> > Since v4:
> > - fix an issue what kernel test bot reported (lkp)
> > - add a list of supported SoC to Kconfig (Bryan)
> > - rename polarity -> hs_polarity (Bryan)
> > 
> > Since v3:
> > - rebase on top of recent tty-next branch
> > - add Reviewed-by tag from Bryan
> > 
> > Since v2:
> > - add tags
> > - rebase on top of new version of [1]
> > 
> > Since v1:
> > - address most of Peter's comments (mostly changelog to patch 8)
> > - add tag to patch 5
> > - drop patch 6 from the series to be separately dealt with
> > 
> > Andy Shevchenko (11):
> >   dmaengine: dw: keep copy of custom slave config in dwc
> >   dmaengine: dw: provide probe(), remove() stubs for users
> >   dmaengine: dw: set polarity of handshake interface
> >   dmaengine: dw: override LLP support if asked in platform data
> >   serial: 8250_dma: switch to new dmaengine_terminate_* API
> >   serial: 8250_dma: adjust DMA address of the UART
> >   serial: 8250: enable AFE on ports where FIFO is 16 bytes
> >   serial: 8250_lpss: split LPSS driver to separate module
> >   serial: 8250_lpss: move Quark code from PCI driver
> >   serial: 8250_lpss: enable MSI for Intel Quark
> >   serial: 8250_lpss: enable DMA on Intel Quark UART
> > 
> >  drivers/dma/dw/core.c                |  43 +++--
> >  drivers/dma/dw/regs.h                |   5 +-
> >  drivers/tty/serial/8250/8250.h       |   5 +
> >  drivers/tty/serial/8250/8250_dma.c   |  14 +-
> >  drivers/tty/serial/8250/8250_lpss.c  | 357
> > +++++++++++++++++++++++++++++++++++
> >  drivers/tty/serial/8250/8250_pci.c   | 257 +------------------------
> >  drivers/tty/serial/8250/8250_port.c  |   9 +-
> >  drivers/tty/serial/8250/Kconfig      |  16 +-
> >  drivers/tty/serial/8250/Makefile     |   1 +
> >  include/linux/dma/dw.h               |   5 +
> >  include/linux/platform_data/dma-dw.h |   4 +
> >  11 files changed, 430 insertions(+), 286 deletions(-)
> >  create mode 100644 drivers/tty/serial/8250/8250_lpss.c
> > 
> 
> -- 
> Andy Shevchenko <andriy.shevchenko@linux.intel.com>
> Intel Finland Oy
> 

[toc] | [prev] | [next] | [standalone]

#1403379

On Thu, 2016-05-12 at 18:06 +0300, Andy Shevchenko wrote:
> On Fri, 2016-05-06 at 18:17 +0300, Andy Shevchenko wrote:
> > This is combined series of two things:
> > - split out the Intel LPSS specific driver from 8250_pci into
> > 8250_lpss
> > - enable DMA support on Intel Quark UART
> > 
> > The patch has been tested on few Intel SoCs / platforms. In any
> > case I
> > would
> > like to ask Bryan to do independent test.

Andy.

It looks as though there's a performance drop when enabling the DMA
here (v) stock PIO mode with the 8250 driver.

Test setup Galileo Gen1 transmitter - Galileo Gen2 receiver. TX/RX
cross-wired between the Arduino UART headers respectively (pins 0 & 1
on the "Digital PWM" header), GND pin14 similarly cross-wired.

User-space:
    Receiver: Stock on-flash yocto-tiny kernel 3.8.7
    Transmitter: Galileo Debian image :
                 https://sourceforge.net/projects/galileodebian/
    3.8.7 kernel: On SPI-flash BSP kernel
    andy-v4: From your tree
             d13ffcf ('serial: 8250: attempt to keep Kconfig and
                       Makefile in order')
    tip-of-tree : 97f0090 ('Merge tag 'rproc-v4.7' of
                            git://github.com/andersson/remoteproc')

Results:

### Test0 ###

galileo dma andy-v4 9600 8n1 

root@galileo:~# time echo "abcdefghijklmnopqrstuvwxyz0123456789" >
/dev/ttyS0

real    0m0.060s
user    0m0.000s
sys     0m0.000s


galileo non-dma Linux 3.8.7 9600 8n1:

9600 8n1
root@galileo:~# time echo "abcdefghijklmnopqrstuvwxyz0123456789" >
/dev/ttyS0

real    0m0.051s
user    0m0.000s
sys     0m0.000s

galileo non-dma tip-of-tree 9600 8n1:

root@galileo:~# time echo "abcdefghijklmnopqrstuvwxyz0123456789" >
/dev/ttyS0

real    0m0.048s
user    0m0.000s
sys     0m0.000s


### Test1 ###

galileo dma andy-v4 9600 8n1:

root@galileo:~# time echo
"abcdefghijklmnopqrstuvwxyz0123456789abcdefghijklmnopqrstuvwxyz01234567
89abcdefghijklmnopqrstuvwxyz0123456789abcdefghijklmnopqrstuvwxyz0123456
789abcdefghijklmnopqrstuvwxyz0123456789" > /dev/ttyS0

real    0m0.209s
user    0m0.000s
sys     0m0.000s

galileo non-dma Linux 3.8.7 9600 8n1:

root@galileo:~# time echo
"abcdefghijklmnopqrstuvwxyz0123456789abcdefghijklmnopqrstuvwxyz01234567
89abcdefghijklmnopqrstuvwxyz0123456789abcdefghijklmnopqrstuvwxyz0123456
789abcdefghijklmnopqrstuvwxyz0123456789" > /dev/ttyS0

real    0m0.206s
user    0m0.000s
sys     0m0.000s


galileo non-dma tip-of-tree 9600 8n1:

root@galileo:~# time echo
"abcdefghijklmnopqrstuvwxyz0123456789abcdefghijklmnopqrstuvwxyz01234567
89abcdefghijklmnopqrstuvwxyz0123456789abcdefghijklmnopqrstuvwxyz0123456
789abcdefghijklmnopqrstuvwxyz0123456789" > /dev/ttyS0

real    0m0.199s
user    0m0.010s
sys     0m0.000s


---
bod

[toc] | [prev] | [next] | [standalone]

#1406305

On Thu, 2016-05-19 at 02:18 +0100, Bryan O'Donoghue wrote:
> On Thu, 2016-05-12 at 18:06 +0300, Andy Shevchenko wrote:
> > On Fri, 2016-05-06 at 18:17 +0300, Andy Shevchenko wrote:
> > > This is combined series of two things:
> > > - split out the Intel LPSS specific driver from 8250_pci into
> > > 8250_lpss
> > > - enable DMA support on Intel Quark UART
> > > 
> > > The patch has been tested on few Intel SoCs / platforms. In any
> > > case I
> > > would
> > > like to ask Bryan to do independent test.
> 
> Andy.
> 
> It looks as though there's a performance drop when enabling the DMA
> here (v) stock PIO mode with the 8250 driver.

Below results doesn't show DMA case (by the way, how did you disable DMA
in 3.8.7?) for stock kernel.

Anyway, I added preliminary support to change DMA mode run-time. Check
my today topic/dw/qrk branch. It would be nice to see all 4 cases: DMA
vs non-DMA on stock kernel (Yocto) and mine.

P.S. What is the point to test arbitrary tip branch?

> 
> Test setup Galileo Gen1 transmitter - Galileo Gen2 receiver. TX/RX
> cross-wired between the Arduino UART headers respectively (pins 0 & 1
> on the "Digital PWM" header), GND pin14 similarly cross-wired.
> 
> User-space:
>     Receiver: Stock on-flash yocto-tiny kernel 3.8.7
>     Transmitter: Galileo Debian image :
>                  https://sourceforge.net/projects/galileodebian/
>     3.8.7 kernel: On SPI-flash BSP kernel
>     andy-v4: From your tree
>              d13ffcf ('serial: 8250: attempt to keep Kconfig and
>                        Makefile in order')
>     tip-of-tree : 97f0090 ('Merge tag 'rproc-v4.7' of
>                             git://github.com/andersson/remoteproc')
> 
> Results:
> 
> ### Test0 ###
> 
> galileo dma andy-v4 9600 8n1 
> 
> root@galileo:~# time echo "abcdefghijklmnopqrstuvwxyz0123456789" >
> /dev/ttyS0
> 
> real    0m0.060s
> user    0m0.000s
> sys     0m0.000s
> 
> 
> galileo non-dma Linux 3.8.7 9600 8n1:
> 
> 9600 8n1
> root@galileo:~# time echo "abcdefghijklmnopqrstuvwxyz0123456789" >
> /dev/ttyS0
> 
> real    0m0.051s
> user    0m0.000s
> sys     0m0.000s
> 
> galileo non-dma tip-of-tree 9600 8n1:
> 
> root@galileo:~# time echo "abcdefghijklmnopqrstuvwxyz0123456789" >
> /dev/ttyS0
> 
> real    0m0.048s
> user    0m0.000s
> sys     0m0.000s
> 
> 
> ### Test1 ###
> 
> galileo dma andy-v4 9600 8n1:
> 
> root@galileo:~# time echo
> "abcdefghijklmnopqrstuvwxyz0123456789abcdefghijklmnopqrstuvwxyz0123456
> 7
> 89abcdefghijklmnopqrstuvwxyz0123456789abcdefghijklmnopqrstuvwxyz012345
> 6
> 789abcdefghijklmnopqrstuvwxyz0123456789" > /dev/ttyS0
> 
> real    0m0.209s
> user    0m0.000s
> sys     0m0.000s
> 
> galileo non-dma Linux 3.8.7 9600 8n1:
> 
> root@galileo:~# time echo
> "abcdefghijklmnopqrstuvwxyz0123456789abcdefghijklmnopqrstuvwxyz0123456
> 7
> 89abcdefghijklmnopqrstuvwxyz0123456789abcdefghijklmnopqrstuvwxyz012345
> 6
> 789abcdefghijklmnopqrstuvwxyz0123456789" > /dev/ttyS0
> 
> real    0m0.206s
> user    0m0.000s
> sys     0m0.000s
> 
> 
> galileo non-dma tip-of-tree 9600 8n1:
> 
> root@galileo:~# time echo
> "abcdefghijklmnopqrstuvwxyz0123456789abcdefghijklmnopqrstuvwxyz0123456
> 7
> 89abcdefghijklmnopqrstuvwxyz0123456789abcdefghijklmnopqrstuvwxyz012345
> 6
> 789abcdefghijklmnopqrstuvwxyz0123456789" > /dev/ttyS0
> 
> real    0m0.199s
> user    0m0.010s
> sys     0m0.000s
> 
> 
> ---
> bod

-- 
Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Intel Finland Oy

[toc] | [prev] | [next] | [standalone]

#1406345

On Tue, 2016-05-24 at 20:37 +0300, Andy Shevchenko wrote:
> On Thu, 2016-05-19 at 02:18 +0100, Bryan O'Donoghue wrote:
> > On Thu, 2016-05-12 at 18:06 +0300, Andy Shevchenko wrote:
> > > On Fri, 2016-05-06 at 18:17 +0300, Andy Shevchenko wrote:
> > > > This is combined series of two things:
> > > > - split out the Intel LPSS specific driver from 8250_pci into
> > > > 8250_lpss
> > > > - enable DMA support on Intel Quark UART
> > > > 
> > > > The patch has been tested on few Intel SoCs / platforms. In any
> > > > case I
> > > > would
> > > > like to ask Bryan to do independent test.
> > 
> > Andy.
> > 
> > It looks as though there's a performance drop when enabling the DMA
> > here (v) stock PIO mode with the 8250 driver.

Wait, the time measurement you done is unrelated to DMA/non-DMA mode.
It's significantly depend to the speed of the port.

It would be better to have 1000(s) of iterations to see any steady
difference out of statistical error. Below result doesn't show anything
like this.

I also tried on my Galileo board, I used longer data stream, i.e. output
of dmesg (since your data pattern makes 0:s in statistics) and got
something quite near to each other:

###
Size of the data (it actually become few dozens of bytes (55 bytes to be
precise) bigger since one message had been printed in the second part of
high speed test
###

% dmesg | wc -c
34965

### Low speed, DMA mode enabled by default ###
% stty -F /dev/ttyS0 9600

% time dmesg > /dev/ttyS0
real    0m 34.22s
user    0m 0.01s
sys     0m 0.02s

NOTE: Transfer speed: 34965b/34.22s ~= 1021.77b/s (9600 baud)

### non-DMA mode ###
% echo 3 > /sys/class/tty/ttyS0/dma_mode 

% time dmesg > /dev/ttyS0
real    0m 36.09s
user    0m 0.01s
sys     0m 0.02s

### Higher speed ###

% stty -F /dev/ttyS0 921600

% time dmesg > /dev/ttyS0
real    0m 0.41s
user    0m 0.01s
sys     0m 0.03s

NOTE: Transfer speed: 34965b/0.41s ~= 85280.49b/s (921600 baud)

### DMA mode enabled ###
% echo 0 > /sys/class/tty/ttyS0/dma_mode 

% time dmesg > /dev/ttyS0
real    0m 0.41s
user    0m 0.01s
sys     0m 0.02s

> 
> Below results doesn't show DMA case (by the way, how did you disable
> DMA
> in 3.8.7?) for stock kernel.
> 
> Anyway, I added preliminary support to change DMA mode run-time. Check
> my today topic/dw/qrk branch. It would be nice to see all 4 cases: DMA
> vs non-DMA on stock kernel (Yocto) and mine.
> 
> P.S. What is the point to test arbitrary tip branch?
> 
> > 
> > Test setup Galileo Gen1 transmitter - Galileo Gen2 receiver. TX/RX
> > cross-wired between the Arduino UART headers respectively (pins 0 &
> > 1
> > on the "Digital PWM" header), GND pin14 similarly cross-wired.
> > 
> > User-space:
> >     Receiver: Stock on-flash yocto-tiny kernel 3.8.7
> >     Transmitter: Galileo Debian image :
> >                  https://sourceforge.net/projects/galileodebian/
> >     3.8.7 kernel: On SPI-flash BSP kernel
> >     andy-v4: From your tree
> >              d13ffcf ('serial: 8250: attempt to keep Kconfig and
> >                        Makefile in order')
> >     tip-of-tree : 97f0090 ('Merge tag 'rproc-v4.7' of
> >                             git://github.com/andersson/remoteproc')
> > 
> > Results:
> > 
> > ### Test0 ###
> > 
> > galileo dma andy-v4 9600 8n1 
> > 
> > root@galileo:~# time echo "abcdefghijklmnopqrstuvwxyz0123456789" >
> > /dev/ttyS0
> > 
> > real    0m0.060s
> > user    0m0.000s
> > sys     0m0.000s
> > 
> > 
> > galileo non-dma Linux 3.8.7 9600 8n1:
> > 
> > 9600 8n1
> > root@galileo:~# time echo "abcdefghijklmnopqrstuvwxyz0123456789" >
> > /dev/ttyS0
> > 
> > real    0m0.051s
> > user    0m0.000s
> > sys     0m0.000s
> > 
> > galileo non-dma tip-of-tree 9600 8n1:
> > 
> > root@galileo:~# time echo "abcdefghijklmnopqrstuvwxyz0123456789" >
> > /dev/ttyS0
> > 
> > real    0m0.048s
> > user    0m0.000s
> > sys     0m0.000s
> > 
> > 
> > ### Test1 ###
> > 
> > galileo dma andy-v4 9600 8n1:
> > 
> > root@galileo:~# time echo
> > "abcdefghijklmnopqrstuvwxyz0123456789abcdefghijklmnopqrstuvwxyz01234
> > 56
> > 7
> > 89abcdefghijklmnopqrstuvwxyz0123456789abcdefghijklmnopqrstuvwxyz0123
> > 45
> > 6
> > 789abcdefghijklmnopqrstuvwxyz0123456789" > /dev/ttyS0
> > 
> > real    0m0.209s
> > user    0m0.000s
> > sys     0m0.000s
> > 
> > galileo non-dma Linux 3.8.7 9600 8n1:
> > 
> > root@galileo:~# time echo
> > "abcdefghijklmnopqrstuvwxyz0123456789abcdefghijklmnopqrstuvwxyz01234
> > 56
> > 7
> > 89abcdefghijklmnopqrstuvwxyz0123456789abcdefghijklmnopqrstuvwxyz0123
> > 45
> > 6
> > 789abcdefghijklmnopqrstuvwxyz0123456789" > /dev/ttyS0
> > 
> > real    0m0.206s
> > user    0m0.000s
> > sys     0m0.000s
> > 
> > 
> > galileo non-dma tip-of-tree 9600 8n1:
> > 
> > root@galileo:~# time echo
> > "abcdefghijklmnopqrstuvwxyz0123456789abcdefghijklmnopqrstuvwxyz01234
> > 56
> > 7
> > 89abcdefghijklmnopqrstuvwxyz0123456789abcdefghijklmnopqrstuvwxyz0123
> > 45
> > 6
> > 789abcdefghijklmnopqrstuvwxyz0123456789" > /dev/ttyS0
> > 
> > real    0m0.199s
> > user    0m0.010s
> > sys     0m0.000s



-- 
Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Intel Finland Oy

[toc] | [prev] | [next] | [standalone]

#1407632

On Tue, 2016-05-24 at 21:08 +0300, Andy Shevchenko wrote:
> On Tue, 2016-05-24 at 20:37 +0300, Andy Shevchenko wrote:
> > On Thu, 2016-05-19 at 02:18 +0100, Bryan O'Donoghue wrote:
> > > On Thu, 2016-05-12 at 18:06 +0300, Andy Shevchenko wrote:
> > > > On Fri, 2016-05-06 at 18:17 +0300, Andy Shevchenko wrote:
> > > > > This is combined series of two things:
> > > > > - split out the Intel LPSS specific driver from 8250_pci into
> > > > > 8250_lpss
> > > > > - enable DMA support on Intel Quark UART
> > > > > 
> > > > > The patch has been tested on few Intel SoCs / platforms. In
> > > > > any
> > > > > case I
> > > > > would
> > > > > like to ask Bryan to do independent test.
> > > 
> > > Andy.
> > > 
> > > It looks as though there's a performance drop when enabling the
> > > DMA
> > > here (v) stock PIO mode with the 8250 driver.
> 
> Wait, the time measurement you done is unrelated to DMA/non-DMA mode.
> It's significantly depend to the speed of the port.

Does it ? The size of the FIFO is what - 16 bytes ? Max burst size is 8
bytes - so for 37 bytes you'd expect to see five DMA transactions {8,
8, 8, 5}, for the second test 183 bytes you'd expect to see 23 eight
byte transactions followed by one 7 byte transaction. The flip side to
add data to the FIFO in PIO mode is a pretty different pattern.

> I also tried on my Galileo board, I used longer data stream, i.e.
> output
> of dmesg (since your data pattern makes 0:s in statistics) and got
> something quite near to each other:

I'll find some time to look at both a short stream and a long stream,
based on the branch you've indicated and maybe against the latest
stable kernel.

---
bod

[toc] | [prev] | [next] | [standalone]

#1407620

On Tue, 2016-05-24 at 20:37 +0300, Andy Shevchenko wrote:
> Below results doesn't show DMA case (by the way, how did you disable
> DMA
> in 3.8.7?) for stock kernel.

The stock kernel on SPI flash for my early Gen1 is a non-DMA 8250
driver, so nothing needs to be done to that, it's basically just the
3.8.7 8250 in PIO mode.

> Anyway, I added preliminary support to change DMA mode run-time.
> Check
> my today topic/dw/qrk branch. It would be nice to see all 4 cases:
> DMA
> vs non-DMA on stock kernel (Yocto) and mine.

I'll have another look at your latest branch and run the same tests.

> 
> P.S. What is the point to test arbitrary tip branch?

Just tip-of-tree, we can benchmark against the 4.6 tag if you prefer.


---
bod

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web