m68k: Add DMA timers support for M5441x

[PATCH 0/2] m68k: Add DMA timers support for M5441x

Posted by Jean-Michel Hautbois 1 year, 2 months ago

This patch series adds support for DMA timers for the M5441x coldfire
family. The aim is to provide finer scheduler resolution and support for
high-resolution timers.

The first patch fixes the clocks and a typo. The second one is the
timers support addition. As there is no device tree, I did not use
TIMER_OF_DECLARE().

Signed-off-by: Jean-Michel Hautbois <jeanmichel.hautbois@yoseli.org>
---
Jean-Michel Hautbois (2):
      m68k: coldfire: Use proper clock rate for timers
      m68k: m5441x: Add DMA timer support

 MAINTAINERS                         |   6 +
 arch/m68k/coldfire/m5441x.c         |  20 +--
 arch/m68k/include/asm/m5441xsim.h   |  18 +++
 drivers/clocksource/Kconfig         |   9 ++
 drivers/clocksource/Makefile        |   1 +
 drivers/clocksource/mcf_dma_timer.c | 240 ++++++++++++++++++++++++++++++++++++
 6 files changed, 284 insertions(+), 10 deletions(-)
---
base-commit: e3f432391d55ec21274bd16a04659b4a24678535
change-id: 20241202-m5441x_dma_tmr-d969f4cc30a8

Best regards,
-- 
Jean-Michel Hautbois <jeanmichel.hautbois@yoseli.org>

Re: [PATCH 0/2] m68k: Add DMA timers support for M5441x

Posted by Jean-Michel Hautbois 1 year, 1 month ago

Hi !

On 02/12/2024 10:29, Jean-Michel Hautbois wrote:
> This patch series adds support for DMA timers for the M5441x coldfire
> family. The aim is to provide finer scheduler resolution and support for
> high-resolution timers.
> 
> The first patch fixes the clocks and a typo. The second one is the
> timers support addition. As there is no device tree, I did not use
> TIMER_OF_DECLARE().
> 
> Signed-off-by: Jean-Michel Hautbois <jeanmichel.hautbois@yoseli.org>
> ---
> Jean-Michel Hautbois (2):
>        m68k: coldfire: Use proper clock rate for timers
>        m68k: m5441x: Add DMA timer support
> 
>   MAINTAINERS                         |   6 +
>   arch/m68k/coldfire/m5441x.c         |  20 +--
>   arch/m68k/include/asm/m5441xsim.h   |  18 +++
>   drivers/clocksource/Kconfig         |   9 ++
>   drivers/clocksource/Makefile        |   1 +
>   drivers/clocksource/mcf_dma_timer.c | 240 ++++++++++++++++++++++++++++++++++++
>   6 files changed, 284 insertions(+), 10 deletions(-)
> ---
> base-commit: e3f432391d55ec21274bd16a04659b4a24678535
> change-id: 20241202-m5441x_dma_tmr-d969f4cc30a8
> 
> Best regards,

Gentle ping, as I don't know if anyone is interested by this patch :-) ?

I think it makes coldfire really better but it certainly has issues !

On my kernel, m54418, I observed a few things.

First, when launching cyclictest I get sometimes a bad irq handle:
[  720.442175] irq 24, desc: 6b62142b, depth: 1, count: 0, unhandled: 0
[  720.442329] ->handle_irq():  e4232a91, handle_bad_irq+0x0/0x1e6
[  720.442498] ->irq_data.chip(): 66a62b77, 0x4183e868
[  720.442613] ->action(): 00000000
[  720.442667] unexpected IRQ trap at vector 18

These IRQ numbers are a bit weird, I suppose it is a spurious IRQ, but 
not sure ?

Next, cyclictest might experience *very* long delay if the CPU is very 
busy and a ping flood burst comes in:
# cyclictest -m -p 98 --secaligned
WARN: stat /dev/cpu_dma_latency failed: No such file or directory
policy: fifo: loadavg: 2.24 1.94 1.03 1/127 252

T: 0 (  251) P:98 I:1000 C:  53113 Min:    116 Act:  327 Avg:  209 Max: 
  60353

Any review would be appreciated, at least to know if it is totally off 
or not that bad and needs a few fixes ?

Thanks,
JM

Re: [PATCH 0/2] m68k: Add DMA timers support for M5441x

Posted by Jean-Michel Hautbois 1 year, 2 months ago

On 02/12/2024 10:29, Jean-Michel Hautbois wrote:
> This patch series adds support for DMA timers for the M5441x coldfire
> family. The aim is to provide finer scheduler resolution and support for
> high-resolution timers.
> 
> The first patch fixes the clocks and a typo. The second one is the
> timers support addition. As there is no device tree, I did not use
> TIMER_OF_DECLARE().

I forgot to mention the result on a real M544118 board !
With PIT timer:
bash-5.2# cyclictest -p 80 -m -q -l 10000
WARN: stat /dev/cpu_dma_latency failed: No such file or directory
WARN: High resolution timers not available
T: 0 (  231) P:80 I:1000 C:  10000 Min:    277 Act: 1965 Avg: 1437 Max: 
   2023

With this driver and the following declaration in m5441x.c:
static struct resource mcf_dmatmr2_resource[] = {
	[0] = {
		.start = MCFDMATIMER_BASE2,
		.end   = MCFDMATIMER_BASE2 + 0xf,
		.flags = IORESOURCE_MEM,
	},
	[1] = {
		.start = MCFDMATIMER_IRQ_DTIM2,
		.end   = MCFDMATIMER_IRQ_DTIM2,
		.flags = IORESOURCE_IRQ,
	},
	[2] = {
		.start = MCFDMATIMER_IRQ_PRIO2,
		.end   = MCFDMATIMER_IRQ_PRIO2,
		.flags = IORESOURCE_REG,
		.name = "prio_reg",
	}
};

static struct platform_device mcf_dmatmr2 = {
	.name			= "mcftmr",
	.id			= 2,
	.num_resources = ARRAY_SIZE(mcf_dmatmr2_resource),
	.resource = mcf_dmatmr2_resource,
};

bash-5.2# cyclictest -p 80 -m -q -l 10000 -R
WARN: stat /dev/cpu_dma_latency failed: No such file or directory
WARN: reported clock resolution: 1 nsec
WARN: measured clock resolution approximately: 2376 nsec
T: 0 (  243) P:80 I:1000 C:  10000 Min:    104 Act:  125 Avg:  129 Max: 
    239

I don't know why the reported resolution is 1ns and not 8ns but it is 
not a big issue I suppose :-) ?

> 
> Signed-off-by: Jean-Michel Hautbois <jeanmichel.hautbois@yoseli.org>
> ---
> Jean-Michel Hautbois (2):
>        m68k: coldfire: Use proper clock rate for timers
>        m68k: m5441x: Add DMA timer support
> 
>   MAINTAINERS                         |   6 +
>   arch/m68k/coldfire/m5441x.c         |  20 +--
>   arch/m68k/include/asm/m5441xsim.h   |  18 +++
>   drivers/clocksource/Kconfig         |   9 ++
>   drivers/clocksource/Makefile        |   1 +
>   drivers/clocksource/mcf_dma_timer.c | 240 ++++++++++++++++++++++++++++++++++++
>   6 files changed, 284 insertions(+), 10 deletions(-)
> ---
> base-commit: e3f432391d55ec21274bd16a04659b4a24678535
> change-id: 20241202-m5441x_dma_tmr-d969f4cc30a8
> 
> Best regards,