media: rc: gpio-ir-recv: add timeout property
Subsystems (3)
v2 cover media: rc: gpio-ir-recv: add timeout property
The default recorder timeout of 125ms is too high for some BPF protocol decoders when a remote sends repeat codes at high rates. This makes the timeout configurable via the devicetree.
Changes in v2:
fix checkpatch.pl warnings
Michael Klein (2):
media: rc: gpio-ir-recv: add recorder timeout property
media: bindings: media: gpio-ir-receiver: add linux,timeout-us
property
.../devicetree/bindings/media/gpio-ir-receiver.txt | 3 +++ Documentation/devicetree/bindings/media/rc.yaml | 6 ++++++ drivers/media/rc/gpio-ir-recv.c | 3 ++- 3 files changed, 11 insertions(+), 1 deletion(-)
Signature · 1 line
2.28.0
PATCH v2 1/2 media: rc: gpio-ir-recv: add recorder timeout property
This optional property allows to set the recorder timeout via the devicetree:
- linux,timeout-us: set the length of a space at which the recorder goes idle, specified in microseconds.
--- drivers/media/rc/gpio-ir-recv.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/media/rc/gpio-ir-recv.c +2 -1
@@ -99,7 +99,8 @@ static int gpio_ir_recv_probe(struct platform_device *pdev) rcdev->dev.parent = dev; rcdev->driver_name =KBUILD_MODNAME ; rcdev->min_timeout =1 ; - rcdev->timeout =IR_DEFAULT_TIMEOUT ; +if (of_property_read_u32 (np,"linux,timeout-us" , &rcdev->timeout)) + rcdev->timeout =IR_DEFAULT_TIMEOUT ; rcdev->max_timeout =10 *IR_DEFAULT_TIMEOUT ; rcdev->allowed_protocols =RC_PROTO_BIT_ALL_IR_DECODER ; rcdev->map_name =of_get_property (np,"linux,rc-map-name" ,NULL );
Signature · 1 line
2.28.0
PATCH v2 2/2 media: bindings: media: gpio-ir-receiver: add linux,timeout-us property
In reply to Michael KleinAdd linux,timeout-us for gpio ir receiver.
--- .../devicetree/bindings/media/gpio-ir-receiver.txt | 3 +++ Documentation/devicetree/bindings/media/rc.yaml | 6 ++++++ 2 files changed, 9 insertions(+)
Documentation/devicetree/bindings/media/gpio-ir-receiver.txt +3 -0
@@ -9,6 +9,8 @@ Optional properties: directory. - linux,autosuspend-period: autosuspend delay time, the unit is milisecond. + - linux,timeout-us: set the length of a space at which + the recorder goes idle, specified in microseconds. Example node:
@@ -17,4 +19,5 @@ Example node: gpios = <&gpio0 19 1>; linux,rc-map-name = "rc-rc6-mce"; linux,autosuspend-period = <125>; + linux,timeout-us = <125000>; };
Documentation/devicetree/bindings/media/rc.yaml +6 -0
@@ -151,4 +151,10 @@ properties: - rc-xbox-dvd - rc-zx-irdec + linux,timeout-us: + description: + Set the length of a space at which the recorder goes idle, specified in + microseconds. + $ref: '/schemas/types.yaml#/definitions/uint32' + additionalProperties: true
Signature · 1 line
2.28.0
Sean Young
maintainerQuoted 3 lines · Michael Klein
The default recorder timeout of 125ms is too high for some BPF protocol decoders when a remote sends repeat codes at high rates. This makes the timeout configurable via the devicetree.
To be honest, 125ms is too much by any measurement. The longest space in any protocol I'm aware of is 40ms in the sharp ir protocol. I think changing IR_DEFAUL_TIMEOUT to something like 50ms would make sense.
Also, when an BPF protocol is loaded, user-space can set the timeout with the LIRC_SET_REC_TIMEOUT ioctl which can depend on the protocol (set to longest space + 10ms error margin). This would mean that the bare minimum timeout can be set, which means decoding is as responsive as can be.
I'm not sure that device tree is really the place for this.
Thanks,
Sean
Quoted 16 lines
Changes in v2:
fix checkpatch.pl warningsMichael Klein (2):
media: rc: gpio-ir-recv: add recorder timeout property
media: bindings: media: gpio-ir-receiver: add linux,timeout-us
property.../devicetree/bindings/media/gpio-ir-receiver.txt | 3 +++ Documentation/devicetree/bindings/media/rc.yaml | 6 ++++++ drivers/media/rc/gpio-ir-recv.c | 3 ++- 3 files changed, 11 insertions(+), 1 deletion(-)Signature · 1 line
2.28.0
Quoted 8 lines · Sean Young
Quoted 3 lines · Michael Klein
The default recorder timeout of 125ms is too high for some BPF protocol decoders when a remote sends repeat codes at high rates. This makes the timeout configurable via the devicetree.
To be honest, 125ms is too much by any measurement. The longest space in any protocol I'm aware of is 40ms in the sharp ir protocol. I think changing
IR_DEFAUL_TIMEOUTto something like 50ms would make sense.
Seconded. I'm happy to prepare a patch if changing the default value is acceptable.
Quoted 3 lines
Also, when an BPF protocol is loaded, user-space can set the timeout with the
LIRC_SET_REC_TIMEOUTioctl which can depend on the protocol (set to longest space + 10ms error margin).
Right, although this is a bit cumbersome with current user-space tools. The BPF is loaded with ir-keytable, while the recorder timeout needs to be set with it-ctl. In the Debian world, those tools are even in different packages.
Quoted 5 lines
This would mean that the
bare minimum timeout can be set, which means decoding is as responsive as can be.I'm not sure that device tree is really the place for this.
Not arguing about this, but IMHO no less than for rc-map-name. So this seems to be at least consistent.
Thanks,
Michael
Sean Young
maintainerQuoted 6 lines · Michael Klein
Quoted 5 lines · Sean Young
Quoted 3 lines · Michael Klein
The default recorder timeout of 125ms is too high for some BPF protocol decoders when a remote sends repeat codes at high rates. This makes the timeout configurable via the devicetree.
Quoted 6 lines
Quoted 3 lines
To be honest, 125ms is too much by any measurement. The longest space in any protocol I'm aware of is 40ms in the sharp ir protocol. I think changing
IR_DEFAUL_TIMEOUTto something like 50ms would make sense.Seconded. I'm happy to prepare a patch if changing the default value is acceptable.
Actually I don't understand why the high timeout is an issue. It means that between ir messages you don't get a LIRC_TIMEOUT, just a LIRC_SPACE. Why is this a problem?
I'm not opposed to such a patch, but we would need to know if it really solves the problem you are having and it would need to sit in linux-next for some time.
Quoted 8 lines
Quoted 3 lines
Also, when an BPF protocol is loaded, user-space can set the timeout with the
LIRC_SET_REC_TIMEOUTioctl which can depend on the protocol (set to longest space + 10ms error margin).Right, although this is a bit cumbersome with current user-space tools. The BPF is loaded with ir-keytable, while the recorder timeout needs to be set with it-ctl. In the Debian world, those tools are even in different packages.
ir-keytable can use the LIRC_SET_REC_TIMEOUT ioctl to adjust the timeout. It has opened the lirc device already to load the bpf program. ir-keytable would need to calculate the minimum timeout needed for all enabled protocols (bpf and non-bpf). Then it can simply do the ioctl.
Quoted 8 lines
Quoted 5 lines
This would mean that the
bare minimum timeout can be set, which means decoding is as responsive as can be.I'm not sure that device tree is really the place for this.
Not arguing about this, but IMHO no less than for rc-map-name. So this seems to be at least consistent.
Well, I guess it can be argued. However, it can also be argued that it is not the best solution for this problem.
Sean
Quoted 11 lines · Sean Young
Quoted 10 lines · Michael Klein
Quoted 8 lines · Sean Young
Quoted 3 lines · Michael Klein
The default recorder timeout of 125ms is too high for some BPF protocol decoders when a remote sends repeat codes at high rates. This makes the timeout configurable via the devicetree.
To be honest, 125ms is too much by any measurement. The longest space in any protocol I'm aware of is 40ms in the sharp ir protocol. I think changing
IR_DEFAUL_TIMEOUTto something like 50ms would make sense.
Quoted 6 lines
Quoted 2 lines
Seconded. I'm happy to prepare a patch if changing the default value is acceptable.
Actually I don't understand why the high timeout is an issue. It means that between ir messages you don't get a
LIRC_TIMEOUT, just aLIRC_SPACE. Why is this a problem?
Never mind; this turned out do be a problem of the BPF protocol decoder, which relied on LIRC_TIMEOUT to terminate each IR message. After overhaul it is quite a bit simpler now and works fine with the long timeout.
Thank you for your insights.
Michael