Re: [PATCH v8 0/11] dmtimer adaptation to platform_driver

[Kevin Hilman <khilman@ti.com>]

Tarun Kanti DebBarma <tarun.kanti@ti.com> writes:

> dmtimer adaptation to platform_driver.
>
> This patch series is adaptation of dmtimer code to platform driver
> using omap_device and omap_hwmod abstraction.
>
> Tested on following platforms:
> OMAP1710 H3 SDP
> OMAP2420 SDP
> OMAP2430 SDP
> OMAP3430 SDP
> OMAP3630 SDP
> OMAP4430 SDP

How exactly was this tested?   Boot tested only?   Do you have a test
that attempts to request/configure/enable all the timers?  The timer-gp
code allows all the way to GPT12 which obviously will not work in this
series.

Also, testing with PM on 34xx/n900, I noticed that this series prevents
PER and CORE from hitting retention during suspend.  I haven't debugged
why yet.

Kevin


> Baseline:
> git://git.kernel.org/pub/scm/linux/kernel/git/tmlind/linux-omap-2.6.git
> Branch: origin/omap-for-linus
>
> v8:
> (1) Baselined on Tony's tree in omap-for-linus branch
>
> (2) The last patch in v7 series has been removed because it is fixed
> by following patch:
> commit: 78f26e872f77b6312273216de1a8f836c6f2e143
> OMAP: hwmod: Set autoidle after smartidle during _sysc_enable
>
>
> TODO:
> (1) OFF Mode support
>
> (2) Upgrade timeout implementation in low-level read/write access to return
> error condition to EXPORT APIs. This is re-frained in the present implementation
> because that would involve change to EXPORTED APIs. Besides, there is no clear
> design as yet which is agreed upon by the community.
>
> v7:
> (1) In omap1_dm_timer_set_src(), the computation of shift value to respective
> dmtimer clock source was corrected:
> From:
> int n = (pdev->id) << 1;
> To:
> int n = (pdev->id - 1) << 1;
>
> This change is needed because dmtimer is indexed from 1 now instead of 0.
>
> (2) In  omap1_dm_timer_init(void) memory resource end address chnaged:
> From:
> res[0].end = base + 0xff;
> To:
> res[0].end = base + 0x46;
>
> This was causing request_mem_region() failure in driver probe().
>
> (3) In the export APIs there are some calls which are not applicable to OMAP1.
> They have been made conditional now. They include following calls:
>
> timer->fclk = clk_get(&timer->pdev->dev, "fck");
> omap_dm_timer_enable()
> omap_dm_timer_disable()
>
> (4) Remove usage of cpu_is_omap16xx() and instead a flag has been added in
> struct dmtimer_platform_data {
> ...
> u32 is_omap16xx:1;
> }
>
> This flag is set to 1 in mach-omap1/dmtimer.c and set to 0 in mach-omap2/dmtimer.c
> This flag is used in plat-omap/dmtimer.c wherever it needs to distiguish omap16xx.
>
> (5) Remove #include <plat/omap_device.h> from mach-omap1/dmtimer.c
>
> (6) Instead of using macros like INT_24XX_GPTIMERx, use the numbers
> directly in OMAP2420, OMAP2430 and OMAP3xxx hwmod database.
>
> (7) pm_runtime_get_sync() and pm_runtime_put_sync() return value check modified
> from positive to negative value:
>
> if (pm_runtime_get_sync(...) < 0) {
> ...
> }
>
> v6:
> (1) Removed reset functions to mach-omap1/dmtimer.c.
> Access to reset function from plat-omap/dmtimer.c is provided by means
> of function pointer.
>
> (2) Remove multiple calls to omap_device_build() for registering timer devices
> during early and regular initialization. Regular device registration is now done
> by reading data from temporary list. This list is populated during early init
> where timer data is read from hwmod database and corresponding memory allocated.
>
> (3) kfree(pdata) under error condition since platform_device_unregister does
> not free its pdata.
>
> (4) Removed extra header inclusion in mach-omap2 and plat-omap
>
> NOTE: omap_dm_timer.<id> field could not be removed because during regular boot
> there is no mechanism to match the current pdev with corresponding entry in the
> timer list which was partially initialized during early boot.
>
> v4:
> (1) clock aliases are renamed as "32k_ck", "sys_ck" and "alt_ck"
> (2) incorporate missing clk_put() for corresponding clk_get()
> (3) modified clk_get()/clk_put() to be called once once in platform driver.
> (4) consistent header for new files
> (5) check return value of omap_hwmod_for_each_by_class() in device init
> routines.
> (6) remove is_abe_timer field in dmtimer_platform_data structure. this is
> no longer needed with new input clock source aliasing.
> (7) proper splitting of patch series
> (8) remove register map from hwmod database.
> (9) remove clock source strings array from hwmod database and associated
> structure declaration from plat/dmtimer.h. this is no longer needed.
> (10) remove dev_attr from hwmod database. this is no longer needed.
> (11) use register offsets to identify OMAP 4 registers instead of register map.
> (12) remove clock source name strings from hwmod database.
> (13) introduce new mechanism for getting struct clk associated with clock source
> names. this is achieved by adding clock alisases for all supported clock sources.
> (14) remove clock setup functions in mach-omap2 for populating struct clk
> associated with all input clock sources because this is no longer needed with
> above implementation.
> (15) device names changed from dmtimer to omap-timer
> (16) device index starts from 1 instead of 0
> (17) remove .init_name from hwmod database. this is not needed.
> (18) introduce separate functions for reading/writing interrupt registers instead of
> doing all operations within a single function.
>
> v3:
> (1) multi-line comment error correction
> (2) provision to allow any of the available dmtimers as early timers
> instead of restricting them to millisecond timers only.
> (3) in 'struct omap_dmtimer{}' is_initialized flag is redundant and
> so must be removed. if the element is found in the list it is already
> initialized.
> (4) remove 'found' flag in omap_dm_timer_request() and
> omap_dm_timer_request_specific() functions.
> this is not needed with alternate implementation.
> (5) use .init_name to initialize device names so that it can be identified
> during early boot as well. This is to avoid duplicate functions for clock
> manipulations during early boot and later.
> (6) remove redundant functions from mach-omap2 which are created just to
> call pm functions like: pm_runtime_get_sync(),pm_runtime_put_sync(),..
> and instead call them directly from plat-omap function api's.
> (7) timer clock source names made part of hwmod database. source_clock[]
> of type 'struct clk' is made part of platform data.
> (8) clockactivity field initialized in hwmod database to preserve fclk
> during idle. code which manipulate OCP config removed since they are
> already taken care by hwmod framework.
> (9) omap2_dm_timer_set_src() is optimized. Clock enable/disbale routines
> moved to plat-omap layer and simplfied to the level so as not to sacrifice
> intended functionality.
> NOTE: During early boot clock management was requested to be placed upon
> client drivers responsibility. this has not been done keeping in mind
> that it would entail (i) multiple modifications of client drivers (ii) it
> would violate the purpose of having a framework (open to debate).
> (10) dmtimer register maps moved to hwmod database
>
> v2:
> (1) removed dedicated functions for early timer clock access.
> instead, now we have common functions for early and normal timers.
> (2) removed usage of clock source strings for reading corresponding
> struct clks. this is now achieved through clock aliases introduced
> for each input clock sources.
> (3) IP revision to distinguish new IP standard and the rest and then
> initialize dmtimer interrupt and functional offsets.
> (4) provision to initialize all dmtimers as early timers.
> (5) remove dm_timer_setup() function because this is no longer needed.
> (6) modify the device index to start from 1 instead of 0.
> (7) device name changed from dmtimer to omap-timer
> (8) extract device ids' from hwmod name and same used for device build.
> (9) additional resource allocation checks and free
> (10) early timer variable initialization
> (11) initialize timer_ip_type and register offsets in platform data structure.
> (12) some more comments/logs
>
> Tarun Kanti DebBarma (11):
>   OMAP2+: dmtimer: add device names to flck nodes
>   OMAP2420: hwmod data: add dmtimer
>   OMAP2430: hwmod data: add dmtimer
>   OMAP3: hwmod data: add dmtimer
>   OMAP4: hwmod data: add dmtimer
>   OMAP1: dmtimer: conversion to platform devices
>   OMAP2+: dmtimer: convert to platform devices
>   OMAP: dmtimer: platform driver
>   OMAP: dmtimer: switch-over to platform device driver
>   OMAP: dmtimer: pm_runtime support
>   OMAP: dmtimer: add timeout to low-level routines
>
>  arch/arm/mach-omap1/Makefile               |    2 +-
>  arch/arm/mach-omap1/dmtimer.c              |  217 +++++++++
>  arch/arm/mach-omap1/timer32k.c             |    4 -
>  arch/arm/mach-omap2/Makefile               |    2 +-
>  arch/arm/mach-omap2/clock2420_data.c       |   60 ++-
>  arch/arm/mach-omap2/clock2430_data.c       |   60 ++-
>  arch/arm/mach-omap2/clock3xxx_data.c       |   48 ++-
>  arch/arm/mach-omap2/clock44xx_data.c       |   45 ++-
>  arch/arm/mach-omap2/dmtimer.c              |  255 +++++++++++
>  arch/arm/mach-omap2/dmtimer.h              |   30 ++
>  arch/arm/mach-omap2/io.c                   |    4 +-
>  arch/arm/mach-omap2/omap_hwmod_2420_data.c |  634 ++++++++++++++++++++++++++
>  arch/arm/mach-omap2/omap_hwmod_2430_data.c |  633 ++++++++++++++++++++++++++
>  arch/arm/mach-omap2/omap_hwmod_3xxx_data.c |  675 ++++++++++++++++++++++++++++
>  arch/arm/mach-omap2/omap_hwmod_44xx_data.c |  622 +++++++++++++++++++++++++
>  arch/arm/mach-omap2/timer-gp.c             |    1 -
>  arch/arm/plat-omap/dmtimer.c               |  608 +++++++++++++------------
>  arch/arm/plat-omap/include/plat/dmtimer.h  |   39 ++-
>  18 files changed, 3584 insertions(+), 355 deletions(-)
>  create mode 100644 arch/arm/mach-omap1/dmtimer.c
>  create mode 100644 arch/arm/mach-omap2/dmtimer.c
>  create mode 100644 arch/arm/mach-omap2/dmtimer.h
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-omap" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html