| Message ID | 20210702225145.2643303-1-martin.blumenstingl@googlemail.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | clk: switch dividers to .determine_rate | expand |
Hi Stephen, On Sat, Jul 3, 2021 at 12:51 AM Martin Blumenstingl <martin.blumenstingl@googlemail.com> wrote: [...] > Martin Blumenstingl (6): > clk: divider: Implement and wire up .determine_rate by default > clk: imx: clk-divider-gate: Switch to clk_divider.determine_rate Abel has already picked this patch (thanks!) Can you please take the rest of the series (that is: patch 1 and patches 3-6)? I can also re-send them if you prefer that. Best regards, Martin
Quoting Martin Blumenstingl (2021-08-03 12:32:41) > Hi Stephen, > > > On Sat, Jul 3, 2021 at 12:51 AM Martin Blumenstingl > <martin.blumenstingl@googlemail.com> wrote: > [...] > > Martin Blumenstingl (6): > > clk: divider: Implement and wire up .determine_rate by default > > clk: imx: clk-divider-gate: Switch to clk_divider.determine_rate > Abel has already picked this patch (thanks!) > > Can you please take the rest of the series (that is: patch 1 and patches 3-6)? > I can also re-send them if you prefer that. > No need to resend. I can pick them up.
This is a continuation of patch #2 ("clk: divider: Switch from .round_rate to .determine_rate by default", [0]) from my other series: "clk: meson: rounding for fast clocks on 32-bit SoCs" [1] That patch caused NULL dereferences in various drivers which are still accessing clk_divider_ops.round_rate directly (which got removed in the mentioned patch). Guenter Roeck found this issue first on Freescale i.MX6 Ultralite and reported it here: [2] Later on Marek Szyprowski reported that BCM2835 is also affected: [3] With this improved version of the patch I am taking a more definsive approach as suggested by Stephen Boyd. Instead of dropping clk_divider_ops.round_rate we're now keeping it and adding clk_divider_ops.determine_rate. CCF core already prefers .determine_rate over .round_rate so the new implementation is used by default. To simplify the transition to .determine_rate in the future this updated series now has five extra patches to port over the drivers which used clk_divider_ops.round_rate over to clk_divider_ops.determine_rate. I have compile-tested all patches. Additionally the imx patch is runtime-tested using Guenter Roeck's suggestion (thanks!): $ qemu-system-arm -M mcimx6ul-evk -m 512M -kernel arch/arm/boot/zImage -dtb arch/arm/boot/dts/imx6ul-14x14-evk.dtb -append "console=ttymxc0 loglevel=8 earlycon earlyprintk" -monitor stdio This series is based on clk-next commit 783d08bd02f5d3 ("Revert "clk: divider: Switch from .round_rate to .determine_rate by default"clk-next"). My suggestion is to let all patches go through clk-next (instead of individual driver trees) as the first patch in this series is a pre- condition for the other ones. [0] https://patchwork.kernel.org/project/linux-clk/patch/20210627223959.188139-3-martin.blumenstingl@googlemail.com/ [1] https://patchwork.kernel.org/project/linux-clk/cover/20210627223959.188139-1-martin.blumenstingl@googlemail.com/ [2] https://lore.kernel.org/linux-clk/20210701202540.GA1085600@roeck-us.net/ [3] https://lore.kernel.org/linux-clk/e21c34a3-2586-057d-013b-6c8ec094d1a8@samsung.com/ Martin Blumenstingl (6): clk: divider: Implement and wire up .determine_rate by default clk: imx: clk-divider-gate: Switch to clk_divider.determine_rate clk: bcm2835: Switch to clk_divider.determine_rate clk: stm32f4: Switch to clk_divider.determine_rate clk: stm32h7: Switch to clk_divider.determine_rate clk: stm32mp1: Switch to clk_divider.determine_rate drivers/clk/bcm/clk-bcm2835.c | 9 ++++----- drivers/clk/clk-divider.c | 23 +++++++++++++++++++++++ drivers/clk/clk-stm32f4.c | 8 ++++---- drivers/clk/clk-stm32h7.c | 8 ++++---- drivers/clk/clk-stm32mp1.c | 10 +++------- drivers/clk/imx/clk-divider-gate.c | 10 +++++----- 6 files changed, 43 insertions(+), 25 deletions(-)