On 29/08/2019 16:12, Philipp Zabel wrote:
On Wed, 2019-08-28 at 10:19 +0300, Tero Kristo wrote:
Add initial PRM (Power and Reset Management) driver for TI OMAP class
SoCs. Initially this driver only supports reset control, but can be
extended to support rest of the functionality, like powerdomain
control, PRCM irq support etc.
Signed-off-by: Tero Kristo <t-kristo@xxxxxx>
---
arch/arm/mach-omap2/Kconfig | 1 +
drivers/soc/ti/Makefile | 1 +
drivers/soc/ti/omap_prm.c | 235 ++++++++++++++++++++++++++++++++++++
3 files changed, 237 insertions(+)
create mode 100644 drivers/soc/ti/omap_prm.c
diff --git a/arch/arm/mach-omap2/Kconfig b/arch/arm/mach-omap2/Kconfig
index fdb6743760a2..ad08d470a2ca 100644
--- a/arch/arm/mach-omap2/Kconfig
+++ b/arch/arm/mach-omap2/Kconfig
@@ -109,6 +109,7 @@ config ARCH_OMAP2PLUS
select TI_SYSC
select OMAP_IRQCHIP
select CLKSRC_TI_32K
+ select ARCH_HAS_RESET_CONTROLLER
help
Systems based on OMAP2, OMAP3, OMAP4 or OMAP5
diff --git a/drivers/soc/ti/Makefile b/drivers/soc/ti/Makefile
index b3868d392d4f..788b5cd1e180 100644
--- a/drivers/soc/ti/Makefile
+++ b/drivers/soc/ti/Makefile
@@ -6,6 +6,7 @@ obj-$(CONFIG_KEYSTONE_NAVIGATOR_QMSS) += knav_qmss.o
knav_qmss-y := knav_qmss_queue.o knav_qmss_acc.o
obj-$(CONFIG_KEYSTONE_NAVIGATOR_DMA) += knav_dma.o
obj-$(CONFIG_AMX3_PM) += pm33xx.o
+obj-$(CONFIG_ARCH_OMAP2PLUS) += omap_prm.o
obj-$(CONFIG_WKUP_M3_IPC) += wkup_m3_ipc.o
obj-$(CONFIG_TI_SCI_PM_DOMAINS) += ti_sci_pm_domains.o
obj-$(CONFIG_TI_SCI_INTA_MSI_DOMAIN) += ti_sci_inta_msi.o
diff --git a/drivers/soc/ti/omap_prm.c b/drivers/soc/ti/omap_prm.c
new file mode 100644
index 000000000000..fd5c431f8736
--- /dev/null
+++ b/drivers/soc/ti/omap_prm.c
@@ -0,0 +1,235 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * OMAP2+ PRM driver
+ *
+ * Copyright (C) 2019 Texas Instruments Incorporated - http://www.ti.com/
+ * Tero Kristo <t-kristo@xxxxxx>
+ */
+
+#include <linux/kernel.h>
+#include <linux/module.h>
Why <linux/module.h>? This is a builtin driver.
Yeah, not anymore. Let me ditch this.
+#include <linux/device.h>
+#include <linux/io.h>
+#include <linux/of.h>
+#include <linux/of_device.h>
+#include <linux/platform_device.h>
+#include <linux/reset-controller.h>
+#include <linux/delay.h>
+
+struct omap_rst_map {
+ s8 rst;
+ s8 st;
+};
+
+struct omap_prm_data {
+ u32 base;
+ const char *name;
+ u16 rstctrl;
+ u16 rstst;
+ const struct omap_rst_map *rstmap;
+ u8 flags;
+};
I wonder if splitting rstctrl/rstst/rstmap out into a separate structure
would make sense. That could be linked from omap_reset_data directly.
That only makes sense if there'd be enough cases where it can be reused
for multiple PRMs instances.
Hmm, splitting these out would make it possible to share the bits for
ipu:s across devices, same for dsp:s and eve:s.
However, adding too many levels of abstraction makes it kind of
difficult to follow what is happening with the code, and it would only
save maybe ~100 bytes of memory at the moment.
+
+struct omap_prm {
+ const struct omap_prm_data *data;
+ void __iomem *base;
+};
+
+struct omap_reset_data {
+ struct reset_controller_dev rcdev;
+ struct omap_prm *prm;
+};
+
+#define to_omap_reset_data(p) container_of((p), struct omap_reset_data, rcdev)
+
+#define OMAP_MAX_RESETS 8
+#define OMAP_RESET_MAX_WAIT 10000
+
+#define OMAP_PRM_HAS_RSTCTRL BIT(0)
+#define OMAP_PRM_HAS_RSTST BIT(1)
+
+#define OMAP_PRM_HAS_RESETS (OMAP_PRM_HAS_RSTCTRL | OMAP_PRM_HAS_RSTST)
+
+static const struct of_device_id omap_prm_id_table[] = {
+ { },
+};
+
+static bool _is_valid_reset(struct omap_reset_data *reset, unsigned long id)
+{
+ const struct omap_rst_map *map = reset->prm->data->rstmap;
+
+ while (map && map->rst >= 0) {
If rstmap is never NULL,
while (map->rst >= 0) {
should be enough.
I'll actually re-write this to use the reset mask.
+ if (map->rst == id)
+ return true;
+ map++;
+ }
+
+ return false;
+}
+
+static int omap_reset_status(struct reset_controller_dev *rcdev,
+ unsigned long id)
+{
+ struct omap_reset_data *reset = to_omap_reset_data(rcdev);
+ u32 v;
+
+ if (!_is_valid_reset(reset, id))
+ return -EINVAL;
Don't check this in the status/(de)assert/reset callbacks. Instead,
implement rcdev.of_xlate and return -EINVAL there, so that invalid ids
can never be requested.
Yeah, let me do that.
+ v = readl_relaxed(reset->prm->base + reset->prm->data->rstst);
+ v &= 1 << id;
+ v >>= id;
omap_get_st_bit below makes it look like the status bit can be in a
different place than the reset control bit, should that be used here as
well?
True, this is a bug.
+
+ return v;
+}
+
+static int omap_reset_assert(struct reset_controller_dev *rcdev,
+ unsigned long id)
+{
+ struct omap_reset_data *reset = to_omap_reset_data(rcdev);
+ u32 v;
+
+ if (!_is_valid_reset(reset, id))
+ return -EINVAL;
Same as above.
Will drop this one.
+ /* assert the reset control line */
+ v = readl_relaxed(reset->prm->base + reset->prm->data->rstctrl);
+ v |= 1 << id;
+ writel_relaxed(v, reset->prm->base + reset->prm->data->rstctrl);
This read-modify-write should be protected with a lock.
Ok, will add a spinlock. I did think about this before but all the cases
where we are sharing a reset register are to be controlled from the same
process / driver, and the events are effectively serialized. Doesn't
hurt adding it for possible future need though.
+
+ return 0;
+}
+
+static int omap_reset_get_st_bit(struct omap_reset_data *reset,
+ unsigned long id)
+{
+ const struct omap_rst_map *map = reset->prm->data->rstmap;
+
+ while (map && map->rst >= 0) {
Same as above.
Yeah, usage of rstmap is now enforced, so no need to check it here.
+ if (map->rst == id)
+ return map->st;
+
+ map++;
+ }
+
+ return id;
+}
+
+/*
+ * Note that status will not change until clocks are on, and clocks cannot be
+ * enabled until reset is deasserted. Consumer drivers must check status
+ * separately after enabling clocks.
+ */
+static int omap_reset_deassert(struct reset_controller_dev *rcdev,
+ unsigned long id)
+{
+ struct omap_reset_data *reset = to_omap_reset_data(rcdev);
+ u32 v;
+ int st_bit;
+ bool has_rstst;
+
+ if (!_is_valid_reset(reset, id))
+ return -EINVAL;
Same as above.
Will drop.
+ /* check the current status to avoid de-asserting the line twice */
+ v = readl_relaxed(reset->prm->base + reset->prm->data->rstctrl);
+ if (!(v & BIT(id)))
+ return -EEXIST;
What is the purpose of this? For shared consumers the core already does
refcounting, and I expect exclusive consumers won't deassert twice.
Since the reset signal is deasserted after this call, this should not
return an error.
This is actually a leftover from legacy code; this driver is mostly a
move of the reset handling from platform codebase to be an actual driver
of its own. But yes, I believe this can be dropped.
+
+ has_rstst = reset->prm->data->rstst ||
+ (reset->prm->data->flags & OMAP_PRM_HAS_RSTST);
+
+ if (has_rstst) {
+ st_bit = omap_reset_get_st_bit(reset, id);
+
+ /* Clear the reset status by writing 1 to the status bit */
+ v = readl_relaxed(reset->prm->base + reset->prm->data->rstst);
+ v |= 1 << st_bit;
+ writel_relaxed(v, reset->prm->base + reset->prm->data->rstst);
What does the value read from the rstst register indicate? Is it the
current state of the reset line? Is it 0 until deassertion is completed,
and then it turns to 1?
Value of 1 indicates that the corresponding IP has been reset
successfully. Writing back 1 to the same bit clears it out, so the
status can be polled later on.
+ }
+
+ /* de-assert the reset control line */
+ v = readl_relaxed(reset->prm->base + reset->prm->data->rstctrl);
Reading the register again seems unnecessary.
I dropped the earlier read for now, so this is again needed.
+ v &= ~(1 << id);
+ writel_relaxed(v, reset->prm->base + reset->prm->data->rstctrl);
As above, the read-modify-write should be locked.
Yep, will protect this.
+
+ return 0;
+}
+
+static const struct reset_control_ops omap_reset_ops = {
+ .assert = omap_reset_assert,
+ .deassert = omap_reset_deassert,
+ .status = omap_reset_status,
+};
+
+static int omap_prm_reset_init(struct platform_device *pdev,
+ struct omap_prm *prm)
+{
+ struct omap_reset_data *reset;
+
+ /*
+ * Check if we have controllable resets. If either rstctrl is non-zero
+ * or OMAP_PRM_HAS_RSTCTRL flag is set, we have reset control register
+ * for the domain.
+ */
+ if (!prm->data->rstctrl && !(prm->data->flags & OMAP_PRM_HAS_RSTCTRL))
+ return 0;
+
+ reset = devm_kzalloc(&pdev->dev, sizeof(*reset), GFP_KERNEL);
+ if (!reset)
+ return -ENOMEM;
+
+ reset->rcdev.owner = THIS_MODULE;
+ reset->rcdev.ops = &omap_reset_ops;
+ reset->rcdev.of_node = pdev->dev.of_node;
+ reset->rcdev.nr_resets = OMAP_MAX_RESETS;
+
+ reset->prm = prm;
+
+ return devm_reset_controller_register(&pdev->dev, &reset->rcdev);
+}
+
+static int omap_prm_probe(struct platform_device *pdev)
+{
+ struct resource *res;
+ const struct omap_prm_data *data;
+ struct omap_prm *prm;
+ const struct of_device_id *match;
+
+ res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+ if (!res)
+ return -ENODEV;
This can be merged with devm_ioremap_resource below.
Well, I actually use the "res" later on to map the DT node to the
corresponding prm_data based on address.
+ match = of_match_device(omap_prm_id_table, &pdev->dev);
+ if (!match)
+ return -ENOTSUPP;
+
+ prm = devm_kzalloc(&pdev->dev, sizeof(*prm), GFP_KERNEL);
+ if (!prm)
+ return -ENOMEM;
+
+ data = match->data;
+
+ while (data->base != res->start) {
+ if (!data->base)
+ return -EINVAL;
+ data++;
+ }
Is this not something that you want to have encoded in the compatible
string? They all have a different register layout.
With the addition of all the prm instances later on, this changes. Most
of the prm instances will have same register layout then. See v1 data
that was posted earlier [1], but which I dropped for now to keep this
series isolated for reset handling only. In this patch, you see that for
DRA7, all the power domain handling related PRM instances have identical
register layout, they just differ based on base address.
[1] https://www.spinics.net/lists/linux-omap/msg149872.html
It would be possible to encode all of this based on different
compatibles, but then the amount of different compatible strings would
explode... DRA7 is just one SoC.
+
+ prm->data = data;
+
+ prm->base = devm_ioremap_resource(&pdev->dev, res);
prm->base = devm_platform_ioremap_resource(pdev, 0);
I still need the "res" pointer as indicated above.
-Tero
+ if (!prm->base)
+ return -ENOMEM;
+
+ return omap_prm_reset_init(pdev, prm);
+}
+
+static struct platform_driver omap_prm_driver = {
+ .probe = omap_prm_probe,
+ .driver = {
+ .name = KBUILD_MODNAME,
+ .of_match_table = omap_prm_id_table,
+ },
+};
+builtin_platform_driver(omap_prm_driver);
regards
Philipp
--
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki