sched_ext: Maintain an online cid mask in the scheduler arenagrafted

Schedulers on the default cid mapping treat [0, nr_online_cids) as the
online set and restart on hotplug. Schedulers that install their own mapping
with scx_bpf_cid_override() have no way to learn which cids are online: the
count no longer identifies members and the CPU-form cpumask is unusable from
cid programs. This is an obvious hole in the cid API.

Add scx_bpf_online_cmask(), a kernel-maintained cmask in the scheduler's
arena, allocated alongside the per-CPU scratch masks and populated after the
cid mapping is finalized and before ops.init(), for child schedulers too.
The pointer stays valid through ops.exit() with no reference to take. It is
the arena offset as a void pointer, the same form struct_ops arena arguments
arrive in. The verifier types the void return as a scalar for the program's
arena cast.

The mask follows the SCX hotplug notifications: seeded from cpu_active_mask
and updated before ops.cid_online/offline() runs, so it lags cpu_online_mask
only inside a hotplug transition. Updates walk the scheduler list under the
lock that also serializes unlinking. Reads are live, not atomic snapshots.
Root initialization excludes hotplug.

v2: Reworded the getter kerneldoc (Andrea Righi).

Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Andrea Righi <arighi@nvidia.com>

1 files changed, 259 insertions, 0 deletions

diff --git a/drivers/memory/dfl-emif.c b/drivers/memory/dfl-emif.c
new file mode 100644
index 000000000..da06cd30a
--- /dev/null
+++ b/drivers/memory/dfl-emif.c
@@ -0,0 +1,259 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * DFL device driver for EMIF private feature
+ *
+ * Copyright (C) 2020 Intel Corporation, Inc.
+ *
+ */
+#include <linux/bitfield.h>
+#include <linux/dfl.h>
+#include <linux/errno.h>
+#include <linux/io.h>
+#include <linux/iopoll.h>
+#include <linux/io-64-nonatomic-lo-hi.h>
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/spinlock.h>
+#include <linux/types.h>
+
+#define FME_FEATURE_ID_EMIF		0x9
+
+#define EMIF_STAT			0x8
+#define EMIF_STAT_INIT_DONE_SFT		0
+#define EMIF_STAT_CALC_FAIL_SFT		8
+#define EMIF_STAT_CLEAR_BUSY_SFT	16
+#define EMIF_CTRL			0x10
+#define EMIF_CTRL_CLEAR_EN_SFT		0
+#define EMIF_CTRL_CLEAR_EN_MSK		GENMASK_ULL(7, 0)
+
+#define EMIF_POLL_INVL			10000 /* us */
+#define EMIF_POLL_TIMEOUT		5000000 /* us */
+
+/*
+ * The Capability Register replaces the Control Register (at the same
+ * offset) for EMIF feature revisions > 0. The bitmask that indicates
+ * the presence of memory channels exists in both the Capability Register
+ * and Control Register definitions. These can be thought of as a C union.
+ * The Capability Register definitions are used to check for the existence
+ * of a memory channel, and the Control Register definitions are used for
+ * managing the memory-clear functionality in revision 0.
+ */
+#define EMIF_CAPABILITY_BASE		0x10
+#define EMIF_CAPABILITY_CHN_MSK_V0	GENMASK_ULL(3, 0)
+#define EMIF_CAPABILITY_CHN_MSK		GENMASK_ULL(7, 0)
+
+struct dfl_emif {
+	struct device *dev;
+	void __iomem *base;
+	spinlock_t lock;	/* Serialises access to EMIF_CTRL reg */
+};
+
+struct emif_attr {
+	struct device_attribute attr;
+	u32 shift;
+	u32 index;
+};
+
+#define to_emif_attr(dev_attr) \
+	container_of(dev_attr, struct emif_attr, attr)
+
+static ssize_t emif_state_show(struct device *dev,
+			       struct device_attribute *attr, char *buf)
+{
+	struct emif_attr *eattr = to_emif_attr(attr);
+	struct dfl_emif *de = dev_get_drvdata(dev);
+	u64 val;
+
+	val = readq(de->base + EMIF_STAT);
+
+	return sysfs_emit(buf, "%u\n",
+			  !!(val & BIT_ULL(eattr->shift + eattr->index)));
+}
+
+static ssize_t emif_clear_store(struct device *dev,
+				struct device_attribute *attr,
+				const char *buf, size_t count)
+{
+	struct emif_attr *eattr = to_emif_attr(attr);
+	struct dfl_emif *de = dev_get_drvdata(dev);
+	u64 clear_busy_msk, clear_en_msk, val;
+	void __iomem *base = de->base;
+
+	if (!sysfs_streq(buf, "1"))
+		return -EINVAL;
+
+	clear_busy_msk = BIT_ULL(EMIF_STAT_CLEAR_BUSY_SFT + eattr->index);
+	clear_en_msk = BIT_ULL(EMIF_CTRL_CLEAR_EN_SFT + eattr->index);
+
+	spin_lock(&de->lock);
+	/* The CLEAR_EN field is WO, but other fields are RW */
+	val = readq(base + EMIF_CTRL);
+	val &= ~EMIF_CTRL_CLEAR_EN_MSK;
+	val |= clear_en_msk;
+	writeq(val, base + EMIF_CTRL);
+	spin_unlock(&de->lock);
+
+	if (readq_poll_timeout(base + EMIF_STAT, val,
+			       !(val & clear_busy_msk),
+			       EMIF_POLL_INVL, EMIF_POLL_TIMEOUT)) {
+		dev_err(de->dev, "timeout, fail to clear\n");
+		return -ETIMEDOUT;
+	}
+
+	return count;
+}
+
+#define emif_state_attr(_name, _shift, _index)				\
+	static struct emif_attr emif_attr_##inf##_index##_##_name =	\
+		{ .attr = __ATTR(inf##_index##_##_name, 0444,		\
+				 emif_state_show, NULL),		\
+		  .shift = (_shift), .index = (_index) }
+
+#define emif_clear_attr(_index)						\
+	static struct emif_attr emif_attr_##inf##_index##_clear =	\
+		{ .attr = __ATTR(inf##_index##_clear, 0200,		\
+				 NULL, emif_clear_store),		\
+		  .index = (_index) }
+
+emif_state_attr(init_done, EMIF_STAT_INIT_DONE_SFT, 0);
+emif_state_attr(init_done, EMIF_STAT_INIT_DONE_SFT, 1);
+emif_state_attr(init_done, EMIF_STAT_INIT_DONE_SFT, 2);
+emif_state_attr(init_done, EMIF_STAT_INIT_DONE_SFT, 3);
+emif_state_attr(init_done, EMIF_STAT_INIT_DONE_SFT, 4);
+emif_state_attr(init_done, EMIF_STAT_INIT_DONE_SFT, 5);
+emif_state_attr(init_done, EMIF_STAT_INIT_DONE_SFT, 6);
+emif_state_attr(init_done, EMIF_STAT_INIT_DONE_SFT, 7);
+
+emif_state_attr(cal_fail, EMIF_STAT_CALC_FAIL_SFT, 0);
+emif_state_attr(cal_fail, EMIF_STAT_CALC_FAIL_SFT, 1);
+emif_state_attr(cal_fail, EMIF_STAT_CALC_FAIL_SFT, 2);
+emif_state_attr(cal_fail, EMIF_STAT_CALC_FAIL_SFT, 3);
+emif_state_attr(cal_fail, EMIF_STAT_CALC_FAIL_SFT, 4);
+emif_state_attr(cal_fail, EMIF_STAT_CALC_FAIL_SFT, 5);
+emif_state_attr(cal_fail, EMIF_STAT_CALC_FAIL_SFT, 6);
+emif_state_attr(cal_fail, EMIF_STAT_CALC_FAIL_SFT, 7);
+
+
+emif_clear_attr(0);
+emif_clear_attr(1);
+emif_clear_attr(2);
+emif_clear_attr(3);
+emif_clear_attr(4);
+emif_clear_attr(5);
+emif_clear_attr(6);
+emif_clear_attr(7);
+
+
+static struct attribute *dfl_emif_attrs[] = {
+	&emif_attr_inf0_init_done.attr.attr,
+	&emif_attr_inf0_cal_fail.attr.attr,
+	&emif_attr_inf0_clear.attr.attr,
+
+	&emif_attr_inf1_init_done.attr.attr,
+	&emif_attr_inf1_cal_fail.attr.attr,
+	&emif_attr_inf1_clear.attr.attr,
+
+	&emif_attr_inf2_init_done.attr.attr,
+	&emif_attr_inf2_cal_fail.attr.attr,
+	&emif_attr_inf2_clear.attr.attr,
+
+	&emif_attr_inf3_init_done.attr.attr,
+	&emif_attr_inf3_cal_fail.attr.attr,
+	&emif_attr_inf3_clear.attr.attr,
+
+	&emif_attr_inf4_init_done.attr.attr,
+	&emif_attr_inf4_cal_fail.attr.attr,
+	&emif_attr_inf4_clear.attr.attr,
+
+	&emif_attr_inf5_init_done.attr.attr,
+	&emif_attr_inf5_cal_fail.attr.attr,
+	&emif_attr_inf5_clear.attr.attr,
+
+	&emif_attr_inf6_init_done.attr.attr,
+	&emif_attr_inf6_cal_fail.attr.attr,
+	&emif_attr_inf6_clear.attr.attr,
+
+	&emif_attr_inf7_init_done.attr.attr,
+	&emif_attr_inf7_cal_fail.attr.attr,
+	&emif_attr_inf7_clear.attr.attr,
+
+	NULL,
+};
+
+static umode_t dfl_emif_visible(struct kobject *kobj,
+				struct attribute *attr, int n)
+{
+	struct dfl_emif *de = dev_get_drvdata(kobj_to_dev(kobj));
+	struct emif_attr *eattr = container_of(attr, struct emif_attr,
+					       attr.attr);
+	struct dfl_device *ddev = to_dfl_dev(de->dev);
+	u64 val;
+
+	/*
+	 * This device supports up to 8 memory interfaces, but not all
+	 * interfaces are used on different platforms. The read out value of
+	 * CAPABILITY_CHN_MSK field (which is a bitmap) indicates which
+	 * interfaces are available.
+	 */
+	if (ddev->revision > 0 && strstr(attr->name, "_clear"))
+		return 0;
+
+	if (ddev->revision == 0)
+		val = FIELD_GET(EMIF_CAPABILITY_CHN_MSK_V0,
+				readq(de->base + EMIF_CAPABILITY_BASE));
+	else
+		val = FIELD_GET(EMIF_CAPABILITY_CHN_MSK,
+				readq(de->base + EMIF_CAPABILITY_BASE));
+
+	return (val & BIT_ULL(eattr->index)) ? attr->mode : 0;
+}
+
+static const struct attribute_group dfl_emif_group = {
+	.is_visible = dfl_emif_visible,
+	.attrs = dfl_emif_attrs,
+};
+
+static const struct attribute_group *dfl_emif_groups[] = {
+	&dfl_emif_group,
+	NULL,
+};
+
+static int dfl_emif_probe(struct dfl_device *ddev)
+{
+	struct device *dev = &ddev->dev;
+	struct dfl_emif *de;
+
+	de = devm_kzalloc(dev, sizeof(*de), GFP_KERNEL);
+	if (!de)
+		return -ENOMEM;
+
+	de->base = devm_ioremap_resource(dev, &ddev->mmio_res);
+	if (IS_ERR(de->base))
+		return PTR_ERR(de->base);
+
+	de->dev = dev;
+	spin_lock_init(&de->lock);
+	dev_set_drvdata(dev, de);
+
+	return 0;
+}
+
+static const struct dfl_device_id dfl_emif_ids[] = {
+	{ FME_ID, FME_FEATURE_ID_EMIF },
+	{ }
+};
+MODULE_DEVICE_TABLE(dfl, dfl_emif_ids);
+
+static struct dfl_driver dfl_emif_driver = {
+	.drv	= {
+		.name       = "dfl-emif",
+		.dev_groups = dfl_emif_groups,
+	},
+	.id_table = dfl_emif_ids,
+	.probe   = dfl_emif_probe,
+};
+module_dfl_driver(dfl_emif_driver);
+
+MODULE_DESCRIPTION("DFL EMIF driver");
+MODULE_AUTHOR("Intel Corporation");
+MODULE_LICENSE("GPL v2");