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, 178 insertions, 0 deletions

diff --git a/drivers/gpu/host1x/context.c b/drivers/gpu/host1x/context.c
new file mode 100644
index 000000000..512f3d189
--- /dev/null
+++ b/drivers/gpu/host1x/context.c
@@ -0,0 +1,178 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (c) 2021, NVIDIA Corporation.
+ */
+
+#include <linux/device.h>
+#include <linux/host1x_context_bus.h>
+#include <linux/kref.h>
+#include <linux/of.h>
+#include <linux/of_device.h>
+#include <linux/pid.h>
+#include <linux/slab.h>
+
+#include "context.h"
+#include "dev.h"
+
+static void host1x_memory_context_release(struct device *dev)
+{
+	/* context device is freed in host1x_memory_context_list_free() */
+}
+
+int host1x_memory_context_list_init(struct host1x *host1x)
+{
+	struct host1x_memory_context_list *cdl = &host1x->context_list;
+	struct device_node *node = host1x->dev->of_node;
+	struct host1x_memory_context *ctx;
+	unsigned int i;
+	int err;
+
+	cdl->devs = NULL;
+	cdl->len = 0;
+	mutex_init(&cdl->lock);
+
+	err = of_property_count_u32_elems(node, "iommu-map");
+	if (err < 0)
+		return 0;
+
+	cdl->len = err / 4;
+	cdl->devs = kzalloc_objs(*cdl->devs, cdl->len);
+	if (!cdl->devs)
+		return -ENOMEM;
+
+	for (i = 0; i < cdl->len; i++) {
+		ctx = &cdl->devs[i];
+
+		ctx->host = host1x;
+
+		device_initialize(&ctx->dev);
+
+		/*
+		 * Due to an issue with T194 NVENC, only 38 bits can be used.
+		 * Anyway, 256GiB of IOVA ought to be enough for anyone.
+		 */
+		ctx->dma_mask = DMA_BIT_MASK(38);
+		ctx->dev.dma_mask = &ctx->dma_mask;
+		ctx->dev.coherent_dma_mask = ctx->dma_mask;
+		dev_set_name(&ctx->dev, "host1x-ctx.%d", i);
+		ctx->dev.bus = &host1x_context_device_bus_type;
+		ctx->dev.parent = host1x->dev;
+		ctx->dev.release = host1x_memory_context_release;
+
+		ctx->dev.dma_parms = &ctx->dma_parms;
+		dma_set_max_seg_size(&ctx->dev, UINT_MAX);
+
+		err = device_add(&ctx->dev);
+		if (err) {
+			dev_err(host1x->dev, "could not add context device %d: %d\n", i, err);
+			put_device(&ctx->dev);
+			goto unreg_devices;
+		}
+
+		err = of_dma_configure_id(&ctx->dev, node, true, &i);
+		if (err) {
+			dev_err(host1x->dev, "IOMMU configuration failed for context device %d: %d\n",
+				i, err);
+			device_unregister(&ctx->dev);
+			goto unreg_devices;
+		}
+
+		if (!tegra_dev_iommu_get_stream_id(&ctx->dev, &ctx->stream_id) ||
+		    !device_iommu_mapped(&ctx->dev)) {
+			dev_err(host1x->dev, "Context device %d has no IOMMU!\n", i);
+			device_unregister(&ctx->dev);
+
+			/*
+			 * This means that if IOMMU is disabled but context devices
+			 * are defined in the device tree, Host1x will fail to probe.
+			 * That's probably OK in this time and age.
+			 */
+			err = -EINVAL;
+
+			goto unreg_devices;
+		}
+	}
+
+	return 0;
+
+unreg_devices:
+	while (i--)
+		device_unregister(&cdl->devs[i].dev);
+
+	kfree(cdl->devs);
+	cdl->devs = NULL;
+	cdl->len = 0;
+
+	return err;
+}
+
+void host1x_memory_context_list_free(struct host1x_memory_context_list *cdl)
+{
+	unsigned int i;
+
+	for (i = 0; i < cdl->len; i++)
+		device_unregister(&cdl->devs[i].dev);
+
+	kfree(cdl->devs);
+	cdl->len = 0;
+}
+
+struct host1x_memory_context *host1x_memory_context_alloc(struct host1x *host1x,
+							  struct device *dev,
+							  struct pid *pid)
+{
+	struct host1x_memory_context_list *cdl = &host1x->context_list;
+	struct host1x_memory_context *free = NULL;
+	int i;
+
+	if (!cdl->len)
+		return ERR_PTR(-EOPNOTSUPP);
+
+	mutex_lock(&cdl->lock);
+
+	for (i = 0; i < cdl->len; i++) {
+		struct host1x_memory_context *cd = &cdl->devs[i];
+
+		if (cd->dev.iommu->iommu_dev != dev->iommu->iommu_dev)
+			continue;
+
+		if (cd->owner == pid) {
+			refcount_inc(&cd->ref);
+			mutex_unlock(&cdl->lock);
+			return cd;
+		} else if (!cd->owner && !free) {
+			free = cd;
+		}
+	}
+
+	if (!free) {
+		mutex_unlock(&cdl->lock);
+		return ERR_PTR(-EBUSY);
+	}
+
+	refcount_set(&free->ref, 1);
+	free->owner = get_pid(pid);
+
+	mutex_unlock(&cdl->lock);
+
+	return free;
+}
+EXPORT_SYMBOL_GPL(host1x_memory_context_alloc);
+
+void host1x_memory_context_get(struct host1x_memory_context *cd)
+{
+	refcount_inc(&cd->ref);
+}
+EXPORT_SYMBOL_GPL(host1x_memory_context_get);
+
+void host1x_memory_context_put(struct host1x_memory_context *cd)
+{
+	struct host1x_memory_context_list *cdl = &cd->host->context_list;
+
+	if (refcount_dec_and_mutex_lock(&cd->ref, &cdl->lock)) {
+		put_pid(cd->owner);
+		cd->owner = NULL;
+		mutex_unlock(&cdl->lock);
+	}
+}
+EXPORT_SYMBOL_GPL(host1x_memory_context_put);