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, 839 insertions, 0 deletions

diff --git a/drivers/firmware/tegra/bpmp-debugfs.c b/drivers/firmware/tegra/bpmp-debugfs.c
new file mode 100644
index 000000000..33c6300af
--- /dev/null
+++ b/drivers/firmware/tegra/bpmp-debugfs.c
@@ -0,0 +1,839 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (c) 2017, NVIDIA CORPORATION.  All rights reserved.
+ */
+
+#include <linux/cleanup.h>
+#include <linux/debugfs.h>
+#include <linux/dma-mapping.h>
+#include <linux/slab.h>
+#include <linux/uaccess.h>
+
+#include <soc/tegra/bpmp.h>
+#include <soc/tegra/bpmp-abi.h>
+
+static DEFINE_MUTEX(bpmp_debug_lock);
+
+struct seqbuf {
+	char *buf;
+	size_t pos;
+	size_t size;
+};
+
+static void seqbuf_init(struct seqbuf *seqbuf, void *buf, size_t size)
+{
+	seqbuf->buf = buf;
+	seqbuf->size = size;
+	seqbuf->pos = 0;
+}
+
+static size_t seqbuf_avail(struct seqbuf *seqbuf)
+{
+	return seqbuf->pos < seqbuf->size ? seqbuf->size - seqbuf->pos : 0;
+}
+
+static size_t seqbuf_status(struct seqbuf *seqbuf)
+{
+	return seqbuf->pos <= seqbuf->size ? 0 : -EOVERFLOW;
+}
+
+static int seqbuf_eof(struct seqbuf *seqbuf)
+{
+	return seqbuf->pos >= seqbuf->size;
+}
+
+static int seqbuf_read(struct seqbuf *seqbuf, void *buf, size_t nbyte)
+{
+	nbyte = min(nbyte, seqbuf_avail(seqbuf));
+	memcpy(buf, seqbuf->buf + seqbuf->pos, nbyte);
+	seqbuf->pos += nbyte;
+	return seqbuf_status(seqbuf);
+}
+
+static int seqbuf_read_u32(struct seqbuf *seqbuf, u32 *v)
+{
+	return seqbuf_read(seqbuf, v, 4);
+}
+
+static int seqbuf_read_str(struct seqbuf *seqbuf, const char **str)
+{
+	*str = seqbuf->buf + seqbuf->pos;
+	seqbuf->pos += strnlen(*str, seqbuf_avail(seqbuf));
+	seqbuf->pos++;
+	return seqbuf_status(seqbuf);
+}
+
+static void seqbuf_seek(struct seqbuf *seqbuf, ssize_t offset)
+{
+	seqbuf->pos += offset;
+}
+
+/* map filename in Linux debugfs to corresponding entry in BPMP */
+static const char *get_filename(struct tegra_bpmp *bpmp,
+				const struct file *file, char *buf, int size)
+{
+	const char *root_path, *filename = NULL;
+	char *root_path_buf;
+	size_t root_len;
+	size_t root_path_buf_len = 512;
+
+	root_path_buf = kzalloc(root_path_buf_len, GFP_KERNEL);
+	if (!root_path_buf)
+		return NULL;
+
+	root_path = dentry_path(bpmp->debugfs_mirror, root_path_buf,
+				root_path_buf_len);
+	if (IS_ERR(root_path))
+		goto out;
+
+	root_len = strlen(root_path);
+
+	filename = dentry_path(file->f_path.dentry, buf, size);
+	if (IS_ERR(filename)) {
+		filename = NULL;
+		goto out;
+	}
+
+	if (strlen(filename) < root_len || strncmp(filename, root_path, root_len)) {
+		filename = NULL;
+		goto out;
+	}
+
+	filename += root_len;
+
+out:
+	kfree(root_path_buf);
+	return filename;
+}
+
+static int mrq_debug_open(struct tegra_bpmp *bpmp, const char *name,
+			  u32 *fd, u32 *len, bool write)
+{
+	struct mrq_debug_request req = {
+		.cmd = write ? CMD_DEBUG_OPEN_WO : CMD_DEBUG_OPEN_RO,
+	};
+	struct mrq_debug_response resp;
+	struct tegra_bpmp_message msg = {
+		.mrq = MRQ_DEBUG,
+		.tx = {
+			.data = &req,
+			.size = sizeof(req),
+		},
+		.rx = {
+			.data = &resp,
+			.size = sizeof(resp),
+		},
+	};
+	ssize_t sz_name;
+	int err = 0;
+
+	sz_name = strscpy(req.fop.name, name, sizeof(req.fop.name));
+	if (sz_name < 0) {
+		pr_err("File name too large: %s\n", name);
+		return -EINVAL;
+	}
+
+	err = tegra_bpmp_transfer(bpmp, &msg);
+	if (err < 0)
+		return err;
+	else if (msg.rx.ret < 0)
+		return -EINVAL;
+
+	*len = resp.fop.datalen;
+	*fd = resp.fop.fd;
+
+	return 0;
+}
+
+static int mrq_debug_close(struct tegra_bpmp *bpmp, u32 fd)
+{
+	struct mrq_debug_request req = {
+		.cmd = CMD_DEBUG_CLOSE,
+		.frd = {
+			.fd = fd,
+		},
+	};
+	struct mrq_debug_response resp;
+	struct tegra_bpmp_message msg = {
+		.mrq = MRQ_DEBUG,
+		.tx = {
+			.data = &req,
+			.size = sizeof(req),
+		},
+		.rx = {
+			.data = &resp,
+			.size = sizeof(resp),
+		},
+	};
+	int err = 0;
+
+	err = tegra_bpmp_transfer(bpmp, &msg);
+	if (err < 0)
+		return err;
+	else if (msg.rx.ret < 0)
+		return -EINVAL;
+
+	return 0;
+}
+
+static int mrq_debug_read(struct tegra_bpmp *bpmp, const char *name,
+			  char *data, size_t sz_data, u32 *nbytes)
+{
+	struct mrq_debug_request req = {
+		.cmd = CMD_DEBUG_READ,
+	};
+	struct mrq_debug_response resp;
+	struct tegra_bpmp_message msg = {
+		.mrq = MRQ_DEBUG,
+		.tx = {
+			.data = &req,
+			.size = sizeof(req),
+		},
+		.rx = {
+			.data = &resp,
+			.size = sizeof(resp),
+		},
+	};
+	u32 fd = 0, len = 0;
+	int remaining, err, close_err;
+
+	mutex_lock(&bpmp_debug_lock);
+	err = mrq_debug_open(bpmp, name, &fd, &len, 0);
+	if (err)
+		goto out;
+
+	if (len > sz_data) {
+		err = -EFBIG;
+		goto close;
+	}
+
+	req.frd.fd = fd;
+	remaining = len;
+
+	while (remaining > 0) {
+		err = tegra_bpmp_transfer(bpmp, &msg);
+		if (err < 0) {
+			goto close;
+		} else if (msg.rx.ret < 0) {
+			err = -EINVAL;
+			goto close;
+		}
+
+		if (resp.frd.readlen > remaining) {
+			pr_err("%s: read data length invalid\n", __func__);
+			err = -EINVAL;
+			goto close;
+		}
+
+		memcpy(data, resp.frd.data, resp.frd.readlen);
+		data += resp.frd.readlen;
+		remaining -= resp.frd.readlen;
+	}
+
+	*nbytes = len;
+
+close:
+	close_err = mrq_debug_close(bpmp, fd);
+	if (!err)
+		err = close_err;
+out:
+	mutex_unlock(&bpmp_debug_lock);
+	return err;
+}
+
+static int mrq_debug_write(struct tegra_bpmp *bpmp, const char *name,
+			   uint8_t *data, size_t sz_data)
+{
+	struct mrq_debug_request req = {
+		.cmd = CMD_DEBUG_WRITE
+	};
+	struct mrq_debug_response resp;
+	struct tegra_bpmp_message msg = {
+		.mrq = MRQ_DEBUG,
+		.tx = {
+			.data = &req,
+			.size = sizeof(req),
+		},
+		.rx = {
+			.data = &resp,
+			.size = sizeof(resp),
+		},
+	};
+	u32 fd = 0, len = 0;
+	size_t remaining;
+	int err;
+
+	mutex_lock(&bpmp_debug_lock);
+	err = mrq_debug_open(bpmp, name, &fd, &len, 1);
+	if (err)
+		goto out;
+
+	if (sz_data > len) {
+		err = -EINVAL;
+		goto close;
+	}
+
+	req.fwr.fd = fd;
+	remaining = sz_data;
+
+	while (remaining > 0) {
+		len = min(remaining, sizeof(req.fwr.data));
+		memcpy(req.fwr.data, data, len);
+		req.fwr.datalen = len;
+
+		err = tegra_bpmp_transfer(bpmp, &msg);
+		if (err < 0) {
+			goto close;
+		} else if (msg.rx.ret < 0) {
+			err = -EINVAL;
+			goto close;
+		}
+
+		data += req.fwr.datalen;
+		remaining -= req.fwr.datalen;
+	}
+
+close:
+	err = mrq_debug_close(bpmp, fd);
+out:
+	mutex_unlock(&bpmp_debug_lock);
+	return err;
+}
+
+static int bpmp_debug_show(struct seq_file *m, void *p)
+{
+	struct file *file = m->private;
+	struct inode *inode = file_inode(file);
+	struct tegra_bpmp *bpmp = inode->i_private;
+	char fnamebuf[256];
+	const char *filename;
+	struct mrq_debug_request req = {
+		.cmd = CMD_DEBUG_READ,
+	};
+	struct mrq_debug_response resp;
+	struct tegra_bpmp_message msg = {
+		.mrq = MRQ_DEBUG,
+		.tx = {
+			.data = &req,
+			.size = sizeof(req),
+		},
+		.rx = {
+			.data = &resp,
+			.size = sizeof(resp),
+		},
+	};
+	u32 fd = 0, len = 0;
+	int remaining, err, close_err;
+
+	filename = get_filename(bpmp, file, fnamebuf, sizeof(fnamebuf));
+	if (!filename)
+		return -ENOENT;
+
+	mutex_lock(&bpmp_debug_lock);
+	err = mrq_debug_open(bpmp, filename, &fd, &len, 0);
+	if (err)
+		goto out;
+
+	req.frd.fd = fd;
+	remaining = len;
+
+	while (remaining > 0) {
+		err = tegra_bpmp_transfer(bpmp, &msg);
+		if (err < 0) {
+			goto close;
+		} else if (msg.rx.ret < 0) {
+			err = -EINVAL;
+			goto close;
+		}
+
+		if (resp.frd.readlen > remaining) {
+			pr_err("%s: read data length invalid\n", __func__);
+			err = -EINVAL;
+			goto close;
+		}
+
+		seq_write(m, resp.frd.data, resp.frd.readlen);
+		remaining -= resp.frd.readlen;
+	}
+
+close:
+	close_err = mrq_debug_close(bpmp, fd);
+	if (!err)
+		err = close_err;
+out:
+	mutex_unlock(&bpmp_debug_lock);
+	return err;
+}
+
+static ssize_t bpmp_debug_store(struct file *file, const char __user *buf,
+		size_t count, loff_t *f_pos)
+{
+	struct inode *inode = file_inode(file);
+	struct tegra_bpmp *bpmp = inode->i_private;
+	char *databuf = NULL;
+	char fnamebuf[256];
+	const char *filename;
+	ssize_t err;
+
+	filename = get_filename(bpmp, file, fnamebuf, sizeof(fnamebuf));
+	if (!filename)
+		return -ENOENT;
+
+	databuf = memdup_user(buf, count);
+	if (IS_ERR(databuf))
+		return PTR_ERR(databuf);
+
+	err = mrq_debug_write(bpmp, filename, databuf, count);
+	kfree(databuf);
+
+	return err ?: count;
+}
+
+static int bpmp_debug_open(struct inode *inode, struct file *file)
+{
+	return single_open_size(file, bpmp_debug_show, file, SZ_256K);
+}
+
+static const struct file_operations bpmp_debug_fops = {
+	.open		= bpmp_debug_open,
+	.read		= seq_read,
+	.llseek		= seq_lseek,
+	.write		= bpmp_debug_store,
+	.release	= single_release,
+};
+
+static int bpmp_populate_debugfs_inband(struct tegra_bpmp *bpmp,
+					struct dentry *parent,
+					char *ppath)
+{
+	const size_t pathlen = SZ_256;
+	const size_t bufsize = SZ_16K;
+	struct dentry *dentry;
+	u32 dsize, attrs = 0;
+	struct seqbuf seqbuf;
+	char *buf, *pathbuf;
+	const char *name;
+	int err = 0;
+
+	if (!bpmp || !parent || !ppath)
+		return -EINVAL;
+
+	buf = kmalloc(bufsize, GFP_KERNEL);
+	if (!buf)
+		return -ENOMEM;
+
+	pathbuf = kzalloc(pathlen, GFP_KERNEL);
+	if (!pathbuf) {
+		kfree(buf);
+		return -ENOMEM;
+	}
+
+	err = mrq_debug_read(bpmp, ppath, buf, bufsize, &dsize);
+	if (err)
+		goto out;
+
+	seqbuf_init(&seqbuf, buf, dsize);
+
+	while (!seqbuf_eof(&seqbuf)) {
+		err = seqbuf_read_u32(&seqbuf, &attrs);
+		if (err)
+			goto out;
+
+		err = seqbuf_read_str(&seqbuf, &name);
+		if (err < 0)
+			goto out;
+
+		if (attrs & DEBUGFS_S_ISDIR) {
+			size_t len;
+
+			dentry = debugfs_create_dir(name, parent);
+			if (IS_ERR(dentry)) {
+				err = PTR_ERR(dentry);
+				goto out;
+			}
+
+			len = snprintf(pathbuf, pathlen, "%s%s/", ppath, name);
+			if (len >= pathlen) {
+				err = -EINVAL;
+				goto out;
+			}
+
+			err = bpmp_populate_debugfs_inband(bpmp, dentry,
+							   pathbuf);
+			if (err < 0)
+				goto out;
+		} else {
+			umode_t mode;
+
+			mode = attrs & DEBUGFS_S_IRUSR ? 0400 : 0;
+			mode |= attrs & DEBUGFS_S_IWUSR ? 0200 : 0;
+			dentry = debugfs_create_file(name, mode, parent, bpmp,
+						     &bpmp_debug_fops);
+			if (IS_ERR(dentry)) {
+				err = PTR_ERR(dentry);
+				goto out;
+			}
+		}
+	}
+
+out:
+	kfree(pathbuf);
+	kfree(buf);
+
+	return err;
+}
+
+static int mrq_debugfs_read(struct tegra_bpmp *bpmp,
+			    dma_addr_t name, size_t sz_name,
+			    dma_addr_t data, size_t sz_data,
+			    size_t *nbytes)
+{
+	struct mrq_debugfs_request req = {
+		.cmd = CMD_DEBUGFS_READ,
+		.fop = {
+			.fnameaddr = (u32)name,
+			.fnamelen = (u32)sz_name,
+			.dataaddr = (u32)data,
+			.datalen = (u32)sz_data,
+		},
+	};
+	struct mrq_debugfs_response resp;
+	struct tegra_bpmp_message msg = {
+		.mrq = MRQ_DEBUGFS,
+		.tx = {
+			.data = &req,
+			.size = sizeof(req),
+		},
+		.rx = {
+			.data = &resp,
+			.size = sizeof(resp),
+		},
+	};
+	int err;
+
+	err = tegra_bpmp_transfer(bpmp, &msg);
+	if (err < 0)
+		return err;
+	else if (msg.rx.ret < 0)
+		return -EINVAL;
+
+	*nbytes = (size_t)resp.fop.nbytes;
+
+	return 0;
+}
+
+static int mrq_debugfs_write(struct tegra_bpmp *bpmp,
+			     dma_addr_t name, size_t sz_name,
+			     dma_addr_t data, size_t sz_data)
+{
+	const struct mrq_debugfs_request req = {
+		.cmd = CMD_DEBUGFS_WRITE,
+		.fop = {
+			.fnameaddr = (u32)name,
+			.fnamelen = (u32)sz_name,
+			.dataaddr = (u32)data,
+			.datalen = (u32)sz_data,
+		},
+	};
+	struct tegra_bpmp_message msg = {
+		.mrq = MRQ_DEBUGFS,
+		.tx = {
+			.data = &req,
+			.size = sizeof(req),
+		},
+	};
+
+	return tegra_bpmp_transfer(bpmp, &msg);
+}
+
+static int mrq_debugfs_dumpdir(struct tegra_bpmp *bpmp, dma_addr_t addr,
+			       size_t size, size_t *nbytes)
+{
+	const struct mrq_debugfs_request req = {
+		.cmd = CMD_DEBUGFS_DUMPDIR,
+		.dumpdir = {
+			.dataaddr = (u32)addr,
+			.datalen = (u32)size,
+		},
+	};
+	struct mrq_debugfs_response resp;
+	struct tegra_bpmp_message msg = {
+		.mrq = MRQ_DEBUGFS,
+		.tx = {
+			.data = &req,
+			.size = sizeof(req),
+		},
+		.rx = {
+			.data = &resp,
+			.size = sizeof(resp),
+		},
+	};
+	int err;
+
+	err = tegra_bpmp_transfer(bpmp, &msg);
+	if (err < 0)
+		return err;
+	else if (msg.rx.ret < 0)
+		return -EINVAL;
+
+	*nbytes = (size_t)resp.dumpdir.nbytes;
+
+	return 0;
+}
+
+static int debugfs_show(struct seq_file *m, void *p)
+{
+	struct file *file = m->private;
+	struct inode *inode = file_inode(file);
+	struct tegra_bpmp *bpmp = inode->i_private;
+	const size_t datasize = m->size;
+	const size_t namesize = SZ_256;
+	void *datavirt, *namevirt;
+	dma_addr_t dataphys, namephys;
+	char buf[256];
+	const char *filename;
+	size_t len, nbytes;
+	int err;
+
+	filename = get_filename(bpmp, file, buf, sizeof(buf));
+	if (!filename)
+		return -ENOENT;
+
+	namevirt = dma_alloc_coherent(bpmp->dev, namesize, &namephys,
+				      GFP_KERNEL | GFP_DMA32);
+	if (!namevirt)
+		return -ENOMEM;
+
+	datavirt = dma_alloc_coherent(bpmp->dev, datasize, &dataphys,
+				      GFP_KERNEL | GFP_DMA32);
+	if (!datavirt) {
+		err = -ENOMEM;
+		goto free_namebuf;
+	}
+
+	len = strlen(filename);
+	strscpy_pad(namevirt, filename, namesize);
+
+	err = mrq_debugfs_read(bpmp, namephys, len, dataphys, datasize,
+			       &nbytes);
+
+	if (!err)
+		seq_write(m, datavirt, nbytes);
+
+	dma_free_coherent(bpmp->dev, datasize, datavirt, dataphys);
+free_namebuf:
+	dma_free_coherent(bpmp->dev, namesize, namevirt, namephys);
+
+	return err;
+}
+
+static int debugfs_open(struct inode *inode, struct file *file)
+{
+	return single_open_size(file, debugfs_show, file, SZ_128K);
+}
+
+static ssize_t debugfs_store(struct file *file, const char __user *buf,
+		size_t count, loff_t *f_pos)
+{
+	struct inode *inode = file_inode(file);
+	struct tegra_bpmp *bpmp = inode->i_private;
+	const size_t datasize = count;
+	const size_t namesize = SZ_256;
+	void *datavirt, *namevirt;
+	dma_addr_t dataphys, namephys;
+	char fnamebuf[256];
+	const char *filename;
+	size_t len;
+	int err;
+
+	filename = get_filename(bpmp, file, fnamebuf, sizeof(fnamebuf));
+	if (!filename)
+		return -ENOENT;
+
+	namevirt = dma_alloc_coherent(bpmp->dev, namesize, &namephys,
+				      GFP_KERNEL | GFP_DMA32);
+	if (!namevirt)
+		return -ENOMEM;
+
+	datavirt = dma_alloc_coherent(bpmp->dev, datasize, &dataphys,
+				      GFP_KERNEL | GFP_DMA32);
+	if (!datavirt) {
+		err = -ENOMEM;
+		goto free_namebuf;
+	}
+
+	len = strlen(filename);
+	strscpy_pad(namevirt, filename, namesize);
+
+	if (copy_from_user(datavirt, buf, count)) {
+		err = -EFAULT;
+		goto free_databuf;
+	}
+
+	err = mrq_debugfs_write(bpmp, namephys, len, dataphys,
+				count);
+
+free_databuf:
+	dma_free_coherent(bpmp->dev, datasize, datavirt, dataphys);
+free_namebuf:
+	dma_free_coherent(bpmp->dev, namesize, namevirt, namephys);
+
+	return err ?: count;
+}
+
+static const struct file_operations debugfs_fops = {
+	.open		= debugfs_open,
+	.read		= seq_read,
+	.llseek		= seq_lseek,
+	.write		= debugfs_store,
+	.release	= single_release,
+};
+
+static int bpmp_populate_dir(struct tegra_bpmp *bpmp, struct seqbuf *seqbuf,
+			     struct dentry *parent, u32 depth)
+{
+	int err;
+	u32 d, t;
+	const char *name;
+	struct dentry *dentry;
+
+	while (!seqbuf_eof(seqbuf)) {
+		err = seqbuf_read_u32(seqbuf, &d);
+		if (err < 0)
+			return err;
+
+		if (d < depth) {
+			seqbuf_seek(seqbuf, -4);
+			/* go up a level */
+			return 0;
+		} else if (d != depth) {
+			/* malformed data received from BPMP */
+			return -EIO;
+		}
+
+		err = seqbuf_read_u32(seqbuf, &t);
+		if (err < 0)
+			return err;
+		err = seqbuf_read_str(seqbuf, &name);
+		if (err < 0)
+			return err;
+
+		if (t & DEBUGFS_S_ISDIR) {
+			dentry = debugfs_create_dir(name, parent);
+			if (IS_ERR(dentry))
+				return PTR_ERR(dentry);
+			err = bpmp_populate_dir(bpmp, seqbuf, dentry, depth+1);
+			if (err < 0)
+				return err;
+		} else {
+			umode_t mode;
+
+			mode = t & DEBUGFS_S_IRUSR ? S_IRUSR : 0;
+			mode |= t & DEBUGFS_S_IWUSR ? S_IWUSR : 0;
+			dentry = debugfs_create_file(name, mode,
+						     parent, bpmp,
+						     &debugfs_fops);
+			if (IS_ERR(dentry))
+				return PTR_ERR(dentry);
+		}
+	}
+
+	return 0;
+}
+
+static int bpmp_populate_debugfs_shmem(struct tegra_bpmp *bpmp)
+{
+	struct seqbuf seqbuf;
+	const size_t sz = SZ_512K;
+	dma_addr_t phys;
+	size_t nbytes;
+	void *virt;
+	int err;
+
+	virt = dma_alloc_coherent(bpmp->dev, sz, &phys,
+				  GFP_KERNEL | GFP_DMA32);
+	if (!virt)
+		return -ENOMEM;
+
+	err = mrq_debugfs_dumpdir(bpmp, phys, sz, &nbytes);
+	if (err < 0) {
+		goto free;
+	} else if (nbytes > sz) {
+		err = -EINVAL;
+		goto free;
+	}
+
+	seqbuf_init(&seqbuf, virt, nbytes);
+	err = bpmp_populate_dir(bpmp, &seqbuf, bpmp->debugfs_mirror, 0);
+free:
+	dma_free_coherent(bpmp->dev, sz, virt, phys);
+
+	return err;
+}
+
+static DEFINE_MUTEX(bpmp_debugfs_root_lock);
+static struct dentry *bpmp_debugfs_root;
+
+static struct dentry *bpmp_debugfs_get_root(void)
+{
+	guard(mutex)(&bpmp_debugfs_root_lock);
+	if (!bpmp_debugfs_root)
+		bpmp_debugfs_root = debugfs_create_dir("bpmp", NULL);
+	return bpmp_debugfs_root;
+}
+
+int tegra_bpmp_init_debugfs(struct tegra_bpmp *bpmp)
+{
+	struct dentry *root, *d;
+	char name[32];
+	bool inband;
+	int err;
+
+	inband = tegra_bpmp_mrq_is_supported(bpmp, MRQ_DEBUG);
+
+	if (!inband && !tegra_bpmp_mrq_is_supported(bpmp, MRQ_DEBUGFS))
+		return 0;
+
+	root = bpmp_debugfs_get_root();
+	if (IS_ERR(root))
+		return PTR_ERR(root);
+
+	if (dev_to_node(bpmp->dev) == NUMA_NO_NODE) {
+		d = root;
+	} else {
+		snprintf(name, sizeof(name), "%d-bpmp", dev_to_node(bpmp->dev));
+		d = debugfs_create_dir(name, root);
+		if (IS_ERR(d)) {
+			err = PTR_ERR(d);
+			goto out;
+		}
+	}
+
+	bpmp->debugfs_mirror = debugfs_create_dir("debug", d);
+	if (IS_ERR(bpmp->debugfs_mirror)) {
+		err = PTR_ERR(bpmp->debugfs_mirror);
+		goto out;
+	}
+
+	if (inband)
+		err = bpmp_populate_debugfs_inband(bpmp, bpmp->debugfs_mirror,
+						   "/");
+	else
+		err = bpmp_populate_debugfs_shmem(bpmp);
+
+out:
+	if (err < 0) {
+		if (!IS_ERR(d))
+			debugfs_remove_recursive(d);
+
+		guard(mutex)(&bpmp_debugfs_root_lock);
+		if (root == d) {
+			bpmp_debugfs_root = NULL;
+		} else if (simple_empty(root)) {
+			debugfs_remove(root);
+			bpmp_debugfs_root = NULL;
+		}
+	}
+
+	return err;
+}