Re: [PATCH 2/2] drivers/misc: Add Aspeed P2A control driver

[Andrew Jeffery]

On Fri, 22 Feb 2019, at 08:55, Patrick Venture wrote:
> The ASPEED AST2400, and AST2500 in some configurations include a
> PCI-to-AHB MMIO bridge.  This bridge allows a server to read and write
> in the BMC's memory space.

Bit of a nit, but I think s/memory space/physical address space/ makes
the power of the interface a bit clearer.

>  This feature is especially useful when using
> this bridge to send large files to the BMC.
> 
> The host may use this to send down a firmware image by staging data at a
> specific memory address, and in a coordinated effort with the BMC's
> software stack and kernel, transmit the bytes.
> 
> This driver enables the BMC to unlock the PCI bridge on demand, and
> configure it via ioctl to allow the host to write bytes to an agreed
> upon location.  In the primary use-case, the region to use is known
> apriori on the BMC, and the host requests this information.  Once this
> request is received, the BMC's software stack will enable the bridge and
> the region and then using some software flow control (possibly via IPMI
> packets), copy the bytes down.  Once the process is complete, the BMC
> will disable the bridge and unset any region involved.

I feel the description is a little subtle. You mention locations and regions
without really defining their relationship. We have the means to prevent
writes via the P2A to following regions in the BMC's physical address space:

* BMC flash MMIO window
* System flash MMIO windows
* SOC IO (peripheral MMIO)
* DRAM

So what I think should be made clear is once we allow the host to write
to e.g. DRAM, it can write to *all* of DRAM, regardless of what location the
BMC recommended, i.e. the BMC is at the mercy of the host wrt
confidentiality once the P2A is enabled (host can always read anywhere)
and integrity when the DRAM write filter is disabled.

There is no way to specify and constrain P2A writes to specific locations
in DRAM.

> 
> The default behavior of this bridge when present is: enabled and all
> regions marked read-write.  This driver will fix the regions to be
> read-only and then disable the bridge entirely.
> 
> Signed-off-by: Patrick Venture <venture@xxxxxxxxxx>
> ---
>  drivers/misc/Kconfig                 |   8 +
>  drivers/misc/Makefile                |   1 +
>  drivers/misc/aspeed-p2a-ctrl.c       | 498 +++++++++++++++++++++++++++
>  include/uapi/linux/aspeed-p2a-ctrl.h |  46 +++
>  4 files changed, 553 insertions(+)
>  create mode 100644 drivers/misc/aspeed-p2a-ctrl.c
>  create mode 100644 include/uapi/linux/aspeed-p2a-ctrl.h
> 
> diff --git a/drivers/misc/Kconfig b/drivers/misc/Kconfig
> index f417b06e11c5..54ed265a26f0 100644
> --- a/drivers/misc/Kconfig
> +++ b/drivers/misc/Kconfig
> @@ -485,6 +485,14 @@ config VEXPRESS_SYSCFG
>  	  bus. System Configuration interface is one of the possible means
>  	  of generating transactions on this bus.
>  
> +config ASPEED_P2A_CTRL
> +	depends on (ARCH_ASPEED || COMPILE_TEST)
> +	tristate "Aspeed ast2400/2500 HOST P2A VGA MMIO to BMC bridge control"
> +	help
> +	  Control Aspeed ast2400/2500 HOST P2A VGA MMIO to BMC mappings 
> through
> +	  ioctl()s, the driver also provides an interface for userspace 
> mappings to
> +	  a pre-defined region.
> +
>  config ASPEED_LPC_CTRL
>  	depends on (ARCH_ASPEED || COMPILE_TEST) && REGMAP && MFD_SYSCON
>  	tristate "Aspeed ast2400/2500 HOST LPC to BMC bridge control"
> diff --git a/drivers/misc/Makefile b/drivers/misc/Makefile
> index e39ccbbc1b3a..57577aee354f 100644
> --- a/drivers/misc/Makefile
> +++ b/drivers/misc/Makefile
> @@ -55,6 +55,7 @@ obj-$(CONFIG_VEXPRESS_SYSCFG)	+= vexpress-syscfg.o
>  obj-$(CONFIG_CXL_BASE)		+= cxl/
>  obj-$(CONFIG_ASPEED_LPC_CTRL)	+= aspeed-lpc-ctrl.o
>  obj-$(CONFIG_ASPEED_LPC_SNOOP)	+= aspeed-lpc-snoop.o
> +obj-$(CONFIG_ASPEED_P2A_CTRL)	+= aspeed-p2a-ctrl.o
>  obj-$(CONFIG_PCI_ENDPOINT_TEST)	+= pci_endpoint_test.o
>  obj-$(CONFIG_OCXL)		+= ocxl/
>  obj-y				+= cardreader/
> diff --git a/drivers/misc/aspeed-p2a-ctrl.c 
> b/drivers/misc/aspeed-p2a-ctrl.c
> new file mode 100644
> index 000000000000..a3cf00de9038
> --- /dev/null
> +++ b/drivers/misc/aspeed-p2a-ctrl.c
> @@ -0,0 +1,498 @@
> +/*
> + * Copyright 2019 Google Inc
> + *
> + * This program is free software; you can redistribute it and/or
> + * modify it under the terms of the GNU General Public License
> + * as published by the Free Software Foundation; either version
> + * 2 of the License, or (at your option) any later version.
> + *
> + * Provides a simple driver to control the ASPEED P2A interface which 
> allows
> + * the host to read and write to various regions of the BMC's memory.
> + */
> +
> +#include <linux/fs.h>
> +#include <linux/io.h>
> +#include <linux/miscdevice.h>
> +#include <linux/mm.h>
> +#include <linux/module.h>
> +#include <linux/mutex.h>
> +#include <linux/of_address.h>
> +#include <linux/of_device.h>
> +#include <linux/platform_device.h>
> +#include <linux/slab.h>
> +#include <linux/uaccess.h>
> +
> +#include <linux/aspeed-p2a-ctrl.h>
> +
> +#define DEVICE_NAME	"aspeed-p2a-ctrl"
> +
> +#define SCU180_ENP2A BIT(0)
> +
> +/* The ast2400/2500 both have six ranges. */
> +#define P2A_REGION_COUNT 6
> +
> +struct region {
> +	u32 min;
> +	u32 max;
> +	u32 bit;
> +};
> +
> +struct aspeed_p2a_model_data {
> +	/* min, max, bit */
> +	struct region regions[P2A_REGION_COUNT];
> +};
> +
> +struct aspeed_p2a_ctrl {
> +	struct miscdevice miscdev;
> +
> +	/* Lock access to the registers so they're is consistent. */
> +	struct mutex lo_mutex;
> +	void __iomem *loregs;
> +	struct mutex hi_mutex;
> +	void __iomem *hiregs;

The naming here seems to have failed to climb the abstraction ladder.
loregs represents the register where the P2A write control filter bits live,
hiregs represents the PCI feature control register. I'd much prefer we name
them as such, that way I'm not jumping through abstractions in my head
whilst trying to understand the code. Where they live relative to eachother
is irrelevant. ASPEED could reverse the register locations on the AST2600
and the names wouldn't make any sense as they are now.

Anyway.

The SCU in general is a mess of bits controlling random parts of the SoC.
State controlling a feature is usually mixed in with state controlling some
other irrelevant features. This is true for the write filter controls, which live
in e.g. SCU2C, which I assume is what you're covering with lo_mutex here.

We already have a solution for this to prevent the proliferation of random
locks - the SCU is described as a "syscon" in the devicetree - this registers
a globally accessible regmap instance which handles all of the locking
for you. There are a bunch of ways you can go about finding the syscon
(see include/linux/mfd/syscon.h):

extern struct regmap *syscon_node_to_regmap(struct device_node *np);
extern struct regmap *syscon_regmap_lookup_by_compatible(const char *s);
extern struct regmap *syscon_regmap_lookup_by_pdevname(const char *s);
extern struct regmap *syscon_regmap_lookup_by_phandle(
					struct device_node *np,
					const char *property);

Please rework the code to use the syscon we already have to avoid drivers
stomping on each-other's SCU state.

> +
> +	const struct aspeed_p2a_model_data *config;

We only hold one of instance of this struct at the struct aspeed_p2a_ctrl
don't we? Why not embed it directly instead of going through a pointer?

> +
> +	/* Access to these needs to be locked, held via probe, mapping ioctl,
> +	 * and release, remove.
> +	 */
> +	struct mutex tracking;
> +	u32 readers;
> +	u32 readerwriters[P2A_REGION_COUNT];

Might be better to use refcount_t here instead of u32?

> +
> +	phys_addr_t mem_base;
> +	resource_size_t mem_size;
> +	/* Because the memory_region is optional, this tracks whether it was
> +	 * set.
> +	 */
> +	bool region_specified;

Can't we imply this from e.g. `mem_base == 0 && mem_size == 0`? Can you
just implement that in a static inline helper and avoid the extra state?

> +};
> +
> +static inline u32 aspeed_p2a_read(void __iomem *base)
> +{
> +	return readl((void *)base);
> +}
> +
> +static inline void aspeed_p2a_write(void __iomem *base, u32 val)
> +{
> +	writel(val, (void *)base);
> +}

I don't think these two are necessary, and you can replace them with the
regmap helpers anyway.

> +
> +struct aspeed_p2a_user {
> +	struct file *file;
> +	struct aspeed_p2a_ctrl *parent;
> +
> +	/** The entire memory space is opened for reading once the bridge is
> +	 * enabled, therefore this needs only to be tracked once per user.
> +	 * If any user has it open for read, the bridge must stay enabled.
> +	 */
> +	u32 read;
> +
> +	/** Each entry of the array corresponds to a P2A Region.  If the user
> +	 * opens for read or readwrite, the reference goes up here.  On 
> release,
> +	 * this array is walked and references adjusted accordingly.
> +	 */
> +	u32 readwrite[P2A_REGION_COUNT];

Do we have a scenario where a user will open the same region multiple
times from the one fd? Supporting that seems to add a fair chunk of
complexity and I'm not convinced it's warranted unless we have a concrete
use case for it.

> +};
> +
> +static void aspeed_p2a_enable_bridge(struct aspeed_p2a_ctrl *p2a_ctrl)
> +{
> +	u32 curr_value;
> +
> +	mutex_lock(&p2a_ctrl->hi_mutex);
> +	curr_value = aspeed_p2a_read(p2a_ctrl->hiregs);
> +	curr_value |= SCU180_ENP2A;
> +	aspeed_p2a_write(p2a_ctrl->hiregs, curr_value);
> +	mutex_unlock(&p2a_ctrl->hi_mutex);

This becomes regmap_update_bits(). Same below.

> +}
> +
> +/** The bridge is controlled by bit 0 of SCU180. */

The P2A is controlled by bit 1, not bit 0. Bit 0 enables/disables the entire VGA
device on the PCI bus. I think just remove the comment? If you want it, put
it next to the #define for SCU180_ENP2A.

> +static void aspeed_p2a_disable_bridge(struct aspeed_p2a_ctrl *p2a_ctrl)
> +{
> +	u32 curr_value;
> +
> +	mutex_lock(&p2a_ctrl->hi_mutex);
> +	curr_value = aspeed_p2a_read(p2a_ctrl->hiregs);
> +	curr_value &= ~SCU180_ENP2A;
> +	aspeed_p2a_write(p2a_ctrl->hiregs, curr_value);
> +	mutex_unlock(&p2a_ctrl->hi_mutex);
> +}
> +
> +static int aspeed_p2a_mmap(struct file *file, struct vm_area_struct 
> *vma)
> +{
> +	struct aspeed_p2a_user *priv = file->private_data;
> +	struct aspeed_p2a_ctrl *ctrl = priv->parent;
> +
> +	if (!ctrl->region_specified)
> +		return -EINVAL;
> +
> +	unsigned long vsize = vma->vm_end - vma->vm_start;
> +	pgprot_t prot = vma->vm_page_prot;
> +
> +	if (vma->vm_pgoff + vsize > ctrl->mem_base + ctrl->mem_size)
> +		return -EINVAL;
> +
> +	/* ast2400/2500 AHB accesses are not cache coherent */
> +	prot = pgprot_noncached(prot);
> +
> +	if (remap_pfn_range(vma, vma->vm_start,
> +		(ctrl->mem_base >> PAGE_SHIFT) + vma->vm_pgoff,
> +		vsize, prot))
> +		return -EAGAIN;
> +
> +	return 0;
> +}
> +
> +static void aspeed_p2a_region_search(struct aspeed_p2a_user *priv,
> +		struct aspeed_p2a_ctrl *ctrl,
> +		struct aspeed_p2a_ctrl_mapping *map)

Bit of a bikeshed, but maybe this could be called
aspeed_p2a_region_update() or aspeed_p2a_region_acquire() instead?
Search implies it won't change any state, but that's not true.

> +{
> +	int i;
> +	u32 base, end;
> +	u32 value;
> +
> +	base = map->addr;
> +	end = map->addr + (map->length - 1);
> +
> +	/* If the value is a legal u32, it will find a match. */
> +	for (i = 0; i < P2A_REGION_COUNT; i++) {
> +		const struct region *curr = &ctrl->config->regions[i];
> +
> +		/* If the top of this region is lower than your base, skip it.
> +		 */
> +		if (curr->max < base)
> +			continue;
> +
> +		/* If the bottom of this region is higher than your end, bail.
> +		 */
> +		if (curr->min > end)
> +			break;
> +
> +		/* Lock this and update it, therefore it someone else is
> +		 * closing their file out, this'll preserve the increment.
> +		 */
> +		mutex_lock(&ctrl->tracking);
> +		ctrl->readerwriters[i] += 1;
> +		mutex_unlock(&ctrl->tracking);
> +
> +		/* Track with the user, so when they close their file, we can
> +		 * decrement properly.
> +		 */
> +		priv->readwrite[i] += 1;
> +
> +		/* Enable the region as read-write. */
> +		mutex_lock(&ctrl->lo_mutex);
> +		value =	aspeed_p2a_read(ctrl->loregs);
> +		value &= ~curr->bit;
> +		aspeed_p2a_write(ctrl->loregs, value);
> +		mutex_unlock(&ctrl->lo_mutex);
> +	}
> +}
> +
> +static long aspeed_p2a_ioctl(struct file *file, unsigned int cmd,
> +		unsigned long data)
> +{
> +	struct aspeed_p2a_user *priv = file->private_data;
> +	struct aspeed_p2a_ctrl *ctrl = priv->parent;
> +	void __user *arg = (void __user *)data;
> +	struct aspeed_p2a_ctrl_mapping map;
> +	int i;
> +
> +	switch (cmd) {
> +	case ASPEED_P2A_CTRL_IOCTL_SET_WINDOW:
> +		if (copy_from_user(&map, arg, sizeof(map)))
> +			return -EFAULT;
> +
> +		/* If they want a region to be read-only, since the entire
> +		 * region is read-only once enabled, we just need to track this
> +		 * user wants to read from the bridge, and if it's not enabled.
> +		 * Enable it.
> +		 */
> +		if (map.flags == ASPEED_P2A_CTRL_READ_ONLY) {
> +			mutex_lock(&ctrl->tracking);
> +			ctrl->readers += 1;
> +			mutex_unlock(&ctrl->tracking);
> +
> +			/* Track with the user, so when they close their file,
> +			 * we can decrement properly.
> +			 */
> +			priv->read += 1;
> +		} else if (map.flags == ASPEED_P2A_CTRL_READWRITE) {
> +			aspeed_p2a_region_search(priv, ctrl, &map);

See, this looks a bit strange with the function called _search(), as you're
not doing anything with a result - and there can't be a result as the function
returns void.

> +		} else {
> +			/* Invalid map flags. */
> +			return -EINVAL;
> +		}
> +
> +		aspeed_p2a_enable_bridge(ctrl);
> +		return 0;
> +	}
> +
> +	return -EINVAL;
> +}
> +
> +
> +/**
> + * When a user opens this file, we create a structure to track their 
> mappings.
> + *
> + * A user can map a region as read-only (bridge enabled), or 
> read-write (bit
> + * flipped, and bridge enabled).  Either way, this tracking is used, 
> s.t. when
> + * they release the device references are handled.
> + *
> + * The bridge is not enabled until a user calls an ioctl to map a 
> region,
> + * simply opening the device does not enable it.
> + */
> +static int aspeed_p2a_open(struct inode *inode, struct file *file)
> +{
> +	struct aspeed_p2a_user *priv;
> +
> +	priv = kmalloc(sizeof(*priv), GFP_KERNEL);
> +	if (!priv)
> +		return -ENOMEM;
> +
> +	priv->file = file;
> +	priv->read = 0;
> +	memset(priv->readwrite, 0, sizeof(priv->readwrite));
> +
> +	/* The file's private_data is initialized to the p2a_ctrl. */
> +	priv->parent = file->private_data;
> +
> +	/* Set the file's private_data to the user's data. */
> +	file->private_data = priv;
> +
> +	return 0;
> +}
> +
> +/**
> + * This will close the users mappings.  It will go through what they 
> had opened
> + * for readwrite, and decrement those counts.  If at the end, this is 
> the last
> + * user, it'll close the bridge.
> + */
> +static int aspeed_p2a_release(struct inode *inode, struct file *file)
> +{
> +	int i;
> +	u32 value;
> +	u32 bits = 0;
> +	bool open_regions = false;
> +	struct aspeed_p2a_user *priv = file->private_data;
> +
> +	mutex_lock(&priv->parent->tracking);
> +	priv->parent->readers -= priv->read;
> +
> +	for (i = 0; i < P2A_REGION_COUNT; i++) {
> +		priv->parent->readerwriters[i] -= priv->readwrite[i];
> +
> +		if (priv->parent->readerwriters[i] > 0)
> +			open_regions = true;
> +		else
> +			bits |= priv->parent->config->regions[i].bit;
> +	}
> +
> +	/* Setting a bit to 1 disables the region, so let's just OR with the
> +	 * above to disable any.
> +	 */
> +
> +	/* Note, if another user is trying to ioctl, they can't grab tracking,
> +	 * and therefore can't grab either register mutex.
> +	 * If another user is trying to close, they can't grab tracking 
> either.
> +	 */
> +	mutex_lock(&priv->parent->lo_mutex);
> +	value = aspeed_p2a_read(priv->parent->loregs);
> +	value |= bits;
> +	aspeed_p2a_write(priv->parent->loregs, value);
> +	mutex_unlock(&priv->parent->lo_mutex);
> +
> +	/* If parent->readers is zero and open windows is 0, disable the
> +	 * bridge.
> +	 */
> +	if (!open_regions && priv->parent->readers == 0)
> +		aspeed_p2a_disable_bridge(priv->parent);
> +
> +	mutex_unlock(&priv->parent->tracking);
> +
> +	kfree(priv);
> +
> +	return 0;
> +}
> +
> +static const struct file_operations aspeed_p2a_ctrl_fops = {
> +	.owner = THIS_MODULE,
> +	.mmap = aspeed_p2a_mmap,
> +	.unlocked_ioctl = aspeed_p2a_ioctl,
> +	.open = aspeed_p2a_open,
> +	.release = aspeed_p2a_release,
> +};
> +
> +/** The regions are controlled by SCU2C */
> +static void aspeed_p2a_disable_all(struct aspeed_p2a_ctrl *p2a_ctrl)
> +{
> +	int i;
> +	u32 value = 0;
> +	u32 curr_value;
> +
> +	for (i = 0; i < P2A_REGION_COUNT; i++)
> +		value |= p2a_ctrl->config->regions[i].bit;
> +
> +	mutex_lock(&p2a_ctrl->lo_mutex);
> +	curr_value = aspeed_p2a_read(p2a_ctrl->loregs);
> +	curr_value |= value;
> +	aspeed_p2a_write(p2a_ctrl->loregs, curr_value);
> +	mutex_unlock(&p2a_ctrl->lo_mutex);
> +
> +	/* Disable the bridge. */
> +	aspeed_p2a_disable_bridge(p2a_ctrl);
> +}
> +
> +static int aspeed_p2a_ctrl_probe(struct platform_device *pdev)
> +{
> +	struct aspeed_p2a_ctrl *p2a_ctrl;
> +	struct device *dev;
> +	struct resource *res, resm;
> +	struct device_node *node;
> +	int rc = 0;
> +	u32 twosee, oneeighty;

How about "misc_ctrl" and "pci_cscr"?

> +
> +	dev = &pdev->dev;
> +
> +	p2a_ctrl = devm_kzalloc(dev, sizeof(*p2a_ctrl), GFP_KERNEL);
> +	if (!p2a_ctrl)
> +		return -ENOMEM;
> +
> +	mutex_init(&p2a_ctrl->tracking);
> +	p2a_ctrl->readers = 0;
> +	memset(p2a_ctrl->readerwriters, 0, sizeof(p2a_ctrl->readerwriters));
> +
> +	p2a_ctrl->mem_size = 0;
> +	p2a_ctrl->mem_base = 0;
> +	p2a_ctrl->region_specified = false;
> +
> +	/* optional. */
> +	node = of_parse_phandle(dev->of_node, "memory-region", 0);
> +	if (node) {
> +		rc = of_address_to_resource(node, 0, &resm);
> +		of_node_put(node);
> +		if (rc) {
> +			dev_err(dev, "Couldn't address to resource for reserved memory\n");
> +			return -ENOMEM;
> +		}
> +
> +		p2a_ctrl->mem_size = resource_size(&resm);
> +		p2a_ctrl->mem_base = resm.start;
> +		p2a_ctrl->region_specified = true;
> +	}
> +
> +	p2a_ctrl->config = of_device_get_match_data(dev);
> +
> +	/* This is the SCU2C handle. */
> +	mutex_init(&p2a_ctrl->lo_mutex);
> +	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
> +	p2a_ctrl->loregs = devm_ioremap_resource(dev, res);
> +	if (IS_ERR(p2a_ctrl->loregs)) {
> +		dev_err(dev, "Unable to remap resource 0\n");
> +		return PTR_ERR(p2a_ctrl->loregs);
> +	}
> +
> +	twosee = aspeed_p2a_read(p2a_ctrl->loregs);

You don't use this value? Why read it?

> +
> +	/* This is the SCU180 handle. */
> +	mutex_init(&p2a_ctrl->hi_mutex);
> +	res = platform_get_resource(pdev, IORESOURCE_MEM, 1);
> +	p2a_ctrl->hiregs = devm_ioremap_resource(dev, res);
> +	if (IS_ERR(p2a_ctrl->hiregs)) {
> +		dev_err(dev, "Unable to remap resource 1\n");
> +		return PTR_ERR(p2a_ctrl->hiregs);
> +	}
> +
> +	oneeighty = aspeed_p2a_read(p2a_ctrl->hiregs);

You don't use this value, why read ti?

> +
> +	dev_set_drvdata(&pdev->dev, p2a_ctrl);
> +
> +	p2a_ctrl->miscdev.minor = MISC_DYNAMIC_MINOR;
> +	p2a_ctrl->miscdev.name = DEVICE_NAME;
> +	p2a_ctrl->miscdev.fops = &aspeed_p2a_ctrl_fops;
> +	p2a_ctrl->miscdev.parent = dev;
> +	rc = misc_register(&p2a_ctrl->miscdev);
> +	if (rc) {
> +		dev_err(dev, "Unable to register device\n");
> +		goto err;
> +	}
> +
> +	aspeed_p2a_disable_all(p2a_ctrl);

Do this before registering the misc device otherwise I think you're at
risk of racing userspace to initialising the P2A state. Maybe that's not
a problem, but we don't have to try reason about it if you move it up.

> +
> +err:
> +	return rc;
> +}
> +
> +static int aspeed_p2a_ctrl_remove(struct platform_device *pdev)
> +{
> +	struct aspeed_p2a_ctrl *p2a_ctrl = dev_get_drvdata(&pdev->dev);
> +
> +	misc_deregister(&p2a_ctrl->miscdev);
> +
> +	return 0;
> +}
> +
> +/*
> + * bit | SCU2C | ast2400
> + *  25 | DRAM  | 0x40000000 - 0x5FFFFFFF
> + *  24 | SPI   | 0x30000000 - 0x3FFFFFFF
> + *  23 | SOC   | 0x18000000 - 0x1FFFFFFF, 0x60000000 - 0xFFFFFFFF
> + *  22 | FLASH | 0x00000000 - 0x17FFFFFF, 0x20000000 - 0x2FFFFFFF
> + *
> + * bit | SCU2C | ast2500
> + *  25 | DRAM  | 0x80000000 - 0xFFFFFFFF
> + *  24 | SPI   | 0x60000000 - 0x7FFFFFFF
> + *  23 | SOC   | 0x10000000 - 0x1FFFFFFF, 0x40000000 - 0x5FFFFFFF
> + *  22 | FLASH | 0x00000000 - 0x0FFFFFFF, 0x20000000 - 0x3FFFFFFF
> + */
> +
> +#define SCU2C_DRAM	BIT(25)
> +#define SCU2C_SPI	BIT(24)
> +#define SCU2C_SOC	BIT(23)
> +#define SCU2C_FLASH	BIT(22)
> +
> +static const struct aspeed_p2a_model_data ast2400_model_data = {
> +	.regions = {
> +		{0x00000000, 0x17FFFFFF, SCU2C_FLASH},
> +		{0x18000000, 0x1FFFFFFF, SCU2C_SOC},
> +		{0x20000000, 0x2FFFFFFF, SCU2C_FLASH},
> +		{0x30000000, 0x3FFFFFFF, SCU2C_SPI},
> +		{0x40000000, 0x5FFFFFFF, SCU2C_DRAM},
> +		{0x60000000, 0xFFFFFFFF, SCU2C_SOC},
> +	}
> +};
> +
> +static const struct aspeed_p2a_model_data ast2500_model_data = {
> +	.regions = {
> +		{0x00000000, 0x0FFFFFFF, SCU2C_FLASH},
> +		{0x10000000, 0x1FFFFFFF, SCU2C_SOC},
> +		{0x20000000, 0x3FFFFFFF, SCU2C_FLASH},
> +		{0x40000000, 0x5FFFFFFF, SCU2C_SOC},
> +		{0x60000000, 0x7FFFFFFF, SCU2C_SPI},
> +		{0x80000000, 0xFFFFFFFF, SCU2C_DRAM},
> +	}
> +};
> +
> +static const struct of_device_id aspeed_p2a_ctrl_match[] = {
> +	{ .compatible = "aspeed,ast2400-p2a-ctrl",
> +	  .data = &ast2400_model_data },
> +	{ .compatible = "aspeed,ast2500-p2a-ctrl",
> +	  .data = &ast2500_model_data },
> +	{ },
> +};
> +
> +static struct platform_driver aspeed_p2a_ctrl_driver = {
> +	.driver = {
> +		.name		= DEVICE_NAME,
> +		.of_match_table = aspeed_p2a_ctrl_match,
> +	},
> +	.probe = aspeed_p2a_ctrl_probe,
> +	.remove = aspeed_p2a_ctrl_remove,
> +};
> +
> +module_platform_driver(aspeed_p2a_ctrl_driver);
> +
> +MODULE_DEVICE_TABLE(of, aspeed_p2a_ctrl_match);
> +MODULE_LICENSE("GPL");
> +MODULE_AUTHOR("Patrick Venture <venture@xxxxxxxxxx>");
> +MODULE_DESCRIPTION("Control for aspeed 2400/2500 P2A VGA HOST to BMC 
> mappings");
> diff --git a/include/uapi/linux/aspeed-p2a-ctrl.h 
> b/include/uapi/linux/aspeed-p2a-ctrl.h
> new file mode 100644
> index 000000000000..473b62efb94c
> --- /dev/null
> +++ b/include/uapi/linux/aspeed-p2a-ctrl.h
> @@ -0,0 +1,46 @@
> +/* SPDX-License-Identifier: GPL-2.0+ WITH Linux-syscall-note */
> +/*
> + * Copyright 2019 Google Inc
> + *
> + * This program is free software; you can redistribute it and/or
> + * modify it under the terms of the GNU General Public License
> + * as published by the Free Software Foundation; either version
> + * 2 of the License, or (at your option) any later version.
> + *
> + * Provides a simple driver to control the ASPEED P2A interface which 
> allows
> + * the host to read and write to various regions of the BMC's memory.
> + */
> +
> +#ifndef _UAPI_LINUX_ASPEED_P2A_CTRL_H
> +#define _UAPI_LINUX_ASPEED_P2A_CTRL_H
> +
> +#include <linux/ioctl.h>
> +#include <linux/types.h>
> +
> +#define ASPEED_P2A_CTRL_READ_ONLY 0
> +#define ASPEED_P2A_CTRL_READWRITE 1
> +
> +/*
> + * This driver provides a mechanism for enabling or disabling the 
> read-write
> + * property of specific windows into the ASPEED BMC's memory.
> + *
> + * A user can map a region of the BMC's memory as read-only or 
> read-write, with
> + * the caveat that once any region is mapped, all regions are unlocked 
> for
> + * reading.
> + */
> +
> +/**
> + * Unlock a region of BMC physical memory for access from the host.
> + */
> +struct aspeed_p2a_ctrl_mapping {
> +	__u32 addr;
> +	__u32 length;
> +	__u32 flags;
> +};
> +
> +#define __ASPEED_P2A_CTRL_IOCTL_MAGIC 0xb3
> +
> +#define ASPEED_P2A_CTRL_IOCTL_SET_WINDOW 
> _IOW(__ASPEED_P2A_CTRL_IOCTL_MAGIC, \
> +		0x00, struct aspeed_p2a_ctrl_mapping)
> +
> +#endif /* _UAPI_LINUX_ASPEED_P2A_CTRL_H */

How does the BMC report to the host what region of memory to write
to? Are you scraping around in sysfs to learn the address of the
reserved memory region? Or are you enforcing some ABI between
the BMC and the host that defines the address? There's no means
exposed here to learn the address.

Andrew