Date:   Sat, 6 Nov 2021 23:14:56 +0800
From:   Chen Yu <yu.c.chen@...el.com>
To:     Andy Shevchenko <andriy.shevchenko@...el.com>
Cc:     linux-acpi@...r.kernel.org,
        Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        "Rafael J. Wysocki" <rafael@...nel.org>,
        Ard Biesheuvel <ardb@...nel.org>, Len Brown <lenb@...nel.org>,
        Ashok Raj <ashok.raj@...el.com>,
        Mike Rapoport <rppt@...nel.org>,
        Aubrey Li <aubrey.li@...el.com>, linux-kernel@...r.kernel.org
Subject: Re: [PATCH v8 2/4] drivers/acpi: Introduce Platform Firmware Runtime
 Update device driver

On Wed, Nov 03, 2021 at 07:32:09PM +0200, Andy Shevchenko wrote:
> On Wed, Nov 03, 2021 at 11:43:50PM +0800, Chen Yu wrote:
> > Introduce the pfru_update driver which can be used for Platform Firmware
> > Runtime code injection and driver update [1]. The user is expected to
> > provide the update firmware in the form of capsule file, and pass it to
> > the driver via ioctl. Then the driver would hand this capsule file to the
> > Platform Firmware Runtime Update via the ACPI device _DSM method. At last
> > the low level Management Mode would do the firmware update.
> > 
> > The corresponding userspace tool and man page will be introduced at
> > tools/power/acpi/pfru.
> 
> ...
> 
> > +#define PFRU_UUID		"ECF9533B-4A3C-4E89-939E-C77112601C6D"
> > +#define PFRU_CODE_INJ_UUID		"B2F84B79-7B6E-4E45-885F-3FB9BB185402"
> > +#define PFRU_DRV_UPDATE_UUID		"4569DD8C-75F1-429A-A3D6-24DE8097A0DF"
> 
> What stops you to have these being binaries?
> GUID_INIT() / EFI_GUID_INIT()
>
Ok, will change to GUID_INIT(). 
> ...
> 
> > +enum cap_index {
> > +	CAP_STATUS_IDX = 0,
> > +	CAP_UPDATE_IDX = 1,
> > +	CAP_CODE_TYPE_IDX = 2,
> > +	CAP_FW_VER_IDX = 3,
> > +	CAP_CODE_RT_VER_IDX = 4,
> > +	CAP_DRV_TYPE_IDX = 5,
> > +	CAP_DRV_RT_VER_IDX = 6,
> > +	CAP_DRV_SVN_IDX = 7,
> > +	CAP_PLAT_ID_IDX = 8,
> > +	CAP_OEM_ID_IDX = 9,
> > +	CAP_OEM_INFO_IDX = 10,
> 
> > +	CAP_NR_IDX = 11
> 
> Assignment here doesn't make any sense (it just adds unneeded churn and
> burden). Same to the rest of similar cases below.
>
Greg mentioned that, we might need to "explicit about the numbers here, because it
is uncerntain this is guaranteed by all C compilers or not."
https://lore.kernel.org/lkml/YXj+QaMcCeV71XbI@kroah.com/
My understanding is that, this applys to both uapi headers and the kernel internal
headers.
> > +};
> 
> ...
> 
> > +struct pfru_device {
> > +	guid_t uuid, code_uuid, drv_uuid;
> 
> You don't need these. At least for now.
> 
Ok, will drop these.
> > +	u32 rev_id, index;
> > +	struct device *parent_dev;
> > +	struct miscdevice miscdev;
> > +};
> 
> ...
> 
> > +	m_hdr = (struct efi_manage_capsule_header *)(data + size);
> 
> Do you need this casting?
> 
Will drop this.
> ...
> 
> > +	m_img_hdr = (struct efi_manage_capsule_image_header *)(data + size);
> 
> Ditto.
> 
> ...
> 
> > +	auth = (struct efi_image_auth *)(data + size);
> 
> Ditto.
> 
> ...
> 
> > +	ACPI_FREE(out_obj);
> 
> Recently with Hans we realised that this (ACPI_FREE() API) is mostly
> for ACPICA use. We may use simple kfree(). Sorry for getting back and
> forward.
> 
> ...
> 
Will change it in next version.
> > +static long pfru_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
> > +{
> > +	struct pfru_update_cap_info cap_hdr;
> > +	struct pfru_device *pfru_dev = to_pfru_dev(file);
> > +	void __user *p = (void __user *)arg;
> > +	u32 rev;
> > +	int ret;
> > +
> > +	switch (cmd) {
> > +	case PFRU_IOC_QUERY_CAP:
> > +		ret = query_capability(&cap_hdr, pfru_dev);
> > +		if (ret)
> > +			return ret;
> > +
> > +		if (copy_to_user(p, &cap_hdr, sizeof(cap_hdr)))
> 
> I'm wondering what will happen if p has less _real data_ than sizeof(cap_hdr)?
> 
Here is my understanding: if the userspace has provided insufficient space,
the userspace might either encounter segfault or data overwrite, and it is up
to the userspace to avoid this situation from happening.
for example:

int my_test(void)
{
	char *cap_on_heap = malloc(insufficient_size);
	char cap_on_stack[insufficient_size];
	int victim;
	...
}

copy_to_user(cap_on_heap) might cause segfault, and copy_to_user(cap_on_stack) might
overwrite victim.
> > +			return -EFAULT;
> > +
> > +		return 0;
> 
> > +	case PFRU_IOC_SET_REV:
> > +		if (copy_from_user(&rev, p, sizeof(u32)))
> 
> sizeof(rev)
>
Ok. 
> > +			return -EFAULT;
> > +
> > +		if (!pfru_valid_revid(rev))
> > +			return -EINVAL;
> > +
> > +		pfru_dev->rev_id = rev;
> > +
> > +		return 0;
> > +	case PFRU_IOC_STAGE:
> > +		return start_acpi_update(START_STAGE, pfru_dev);
> > +	case PFRU_IOC_ACTIVATE:
> > +		return start_acpi_update(START_ACTIVATE, pfru_dev);
> > +	case PFRU_IOC_STAGE_ACTIVATE:
> > +		return start_acpi_update(START_STAGE_ACTIVATE, pfru_dev);
> > +	default:
> > +		return -ENOTTY;
> > +	}
> > +}
> 
> ...
> 
> > +	/* map the communication buffer */
> > +	phy_addr = (phys_addr_t)(buf_info.addr_lo | (buf_info.addr_hi << 32));
> 
> It's better to read if you start from MSB part to LSB.
>
Ok, will do. 
> ...
> 
> > +	ret = ida_alloc(&pfru_ida, GFP_KERNEL);
> > +	if (ret < 0)
> > +		return ret;
> 
> (1)
> 
> ...
> 
> > +	pfru_dev->miscdev.name = kasprintf(GFP_KERNEL,
> > +					   "pfru%d", pfru_dev->index);
> 
> devm_kasprinf()
> 
> ...
> 
> > +	pfru_dev->miscdev.nodename = kasprintf(GFP_KERNEL,
> > +					       "acpi_pfru%d", pfru_dev->index);
> 
> Ditto.
> 
> Yep, I know about (1), but do your homework and see how you can satisfy both
> comments.
>
I did not realize devm_add_action_or_reset() could be used in (1) to deal with
this situation, will do in next version. 
> ...
> 
> > +static const struct acpi_device_id acpi_pfru_ids[] = {
> > +	{"INTC1080", 0},
> 
> 0 is redundant.
>
Ok. 
> > +	{}
> > +};
> 
> ...
> 
> > +#include <linux/types.h>
> > +#include <linux/ioctl.h>
> 
> Order?
> 
Will adjust it.
> ...
> 
> > +#define PFRU_MAGIC 0xEE
> 
> Perhaps PFRU_MAGIC_FOR_IOCTL.
> 
Ok.

Thanks,
Chenyu

Powered by blists - more mailing lists