Re: [PATCH v9 2/5] firmware: psci: Read and use vendor reset types

From: Dmitry Baryshkov
Date: Thu Jun 19 2025 - 07:05:23 EST

On 19/06/2025 12:00, Shivendra Pratap wrote:


On 6/18/2025 6:44 PM, Dmitry Baryshkov wrote:
On Tue, May 06, 2025 at 11:03:55PM +0530, Shivendra Pratap wrote:


On 4/16/2025 5:35 PM, Lorenzo Pieralisi wrote:
On Wed, Apr 09, 2025 at 11:48:24PM +0530, Shivendra Pratap wrote:


On 4/8/2025 8:46 PM, Lorenzo Pieralisi wrote:
On Tue, Mar 25, 2025 at 07:33:36PM +0530, Mukesh Ojha wrote:
On Fri, Mar 14, 2025 at 12:19:31PM +0100, Lorenzo Pieralisi wrote:
On Mon, Mar 03, 2025 at 01:08:31PM -0800, Elliot Berman wrote:
From: Elliot Berman <elliot.berman@xxxxxxxxxxxxxxxx>

SoC vendors have different types of resets and are controlled through
various registers. For instance, Qualcomm chipsets can reboot to a
"download mode" that allows a RAM dump to be collected. Another example
is they also support writing a cookie that can be read by bootloader
during next boot. PSCI offers a mechanism, SYSTEM_RESET2, for these
vendor reset types to be implemented without requiring drivers for every
register/cookie.

Add support in PSCI to statically map reboot mode commands from
userspace to a vendor reset and cookie value using the device tree.

I have managed to discuss a little bit this patchset over the last
few days and I think we have defined a plan going forward.

A point that was raised is:

https://man7.org/linux/man-pages/man2/reboot.2.html

LINUX_REBOOT_CMD_RESTART2 *arg command, what is it supposed to
represent ?

Is it the mode the system should reboot into OR it is the
actual command to be issued (which is what this patchset
implements) ?

LINUX_REBOOT_CMD_RESTART "..a default restart..."

It is unclear what "default" means. We wonder whether the
reboot_mode variable was introduced to _define_ that "default".

So, in short, my aim is trying to decouple reboot_mode from the
LINUX_REBOOT_CMD_RESTART2 *arg command.

I believe that adding a sysfs interface to reboot-mode driver
infrastructure would be useful, so that the commands would
be exposed to userspace and userspace can set the *arg command
specifically to issue a given reset/mode.

I wonder why this is not already in place for eg syscon-reboot-mode
resets, how does user space issue a command in those systems if the
available commands aren't exposed to userspace ?

Is there a kernel entity exposing those "modes" to userspace, somehow ?

A separate initcall is needed to parse the devicetree, instead of using
psci_dt_init because mm isn't sufficiently set up to allocate memory.

Reboot mode framework is close but doesn't quite fit with the
design and requirements for PSCI SYSTEM_RESET2. Some of these issues can
be solved but doesn't seem reasonable in sum:
1. reboot mode registers against the reboot_notifier_list, which is too
early to call SYSTEM_RESET2. PSCI would need to remember the reset
type from the reboot-mode framework callback and use it
psci_sys_reset.
2. reboot mode assumes only one cookie/parameter is described in the
device tree. SYSTEM_RESET2 uses 2: one for the type and one for
cookie.

This can be changed and I think it should, so that the reboot modes
are exposed to user space and PSCI can use that.

In the case of a regular reboot or panic, the reboot/panic notifiers run
first, followed by the restart notifiers. The PSCI reset/reset2 should
be the last call from Linux, and ideally, this call should not fail.

Reboot mode notifiers => restart notifiers or Panic notifiers => restart
notifiers

So, if I understand correctly, you mean that we can change the reboot
mode framework to expose the arguments available to user space. We can
extend it to accept magic and cookies, save them in the reboot
framework, and retrieve them via a call from PSCI during a regular
reboot or panic based on the current arguments. Is this leading towards
writing an ARM-specific PSCI-reboot-mode driver, which in its reboot
notifier callback saves the magic and cookies, and these magic and
cookies will be used during psci_sys_reset2()? Or is there something
wrong with my understanding?

No, you got it right (apologies for the delay in replying) - if the
case for making reboot mode available to user space is accepted.

While moving this into reboot-mode framework, one more query came up.
The "ARM-specific PSCI-reboot-mode driver" that we are going to write needs
to be a Platform device driver for using reboot-mode framework.

No, it doesn't. It rqeuires struct device, but there is no requirement
for struct platform_device at any place.
yes, it can be struct device so may be create a virtual device
using reset-type node?

It can be created, but I don't see a strong need for it.


As psci is not a platform device driver, a subdevice under it may not probe as a
platform driver. Is it ok to implement the "PSCI-reboot-mode driver" as a
early_initcall("psci_xyz") and then create a platform device something as
below or any other suggestions for this?

Change struct reboot_mode_driver to pass corresponding of_node (or
better fwnode) directly. Corresponding device is used only in the
reboot_mode_register() and only to access of-node or to print error
messages.
struct reboot_mode_driver can be changed just to pass of_node. But then the other
suggestion was to expose sysfs from reboot-mode to show available commands.
For that we need a device. Any suggestion? A virtual device with reset-types node
passed to reboot-mode framework looks fine?

You still don't need it. You'll create a new device, belonging to the new 'reboot' or 'reset' class to hold corresponding attributes.



power:reset:<psci-vendor-reset-driver>:
-----
static int __init psci_vendor_reset_init(void) {
..
..
np = of_find_node_by_name(NULL, "psci-vendor-reset");
if(!np)
return -ENODEV;
pdev = of_platform_device_create(np, "psci-vendor-reset", NULL);
..
..
}
-------

the sysfs we will expose from reboot-mode may show like below in above
implementation:

######
/ # cat ./sys/devices/platform/psci-vendor-reset/available_modes
bootloader edl
######

thanks,
Shivendra


Agree that the available modes should be exposed to usespace via sysfs interface
and we should implement it. Also #1 and #2 can be handled via some
changes in the design as mentioned in above discussion.

I have one doubt though when we implement this via reboot-mode framework.
The current patch implements PSCI ARM PSCI SYSTEM RESET2 vendor reset types.
psci driver is initialized very early at boot but potential ARM psci reboot-mode
driver will not probe at that stage and the ARM PSCI SYSTEM RESET2 vendor reset
types functionality will not be available in psci reset path until the reboot-mode
driver probes. Will this cause any limitation on usage of ARM's PSCI vendor-reset
types for early device resets?

One use-case may be an early device crash or a early reset where a vendor
wants to use PSCI SYSTEM RESET2 vendor reset type to a reset the device to a
specific state but may not be able to use this driver.
(eg: a kernel panic at early boot where a vendor wants to reset device
to a specific state using vendor reset. Currently panic passes a NULL
(*arg command) while device reset but it may be explored for vendor specific
reset).

As you said, that would not be a PSCI only issue - *if* we wanted to
plug in this use case we should find a way to do it at reboot mode
driver level.

As a matter of fact, this is not a mainline issue AFAICS.

Even if we did not design this as a reboot mode driver there would be a
time window where you would not be able to use vendor resets on panic.

I don't see it as a major roadblock at the moment.
Got it.

Thanks,
Lorenzo


- Shivendra

P.S. We appreciate Elliot for his work and follow-up on this while being
employed at Qualcomm.

Yes I sincerely do for his patience, thank you.

Lorenzo

3. psci cpuidle driver already registers a driver against the
arm,psci-1.0 compatible. Refactoring would be needed to have both a
cpuidle and reboot-mode driver.

Signed-off-by: Elliot Berman <elliot.berman@xxxxxxxxxxxxxxxx>
---
drivers/firmware/psci/psci.c | 105 +++++++++++++++++++++++++++++++++++++++++++
1 file changed, 105 insertions(+)

diff --git a/drivers/firmware/psci/psci.c b/drivers/firmware/psci/psci.c
index a1ebbe9b73b136218e9d9f9b8daa7756b3ab2fbe..6f8c47deaec0225f26704e1f3bcad52603127a85 100644
--- a/drivers/firmware/psci/psci.c
+++ b/drivers/firmware/psci/psci.c
@@ -80,6 +80,14 @@ static u32 psci_cpu_suspend_feature;
static bool psci_system_reset2_supported;
static bool psci_system_off2_hibernate_supported;
+struct psci_reset_param {
+ const char *mode;
+ u32 reset_type;
+ u32 cookie;
+};
+static struct psci_reset_param *psci_reset_params __ro_after_init;
+static size_t num_psci_reset_params __ro_after_init;
+
static inline bool psci_has_ext_power_state(void)
{
return psci_cpu_suspend_feature &
@@ -306,9 +314,39 @@ static int get_set_conduit_method(const struct device_node *np)
return 0;
}
+static int psci_vendor_system_reset2(const char *cmd)
+{
+ unsigned long ret;
+ size_t i;
+
+ for (i = 0; i < num_psci_reset_params; i++) {
+ if (!strcmp(psci_reset_params[i].mode, cmd)) {
+ ret = invoke_psci_fn(PSCI_FN_NATIVE(1_1, SYSTEM_RESET2),
+ psci_reset_params[i].reset_type,
+ psci_reset_params[i].cookie, 0);
+ /*
+ * if vendor reset fails, log it and fall back to
+ * architecture reset types

That's not what the code does.

Ack.

-Mukesh

+ */
+ pr_err("failed to perform reset \"%s\": %ld\n", cmd,
+ (long)ret);
+ return 0;
+ }
+ }
+
+ return -ENOENT;
+}
+
static int psci_sys_reset(struct notifier_block *nb, unsigned long action,
void *data)
{
+ /*
+ * try to do the vendor system_reset2
+ * If there wasn't a matching command, fall back to architectural resets
+ */
+ if (data && !psci_vendor_system_reset2(data))
+ return NOTIFY_DONE;
+
if ((reboot_mode == REBOOT_WARM || reboot_mode == REBOOT_SOFT) &&
psci_system_reset2_supported) {
/*
@@ -795,6 +833,73 @@ static const struct of_device_id psci_of_match[] __initconst = {
{},
};
+#define REBOOT_PREFIX "mode-"
+
+static int __init psci_init_system_reset2_modes(void)
+{
+ const size_t len = strlen(REBOOT_PREFIX);
+ struct psci_reset_param *param;
+ struct device_node *psci_np __free(device_node) = NULL;
+ struct device_node *np __free(device_node) = NULL;
+ struct property *prop;
+ size_t count = 0;
+ u32 magic[2];
+ int num;
+
+ if (!psci_system_reset2_supported)
+ return 0;
+
+ psci_np = of_find_matching_node(NULL, psci_of_match);
+ if (!psci_np)
+ return 0;
+
+ np = of_find_node_by_name(psci_np, "reset-types");
+ if (!np)
+ return 0;

Related to my initial question above. If LINUX_REBOOT_CMD_RESTART2 *arg command,
is the actual reset to be issued, should we add a default mode "cold"
and, if SYSTEM_RESET2 is supported, a "warm" reset mode too ?

It all boils down to what *arg represents - adding "cold" and "warm"
modes would remove the dependency on reboot_mode for resets issued
through LINUX_REBOOT_CMD_RESTART2, the question is whether this
is the correct thing to do.

Comments very welcome.

Thanks,
Lorenzo

+
+ for_each_property_of_node(np, prop) {
+ if (strncmp(prop->name, REBOOT_PREFIX, len))
+ continue;
+ num = of_property_count_u32_elems(np, prop->name);
+ if (num != 1 && num != 2)
+ continue;
+
+ count++;
+ }
+
+ param = psci_reset_params =
+ kcalloc(count, sizeof(*psci_reset_params), GFP_KERNEL);
+ if (!psci_reset_params)
+ return -ENOMEM;
+
+ for_each_property_of_node(np, prop) {
+ if (strncmp(prop->name, REBOOT_PREFIX, len))
+ continue;
+
+ num = of_property_read_variable_u32_array(np, prop->name, magic,
+ 1, ARRAY_SIZE(magic));
+ if (num < 0) {
+ pr_warn("Failed to parse vendor reboot mode %s\n",
+ param->mode);
+ kfree_const(param->mode);
+ continue;
+ }
+
+ param->mode = kstrdup_const(prop->name + len, GFP_KERNEL);
+ if (!param->mode)
+ continue;
+
+ /* Force reset type to be in vendor space */
+ param->reset_type = PSCI_1_1_RESET_TYPE_VENDOR_START | magic[0];
+ param->cookie = num > 1 ? magic[1] : 0;
+ param++;
+ num_psci_reset_params++;
+ }
+
+ return 0;
+}
+arch_initcall(psci_init_system_reset2_modes);
+
int __init psci_dt_init(void)
{
struct device_node *np;

--
2.34.1




--
With best wishes
Dmitry