Skip to contentSkip to conversation

firmware loader: log path to loaded firmwares

PATCH In review Drew DeVault 4 participants 8 messages Nov 3, 2019 – Nov 18, 2019

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 firmware loader: log path to loaded firmwares

This is useful for users who are trying to identify the firmwares in use on their system.

---
v2 uses dev_dbg instead of printk(KERN_INFO)

 drivers/base/firmware_loader/main.c | 1 +
 1 file changed, 1 insertion(+)
drivers/base/firmware_loader/main.c +1 -0
index bf44c79beae9..2537da43a572 100644
@@ -504,6 +504,7 @@ fw_get_filesystem_firmware(struct device *device, struct fw_priv *fw_priv,
 					 path);
 			continue;
 		}
+		dev_dbg(device, "Loading firmware from %s\n", path);
 		if (decompress) {
 			dev_dbg(device, "f/w decompressing %s\n",
 				fw_priv->fw_name);
Signature · 1 line
2.23.0
maintainer
#1 lore raw
Quoted 16 lines · Drew DeVault

This is useful for users who are trying to identify the firmwares in use on their system.

---
v2 uses dev_dbg instead of printk(KERN_INFO)

 drivers/base/firmware_loader/main.c | 1 +
 1 file changed, 1 insertion(+)
drivers/base/firmware_loader/main.c +0 -0
index bf44c79beae9..2537da43a572 100644

drivers/base/firmware_loader/main.c fw_get_filesystem_firmware(struct device *device, struct fw_priv *fw_priv, · line 507

Quoted 4 lines
drivers/base/firmware_loader/main.c +1 -0
@@ -504,6 +504,7 @@ fw_get_filesystem_firmware(struct device *device, struct fw_priv *fw_priv,
 					 path);
 			continue;
 		}
+		dev_dbg(device, "Loading firmware from %s\n", path);

Because this is dev_dbg() I'm willing to consider it, so that its not always enabled. However its not in the right place, the code path you are addressing is only for direct filesystem lookups. If that fails some systems do a fallback call out to userspace. To cover both cases, you want it at the end of _request_firmware() on the success path. Can you send a new patch?

Luis

Quoted 6 lines
drivers/base/firmware_loader/main.c +0 -0
 		if (decompress) {
 			dev_dbg(device, "f/w decompressing %s\n",
 				fw_priv->fw_name);
Signature · 1 line
2.23.0
#3 lore raw
In reply to Luis Chamberlain
Quoted 12 lines · Luis Chamberlain
Quoted 11 lines · Drew DeVault

This is useful for users who are trying to identify the firmwares in use on their system.

---
v2 uses dev_dbg instead of printk(KERN_INFO)

 drivers/base/firmware_loader/main.c | 1 +
 1 file changed, 1 insertion(+)
Quoted 16 lines

drivers/base/firmware_loader/main.c fw_get_filesystem_firmware(struct device *device, struct fw_priv *fw_priv, · line 507

Quoted 9 lines
drivers/base/firmware_loader/main.c +1 -0
index bf44c79beae9..2537da43a572 100644
@@ -504,6 +504,7 @@ fw_get_filesystem_firmware(struct device *device, struct fw_priv *fw_priv,
 					 path);
 			continue;
 		}
+		dev_dbg(device, "Loading firmware from %s\n", path);

Because this is dev_dbg() I'm willing to consider it, so that its not always enabled. However its not in the right place, the code path you are addressing is only for direct filesystem lookups. If that fails some systems do a fallback call out to userspace. To cover both cases, you want it at the end of _request_firmware() on the success path. Can you send a new patch?

As the author of a separate patch that predates Drew's patch (originally in July, with a later version sent to the list last week):
https://pastebin.com/Tf09x3ed (v1)
https://lkml.org/lkml/2019/11/7/800 (v2)

This already uses the _request_firmware path to fire after the firmware was successfully loaded. The commit message is also specific that it's to cover early boot situations, before UEVENT can be logged.

dev_dbg means that the loglevel must have been set to debug BEFORE the firmware load took place, but this means either setting system-wide debug spam or requiring debugfs, which is annoying for boot stuff (not impossible, just annoying).

I have two uses cases overall:
- log so you know exactly when it's loaded successfully (great if loading a firmware causes your system to lock up a few seconds later)
- at some point in the future, being able to query what firmware was loaded in the past, and esp. exactly what version/data was in that firmware file.

Signature · 5 lines
Robin Hugh Johnson
Gentoo Linux: Dev, Infra Lead, Foundation Treasurer
E-Mail   : [email protected]
GnuPG FP : 11ACBA4F 4778E3F6 E4EDF38E B27B944E 34884E85
GnuPG FP : 7D0B3CEB E9B85B1F 825BCECF EE05E6F6 A48F6136
maintainer
#4 lore raw
Quoted 3 lines · Robin H. Johnson

I have two uses cases overall:
- log so you know exactly when it's loaded successfully (great if loading a firmware causes your system to lock up a few seconds later)

Then you can change the driver to confirm this, not impose every driver to do the same.

Quoted 3 lines

- at some point in the future, being able to query what firmware was loaded in the past, and esp. exactly what version/data was in that firmware file.

Firmware data is opaque to the firmware loader, as such details to extract generic information about firmware details can only be done by the driver, which could decode the firmware information. Many drivers print these details themselves already, if they want it.

A generic interface to let us query *all* devices and currently loaded firmware through the firmware loader would only be possible today for firmware which requests (the default) caching of firmware upon suspend/resume given that we keep the device / firmware name pair around prior to suspend. For those devices it could be possible to extend the firmware loader with a driver callback which can extract firmware details in a generic codified way. To support *all* drivers though, in a more clean way for this, a separate but similar list could be kept which enables one to do this. Such items would be torn down upon driver removal. But that would then be an opt-in new mechanism.

Luis

#5 lore raw

I have a new patch prepared which moves my logging into _request_firmware. Would you prefer I send this along or would you like to take Robin's patch?

#6 lore raw

PATCH v3 1/1 firmware: log name & outcome of loaded firmware

In reply to Robin H. Johnson

It's non-trivial to figure out names of firmware that was actually loaded, add a debug statement at the end of _request_firmware that logs the name & result of each firmware.

This is esp. valuable early in boot, before logging of UEVENT is available.

v3:
- Log at dev_dbg level per maintainer.
- HOWTO: Enable at boot via kernel boot param
dyndbg="func _request_firmware +p"
- Credit to Drew DeVault for parallel creation and help promoting the idea.

Alternate-Creation: Drew DeVault <[email protected]>

---
 drivers/base/firmware_loader/main.c | 7 +++++++
 1 file changed, 7 insertions(+)
drivers/base/firmware_loader/main.c +7 -0
index bf44c79beae9..84a879608ca4 100644
@@ -791,6 +791,13 @@ _request_firmware(const struct firmware **firmware_p, const char *name,
 		fw = NULL;
 	}
 
+	/* Provide a consistent way to capture the result of trying to load any
+	 * firmware. As a potential future improvement, this might include
+	 * persistent state that firmware is loaded (or failed to load for some
+	 * reason). See Message-ID: <[email protected]>
+	 * for background */
+	dev_dbg(device, "%s %s ret=%d\n", __func__, name, ret);
+
 	*firmware_p = fw;
 	return ret;
 }
Signature · 1 line
2.24.0
maintainer
#7 lore raw

Re: [PATCH v3] firmware: log name & outcome of loaded firmware

Quoted 7 lines · Robin H. Johnson

It's non-trivial to figure out names of firmware that was actually loaded, add a debug statement at the end of _request_firmware that logs the name & result of each firmware.

This is esp. valuable early in boot, before logging of UEVENT is available.

Quoted 6 lines

v3:
- Log at dev_dbg level per maintainer.
- HOWTO: Enable at boot via kernel boot param
dyndbg="func _request_firmware +p"
- Credit to Drew DeVault for parallel creation and help promoting the idea.

These "v3.." lines need to go below the --- line.

Quoted 2 lines

Alternate-Creation: Drew DeVault <[email protected]>

Is that a valid tag? You can have co-developed-by (or something like that, read the documentation for the real name), but I have never seen this one before.

Quoted 10 lines
---
 drivers/base/firmware_loader/main.c | 7 +++++++
 1 file changed, 7 insertions(+)
drivers/base/firmware_loader/main.c +0 -0
index bf44c79beae9..84a879608ca4 100644

drivers/base/firmware_loader/main.c _request_firmware(const struct firmware **firmware_p, const char *name, · line 797

Quoted 7 lines
drivers/base/firmware_loader/main.c +4 -0
@@ -791,6 +791,13 @@ _request_firmware(const struct firmware **firmware_p, const char *name,
 		fw = NULL;
 	}
 
+	/* Provide a consistent way to capture the result of trying to load any
+	 * firmware. As a potential future improvement, this might include
+	 * persistent state that firmware is loaded (or failed to load for some
+	 * reason). See Message-ID: <[email protected]>

Just provide a lore link with the message id if you really want this.

But really, this type of thing belongs in the changelog text, not in a comment, right?

drivers/base/firmware_loader/main.c _request_firmware(const struct firmware **firmware_p, const char *name, · line 799

Quoted 2 lines
drivers/base/firmware_loader/main.c +2 -0
+	 * for background */
+	dev_dbg(device, "%s %s ret=%d\n", __func__, name, ret);

That does not provide any real information as to what is going on, why doesn't ftrace suffice for this?

thanks,

greg k-h