Skip to contentSkip to conversation

usb: core: add USB_QUIRK_CONFIG_DESC_READ_255 for Razer BlackShark V3 Pro

PATCH In review Charles Daoust 3 participants 5 messages Jul 20 – Aug 5

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 usb: core: add USB_QUIRK_CONFIG_DESC_READ_255 for Razer BlackShark V3 Pro

From: Charles D'Aoust <[email protected]>

The Razer BlackShark V3 Pro wireless headset dongle (1532:0577) fingerprints its host during the first enumeration after power-on: if the first GET_DESCRIPTOR(CONFIGURATION) request asks for 255 bytes, as Windows does, the dongle enables its vendor HID command channel; if it sees the 9-byte header-only read that Linux issues, it disables that channel for the rest of the power session. Audio still works in that state, but battery reporting and all vendor commands are silently ignored, and nothing short of removing power recovers the device:
resets, re-enumerations and byte-exact replays of complete Windows control sessions were all verified not to help.

Both read lengths are spec-compliant (the device truncates the reply to wLength); the firmware was evidently only validated against the larger request. The kernel already accommodates this class of firmware assumption during enumeration: hub_port_init() reads the device descriptor with a 64-byte request because that is what Windows does and what many devices expect.

Add USB_QUIRK_CONFIG_DESC_READ_255, which makes usb_get_configuration() request 255 bytes for the initial configuration descriptor read rather than USB_DT_CONFIG_SIZE, apply it to 1532:0577, and expose it as runtime quirk letter 'r'. Devices without the quirk are unaffected.

The trigger was isolated by single-variable bisection on otherwise unmodified kernels: with only the widened initial read, the dongle's vendor channel comes up enabled on a cold plug with no interface drivers bound (bare enumeration only); without it, it never does.

---
 Documentation/admin-guide/kernel-parameters.txt |  3 +++
 drivers/usb/core/config.c                       | 13 ++++++++++---
 drivers/usb/core/quirks.c                       |  6 ++++++
 include/linux/usb/quirks.h                      |  3 +++
 4 files changed, 22 insertions(+), 3 deletions(-)
Documentation/admin-guide/kernel-parameters.txt +3 -0
index b5493a7f8..1bd138ba6 100644
@@ -8169,6 +8169,9 @@ Kernel parameters
 				q = USB_QUIRK_FORCE_ONE_CONFIG (Device
 					claims zero configurations,
 					forcing to 1);
+				r = USB_QUIRK_CONFIG_DESC_READ_255 (initial
+					configuration descriptor read must
+					request 255 bytes, as Windows does);
 			Example: quirks=0781:5580:bk,0a5c:5834:gij
 
 	usbhid.mousepoll=
drivers/usb/core/config.c +10 -3
index cd3231d21..c7e21cc76 100644
@@ -908,7 +908,7 @@ int usb_get_configuration(struct usb_device *dev)
 {
 	struct device *ddev = &dev->dev;
 	int ncfg = dev->descriptor.bNumConfigurations;
-	unsigned int cfgno, length;
+	unsigned int cfgno, length, first_read_len;
 	unsigned char *bigbuffer;
 	struct usb_config_descriptor *desc;
 	int result;
@@ -938,7 +938,14 @@ int usb_get_configuration(struct usb_device *dev)
 	if (!dev->rawdescriptors)
 		return -ENOMEM;
 
-	desc = kmalloc(USB_DT_CONFIG_SIZE, GFP_KERNEL);
+	/*
+	 * Some devices (see USB_QUIRK_CONFIG_DESC_READ_255) malfunction unless
+	 * the initial configuration descriptor read requests 255 bytes, as
+	 * Windows does.  Only the header is consumed here either way.
+	 */
+	first_read_len = (dev->quirks & USB_QUIRK_CONFIG_DESC_READ_255) ?
+			 255 : USB_DT_CONFIG_SIZE;
+	desc = kmalloc(first_read_len, GFP_KERNEL);
 	if (!desc)
 		return -ENOMEM;
 
@@ -946,7 +953,7 @@ int usb_get_configuration(struct usb_device *dev)
 		/* We grab just the first descriptor so we know how long
 		 * the whole configuration is */
 		result = usb_get_descriptor(dev, USB_DT_CONFIG, cfgno,
-		    desc, USB_DT_CONFIG_SIZE);
+		    desc, first_read_len);
 		if (result < 0) {
 			dev_err(ddev, "unable to read config index %d "
 			    "descriptor/%s: %d\n", cfgno, "start", result);
drivers/usb/core/quirks.c +6 -0
index 87ee2d938..c8f71b825 100644
@@ -142,6 +142,9 @@ static int quirks_param_set(const char *value, const struct kernel_param *kp)
 				break;
 			case 'q':
 				flags |= USB_QUIRK_FORCE_ONE_CONFIG;
+				break;
+			case 'r':
+				flags |= USB_QUIRK_CONFIG_DESC_READ_255;
 			/* Ignore unrecognized flag characters */
 			}
 		}
@@ -496,6 +499,9 @@ static const struct usb_device_id usb_quirk_list[] = {
 	/* Razer - Razer Blade Keyboard */
 	{ USB_DEVICE(0x1532, 0x0116), .driver_info =
 			USB_QUIRK_LINEAR_UFRAME_INTR_BINTERVAL },
+	/* Razer - Razer BlackShark V3 Pro wireless headset dongle */
+	{ USB_DEVICE(0x1532, 0x0577), .driver_info =
+			USB_QUIRK_CONFIG_DESC_READ_255 },
 	/* Razer - Razer Kiyo Pro Webcam */
 	{ USB_DEVICE(0x1532, 0x0e05), .driver_info = USB_QUIRK_NO_LPM },
 
include/linux/usb/quirks.h +3 -0
index b3cc7beab..02de5ed48 100644
@@ -81,4 +81,7 @@
 /* Device claims zero configurations, forcing to 1 */
 #define USB_QUIRK_FORCE_ONE_CONFIG		BIT(18)
 
+/* initial configuration descriptor read must request 255 bytes (Windows) */
+#define USB_QUIRK_CONFIG_DESC_READ_255		BIT(19)
+
 #endif /* __LINUX_USB_QUIRKS_H */
Signature · 1 line
2.55.0
#1 lore raw
Quoted 25 lines · Charles Daoust

From: Charles D'Aoust <[email protected]>

The Razer BlackShark V3 Pro wireless headset dongle (1532:0577) fingerprints its host during the first enumeration after power-on: if the first GET_DESCRIPTOR(CONFIGURATION) request asks for 255 bytes, as Windows does, the dongle enables its vendor HID command channel; if it sees the 9-byte header-only read that Linux issues, it disables that channel for the rest of the power session. Audio still works in that state, but battery reporting and all vendor commands are silently ignored, and nothing short of removing power recovers the device:
resets, re-enumerations and byte-exact replays of complete Windows control sessions were all verified not to help.

Both read lengths are spec-compliant (the device truncates the reply to wLength); the firmware was evidently only validated against the larger request. The kernel already accommodates this class of firmware assumption during enumeration: hub_port_init() reads the device descriptor with a 64-byte request because that is what Windows does and what many devices expect.

Add USB_QUIRK_CONFIG_DESC_READ_255, which makes usb_get_configuration() request 255 bytes for the initial configuration descriptor read rather than USB_DT_CONFIG_SIZE, apply it to 1532:0577, and expose it as runtime quirk letter 'r'. Devices without the quirk are unaffected.

Quoted 8 lines

The trigger was isolated by single-variable bisection on otherwise unmodified kernels: with only the widened initial read, the dongle's vendor channel comes up enabled on a cold plug with no interface drivers bound (bare enumeration only); without it, it never does.

---

Is this not the same as

https://lore.kernel.org/linux-usb/[email protected]/

?

Alan Stern

Quoted 98 lines
 Documentation/admin-guide/kernel-parameters.txt |  3 +++
 drivers/usb/core/config.c                       | 13 ++++++++++---
 drivers/usb/core/quirks.c                       |  6 ++++++
 include/linux/usb/quirks.h                      |  3 +++
 4 files changed, 22 insertions(+), 3 deletions(-)
Documentation/admin-guide/kernel-parameters.txt +3 -0
index b5493a7f8..1bd138ba6 100644
@@ -8169,6 +8169,9 @@ Kernel parameters
 				q = USB_QUIRK_FORCE_ONE_CONFIG (Device
 					claims zero configurations,
 					forcing to 1);
+				r = USB_QUIRK_CONFIG_DESC_READ_255 (initial
+					configuration descriptor read must
+					request 255 bytes, as Windows does);
 			Example: quirks=0781:5580:bk,0a5c:5834:gij
 
 	usbhid.mousepoll=
drivers/usb/core/config.c +10 -3
index cd3231d21..c7e21cc76 100644
@@ -908,7 +908,7 @@ int usb_get_configuration(struct usb_device *dev)
 {
 	struct device *ddev = &dev->dev;
 	int ncfg = dev->descriptor.bNumConfigurations;
-	unsigned int cfgno, length;
+	unsigned int cfgno, length, first_read_len;
 	unsigned char *bigbuffer;
 	struct usb_config_descriptor *desc;
 	int result;
@@ -938,7 +938,14 @@ int usb_get_configuration(struct usb_device *dev)
 	if (!dev->rawdescriptors)
 		return -ENOMEM;
 
-	desc = kmalloc(USB_DT_CONFIG_SIZE, GFP_KERNEL);
+	/*
+	 * Some devices (see USB_QUIRK_CONFIG_DESC_READ_255) malfunction unless
+	 * the initial configuration descriptor read requests 255 bytes, as
+	 * Windows does.  Only the header is consumed here either way.
+	 */
+	first_read_len = (dev->quirks & USB_QUIRK_CONFIG_DESC_READ_255) ?
+			 255 : USB_DT_CONFIG_SIZE;
+	desc = kmalloc(first_read_len, GFP_KERNEL);
 	if (!desc)
 		return -ENOMEM;
 
@@ -946,7 +953,7 @@ int usb_get_configuration(struct usb_device *dev)
 		/* We grab just the first descriptor so we know how long
 		 * the whole configuration is */
 		result = usb_get_descriptor(dev, USB_DT_CONFIG, cfgno,
-		    desc, USB_DT_CONFIG_SIZE);
+		    desc, first_read_len);
 		if (result < 0) {
 			dev_err(ddev, "unable to read config index %d "
 			    "descriptor/%s: %d\n", cfgno, "start", result);
drivers/usb/core/quirks.c +6 -0
index 87ee2d938..c8f71b825 100644
@@ -142,6 +142,9 @@ static int quirks_param_set(const char *value, const struct kernel_param *kp)
 				break;
 			case 'q':
 				flags |= USB_QUIRK_FORCE_ONE_CONFIG;
+				break;
+			case 'r':
+				flags |= USB_QUIRK_CONFIG_DESC_READ_255;
 			/* Ignore unrecognized flag characters */
 			}
 		}
@@ -496,6 +499,9 @@ static const struct usb_device_id usb_quirk_list[] = {
 	/* Razer - Razer Blade Keyboard */
 	{ USB_DEVICE(0x1532, 0x0116), .driver_info =
 			USB_QUIRK_LINEAR_UFRAME_INTR_BINTERVAL },
+	/* Razer - Razer BlackShark V3 Pro wireless headset dongle */
+	{ USB_DEVICE(0x1532, 0x0577), .driver_info =
+			USB_QUIRK_CONFIG_DESC_READ_255 },
 	/* Razer - Razer Kiyo Pro Webcam */
 	{ USB_DEVICE(0x1532, 0x0e05), .driver_info = USB_QUIRK_NO_LPM },
 
include/linux/usb/quirks.h +3 -0
index b3cc7beab..02de5ed48 100644
@@ -81,4 +81,7 @@
 /* Device claims zero configurations, forcing to 1 */
 #define USB_QUIRK_FORCE_ONE_CONFIG		BIT(18)
 
+/* initial configuration descriptor read must request 255 bytes (Windows) */
+#define USB_QUIRK_CONFIG_DESC_READ_255		BIT(19)
+
 #endif /* __LINUX_USB_QUIRKS_H */
Signature · 1 line
2.55.0
#2 lore raw

You are correct, they are indeed the same. I apologize, I missed it.
---
Charles D’Aoust

---
Charles D’Aoust

Quoted 142 lines · Alan Stern
Quoted 25 lines · Charles Daoust

From: Charles D'Aoust <[email protected]>

The Razer BlackShark V3 Pro wireless headset dongle (1532:0577) fingerprints its host during the first enumeration after power-on: if the first GET_DESCRIPTOR(CONFIGURATION) request asks for 255 bytes, as Windows does, the dongle enables its vendor HID command channel; if it sees the 9-byte header-only read that Linux issues, it disables that channel for the rest of the power session. Audio still works in that state, but battery reporting and all vendor commands are silently ignored, and nothing short of removing power recovers the device:
resets, re-enumerations and byte-exact replays of complete Windows control sessions were all verified not to help.

Both read lengths are spec-compliant (the device truncates the reply to wLength); the firmware was evidently only validated against the larger request. The kernel already accommodates this class of firmware assumption during enumeration: hub_port_init() reads the device descriptor with a 64-byte request because that is what Windows does and what many devices expect.

Add USB_QUIRK_CONFIG_DESC_READ_255, which makes usb_get_configuration() request 255 bytes for the initial configuration descriptor read rather than USB_DT_CONFIG_SIZE, apply it to 1532:0577, and expose it as runtime quirk letter 'r'. Devices without the quirk are unaffected.

Quoted 8 lines

The trigger was isolated by single-variable bisection on otherwise unmodified kernels: with only the widened initial read, the dongle's vendor channel comes up enabled on a cold plug with no interface drivers bound (bare enumeration only); without it, it never does.

---

Is this not the same as

https://lore.kernel.org/linux-usb/[email protected]/

?

Alan Stern

Quoted 98 lines
 Documentation/admin-guide/kernel-parameters.txt |  3 +++
 drivers/usb/core/config.c                       | 13 ++++++++++---
 drivers/usb/core/quirks.c                       |  6 ++++++
 include/linux/usb/quirks.h                      |  3 +++
 4 files changed, 22 insertions(+), 3 deletions(-)
Documentation/admin-guide/kernel-parameters.txt +3 -0
index b5493a7f8..1bd138ba6 100644
@@ -8169,6 +8169,9 @@ Kernel parameters
                              q = USB_QUIRK_FORCE_ONE_CONFIG (Device
                                      claims zero configurations,
                                      forcing to 1);
+                             r = USB_QUIRK_CONFIG_DESC_READ_255 (initial
+                                     configuration descriptor read must
+                                     request 255 bytes, as Windows does);
                      Example: quirks=0781:5580:bk,0a5c:5834:gij
 
      usbhid.mousepoll=
drivers/usb/core/config.c +10 -3
index cd3231d21..c7e21cc76 100644
@@ -908,7 +908,7 @@ int usb_get_configuration(struct usb_device *dev)
 {
      struct device *ddev = &dev->dev;
      int ncfg = dev->descriptor.bNumConfigurations;
-     unsigned int cfgno, length;
+     unsigned int cfgno, length, first_read_len;
      unsigned char *bigbuffer;
      struct usb_config_descriptor *desc;
      int result;
@@ -938,7 +938,14 @@ int usb_get_configuration(struct usb_device *dev)
      if (!dev->rawdescriptors)
              return -ENOMEM;
 
-     desc = kmalloc(USB_DT_CONFIG_SIZE, GFP_KERNEL);
+     /*
+      * Some devices (see USB_QUIRK_CONFIG_DESC_READ_255) malfunction unless
+      * the initial configuration descriptor read requests 255 bytes, as
+      * Windows does.  Only the header is consumed here either way.
+      */
+     first_read_len = (dev->quirks & USB_QUIRK_CONFIG_DESC_READ_255) ?
+                      255 : USB_DT_CONFIG_SIZE;
+     desc = kmalloc(first_read_len, GFP_KERNEL);
      if (!desc)
              return -ENOMEM;
 
@@ -946,7 +953,7 @@ int usb_get_configuration(struct usb_device *dev)
              /* We grab just the first descriptor so we know how long
               * the whole configuration is */
              result = usb_get_descriptor(dev, USB_DT_CONFIG, cfgno,
-                 desc, USB_DT_CONFIG_SIZE);
+                 desc, first_read_len);
              if (result < 0) {
                      dev_err(ddev, "unable to read config index %d "
                          "descriptor/%s: %d\n", cfgno, "start", result);
drivers/usb/core/quirks.c +6 -0
index 87ee2d938..c8f71b825 100644
@@ -142,6 +142,9 @@ static int quirks_param_set(const char *value, const struct kernel_param *kp)
                              break;
                      case 'q':
                              flags |= USB_QUIRK_FORCE_ONE_CONFIG;
+                             break;
+                     case 'r':
+                             flags |= USB_QUIRK_CONFIG_DESC_READ_255;
                      /* Ignore unrecognized flag characters */
                      }
              }
@@ -496,6 +499,9 @@ static const struct usb_device_id usb_quirk_list[] = {
      /* Razer - Razer Blade Keyboard */
      { USB_DEVICE(0x1532, 0x0116), .driver_info =
                      USB_QUIRK_LINEAR_UFRAME_INTR_BINTERVAL },
+     /* Razer - Razer BlackShark V3 Pro wireless headset dongle */
+     { USB_DEVICE(0x1532, 0x0577), .driver_info =
+                     USB_QUIRK_CONFIG_DESC_READ_255 },
      /* Razer - Razer Kiyo Pro Webcam */
      { USB_DEVICE(0x1532, 0x0e05), .driver_info = USB_QUIRK_NO_LPM },
 
include/linux/usb/quirks.h +3 -1
index b3cc7beab..02de5ed48 100644
@@ -81,4 +81,7 @@
 /* Device claims zero configurations, forcing to 1 */
 #define USB_QUIRK_FORCE_ONE_CONFIG           BIT(18)
 
+/* initial configuration descriptor read must request 255 bytes (Windows) */
+#define USB_QUIRK_CONFIG_DESC_READ_255               BIT(19)
+
 #endif /* __LINUX_USB_QUIRKS_H */
--

2.55.0

#3 lore raw
In reply to Charles Daoust
Quoted 13 lines · Charles Daoust

From: Charles D'Aoust <[email protected]>

The Razer BlackShark V3 Pro wireless headset dongle (1532:0577) fingerprints its host during the first enumeration after power-on: if the first GET_DESCRIPTOR(CONFIGURATION) request asks for 255 bytes, as Windows does, the dongle enables its vendor HID command channel; if it sees the 9-byte header-only read that Linux issues, it disables that channel for the rest of the power session. Audio still works in that state, but battery reporting and all vendor commands are silently ignored, and nothing short of removing power recovers the device:
resets, re-enumerations and byte-exact replays of complete Windows control sessions were all verified not to help.

Quoted 3 lines

Both read lengths are spec-compliant (the device truncates the reply to wLength); the firmware was evidently only validated against the larger request.

Hmm, so is it fingerprinting of the host or sloppy QA? ;) Out of curiosity, was this patch AI-generated?

Quoted 10 lines

The kernel already accommodates this class of
firmware assumption during enumeration: hub_port_init() reads the device descriptor with a 64-byte request because that is what Windows does and what many devices expect.

Add USB_QUIRK_CONFIG_DESC_READ_255, which makes usb_get_configuration() request 255 bytes for the initial configuration descriptor read rather than USB_DT_CONFIG_SIZE, apply it to 1532:0577, and expose it as runtime quirk letter 'r'. Devices without the quirk are unaffected.

Quoted 7 lines

The trigger was isolated by single-variable bisection on otherwise unmodified kernels: with only the widened initial read, the dongle's vendor channel comes up enabled on a cold plug with no interface drivers bound (bare enumeration only); without it, it never does.

But anyway, FYI an equivalent quirk is now sitting in Greg KH's queue of fixes for (presumably) the 7.2 release.

I suspect that Greg would accept one more patch to enable the quirk by default on your device ID.

Regards,
Michal

#4 lore raw

It was indeed AI assisted. It's hard to tell if this is sloppy QA, obviously, but initially I thought it was an actual lockdown in the firmware to prevent non-Razer software from accessing the device. Turns out you're probably right.
---
Charles D’Aoust

Quoted 48 lines · Michal Pecio
Quoted 13 lines · Charles Daoust

From: Charles D'Aoust <[email protected]>

The Razer BlackShark V3 Pro wireless headset dongle (1532:0577) fingerprints its host during the first enumeration after power-on: if the first GET_DESCRIPTOR(CONFIGURATION) request asks for 255 bytes, as Windows does, the dongle enables its vendor HID command channel; if it sees the 9-byte header-only read that Linux issues, it disables that channel for the rest of the power session. Audio still works in that state, but battery reporting and all vendor commands are silently ignored, and nothing short of removing power recovers the device:
resets, re-enumerations and byte-exact replays of complete Windows control sessions were all verified not to help.

Quoted 3 lines

Both read lengths are spec-compliant (the device truncates the reply to wLength); the firmware was evidently only validated against the larger request.

Hmm, so is it fingerprinting of the host or sloppy QA? ;) Out of curiosity, was this patch AI-generated?

Quoted 10 lines

The kernel already accommodates this class of
firmware assumption during enumeration: hub_port_init() reads the device descriptor with a 64-byte request because that is what Windows does and what many devices expect.

Add USB_QUIRK_CONFIG_DESC_READ_255, which makes usb_get_configuration() request 255 bytes for the initial configuration descriptor read rather than USB_DT_CONFIG_SIZE, apply it to 1532:0577, and expose it as runtime quirk letter 'r'. Devices without the quirk are unaffected.

Quoted 7 lines

The trigger was isolated by single-variable bisection on otherwise unmodified kernels: with only the widened initial read, the dongle's vendor channel comes up enabled on a cold plug with no interface drivers bound (bare enumeration only); without it, it never does.

But anyway, FYI an equivalent quirk is now sitting in Greg KH's queue of fixes for (presumably) the 7.2 release.

I suspect that Greg would accept one more patch to enable the quirk by default on your device ID.

Regards,
Michal