Skip to contentSkip to conversation

USB: usblp: fix DMA to stack

PATCH In review Johan Hovold 4 participants 7 messages Jan 4, 2021 – Jan 6, 2021

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

PATCH 1/1 USB: usblp: fix DMA to stack

Stack-allocated buffers cannot be used for DMA (on all architectures).

Replace the HP-channel macro with a helper function that allocates a dedicated transfer buffer so that it can continue to be used with arguments from the stack.

Note that the buffer is cleared on allocation as usblp_ctrl_msg() returns success also on short transfers (the buffer is only used for debugging).

---
 drivers/usb/class/usblp.c | 21 +++++++++++++++++++--
 1 file changed, 19 insertions(+), 2 deletions(-)
drivers/usb/class/usblp.c +19 -2
index 67cbd42421be..134dc2005ce9 100644
@@ -274,8 +274,25 @@ static int usblp_ctrl_msg(struct usblp *usblp, int request, int type, int dir, i
 #define usblp_reset(usblp)\
 	usblp_ctrl_msg(usblp, USBLP_REQ_RESET, USB_TYPE_CLASS, USB_DIR_OUT, USB_RECIP_OTHER, 0, NULL, 0)
 
-#define usblp_hp_channel_change_request(usblp, channel, buffer) \
-	usblp_ctrl_msg(usblp, USBLP_REQ_HP_CHANNEL_CHANGE_REQUEST, USB_TYPE_VENDOR, USB_DIR_IN, USB_RECIP_INTERFACE, channel, buffer, 1)
+static int usblp_hp_channel_change_request(struct usblp *usblp, int channel, u8 *new_channel)
+{
+	u8 *buf;
+	int ret;
+
+	buf = kzalloc(1, GFP_KERNEL);
+	if (!buf)
+		return -ENOMEM;
+
+	ret = usblp_ctrl_msg(usblp, USBLP_REQ_HP_CHANNEL_CHANGE_REQUEST,
+			USB_TYPE_VENDOR, USB_DIR_IN, USB_RECIP_INTERFACE,
+			channel, buf, 1);
+	if (ret == 0)
+		*new_channel = buf[0];
+
+	kfree(buf);
+
+	return ret;
+}
 
 /*
  * See the description for usblp_select_alts() below for the usage
Signature · 1 line
2.26.2
#1 lore raw
Quoted 35 lines · Johan Hovold

Stack-allocated buffers cannot be used for DMA (on all architectures).

Replace the HP-channel macro with a helper function that allocates a dedicated transfer buffer so that it can continue to be used with arguments from the stack.

Note that the buffer is cleared on allocation as usblp_ctrl_msg() returns success also on short transfers (the buffer is only used for debugging).

---
 drivers/usb/class/usblp.c | 21 +++++++++++++++++++--
 1 file changed, 19 insertions(+), 2 deletions(-)
drivers/usb/class/usblp.c +9 -2
index 67cbd42421be..134dc2005ce9 100644
@@ -274,8 +274,25 @@ static int usblp_ctrl_msg(struct usblp *usblp, int request, int type, int dir, i
 #define usblp_reset(usblp)\
 	usblp_ctrl_msg(usblp, USBLP_REQ_RESET, USB_TYPE_CLASS, USB_DIR_OUT, USB_RECIP_OTHER, 0, NULL, 0)
 
-#define usblp_hp_channel_change_request(usblp, channel, buffer) \
-	usblp_ctrl_msg(usblp, USBLP_REQ_HP_CHANNEL_CHANGE_REQUEST, USB_TYPE_VENDOR, USB_DIR_IN, USB_RECIP_INTERFACE, channel, buffer, 1)
+static int usblp_hp_channel_change_request(struct usblp *usblp, int channel, u8 *new_channel)
+{
+	u8 *buf;
+	int ret;
+
+	buf = kzalloc(1, GFP_KERNEL);
+	if (!buf)
+		return -ENOMEM;
+

drivers/usb/class/usblp.c static int usblp_ctrl_msg(struct usblp *usblp, int request, int type, int dir, i · line 295

Quoted 10 lines
drivers/usb/class/usblp.c +10 -0
+	ret = usblp_ctrl_msg(usblp, USBLP_REQ_HP_CHANNEL_CHANGE_REQUEST,
+			USB_TYPE_VENDOR, USB_DIR_IN, USB_RECIP_INTERFACE,
+			channel, buf, 1);
+	if (ret == 0)
+		*new_channel = buf[0];
+
+	kfree(buf);
+
+	return ret;
+}

Wow, no one uses this driver anymore it seems, this should have triggered a runtime warning on newer kernels :(

Thanks for this, will queue it up soon.

greg k-h

#2 lore raw
Quoted 9 lines · Greg Kroah-Hartman
Quoted 5 lines · Johan Hovold

Stack-allocated buffers cannot be used for DMA (on all architectures).

Replace the HP-channel macro with a helper function that allocates a dedicated transfer buffer so that it can continue to be used with arguments from the stack.

Wow, no one uses this driver anymore it seems, this should have triggered a runtime warning on newer kernels :(

This helper is only used by the IOCNR_HP_SET_CHANNEL ioctl so perhaps it's just that one which isn't used much.

Johan

#3 lore raw

Johan/Greg,

Since CUPS uses libusb to communicate with printers these days (well, for over a decade now) so that printing and scanning can coexist, the usblp driver really doesn't get any usage anymore.

Quoted 10 lines · Johan Hovold
Quoted 6 lines · Greg Kroah-Hartman
Quoted 5 lines · Johan Hovold

Stack-allocated buffers cannot be used for DMA (on all architectures).

Replace the HP-channel macro with a helper function that allocates a dedicated transfer buffer so that it can continue to be used with arguments from the stack.

Quoted 8 lines
Quoted 2 lines

Wow, no one uses this driver anymore it seems, this should have triggered a runtime warning on newer kernels :(

This helper is only used by the IOCNR_HP_SET_CHANNEL ioctl so perhaps it's just that one which isn't used much.

Johan

________________________
Michael Sweet

#4 lore raw
Quoted 5 lines · Michael Sweet

Johan/Greg,

Since CUPS uses libusb to communicate with printers these days (well, for over a decade now) so that printing and scanning can coexist, the usblp driver really doesn't get any usage anymore.

Ah, ok. I still seem to be using usblp.ko for my printer with CUPS, though.

Johan

maintainer Acked-by on v1 1/1
#5 lore raw
In reply to Johan Hovold

drivers/usb/class/usblp.c static int usblp_ctrl_msg(struct usblp *usblp, int request, int type, int dir, i · line 277

Quoted 4 lines · On Mon, 4 Jan 2021 15:53:02 +0100 Johan Hovold <[email protected]> wrote:

+++ b/drivers/usb/class/usblp.c

drivers/usb/class/usblp.c +1 -2
-#define usblp_hp_channel_change_request(usblp, channel, buffer) \
-	usblp_ctrl_msg(usblp, USBLP_REQ_HP_CHANNEL_CHANGE_REQUEST, USB_TYPE_VENDOR, USB_DIR_IN, USB_RECIP_INTERFACE, channel, buffer, 1)
+static int usblp_hp_channel_change_request(struct usblp *usblp, int channel, u8 *new_channel)

I would probably get rid of the buffer pointer and return new_channel & 0xFF in case of success. That would kill the newChannel too, and there's no need to debage u8 versus unsigned char. But this is good enough. A function is better than trying to cram the kfree() into the clause of the switch.

-- Pete

#6 lore raw
Quoted 10 lines · Pete Zaitcev

drivers/usb/class/usblp.c static int usblp_ctrl_msg(struct usblp *usblp, int request, int type, int dir, i · line 277

Quoted 4 lines · On Mon, 4 Jan 2021 15:53:02 +0100 Johan Hovold <[email protected]> wrote:

+++ b/drivers/usb/class/usblp.c

drivers/usb/class/usblp.c +1 -2
-#define usblp_hp_channel_change_request(usblp, channel, buffer) \
-	usblp_ctrl_msg(usblp, USBLP_REQ_HP_CHANNEL_CHANGE_REQUEST, USB_TYPE_VENDOR, USB_DIR_IN, USB_RECIP_INTERFACE, channel, buffer, 1)
+static int usblp_hp_channel_change_request(struct usblp *usblp, int channel, u8 *new_channel)
Quoted 5 lines

I would probably get rid of the buffer pointer and return new_channel & 0xFF in case of success. That would kill the newChannel too, and there's no need to debage u8 versus unsigned char. But this is good enough. A function is better than trying to cram the kfree() into the clause of the switch.

Yeah, I wanted a minimal change suitable for stable and the helper was already there to be used for this.

Johan