[PATCH 0/1] Fix DMA_API_DEBUG warning in hid-lenovo

[PATCH 0/1] Fix DMA_API_DEBUG warning in hid-lenovo

[Josh Boyer]

We've had a report[1] in Fedora of the hid-lenovo driver throwing
the backtrace below when CONFIG_DMA_API_DEBUG is set.  The following
patch should fix this.

WARNING: CPU: 3 PID: 385 at lib/dma-debug.c:1169 check_for_stack+0x90/0xd0
ohci-pci 0000:00:13.0: DMA-API: device driver maps memory from stack [addr=ffff880426cafa7d]
Modules linked in: serio_raw amdkfd amd_iommu_v2 alx mdio radeon(+) hid_lenovo(+) ax88179_178a(+) pata_atiixp usbnet i2c_algo_bit drm_kms_helper ttm drm r8169 mii fjes
CPU: 3 PID: 385 Comm: systemd-udevd Not tainted 4.6.0-0.rc0.git20.1.fc25.x86_64 #1
Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./FM2A88X-ITX+, BIOS P3.20 12/03/2015
 0000000000000286 00000000a490d2cf ffff880426caf670 ffffffff8145b265
 ffff880426caf6c0 0000000000000000 ffff880426caf6b0 ffffffff810b2f2b
 0000049126caf708 ffff880426cafa7d ffff88042d6a9098 ffff88042d6277a0
Call Trace:
 [<ffffffff8145b265>] dump_stack+0x86/0xc1
 [<ffffffff810b2f2b>] __warn+0xcb/0xf0
 [<ffffffff810b2faf>] warn_slowpath_fmt+0x5f/0x80
 [<ffffffff8148d2a0>] check_for_stack+0x90/0xd0
 [<ffffffff8148d899>] debug_dma_map_page+0xf9/0x150
 [<ffffffff816678db>] usb_hcd_map_urb_for_dma+0x5db/0x780
 [<ffffffff8166808b>] usb_hcd_submit_urb+0x37b/0xb60
 [<ffffffff815bcbef>] ? dev_vprintk_emit+0xbf/0x230
 [<ffffffff81118771>] ? __raw_spin_lock_init+0x21/0x60
 [<ffffffff8110f034>] ? lockdep_init_map+0x64/0x6c0
 [<ffffffff811121a5>] ? trace_hardirqs_on_caller+0xf5/0x1b0
 [<ffffffff81669a74>] usb_submit_urb+0x2f4/0x550
 [<ffffffff8166a3f4>] usb_start_wait_urb+0x74/0x180
 [<ffffffff8166a5dc>] usb_control_msg+0xdc/0x120
 [<ffffffff8173dad9>] usbhid_raw_request+0xa9/0x180
 [<ffffffffc02930a1>] lenovo_send_cmd_cptkbd+0xa1/0xc0 [hid_lenovo]
 [<ffffffffc02941c3>] lenovo_probe+0x313/0x460 [hid_lenovo]
 [<ffffffff81730e30>] ? hid_match_device+0xa0/0xb0
 [<ffffffff81733207>] hid_device_probe+0xd7/0x160
 [<ffffffff815c18cc>] driver_probe_device+0x22c/0x440
 [<ffffffff815c1bb5>] __driver_attach+0xd5/0x100
 [<ffffffff815c1ae0>] ? driver_probe_device+0x440/0x440
 [<ffffffff815bf113>] bus_for_each_dev+0x73/0xc0
 [<ffffffff815c0f9e>] driver_attach+0x1e/0x20
 [<ffffffff815c09c6>] bus_add_driver+0x1c6/0x290
 [<ffffffffc00bf000>] ? 0xffffffffc00bf000
 [<ffffffff815c2880>] driver_register+0x60/0xe0
 [<ffffffff81732106>] __hid_register_driver+0x66/0xa0
 [<ffffffffc00bf01e>] lenovo_driver_init+0x1e/0x1000 [hid_lenovo]
 [<ffffffff81002123>] do_one_initcall+0xb3/0x1f0
 [<ffffffff81131df5>] ? rcu_read_lock_sched_held+0x45/0x80
 [<ffffffff8126b6fc>] ? kmem_cache_alloc_trace+0x2ac/0x310
 [<ffffffff811f3f4e>] ? do_init_module+0x27/0x1da
 [<ffffffff811f3f86>] do_init_module+0x5f/0x1da
 [<ffffffff8115e588>] load_module+0x21e8/0x2950
 [<ffffffff8115ab30>] ? __symbol_put+0x70/0x70
 [<ffffffff81159b30>] ? show_coresize+0x30/0x30
 [<ffffffff8103fd19>] ? sched_clock+0x9/0x10
 [<ffffffff8115ee62>] SYSC_init_module+0x172/0x1b0
 [<ffffffff8115efce>] SyS_init_module+0xe/0x10
 [<ffffffff81004039>] do_syscall_64+0x69/0x190
 [<ffffffff818ce5ff>] entry_SYSCALL64_slow_path+0x25/0x25

josh

[1] https://bugzilla.redhat.com/show_bug.cgi?id=1321421

[PATCH] HID: lenovo: Don't use stack variables for DMA buffers

[Josh Boyer]

The lenovo_send_cmd_cptkbd function uses a stack variable to submit
commands via hid_hw_raw_request.  Eventually this gets to the
usb_hcd_map_urb_for_dma function, which causes a warning to be thrown
if the CONFIG_DMA_API_DEBUG option is enabled.

Fix this by allocating a temporary buffer instead.

Reported-by: lejeczek <peljasz@yahoo.co.uk>
Signed-off-by: Josh Boyer <jwboyer@fedoraproject.org>
---
 drivers/hid/hid-lenovo.c | 16 +++++++++++++---
 1 file changed, 13 insertions(+), 3 deletions(-)

diff --git a/drivers/hid/hid-lenovo.c b/drivers/hid/hid-lenovo.c
index 0125e356bd8d..2ba5e8ea66a4 100644
--- a/drivers/hid/hid-lenovo.c
+++ b/drivers/hid/hid-lenovo.c
@@ -184,21 +184,31 @@ static int lenovo_send_cmd_cptkbd(struct hid_device *hdev,
 			unsigned char byte2, unsigned char byte3)
 {
 	int ret;
-	unsigned char buf[] = {0x18, byte2, byte3};
+	unsigned char *buf = NULL;
+
+	buf = kzalloc(3, GFP_KERNEL);
+	if (!buf)
+		return -ENOMEM;
+
+	buf[0] = 0x18;
+	buf[1] = byte2;
+	buf[2] = byte3;
 
 	switch (hdev->product) {
 	case USB_DEVICE_ID_LENOVO_CUSBKBD:
-		ret = hid_hw_raw_request(hdev, 0x13, buf, sizeof(buf),
+		ret = hid_hw_raw_request(hdev, 0x13, buf, 3,
 					HID_FEATURE_REPORT, HID_REQ_SET_REPORT);
 		break;
 	case USB_DEVICE_ID_LENOVO_CBTKBD:
-		ret = hid_hw_output_report(hdev, buf, sizeof(buf));
+		ret = hid_hw_output_report(hdev, buf, 3);
 		break;
 	default:
 		ret = -EINVAL;
 		break;
 	}
 
+	kfree(buf);
+
 	return ret < 0 ? ret : 0; /* BT returns 0, USB returns sizeof(buf) */
 }
 
-- 
2.5.5

Re: [PATCH] HID: lenovo: Don't use stack variables for DMA buffers

[Benjamin Tissoires]

On Mar 28 2016 or thereabouts, Josh Boyer wrote:
> The lenovo_send_cmd_cptkbd function uses a stack variable to submit
> commands via hid_hw_raw_request.  Eventually this gets to the
> usb_hcd_map_urb_for_dma function, which causes a warning to be thrown
> if the CONFIG_DMA_API_DEBUG option is enabled.
> 
> Fix this by allocating a temporary buffer instead.
> 
> Reported-by: lejeczek <peljasz@yahoo.co.uk>
> Signed-off-by: Josh Boyer <jwboyer@fedoraproject.org>
> ---

The patch is
Reviewed-by: Benjamin Tissoires <benjamin.tissoires@redhat.com>

There is just one nitpick that can be solved while committing (or not,
depending on how Jiri handles it :-P ).

>  drivers/hid/hid-lenovo.c | 16 +++++++++++++---
>  1 file changed, 13 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/hid/hid-lenovo.c b/drivers/hid/hid-lenovo.c
> index 0125e356bd8d..2ba5e8ea66a4 100644
> --- a/drivers/hid/hid-lenovo.c
> +++ b/drivers/hid/hid-lenovo.c
> @@ -184,21 +184,31 @@ static int lenovo_send_cmd_cptkbd(struct hid_device *hdev,
>  			unsigned char byte2, unsigned char byte3)
>  {
>  	int ret;
> -	unsigned char buf[] = {0x18, byte2, byte3};
> +	unsigned char *buf = NULL;

No need to initialized buf here :)

Cheers,
Benjamin

> +
> +	buf = kzalloc(3, GFP_KERNEL);
> +	if (!buf)
> +		return -ENOMEM;
> +
> +	buf[0] = 0x18;
> +	buf[1] = byte2;
> +	buf[2] = byte3;
>  
>  	switch (hdev->product) {
>  	case USB_DEVICE_ID_LENOVO_CUSBKBD:
> -		ret = hid_hw_raw_request(hdev, 0x13, buf, sizeof(buf),
> +		ret = hid_hw_raw_request(hdev, 0x13, buf, 3,
>  					HID_FEATURE_REPORT, HID_REQ_SET_REPORT);
>  		break;
>  	case USB_DEVICE_ID_LENOVO_CBTKBD:
> -		ret = hid_hw_output_report(hdev, buf, sizeof(buf));
> +		ret = hid_hw_output_report(hdev, buf, 3);
>  		break;
>  	default:
>  		ret = -EINVAL;
>  		break;
>  	}
>  
> +	kfree(buf);
> +
>  	return ret < 0 ? ret : 0; /* BT returns 0, USB returns sizeof(buf) */
>  }
>  
> -- 
> 2.5.5
> 

Re: [PATCH] HID: lenovo: Don't use stack variables for DMA buffers

[Jiri Kosina]

On Tue, 29 Mar 2016, Benjamin Tissoires wrote:

> > The lenovo_send_cmd_cptkbd function uses a stack variable to submit
> > commands via hid_hw_raw_request.  Eventually this gets to the
> > usb_hcd_map_urb_for_dma function, which causes a warning to be thrown
> > if the CONFIG_DMA_API_DEBUG option is enabled.
> > 
> > Fix this by allocating a temporary buffer instead.
> > 
> > Reported-by: lejeczek <peljasz@yahoo.co.uk>
> > Signed-off-by: Josh Boyer <jwboyer@fedoraproject.org>
> > ---
> 
> The patch is
> Reviewed-by: Benjamin Tissoires <benjamin.tissoires@redhat.com>
> 
> There is just one nitpick that can be solved while committing (or not,
> depending on how Jiri handles it :-P ).

I've nuked the NULL-initialization and applied, thanks!

-- 
Jiri Kosina
SUSE Labs