- 10 Oct, 2019 40 commits
-
-
Benoit Parrot authored
For each queue we need to maintain resolutions, pixel format, bytesperline, sizeimage, colorspace, etc. Instead of manually adding more entries in the vpe_q_data struct, it is better to just add a "struct v4l2_format" member and use that to store all needed information. Signed-off-by:
Benoit Parrot <bparrot@ti.com> [hverkuil@xs4all.nl: fix checkpatch warning] Signed-off-by:
Hans Verkuil <hverkuil-cisco@xs4all.nl> Signed-off-by:
Mauro Carvalho Chehab <mchehab+samsung@kernel.org>
-
Benoit Parrot authored
VPE uses VPDMA (built-in dma engine) to transfer data to and from the IP and memory. VPDMA expect 32 bits addresses. To make sure that is always the case set the DMA mask and coherent mask for the device. Signed-off-by:
-
Benoit Parrot authored
Using dma_addr_t as the type to hold address inside of a fix sized descriptor used by the vpdma firmware is prone to fail when the expected width is 32 bits and suddenly when CONFIG_LPAE is enabled the data size is now 64 bits shifted the remaining members of the descriptor in memory which confuses the firmware. Signed-off-by:
-
Benoit Parrot authored
v4l2-compliance fails with this message: fail: v4l2-test-buffers.cpp(691): ret == 0 fail: v4l2-test-buffers.cpp(974): captureBufs(node, q, m2m_q, frame_count, true) test MMAP: FAIL This caused the following Kernel Warning: WARNING: CPU: 0 PID: 961 at drivers/media/v4l2-core/videobuf2-core.c:1658 __vb2_queue_cancel+0x174/0x1d8 ... CPU: 0 PID: 961 Comm: v4l2-compliance Not tainted 4.14.62-01720-g20ecd717e87a #6 Hardware name: Generic DRA72X (Flattened Device Tree) Backtrace: [<c020b5bc>] (dump_backtrace) from [<c020b8a0>] (show_stack+0x18/0x1c) r7:00000009 r6:60070013 r5:00000000 r4:c1053824 [<c020b888>] (show_stack) from [<c09232e8>] (dump_stack+0x90/0xa4) [<c0923258>] (dump_stack) from [<c022b740>] (__warn+0xec/0x104) r7:00000009 r6:c0c0ad50 r5:00000000 r4:00000000 [<c022b654>] (__warn) from [<c022b810>] (warn_slowpath_null+0x28/0x30) r9:00000008 r8:00000000 r7:eced4808 r6:edbc9bac r5:eced4844 r4:eced4808 [<c022b7e8>] (warn_slowpath_null) from [<c0726f48>] (__vb2_queue_cancel+0x174/0x1d8) [<c0726dd4>] (__vb2_queue_cancel) from [<c0727648>] (vb2_core_queue_release+0x20/0x40) r10:ecc7bd70 r9:00000008 r8:00000000 r7:edb73010 r6:edbc9bac r5:eced4844 r4:eced4808 r3:00000004 [<c0727628>] (vb2_core_queue_release) from [<c0729528>] (vb2_queue_release+0x10/0x14) r5:edbc9810 r4:eced4800 [<c0729518>] (vb2_queue_release) from [<c0724d08>] (v4l2_m2m_ctx_release+0x1c/0x30) [<c0724cec>] (v4l2_m2m_ctx_release) from [<bf0e8f28>] (vpe_release+0x74/0xb0 [ti_vpe]) r5:edbc9810 r4:ed67a400 [<bf0e8eb4>] (vpe_release [ti_vpe]) from [<c070fccc>] (v4l2_release+0x3c/0x80) r7:edb73010 r6:ed176aa0 r5:edbc9868 r4:ed5119c0 [<c070fc90>] (v4l2_release) from [<c033cf1c>] (__fput+0x8c/0x1dc) r5:ecc7bd70 r4:ed5119c0 [<c033ce90>] (__fput) from [<c033d0cc>] (____fput+0x10/0x14) r10:00000000 r9:ed5119c0 r8:ece392d0 r7:c1059544 r6:ece38d80 r5:ece392b4 r4:00000000 [<c033d0bc>] (____fput) from [<c0246e00>] (task_work_run+0x98/0xb8) [<c0246d68>] (task_work_run) from [<c022f1d8>] (do_exit+0x170/0xa80) r9:ece351fc r8:00000000 r7:ecde3f58 r6:ffffe000 r5:ece351c0 r4:ece38d80 [<c022f068>] (do_exit) from [<c022fb6c>] (do_group_exit+0x48/0xc4) r7:000000f8 [<c022fb24>] (do_group_exit) from [<c022fc00>] (__wake_up_parent+0x0/0x28) r7:000000f8 r6:b6c6a798 r5:00000001 r4:00000001 [<c022fbe8>] (SyS_exit_group) from [<c0207c80>] (ret_fast_syscall+0x0/0x4c) These warnings are caused by buffers which not properly cleaned up/release during an abort use case. In the abort cases the VPDMA desc buffers would still be mapped and the in-flight VB2 buffers would not be released properly causing a kernel warning from being generated by the videobuf2-core level. Signed-off-by:
Tomi Valkeinen <tomi.valkeinen@ti.com> Signed-off-by:
-
Benoit Parrot authored
v4l2-compliance fails with this message: fail: v4l2-test-buffers.cpp(294): \ (int)g_sequence() < seq.last_seq + 1 fail: v4l2-test-buffers.cpp(740): \ buf.check(m2m_q, last_m2m_seq) fail: v4l2-test-buffers.cpp(974): \ captureBufs(node, q, m2m_q, frame_count, true) test MMAP: FAIL The driver is failing to update the source frame sequence number in the vb2 buffer object. Only the destination frame sequence was being updated. This is only a reporting issue if the user space app actually cares about the frame sequence number. But it is fixed nonetheless. Signed-off-by:
-
Benoit Parrot authored
v4l2-compliance fails with this message: fail: v4l2-test-formats.cpp(463): !pfmt.sizeimage fail: v4l2-test-formats.cpp(736): \ Video Capture Multiplanar is valid, \ but TRY_FMT failed to return a format test VIDIOC_TRY_FMT: FAIL This failure is causd by the driver failing to handle out range 'bytesperline' values from user space applications. VPDMA hardware is limited to 64k line stride (16 bytes aligned, so 65520 bytes). So make sure the provided or calculated 'bytesperline' is smaller than the maximum value. Signed-off-by:
-
Benoit Parrot authored
v4l2-compliance fails with this message: fail: v4l2-test-formats.cpp(672): \ Video Capture Multiplanar: TRY_FMT(G_FMT) != G_FMT fail: v4l2-test-formats.cpp(672): \ Video Output Multiplanar: TRY_FMT(G_FMT) != G_FMT ... test VIDIOC_TRY_FMT: FAIL The default pixel format was setup as pointing to a specific offset in the vpe_formats table assuming it was pointing to the V4L2_PIX_FMT_YUYV entry. This became false after the addition on the NV21 format (see above commid-id) So instead of hard-coding an offset which might change over time we need to use a lookup helper instead so we know the default will always be what we intended. Signed-off-by:
-
Benoit Parrot authored
v4l2-compliance warns with this message: warn: v4l2-test-formats.cpp(717): \ TRY_FMT cannot handle an invalid pixelformat. warn: v4l2-test-formats.cpp(718): \ This may or may not be a problem. For more information see: warn: v4l2-test-formats.cpp(719): \ http://www.mail-archive.com/linux-media@vger.kernel.org/msg56550.html ... test VIDIOC_TRY_FMT: FAIL We need to make sure that the returns a valid pixel format in all instance. Based on the v4l2 framework convention drivers must return a valid pixel format when the requested pixel format is either invalid or not supported. Signed-off-by:
-
Benoit Parrot authored
v4l2-compliance fails with this message: warn: v4l2-test-formats.cpp(717): \ TRY_FMT cannot handle an invalid pixelformat. test VIDIOC_TRY_FMT: FAIL This causes the following kernel panic: Unable to handle kernel paging request at virtual address 56595561 pgd = ecd80e00 *pgd=00000000 Internal error: Oops: 205 [#1] PREEMPT SMP ARM ... CPU: 0 PID: 930 Comm: v4l2-compliance Not tainted \ 4.14.62-01715-gc8cd67f49a19 #1 Hardware name: Generic DRA72X (Flattened Device Tree) task: ece44d80 task.stack: ecc6e000 PC is at __vpe_try_fmt+0x18c/0x2a8 [ti_vpe] LR is at 0x8 Because the driver fails to properly check the 'num_planes' values for proper ranges it ends up accessing out of bound data causing the kernel panic. Since this driver only handle single or dual plane pixel format, make sure the provided value does not exceed 2 planes. Signed-off-by:
-
Ram Prasad authored
VPE's max height supported MAX_H is set to 1184 which is the padded height from VC1 decoder output. In case of 90, 270 degree rotated video processing, input to VPE will be 1080x1920, 720x1280 etc and MAX_H needs to be set correct value. Setting MAX_H to 2048 as worst case height. Signed-off-by:
Ram Prasad <x0038811@ti.com> Signed-off-by:
-
Nikhil Devshatwar authored
In NV21 format, the chroma plane is written to memory such that the U and V components are swapped for NV12. Create a new entry in the VPDMA formats to describe the correct data types used in the data descriptors. Update all checks for NV12 and add NV21 there as well. Add support for V4L2_PIX_FMT_NV21 format for both capture and output streams. Signed-off-by:
Nikhil Devshatwar <nikhil.nd@ti.com> Signed-off-by:
-
Nikhil Devshatwar authored
SEQ_BT indicates the buffer for bottom field needs to be processed before the top field. Simplify the field selection logic to support SEQ_BT as well. Modify the interlace flags to include any of alternate, SEQ_TB, SEQ_BT. Update other format error checking to consider SEQ_BT. Replace SEQ_TB with SEQ_XX wherever applicable. Signed-off-by:
-
Benoit Parrot authored
Instead of saving a pointer to the 'fh' member of struct vpe_ctx to later have to use container_of to retrieve the actual pointer to the context structure, which seems to confuse static code analysis tool anyways, just save the pointer to the actual structure and then retrieve it directly. Signed-off-by:
-
Benoit Parrot authored
A few NULL pointer checks were missing. Add check with appropriate return code. Signed-off-by:
-
Benoit Parrot authored
commit 3dc2046c ("[media] media: ti-vpe: vpe: allow use of user specified stride") and commit da4414ea ("[media] media: ti-vpe: vpdma: add support for user specified stride") resulted in the Motion Vector stride to be the same as the image stride. This caused memory corruption in the output image as mentioned in commit 00db9699 ("[media] media: ti-vpe: vpe: Fix line stride for output motion vector"). Fixes: 3dc2046c ("[media] media: ti-vpe: vpe: allow use of user specified stride") Fixes: da4414ea ("[media] media: ti-vpe: vpdma: add support for user specified stride") Signed-off-by:
-
Benoit Parrot authored
When TI CAL was introduce as another driver under platform/ti-vpe adding a second entry into the ti-vpe directory in the platform Makefile caused issues during parallel build. Signed-off-by:
Jyri Sarha <jsarha@ti.com> Signed-off-by:
-
Ricardo Ribalda Delgado authored
According to the product brief, the unit cell size is 1120 nanometers^2. https://www.sony-semicon.co.jp/products_en/IS/sensor1/img/products/ProductBrief_IMX214_20150428.pdfSigned-off-by:
Ricardo Ribalda Delgado <ribalda@kernel.org> Signed-off-by:
-
Ricardo Ribalda Delgado authored
This helper function simplifies the code by not needing a union v4l2_ctrl_ptr and an assignment every time we need to use a ctrl_ptr. Suggested-by:
-
Ricardo Ribalda Delgado authored
New control to pass to userspace the width/height of a pixel. Which is needed for calibration and lens selection. Reviewed-by:
Philipp Zabel <p.zabel@pengutronix.de> Signed-off-by:
-
Ricardo Ribalda Delgado authored
This control returns the unit cell size in nanometres. The struct provides the width and the height in separated fields to take into consideration asymmetric pixels and/or hardware binning. This control is required for automatic calibration of sensors/cameras. Reviewed-by:
-
Ricardo Ribalda Delgado authored
A struct v4l2_area containing the width and the height of a rectangular area. Reviewed-by:
Jacopo Mondi <jacopo@jmondi.org> Reviewed-by:
-
Ricardo Ribalda Delgado authored
This type contains the width and the height of a rectangular area. Reviewed-by:
-
Ricardo Ribalda Delgado authored
Function for initializing compound controls with a default value. Suggested-by:
-
Ricardo Ribalda Delgado authored
Currently compound controls do not have a simple way of initializing its values. This results in ofuscated code with type_ops init. This patch introduces a new field on the control with the default value for the compound control that can be set with the brand new v4l2_ctrl_new_std_compound function Suggested-by:
-
Markus Elfring authored
Simplify these function implementations by using a known wrapper function. This issue was detected by using the Coccinelle software. Signed-off-by:
Markus Elfring <elfring@users.sourceforge.net> Reviewed-by:
Geert Uytterhoeven <geert+renesas@glider.be> Tested-by:
Niklas Söderlund <niklas.soderlund@ragnatech.se> Signed-off-by:
-
Niklas Söderlund authored
The image size is doubled for NV16 and is calculated as bytesperline * height * 2 to accommodate the split of UV data. When writing the offset to hardware, the width is used instead of bytesperline, fix this. Signed-off-by:
Niklas Söderlund <niklas.soderlund+renesas@ragnatech.se> Reviewed-by:
Kieran Bingham <kieran.bingham+renesas@ideasonboard.com> Signed-off-by:
-
Steve Longerbeam authored
If a subdevice is unregistered and then registered again without the driver being removed and re-probed (which will happen when the media device is removed and re-probed without also removing/re-probing the subdevice), media_device_register_entity() is called with a non-zero entity->num_pads, and then the subdevice's .registered callback calls media_entity_pads_init(). Thus the subdevice's pad objects are added to the media device pad list twice, causing list corruption. One way to fix this would be to create media_entity_pads_destroy(), and call it in the subdevice's .unregistered callback. But calling media_entity_pads_init() in the .registered callbacks was done for legacy reasons and is no longer necessary, so move the call to media_entity_pads_init() into the subdevice's probe functions. This fixes the duplicate pad obejcts in the media device pad list. Signed-off-by:
Steve Longerbeam <slongerbeam@gmail.com> Signed-off-by:
-
Steve Longerbeam authored
If the CSI is unregistered and then registered again without the driver being removed and re-probed (which will happen when the media device is removed and re-probed without also removing/re-probing the CSI), the result is the kobject error and backtrace "tried to init an initialized object". This is because the video device is left in an initialized state after being unregistered, thus the video device's underlying kobject is also left in an initialized state when the device is registered again. Fix this by moving imx_media_capture_device_init() and _remove() into csi_registered() and csi_unregistered(). This will create a new un-initialized video device when the CSI is re-registered. Do this for all the subdevices that register a capture device. Reported-by:
Russell King <linux@armlinux.org.uk> Signed-off-by:
-
Niklas Söderlund authored
The rectangle used to correct the compose settings when changing the format was created inside a helper function and not where it was used. This is confusing and makes the code harder to read, fix this. This cleanup is made possible due to refactoring elsewhere and there is no functional change. Signed-off-by:
Laurent Pinchart <laurent.pinchart@ideasonboard.com> Signed-off-by:
-
Niklas Söderlund authored
After recent refactoring the rectangle named crop no longer reflects it usage, to contain the source rectangle. Fix this by renaming it. There is no functional change. Signed-off-by:
-
Wei Yongjun authored
Fix to return a negative error code from the error handling case instead of 0, as done elsewhere in this function. Fixes: f13d5f36 ("media: vimc: Collapse component structure into a single monolithic driver") Signed-off-by:
Wei Yongjun <weiyongjun1@huawei.com> Acked-by:
Helen Koike <helen.koike@collabora.com> Signed-off-by:
-
Alan Stern authored
Visual inspection of the usbvision driver shows that it suffers from three races between its open, close, and disconnect handlers. In particular, the driver is careful to update its usbvision->user and usbvision->remove_pending flags while holding the private mutex, but: usbvision_v4l2_close() and usbvision_radio_close() don't hold the mutex while they check the value of usbvision->remove_pending; usbvision_disconnect() doesn't hold the mutex while checking the value of usbvision->user; and also, usbvision_v4l2_open() and usbvision_radio_open() don't check whether the device has been unplugged before allowing the user to open the device files. Each of these can potentially lead to usbvision_release() being called twice and use-after-free errors. This patch fixes the races by reading the flags while the mutex is still held and checking for pending removes before allowing an open to succeed. Signed-off-by:
Alan Stern <stern@rowland.harvard.edu> CC: <stable@vger.kernel.org> Signed-off-by:
-
Alan Stern authored
The syzbot fuzzer found two invalid-access bugs in the usbvision driver. These bugs occur when userspace keeps the device file open after the device has been disconnected and usbvision_disconnect() has set usbvision->dev to NULL: When the device file is closed, usbvision_radio_close() tries to issue a usb_set_interface() call, passing the NULL pointer as its first argument. If userspace performs a querycap ioctl call, vidioc_querycap() calls usb_make_path() with the same NULL pointer. This patch fixes the problems by making the appropriate tests beforehand. Note that vidioc_querycap() is protected by usbvision->v4l2_lock, acquired in a higher layer of the V4L2 subsystem. Reported-and-tested-by: syzbot+7fa38a608b1075dfd634@syzkaller.appspotmail.com Signed-off-by:
-
Colin Ian King authored
Don't populate the array st6422_bridge_init on the stack but instead make it static. Makes the object code smaller by 231 bytes. Before: text data bss dec hex filename 3419 752 64 4235 108b gspca/stv06xx/stv06xx_st6422.o After: text data bss dec hex filename 3124 816 64 4004 fa4 gspca/stv06xx/stv06xx_st6422.o (gcc version 9.2.1, amd64) Signed-off-by:
Colin Ian King <colin.king@canonical.com> Signed-off-by:
-
Colin Ian King authored
Don't populate the array mapping on the stack but instead make it static. Makes the object code smaller by 79 bytes. Before: text data bss dec hex filename 27572 2096 0 29668 73e4 drivers/media/pci/cx18/cx18-ioctl.o After: text data bss dec hex filename 27429 2160 0 29589 7395 drivers/media/pci/cx18/cx18-ioctl.o (gcc version 9.2.1, amd64) Signed-off-by:
-
YueHaibing authored
Remove duplicated include. Signed-off-by:
YueHaibing <yuehaibing@huawei.com> Signed-off-by:
-
Jeeeun Evans authored
This patch fixes a warning by coccicheck: drivers/staging/media/imx/imx7-mipi-csis.c:973:1-12: WARNING: Use devm_platform_ioremap_resource for state -> regs Use devm_platform_ioremap_resource helper which wraps platform_get_resource() and devm_ioremap_resource() together. Signed-off-by:
Jeeeun Evans <jeeeunevans@gmail.com> Reviewed-by:
Rui Miguel Silva <rmfrfs@gmail.com> Signed-off-by:
-
Christophe JAILLET authored
A call to 'pci_disable_device()' is missing in the error handling path. In some cases, a call to 'free_irq()' may also be missing. Reorder the error handling path, add some new labels and fix the 2 issues mentionned above. This way, the error handling path in more in line with 'cx8800_finidev()' (i.e. the remove function) Signed-off-by:
Christophe JAILLET <christophe.jaillet@wanadoo.fr> Signed-off-by:
-
Christophe JAILLET authored
The 2nd parameter of 'find_first_zero_bit()' is a number of bits, not of bytes. So use 'csi2rx->max_lanes' instead of 'sizeof(lanes_used)'. Fixes: 1fc3b37f ("media: v4l: cadence: Add Cadence MIPI-CSI2 RX driver") Signed-off-by:
-
Navid Emamdoost authored
In sd_start return value of create_singlethread_workqueue needs null check. Signed-off-by:
Navid Emamdoost <navid.emamdoost@gmail.com> Signed-off-by:
-
