1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
|
NOTE:
=====
While the driver probes the hardware and reports itself as a
V4L2 driver, there are still some issues preventing it to
stream (at least it doesn't with the standard V4L2 applications.
Didn't test yet with some custom-made app for this driver).
Solving the related bugs and issues preventing it to work is
needed (items 6 and 7 from the list below).
TODO
====
1. The atomisp doesn't rely at the usual i2c stuff to discover the
sensors. Instead, it calls a function from atomisp_gmin_platform.c.
There are some hacks added there for it to wait for sensors to be
probed (with a timeout of 2 seconds or so).
This should be converted to the usual way, using V4L2 async subdev
framework to wait for cameras to be probed;
2. Use ACPI _DSM table - DONE!
3. Switch the driver to use pm_runtime stuff. Right now, it probes the
existing PMIC code and sensors call it directly.
4. There's a problem at the sensor drivers: when trying to set a video
format, the atomisp main driver calls the sensor drivers with the
sensor turned off. This causes them to fail.
The only exception is the atomisp-ov2880, which has a hack inside it
to turn it on when VIDIOC_S_FMT is called.
The right fix seems to power on the sensor when a video device is
opened (or at the first VIDIOC_ ioctl - except for VIDIOC_QUERYCAP),
powering it down at close() syscall.
Such kind of control would need to be done inside the atomisp driver,
not at the sensors code.
5. There are several issues related to memory management, causing
crashes. The atomisp splits the memory management on three separate
regions:
- dynamic pool;
- reserved pool;
- generic pool
The code implementing it is at:
drivers/staging/media/atomisp/pci/hmm/
It also has a separate code for managing DMA buffers at:
drivers/staging/media/atomisp/pci/mmu/
The code there is really dirty, ugly and probably wrong. I fixed
one bug there already, but the best would be to just trash it and use
something else. Maybe the code from the newer intel driver could
serve as a model:
drivers/staging/media/ipu3/ipu3-mmu.c
But converting it to use something like that is painful and may
cause some breakages.
6. There is some issues at the frame receive logic, causing the
DQBUF ioctls to fail.
7. A single AtomISP driver needs to be implemented to support both
Baytrail (BYT) and Cherrytail (CHT) platforms at the same time.
The current driver is a mechanical and hand combined merge of the
two using several runtime macros, plus some ifdef ISP2401 to select the
CHT version. Yet, there are some ISP-specific headers that change the
driver's behavior during compile time.
8. The file structure needs to get tidied up to resemble a normal Linux
driver.
9. Lots of the midlayer glue. unused code and abstraction needs removing.
10. The AtomISP driver includes some special IOCTLS (ATOMISP_IOC_XXXX_XXXX)
and controls that require some cleanup. Some of those code may have
been removed during the cleanups. They could be needed in order to
properly support 3A algorithms
Such IOCTL interface needs more documentation. The better would
be to use something close to the interface used by the IPU3 IMGU driver.
11. The ISP code has some dependencies of the exact FW version.
The version defined in pci/sh_css_firmware.c:
BYT (isp2400): "irci_stable_candrpv_0415_20150521_0458"
CHT (isp2401): "irci_ecr - master_20150911_0724"
Those versions don't seem to be available anymore. On the tests we've
done so far, this version also seems to work for CHT:
"irci_stable_candrpv_0415_20150521_0458"
Which can be obtainable from Yocto Atom ISP respository.
but this was not thoroughly tested.
At some point we may need to round up a few driver versions and see if
there are any specific things that can be done to fold in support for
multiple firmware versions.
12. Switch to standard V4L2 sub-device API for sensor and lens. In
particular, the user space API needs to support V4L2 controls as
defined in the V4L2 spec and references to atomisp must be removed from
these drivers.
13. Use LED flash API for flash LED drivers such as LM3554 (which already
has a LED class driver).
14. Switch from videobuf1 to videobuf2. Videobuf1 is being removed!
15. Correct Coding Style. Please refrain sending coding style patches
for this driver until the other work is done, as there will be a lot
of code churn until this driver becomes functional again.
16. Fix private ioctls to not need a compat_ioctl handler for running
32-bit tasks. The compat code has been removed because of bugs,
and should not be needed for modern drivers. Fixing this properly
unfortunately means an incompatible ABI change.
Limitations
===========
1. To test the patches, you also need the ISP firmware
for BYT: /lib/firmware/shisp_2400b0_v21.bin
for CHT: /lib/firmware/shisp_2401a0_v21.bin
The firmware files will usually be found in /etc/firmware on an Android
device but can also be extracted from the upgrade kit if you've managed
to lose them somehow.
2. Without a 3A library the capture behaviour is not very good. To take a good
picture, you need tune ISP parameters by IOCTL functions or use a 3A library
such as libxcam.
3. The driver is intended to drive the PCI exposed versions of the device.
It will not detect those devices enumerated via ACPI as a field of the
i915 GPU driver.
There are some patches adding i915 GPU support floating at the Yocto's
Aero repository (so far, untested upstream).
4. The driver supports only v2 of the IPU/Camera. It will not work with the
versions of the hardware in other SoCs.
|
