* DMA Bulk IN (ie our TX) results in sequential transfers 33-68% faster.
* DMA Bulk OUT (ie RX) is mostly stripped out due to complete brokenness.
* Interrupt and control endpoints remain PIO-driven.
Other improvements:
1) Use consistent endpoint references (no magic numbers)
2) Greatly enhanced logging
3) DMA support can be compiled out completely
4) Setting lockswitch will disable all DMA operations at runtime
5) Much more robust error checking and recovery
Change-Id: I57b82e655e55ced0dfe289e379b0b61d8fe443b4
swr/swl instructions used for word aligning were wrong. This
made memset() terribly broken. I can't imagine how it went
uncaught for soooo long. Spotted by Solomon Peachy.
I run unit tests for alignments 0,1,2,3
size 1, 2, 3, 4, 5, 63, 64, 65, 127, 128, 129;
and fill pattern 0x00 and other (since 0 is special case in this
implementation).
Change-Id: I513a10734335fe97734c10ab5a6c3e3fb3f4687a
Previously only atomic read/write 8/16/32 were exposed. But it is useful to
be able to read a whole buffer at once, this is more efficient than N times
read8.
Change-Id: I06e331641e1ab1f74c0e16e8c432eafb398e8e6d
Add lua code to check whether ei/di and ext instructions are supported. This
is unclear since xburst is somewhere between mips32r1 and mips32r2. Details
results are below, but in summary: they don't work (ei has no effect, di/ext
cause illegal instruction exceptions)
> ./hwstub_shell -q -b -e 'require("jz/misc"); JZ.misc.enable_sram()' \
-f lua/xburst.lua -e "XBURST.test_ext_inst(0xb32d0000)"
[...]
Selecting soc jz4760b. Redirecting HW to hwstub.soc.jz4760b
data: d7168acf
error: lua/xburst.lua:209: call failed
trapped exception in call
> ./hwstub_shell -q -b -e 'require("jz/misc"); JZ.misc.enable_sram()' \
-f lua/xburst.lua -e "XBURST.test_ei_di_inst(0xb32d0000)"
[...]
Selecting soc jz4760b. Redirecting HW to hwstub.soc.jz4760b
Testing ei
Test SR
Enable interrupts with CP0
SR: 0x1
Disable interrupts with CP0
SR: 0x0
Test ei/di
Enable interrupts with ei
SR: 0x0
Disable interrupts with di
error: lua/xburst.lua:244: call failed
trapped exception in call
Change-Id: I2e162b5dd5e70488bcd8b58f3ca401a3ecab3c4b
Since we can catch exceptions like data aborts on read/write, it takes very
little to also catch exceptions in calls. When extending this with the catching
of illegal instructions, the call instruction now becomes much more robust and
also for address and instruction probing. Since we can catch several types of
exception, rename set_data_abort_jmp to set_exception_jmp. At the same time,
simplify the logic in read/write request handlers. Also fix a bug in ARM
jump code: it was using
stmia r1, {..., pc}
as if pc would get current pc + 8 but this is actually implementation defined
on older ARMs (typically pc + 12) and deprecated on newer ARMs, so rewrite the
code avoid that. The set_exception_jmp() function now also reports the exception
type.
Change-Id: Icd0dd52d2456b361b27c4776be09c3d13528ed93
Now that we now that jz4760b implements EBASE, we can use it to rebase
exceptions to use a k1seg address, that maps to the physical address of the
TCSM0. It requires to enable HAB1 to have this translation. This most the most
inefficient way to access tighly coupled memory ever, but it works.
Change-Id: I894ca929c9835696102eb2fef44b06e6eaf96d44
The JZ misc allows to enable and test SRAM.
The XBurst code uses the coprocessor interface to analyse the cpu. It also
provides a test platform for various features like EBASE and exceptions.
I was able to test and confirm that on jz4760b (thus xburst), EBASE works
(but top 2 bits are not controllable and always 01). The processor claims
to support vector interrupts but this is untested. The values in ConfigX
are not to be trusted blindly, clearly some are wrong. I tried to use the
JZ4780 Config7 "ebase gate" to change bit 30 of EBASE but it does not work,
which suggests that JZ480 uses a newer version of XBurst. Detailled log below:
> ./hwstub_shell -q -f lua/xburst.lua -e "XBURST.init()"
[...]
XBurst:
PRId: 0x2ed0024f
CPU: JZ4760(B)
Config: 0x80000483
Architecture Type: MIPS32
Architecture Level: Release 2 (or more)
MMU Type: Standard TLB
Config1: 0x3e63318a
MMU Size: 32
ICache
Sets per way: 128
Ways: 4
Line size: 32
DCache
Sets per way: 128
Ways: 4
Line size: 32
FPU: no
Config2: 0x80000000
Config3: 0x20
Vectored interrupt: yes
Config7: 0x0
> ./hwstub_shell -q -e 'require("jz/misc"); JZ.misc.enable_sram()' \
-f lua/xburst.lua -e "XBURST.test_ebase(0x80000000);XBURST.test_ebase(0xb32d0000)
[...]
Testing EBASE...
Disable BEV
SR value: 0x2000fc00
EBASE value: 0x80000000
Value after writing 0x80000000: 0x80000000
Value after writing 0x80040000: 0x80040000
Test result: EBase seems to work
Disable config7 gate: write 0x0 to Config7
Value after writing 0xfffff000: 0xbffff000
Enable config7 gate: write 0x80 to Config7
Value after writing 0xc0000000: 0x80000000
Config7 result: Config7 gate does not work
Exception test with EBASE at 0x80000000...
Writing instructions to memory
Old SR: 0x2000fc00
New SR: 0xfc00
EBASE: 80000000
Before: cafebabe
After: deadbeef
Exception result: Exception and EBASE are working
Testing EBASE...
Disable BEV
SR value: 0x2000fc00
EBASE value: 0x80000000
Value after writing 0x80000000: 0x80000000
Value after writing 0x80040000: 0x80040000
Test result: EBase seems to work
Disable config7 gate: write 0x0 to Config7
Value after writing 0xfffff000: 0xbffff000
Enable config7 gate: write 0x80 to Config7
Value after writing 0xc0000000: 0x80000000
Config7 result: Config7 gate does not work
Exception test with EBASE at 0xb32d0000...
Writing instructions to memory
Old SR: 0x2000fc00
New SR: 0xfc00
EBASE: b32d0000
Before: cafebabe
After: deadbeef
Exception result: Exception and EBASE are working
Change-Id: I894227981a141a8c14419b36ed9f519baf145ad1
Apparently I completely forgot to implement it so using hwstub over net would
just fail all EXEC commands :-s
Change-Id: I0d0506cbbce9b86c9a4f19036dacc922d1e51338
This is needed on the jz4760b because if some data is loaded to DRAM, then it
is cached and a disaster lurks if dcaches/icache are not flushed. Targets that
needs this must define CONFIG_FLUSH_CACHES in target-config.h and implement
target_flush_caches(). Currently MIPS has some generic code for mips32r1 that
requires to define {D,I}CACHE_SIZE and {D,I}CACHE_LINE_SIZE in target-config.h
Change-Id: I5a3fc085de9445d8c8a2eb61ae4e2dc9bb6b4e8e
The stub is quite versatile: it can be loaded using bootrom or another other
means (like factory boot on Fiio X1). It relocates itself to TCSM0 and provides
basic functionality (it does not recover from failed read/writes at the moment).
Change-Id: Ib646a4b43fba9358d6f93f0f73a5c2e9bcd775a7
The jz code can do several useful things like dumping the IPL and SPL.
The Fiio code can play with backlight and has code do dump the IPL
and SPL with the correct parameters (extracted by reverse engineering).
Change-Id: I317b3174f5db8d38c9a56670c1d45565142ec208
- hwstub load now properly stops reading the log when the device returns a 0
size buffer instead of STALLing
- add debug output option to hwstub_load
- correctly report transfered size on write error
- add some debug error message in usb code so that some errors can be diagnosed
more easily
- add a batch mode to hwstub_shell to disable the interactive shell
- increase usb control timeout to 1sec, 100ms was really tight
- cap usb buffer size to ~4000 bytes because libusb has a hardwired limit of
4096 bytes for control transfers
Change-Id: Id3200ab99ce70a7a3b09ce7faeaafa4a0fac64c7
The unsigned function were deprecated and hidden begind a compat flag. Since
they will get removed in the next version, I simply reimplemented them, those
are mostly casts anyway.
Change-Id: Ie5db81d7c8589b1e5e9b45a2a4ed41919aa07387
0e2b490 introduced rework of usb driver which was broken. It was reverted
in f2da975 to restore hwstub functionality on ATJ.
This commit reenables usb rework AND fixes remining issues.
The problem was with 0 length OUT thransfers. Additionally
a few cleanups were made.
Change-Id: I529ea9ad6540509e9287ca7e1cd2b44369b03cbb
This reverts commit 0e2b4908d0.
Although I swear it was tested it apparently broke hwstub on atj.
I will need to investigate more whats going on. Revert for now.
Change-Id: I2ff3adf8c72bb0e53be7d81b975382adfb700eab
Also use this opportunity to cleanup support for multiple devices: the shell
now supports dynamic changes in the device and will call init() everytime
a new device is selected, to prepare a new environment. The shell now
honors register width on register read/write. The shell also provides access
to variants as follows by creating a subtable under the register using the
variant type in UPPER case and having the same layout as a register.
For example if register HW.GPIO.DIR has variants "set" and "clr", those can
be used like this:
HW.GPIO.DIR.SET.write(0xff)
HW.GPIO.DIR.CLR.write(0xff00)
Change-Id: I943947fa98bce875de0cba4338e8b7196a4c1165
Rewrite the hwstub library in C++, with a clean and modular design.
The library was designed from the ground up to be aware of multithreading
issues and to handle memory allocation nicely with shared pointers.
Compared to the original library, it brings the following major features:
- support for JZ boot devices, it is very easy to add support for others
- support for network transparent operations (through sockets): both tcp
and unix domains are support
Change-Id: I75899cb9c7aa938c17ede2bb3f468e7a55d625b4
With this you can upload and run code on DSP core in atj213x.
The files can be produced using as2181. You can download
this assembler from https://github.com/wodz/as2181
You should use extended mode (-x switch) since DSP core
in atj is non standard and uses 24bit operands.
PX register has different meaning as well and is used as MSB
when loading other registers with immediates (immediate field
is 16bit in instruction so to set register to 24bit value
you need to store MSB in PX prior to this).
MAC MR is 56bit accordingly.
HIP interface seems to be mapped at standard addresses
(except that regular 218x doesn't have HIP).
Have a fun!
Change-Id: I9a80ca0dd3718ba8435ae8579bfffa66e067e022
This uses slightly hacked luaprompt to provide all the goodis.
See https://github.com/dpapavas/luaprompt for original.
Change-Id: Iedddb79abae5809299322bc215722dd928c35cca
