hifive1: Upgrade to the revB board #1847
Conversation
|
I got uart working by:
|
Now that I think about it maybe I'm always getting some other non-UART interrupt. I'll dig more into that today. |
|
Yeah that is definitely an interrupt-related issue. |
|
Ah, figured it out. It looks like interrupts aren't correctly re-enabled after being globally disabled. |
7c3dc5b to
2867f8c
Compare
|
The Travis failure is because it's QEMU cache is out of date. |
|
I tried running libtock-c/blink. I get: with this libtock-c diff from the riscv branch: diff --git a/Configuration.mk b/Configuration.mk
index 6a9fee7..786d1e9 100644
--- a/Configuration.mk
+++ b/Configuration.mk
@@ -39,6 +39,7 @@ ifndef ELF2TAB_EXISTS
endif
ELF2TAB_ARGS += -n $(PACKAGE_NAME)
ELF2TAB_ARGS += --stack $(STACK_SIZE) --app-heap $(APP_HEAP_SIZE) --kernel-heap $(KERNEL_HEAP_SIZE)
+ELF2TAB_ARGS += --fixed-address-ram 0x80002400 --fixed-address-flash 0x20040040
# Special extra flag for RISC-V since it does not have PIC support.
ELF2TAB_ARGS += --protected-region-size=64
diff --git a/userland_generic.ld b/userland_generic.ld
index bf69caa..bd12da1 100644
--- a/userland_generic.ld
+++ b/userland_generic.ld
@@ -15,8 +15,8 @@ ENTRY(_start)
* actual location in flash where the app is placed.
*/
MEMORY {
- FLASH (rx) : ORIGIN = 0x4043003c, LENGTH = PROG_LENGTH
- SRAM (RWX) : ORIGIN = 0x80004000, LENGTH = RAM_LENGTH
+ FLASH (rx) : ORIGIN = 0x20040040, LENGTH = PROG_LENGTH
+ SRAM (RWX) : ORIGIN = 0x80002400, LENGTH = RAM_LENGTH
}
SECTIONS {this make command: and this tock diff: diff --git a/boards/hifive1/src/main.rs b/boards/hifive1/src/main.rs
index e38b29c99..268dc7187 100644
--- a/boards/hifive1/src/main.rs
+++ b/boards/hifive1/src/main.rs
@@ -36,7 +36,7 @@ const FAULT_RESPONSE: kernel::procs::FaultResponse = kernel::procs::FaultRespons
// RAM to be shared by all application processes.
#[link_section = ".app_memory"]
-static mut APP_MEMORY: [u8; 5 * 1024] = [0; 5 * 1024];
+static mut APP_MEMORY: [u8; (10* 1024)-3072] = [0; (10 * 1024)-3072];
/// Dummy buffer that causes the linker to reserve enough space for the stack.
#[no_mangle]The fixed-addresses stuff if WIP, but it's just for checking and can be ignored. I also merged pr #1845 into this PR branch. |
|
Thanks @bradjc . Just to clarify are you saying that apps work (with libtock changes)? If so I think this is ready to merge. On the OpenOCD front I have sent patches to OpenOCD, the RISC-V fork of OpenOCD and libjaylink to add support. Once they are merged OpenOCD should be working. |
|
I can't get the apps to fit. Event the blinky or hello_world libtock-rs apps are too big. |
ba32b30 to
2de9723
Compare
|
Got it! I have pushed the working version.
I have tested on libtock-rs and will send the patches once this is merged. All I need is an address change and reduce the app stack size to fit. @bradjc I forced pushed over your readme update. Can you apply it again. Sorry about that. |
|
My status:
If I switch to today's nightly, I get: And if I make a custom syscall formatter I can get the full print out (sometimes): Othertimes it gets stuck here: I'm not sure what is going on there. |
|
Also I seem to get different behavior sometimes, or immediately after flashing the kernel, or after unplugging and replugging the board. |
|
Hmm... It seems like you are getting strange interrupt as well. |
|
Maybe try disabling PMP? See if that helps |
Signed-off-by: Alistair Francis <[email protected]>
Signed-off-by: Alistair Francis <[email protected]>
Signed-off-by: Alistair Francis <[email protected]>
Signed-off-by: Alistair Francis <[email protected]>
Re-write the interrupt handling logic to follow what the Ibex chip does. This is a simpler implemenation and fixes cases where interrupts aren't re-enabled after being disabled. Signed-off-by: Alistair Francis <[email protected]>
This is already turned off in OpenOCD, plus the current way in Tock doesn't work, so let's remove it. Signed-off-by: Alistair Francis <[email protected]>
Signed-off-by: Alistair Francis <[email protected]>
Signed-off-by: Alistair Francis <[email protected]>
The SiFive FE310-G002 chip supports PMP with 8 regions and a 4 byte granularity, enable it for the chip. Signed-off-by: Alistair Francis <[email protected]>
Signed-off-by: Alistair Francis <[email protected]>
Use a fork of QEMU (patches are pending on the mailing list) that supports the hifve1 and OpenTitan. Signed-off-by: Alistair Francis <[email protected]>
afe47d3 to
6f349c4
Compare
|
Draft PR send to @bradjc Two things I noticed.
|
|
It's actually blinking on the board! Using the debug build was the trick. |
|
Awesome!
Then we should be able to merge this |
6f349c4 to
82b16d5
Compare
Increase the app memory size to allow more/bigger apps to fit. Signed-off-by: Alistair Francis <[email protected]>
82b16d5 to
efd0489
Compare
|
Cleared the Travis cache, the build is passing now :) |
|
Should we change the makefile to |
|
I don't think so, the release version works with libtock-rs and it's so space constrained that the debug build makes a substantial difference. |
|
I wonder why libtock-rs would be different. I can't get libtock-rs to work at the moment. Is it easy to trigger a fault? Does the fault message print with the release kernel build? |
|
If you set a stack size that is too large it will fault. The fault messages aren't printed fully in the release build. |
There was a problem hiding this comment.
There's definitely still something wrong that ends up with print formatting breaking and other weird issues, but this is overall a good step towards hifive1b support.
One other note: the 16k RAM is very limiting. This isn't really a suitable Tock board, and hopefully other easily available RISC-V hardware will continue to come out and we can replace this with something.
|
bors r+ |
|
Build failed: |
|
Argh! It looks like Bors has it's own Travis cache. Can that be cleared? |
|
bors retry just did |
Pull Request Overview
This PR updates the current HiFive1 to board in tock to use the revB version instead.
Testing Strategy
Tested by running on a revB board. UART does work, but only when GDB is connected, so something still seems wrong with the UART.
I tried to run an app and it looks like it didn't work.
TODO or Help Wanted
Documentation Updated
/docs, or no updates are required.Formatting
make formatall.