repo_name
stringlengths
1
62
dataset
stringclasses
1 value
lang
stringclasses
11 values
pr_id
int64
1
20.1k
owner
stringlengths
2
34
reviewer
stringlengths
2
39
diff_hunk
stringlengths
15
262k
code_review_comment
stringlengths
1
99.6k
cheriot-rtos
github_2023
others
264
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,125 @@ +Safe Configuration Management +============================= + +This example shows how dynamic configuration changes can be made to compartments using the ChERIoT features of static sealed capabilities, memory claims, a futex, and a sandbox compartment for validating untrusted data. + +In this model ...
We try to follow a convention in Markdown of one sentence per line. It makes it easier to review diffs because changing one sentence doesn't affect the rest of the paragraph.
cheriot-rtos
github_2023
others
264
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,125 @@ +Safe Configuration Management +============================= + +This example shows how dynamic configuration changes can be made to compartments using the ChERIoT features of static sealed capabilities, memory claims, a futex, and a sandbox compartment for validating untrusted data. + +In this model ...
Note that this still means that subscribers *are* trusted for availability. They can infinite loop in the callback and that will prevent any other notifications being delivered. You could avoid this by exposing a read-only capability to a `uint32_t` that they then did a futex wait on (for C++, using `std::atomic<uint...
cheriot-rtos
github_2023
others
264
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,125 @@ +Safe Configuration Management +============================= + +This example shows how dynamic configuration changes can be made to compartments using the ChERIoT features of static sealed capabilities, memory claims, a futex, and a sandbox compartment for validating untrusted data. + +In this model ...
You probably don't need the error handler in the validator to do anything. If it returns `int`, errors will be reported as `-ECOMPARTMENTFAIL`.
cheriot-rtos
github_2023
others
264
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,125 @@ +Safe Configuration Management +============================= + +This example shows how dynamic configuration changes can be made to compartments using the ChERIoT features of static sealed capabilities, memory claims, a futex, and a sandbox compartment for validating untrusted data. + +In this model ...
Both are equally easy in the auditing framework.
cheriot-rtos
github_2023
cpp
264
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,78 @@ +// Copyright Microsoft and CHERIoT Contributors. +// SPDX-License-Identifier: MIT + +// Contributed by Configured Things Ltd + +#include <compartment.h> +#include <debug.hh> +#include <fail-simulator-on-error.h> +#include <thread.h> + +// Define a sealed capability that gives this compartment +// read...
It would be nice to have slightly more descriptive names for these.
cheriot-rtos
github_2023
cpp
264
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,78 @@ +// Copyright Microsoft and CHERIoT Contributors. +// SPDX-License-Identifier: MIT + +// Contributed by Configured Things Ltd + +#include <compartment.h> +#include <debug.hh> +#include <fail-simulator-on-error.h> +#include <thread.h> + +// Define a sealed capability that gives this compartment +// read...
Since we have a claim on this, it is probably better to use `std::unique_ptr`, which will free on assignment automatically.
cheriot-rtos
github_2023
cpp
258
CHERIoT-Platform
nwf-msr
@@ -0,0 +1,211 @@ +// Copyright Microsoft and CHERIoT Contributors. +// SPDX-License-Identifier: MIT + +#define TEST_NAME "List" +#include "tests.hh" +#include <ds/linked_list.h> + +namespace +{ + using ObjectRing = ds::linked_list::cell::Pointer; + + /** + * Example class we want to link into a doubly linked list. + ...
This bit of the comment might be more naturally associated with the `using ObjectRing` above? That `using` could, I think, move into the `LinkedObject` class were that to be a more natural place for it (the `Sentinel<ObjectRing>` below would have to become `Sentinel<LinkedObject::ObjectRing>`, but otherwise nothing ch...
cheriot-rtos
github_2023
cpp
258
CHERIoT-Platform
nwf-msr
@@ -0,0 +1,211 @@ +// Copyright Microsoft and CHERIoT Contributors. +// SPDX-License-Identifier: MIT + +#define TEST_NAME "List" +#include "tests.hh" +#include <ds/linked_list.h> + +namespace +{ + using ObjectRing = ds::linked_list::cell::Pointer; + + /** + * Example class we want to link into a doubly linked list. + ...
This is all true, but it might skip over some steps. Perhaps something like... `ds::linked_list`-s are circular, doubly-linked collections. While they can stand on their own as rings of objects, it is sometimes convenient to create a designated "sentinel" node that participates in the collection without being part...
cheriot-rtos
github_2023
cpp
258
CHERIoT-Platform
nwf-msr
@@ -0,0 +1,211 @@ +// Copyright Microsoft and CHERIoT Contributors. +// SPDX-License-Identifier: MIT + +#define TEST_NAME "List" +#include "tests.hh" +#include <ds/linked_list.h> + +namespace +{ + using ObjectRing = ds::linked_list::cell::Pointer; + + /** + * Example class we want to link into a doubly linked list. + ...
This is a really good point and is more general than just the sentinel: don't mix heap and stack allocated objects in a `ds::linked_list` in general! All of our `cell` implementations will crash in one way or another if you try.
cheriot-rtos
github_2023
cpp
258
CHERIoT-Platform
nwf-msr
@@ -0,0 +1,211 @@ +// Copyright Microsoft and CHERIoT Contributors. +// SPDX-License-Identifier: MIT + +#define TEST_NAME "List" +#include "tests.hh" +#include <ds/linked_list.h> + +namespace +{ + using ObjectRing = ds::linked_list::cell::Pointer; + + /** + * Example class we want to link into a doubly linked list. + ...
`==` on pointers in CHERI C/C++ presently defaults to address-only equality. You might use `__builtin_cheri_equal_exact`, or wrap with `Capability` first (since its `operator==` uses that primitive), to be more stringent.
cheriot-rtos
github_2023
cpp
258
CHERIoT-Platform
rmn30
@@ -0,0 +1,211 @@ +// Copyright Microsoft and CHERIoT Contributors. +// SPDX-License-Identifier: MIT + +#define TEST_NAME "List" +#include "tests.hh" +#include <ds/linked_list.h> + +namespace +{ + using ObjectRing = ds::linked_list::cell::Pointer; + + /** + * Example class we want to link into a doubly linked list. + ...
```suggestion TEST(!objects.is_empty(), "The list is empty after adding objects."); ```
cheriot-rtos
github_2023
cpp
258
CHERIoT-Platform
nwf-msr
@@ -0,0 +1,211 @@ +// Copyright Microsoft and CHERIoT Contributors. +// SPDX-License-Identifier: MIT + +#define TEST_NAME "List" +#include "tests.hh" +#include <ds/linked_list.h> + +namespace +{ + using ObjectRing = ds::linked_list::cell::Pointer; + + /** + * Example class we want to link into a doubly linked list. + ...
More idiomatically, there's `.search()` on the `Sentinel`, though it's probably not bad to demonstrate the raw form as well.
cheriot-rtos
github_2023
cpp
258
CHERIoT-Platform
rmn30
@@ -0,0 +1,211 @@ +// Copyright Microsoft and CHERIoT Contributors. +// SPDX-License-Identifier: MIT + +#define TEST_NAME "List" +#include "tests.hh" +#include <ds/linked_list.h> + +namespace +{ + using ObjectRing = ds::linked_list::cell::Pointer; + + /** + * Example class we want to link into a doubly linked list. + ...
Aside: would be nice to have a `TEST_EQUAL` to make this pattern less repetitive.
cheriot-rtos
github_2023
cpp
258
CHERIoT-Platform
nwf-msr
@@ -0,0 +1,211 @@ +// Copyright Microsoft and CHERIoT Contributors. +// SPDX-License-Identifier: MIT + +#define TEST_NAME "List" +#include "tests.hh" +#include <ds/linked_list.h> + +namespace +{ + using ObjectRing = ds::linked_list::cell::Pointer; + + /** + * Example class we want to link into a doubly linked list. + ...
Speaking of `Sentinel.search`, `cell::remove` makes the claim that it can be used within if one is careful. It'd be good to both validate that that isn't a stale claim and, if all's good, demonstrate that somewhere in here, too.
cheriot-rtos
github_2023
cpp
258
CHERIoT-Platform
nwf-msr
@@ -0,0 +1,211 @@ +// Copyright Microsoft and CHERIoT Contributors. +// SPDX-License-Identifier: MIT + +#define TEST_NAME "List" +#include "tests.hh" +#include <ds/linked_list.h> + +namespace +{ + using ObjectRing = ds::linked_list::cell::Pointer; + + /** + * Example class we want to link into a doubly linked list. + ...
It might be worth pointing out that `middle` is, after this, still a valid handle to a valid collection of objects that have become detached from the sentinel.
cheriot-rtos
github_2023
cpp
258
CHERIoT-Platform
nwf-msr
@@ -0,0 +1,211 @@ +// Copyright Microsoft and CHERIoT Contributors. +// SPDX-License-Identifier: MIT + +#define TEST_NAME "List" +#include "tests.hh" +#include <ds/linked_list.h> + +namespace +{ + using ObjectRing = ds::linked_list::cell::Pointer; + + /** + * Example class we want to link into a doubly linked list. + ...
It may be worth expanding on this comment. It's typical that `T *remove(T*)` methods return the thing being removed, but `ds::linked_list::cell::remove` breaks that expectation (for reasons that I think are good reasons, but YMMV), which might be surprising to users.
cheriot-rtos
github_2023
cpp
258
CHERIoT-Platform
nwf-msr
@@ -0,0 +1,211 @@ +// Copyright Microsoft and CHERIoT Contributors. +// SPDX-License-Identifier: MIT + +#define TEST_NAME "List" +#include "tests.hh" +#include <ds/linked_list.h> + +namespace +{ + using ObjectRing = ds::linked_list::cell::Pointer; + + /** + * Example class we want to link into a doubly linked list. + ...
This might be a good place to demonstrate some of why the interface is as quirky as it is. You could showcase in-place mutation and traversal without a sentinel with something like (not compile tested, but hopefully close): ``` ds::linked_list::search(middle, [](ObjectRing &cell) { auto l = ds::linked_list::uns...
cheriot-rtos
github_2023
cpp
248
CHERIoT-Platform
davidchisnall
@@ -143,6 +143,30 @@ void __cheri_libcall flaglock_unlock(struct FlagLockState *lock); void __cheri_libcall flaglock_upgrade_for_destruction(struct FlagLockState *lock); +/** + * Return the thread ID of the owner of the lock. + * + * This is only available for priority inherited locks, as this is the only + * case ...
```suggestion // The lock must be held at this point for the value to be stable so do a non-atomic read. return ((*(uint32_t*)&(lock->lockWord)) & 0x0000ffff); ```
cheriot-rtos
github_2023
others
244
CHERIoT-Platform
nwf-msr
@@ -80,9 +80,14 @@ _Z16token_obj_unsealP10SKeyStructP10SObjStruct: bne t0, t1, .Lexit_failure /* Subset bounds to ->data */ + // Get the top into t1 cgetlen t1, ca0 + cgetbase t0, ca0 + add t1, t1, t0 + // Move the address to the start of the data cincoffset ca0, ca0, T...
Removing the `addi` seemed odd on first read, but of course the as-integer access to the `incoffset`-ed `ca0` does the right thing. Maybe something like this would have avoided my confusion. ```suggestion // Subtract the address of the data from the top to give the data length ```
cheriot-rtos
github_2023
cpp
244
CHERIoT-Platform
nwf-msr
@@ -1093,13 +1090,20 @@ namespace Debug::log("Underlying allocation failed for sealed object"); return {nullptr, nullptr}; } + obj.address() = obj.top() - (sz + ObjHdrSize); + // Round down the base to the heap alignment size. + // This ensures that the header is aligned and gives the same alignment + //...
This probably deserves more commentary, something like... we're guaranteed that `obj` is still in-bounds because it is the result of `malloc_internal` for `sz + ObjHdrSize` bytes and so it is at least that big and its base must already be `MallocAlignment`-aligned, so `obj`'s address is at least its base. Moreover,...
cheriot-rtos
github_2023
others
246
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,83 @@ +#pragma once +#include <cdefs.h> +#include <stdint.h> + +/** + * An enum representing each of the Sonata's RGB LEDs. + */ +enum class SonataRgbLed +{ + Led0, + Led1, +}; + +/** + * A driver for the Sonata's RGB LED Controller + */ +struct SonataRgbLedController +{ + /** + * Registers for setting the ...
Should this be `uint32_t ledColours[2]` or similar? You can then use the value of `SonataRgbLed` (0 or 1) to index into them.
cheriot-rtos
github_2023
others
246
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,83 @@ +#pragma once +#include <cdefs.h> +#include <stdint.h> + +/** + * An enum representing each of the Sonata's RGB LEDs. + */ +enum class SonataRgbLed +{ + Led0, + Led1, +}; + +/** + * A driver for the Sonata's RGB LED Controller + */ +struct SonataRgbLedController +{ + /** + * Registers for setting the ...
It would be nice to document the bits here. You've done that properly with the enums below, so just a forward ref saying that these are set with the values from the control enum below would be nice.
cheriot-rtos
github_2023
others
246
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,83 @@ +#pragma once +#include <cdefs.h> +#include <stdint.h> + +/** + * An enum representing each of the Sonata's RGB LEDs. + */ +enum class SonataRgbLed +{ + Led0, + Led1, +}; + +/** + * A driver for the Sonata's RGB LED Controller + */ +struct SonataRgbLedController +{ + /** + * Registers for setting the ...
What does idle mean for an LED? I vaguely remember you or Greg saying that these are surprisingly stateful devices and you're sending them update commands, but a comment to that effect would be nice.
cheriot-rtos
github_2023
others
246
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,83 @@ +#pragma once +#include <cdefs.h> +#include <stdint.h> + +/** + * An enum representing each of the Sonata's RGB LEDs. + */ +enum class SonataRgbLed +{ + Led0, + Led1, +}; + +/** + * A driver for the Sonata's RGB LED Controller + */ +struct SonataRgbLedController +{ + /** + * Registers for setting the ...
```suggestion wait_for_idle(); ledColours[led] = static_cast<uint32_t>(blue) << 16) | (static_cast<uint32_t>(green) << 8) | static_cast<uint32_t>(red); ```
cheriot-rtos
github_2023
others
246
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,83 @@ +#pragma once +#include <cdefs.h> +#include <stdint.h> + +/** + * An enum representing each of the Sonata's RGB LEDs. + */ +enum class SonataRgbLed +{ + Led0, + Led1, +}; + +/** + * A driver for the Sonata's RGB LED Controller + */ +struct SonataRgbLedController +{ + /** + * Registers for setting the ...
From the existence of the `update` method, I presume that this does not actually change the colour? A comment to that effect would be nice.
cheriot-rtos
github_2023
others
246
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,83 @@ +#pragma once +#include <cdefs.h> +#include <stdint.h> + +/** + * An enum representing each of the Sonata's RGB LEDs. + */ +enum class SonataRgbLed +{ + Led0, + Led1, +}; + +/** + * A driver for the Sonata's RGB LED Controller + */ +struct SonataRgbLedController +{ + /** + * Registers for setting the ...
Does this need to be called after `rgb`? What happens if you call it first (LEDs come on in some undefined colour?)
cheriot-rtos
github_2023
others
246
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,83 @@ +#pragma once +#include <cdefs.h> +#include <stdint.h> + +/** + * An enum representing each of the Sonata's RGB LEDs. + */ +enum class SonataRgbLed +{ + Led0, + Led1, +}; + +/** + * A driver for the Sonata's RGB LED Controller + */ +struct SonataRgbLedController +{ + /** + * Registers for setting the ...
Does this turn the LEDs off? Both of them?
cheriot-rtos
github_2023
others
246
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,89 @@ +#pragma once +#include <cdefs.h> +#include <stdint.h> + +/** + * An enum representing each of the Sonata's RGB LEDs. + */ +enum class SonataRgbLed +{ + Led0 = 0, + Led1 = 1, +}; + +/** + * A driver for the Sonata's RGB LED Controller + */ +struct SonataRgbLedController +{ + /** + * Registers for sett...
'Above' isn't great because the 'above' bit isn't visible in the mouse-over pop-up in IDEs (which is the main thing that we write doc comments for). If you point at another enum by name, people can refer to that.
cheriot-rtos
github_2023
cpp
222
CHERIoT-Platform
davidchisnall
@@ -247,6 +247,10 @@ namespace HeapObject(struct SObjStruct *heapCapability, T *allocatedObject) : pointer(allocatedObject, heapCapability) { + if (__builtin_cheri_tag_get(allocatedObject) == 0)
This built in should return a bool, so comparing it against zero is confusing.
cheriot-rtos
github_2023
cpp
222
CHERIoT-Platform
davidchisnall
@@ -247,6 +247,10 @@ namespace HeapObject(struct SObjStruct *heapCapability, T *allocatedObject) : pointer(allocatedObject, heapCapability) { + if (__builtin_cheri_tag_get(allocatedObject) == false)
Why not just ! __builtin_cheri_tag_get(allocatedObject)?
cheriot-rtos
github_2023
cpp
241
CHERIoT-Platform
davidchisnall
@@ -85,6 +85,15 @@ DEFINE_ALLOCATOR_CAPABILITY(__default_malloc_capability, MALLOC_QUOTA) */ #define MALLOC_CAPABILITY STATIC_SEALED_VALUE(__default_malloc_capability) +/** + * Define how long a call to `malloc` and `calloc` can block to fulfil an + * allocation. Regardless of this value, `malloc` and `calloc` wil...
Maybe wrap this in an `#ifndef` so that users can override it? Since our malloc is always-inline, it's safe to make this a per-compilation-unit thing.
cheriot-rtos
github_2023
others
239
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,385 @@ +#pragma once +#include <cdefs.h> +#include <stdint.h> + +static constexpr uint32_t I2cDevClockHz = CPU_TIMER_HZ;
Is there a reason that this and the things below aren't inside the scope of the device class?
cheriot-rtos
github_2023
others
239
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,385 @@ +#pragma once +#include <cdefs.h> +#include <stdint.h> + +static constexpr uint32_t I2cDevClockHz = CPU_TIMER_HZ; + +/** + * Performs a 32-bit integer unsigned division, rounding up. The bottom + * 16 bits of the result are then returned. + * + * As usual, a divisor of 0 is still Undefined Behavior. +...
Missing braces.
cheriot-rtos
github_2023
others
239
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,385 @@ +#pragma once +#include <cdefs.h> +#include <stdint.h> + +static constexpr uint32_t I2cDevClockHz = CPU_TIMER_HZ; + +/** + * Performs a 32-bit integer unsigned division, rounding up. The bottom + * 16 bits of the result are then returned. + * + * As usual, a divisor of 0 is still Undefined Behavior. +...
Please make these proper sentences and doc comments.
cheriot-rtos
github_2023
others
239
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,385 @@ +#pragma once +#include <cdefs.h> +#include <stdint.h> + +static constexpr uint32_t I2cDevClockHz = CPU_TIMER_HZ; + +/** + * Performs a 32-bit integer unsigned division, rounding up. The bottom + * 16 bits of the result are then returned. + * + * As usual, a divisor of 0 is still Undefined Behavior. +...
What is a Spc Thigh? This kind of naming is why we avoid abbreviations in names.
cheriot-rtos
github_2023
others
239
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,385 @@ +#pragma once +#include <cdefs.h> +#include <stdint.h> + +static constexpr uint32_t I2cDevClockHz = CPU_TIMER_HZ; + +/** + * Performs a 32-bit integer unsigned division, rounding up. The bottom + * 16 bits of the result are then returned. + * + * As usual, a divisor of 0 is still Undefined Behavior. +...
We normally name things noun-verb, so that similar things are sorted together when autocompleting or reading generated docs. Ideally avoid starting things with `set` because it ends up with a load of unrelated `set_` things in autocomplete. My preferred style here (which we haven't done uniformly) is to have two ov...
cheriot-rtos
github_2023
others
239
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,385 @@ +#pragma once +#include <cdefs.h> +#include <stdint.h> + +static constexpr uint32_t I2cDevClockHz = CPU_TIMER_HZ; + +/** + * Performs a 32-bit integer unsigned division, rounding up. The bottom + * 16 bits of the result are then returned. + * + * As usual, a divisor of 0 is still Undefined Behavior. +...
Is this doing the same round up thing as your helper?
cheriot-rtos
github_2023
others
239
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,385 @@ +#pragma once +#include <cdefs.h> +#include <stdint.h> + +static constexpr uint32_t I2cDevClockHz = CPU_TIMER_HZ; + +/** + * Performs a 32-bit integer unsigned division, rounding up. The bottom + * 16 bits of the result are then returned. + * + * As usual, a divisor of 0 is still Undefined Behavior. +...
I can't quite parse this comment.
cheriot-rtos
github_2023
others
239
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,385 @@ +#pragma once +#include <cdefs.h> +#include <stdint.h> + +static constexpr uint32_t I2cDevClockHz = CPU_TIMER_HZ; + +/** + * Performs a 32-bit integer unsigned division, rounding up. The bottom + * 16 bits of the result are then returned. + * + * As usual, a divisor of 0 is still Undefined Behavior. +...
There's a lot of code duplication here. Would it be cleaner if you created an array of these and iterated over them? Similarly, if `timing0` to `timing4` were exposed as an array, you'd be able to set them all in the same loop.
cheriot-rtos
github_2023
others
239
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,385 @@ +#pragma once +#include <cdefs.h> +#include <stdint.h> + +static constexpr uint32_t I2cDevClockHz = CPU_TIMER_HZ; + +/** + * Performs a 32-bit integer unsigned division, rounding up. The bottom + * 16 bits of the result are then returned. + * + * As usual, a divisor of 0 is still Undefined Behavior. +...
Predicates should normally include `is` or `are` in their names. This one is particularly bad. We have an explicit style rule to avoid words that are both verbs and nouns and, in this case, format and empty are *both* verbs and nouns and I don't know if this is a thing that formats an empty thing, empties a format, o...
cheriot-rtos
github_2023
others
239
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,385 @@ +#pragma once +#include <cdefs.h> +#include <stdint.h> + +static constexpr uint32_t I2cDevClockHz = CPU_TIMER_HZ; + +/** + * Performs a 32-bit integer unsigned division, rounding up. The bottom + * 16 bits of the result are then returned. + * + * As usual, a divisor of 0 is still Undefined Behavior. +...
Missing braces.
cheriot-rtos
github_2023
others
239
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,385 @@ +#pragma once +#include <cdefs.h> +#include <stdint.h> + +static constexpr uint32_t I2cDevClockHz = CPU_TIMER_HZ; + +/** + * Performs a 32-bit integer unsigned division, rounding up. The bottom + * 16 bits of the result are then returned. + * + * As usual, a divisor of 0 is still Undefined Behavior. +...
Consider `interrupt_is_asserted`. Avoid `const` for value parameters. This could also do with some explanation of how these relate to interrupts in the interrupt controller.
cheriot-rtos
github_2023
others
239
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,385 @@ +#pragma once +#include <cdefs.h> +#include <stdint.h> + +static constexpr uint32_t I2cDevClockHz = CPU_TIMER_HZ; + +/** + * Performs a 32-bit integer unsigned division, rounding up. The bottom + * 16 bits of the result are then returned. + * + * As usual, a divisor of 0 is still Undefined Behavior. +...
Verb-noun again.
cheriot-rtos
github_2023
others
234
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,92 @@ +#pragma once +#include <cdefs.h> +#include <stdint.h> + +/** + * A Simple Driver for the Sonata's GPIO. + * + * Documentation source can be found at: + * https://github.com/lowRISC/sonata-system/blob/1a59633d2515d4fe186a07d53e49ff95c18d9bbf/doc/ip/spi.md + * + * Rendered documentation is served from: ...
Please use full words for variable and function names. Our coding guidelines to avoid abbreviations. You know what intr means but the next person looking at this file or seeing it pop up in autocomplete may not. Same for most of the other fields and methods here.
cheriot-rtos
github_2023
others
234
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,92 @@ +#pragma once +#include <cdefs.h> +#include <stdint.h> + +/** + * A Simple Driver for the Sonata's GPIO. + * + * Documentation source can be found at: + * https://github.com/lowRISC/sonata-system/blob/1a59633d2515d4fe186a07d53e49ff95c18d9bbf/doc/ip/spi.md + * + * Rendered documentation is served from: ...
Please make these doc comments (three slashes).
cheriot-rtos
github_2023
others
234
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,92 @@ +#pragma once +#include <cdefs.h> +#include <stdint.h> + +/** + * A Simple Driver for the Sonata's GPIO. + * + * Documentation source can be found at: + * https://github.com/lowRISC/sonata-system/blob/1a59633d2515d4fe186a07d53e49ff95c18d9bbf/doc/ip/spi.md + * + * Rendered documentation is served from: ...
What are all of these magic numbers?
cheriot-rtos
github_2023
others
234
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,92 @@ +#pragma once +#include <cdefs.h> +#include <stdint.h> + +/** + * A Simple Driver for the Sonata's GPIO. + * + * Documentation source can be found at: + * https://github.com/lowRISC/sonata-system/blob/1a59633d2515d4fe186a07d53e49ff95c18d9bbf/doc/ip/spi.md + * + * Rendered documentation is served from: ...
That’s fine (though please name these ‘transmit’ and ‘receive’ or similar). We probably should also have a concept covering this interface.
cheriot-rtos
github_2023
others
234
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,92 @@ +#pragma once +#include <cdefs.h> +#include <stdint.h> + +/** + * A Simple Driver for the Sonata's GPIO. + * + * Documentation source can be found at: + * https://github.com/lowRISC/sonata-system/blob/1a59633d2515d4fe186a07d53e49ff95c18d9bbf/doc/ip/spi.md + * + * Rendered documentation is served from: ...
What is the behaviour of this if there are not the required number of bytes?
cheriot-rtos
github_2023
others
234
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,92 @@ +#pragma once +#include <cdefs.h> +#include <stdint.h> + +/** + * A Simple Driver for the Sonata's GPIO. + * + * Documentation source can be found at: + * https://github.com/lowRISC/sonata-system/blob/1a59633d2515d4fe186a07d53e49ff95c18d9bbf/doc/ip/spi.md + * + * Rendered documentation is served from: ...
I don’t really like this spinning without providing an opportunity to yield. Is there not an interrupt that this can block on?
cheriot-rtos
github_2023
others
234
CHERIoT-Platform
nwf
@@ -0,0 +1,92 @@ +#pragma once +#include <cdefs.h> +#include <stdint.h> + +/** + * A Simple Driver for the Sonata's GPIO.
Copy-paste-o ```suggestion * A Simple Driver for the Sonata's SPI. ```
cheriot-rtos
github_2023
others
234
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,171 @@ +#pragma once +#include <cdefs.h> +#include <stdint.h> + +/** + * A Simple Driver for the Sonata's SPI. + * + * Documentation source can be found at: + * https://github.com/lowRISC/sonata-system/blob/1a59633d2515d4fe186a07d53e49ff95c18d9bbf/doc/ip/spi.md + * + * Rendered documentation is served from: ...
If the code below can use the symbolic constants then that's fine. I would like the doc comment t explain what the arguments mean.
cheriot-rtos
github_2023
others
234
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,175 @@ +#pragma once +#include <cdefs.h> +#include <stdint.h> + +/** + * A Simple Driver for the Sonata's SPI. + * + * Documentation source can be found at: + * https://github.com/lowRISC/sonata-system/blob/1a59633d2515d4fe186a07d53e49ff95c18d9bbf/doc/ip/spi.md + * + * Rendered documentation is served from: ...
Maybe a `Debug::Assert` that `Len <= 0x7ff`?
cheriot-rtos
github_2023
others
234
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,175 @@ +#pragma once +#include <cdefs.h> +#include <stdint.h> + +/** + * A Simple Driver for the Sonata's SPI. + * + * Documentation source can be found at: + * https://github.com/lowRISC/sonata-system/blob/1a59633d2515d4fe186a07d53e49ff95c18d9bbf/doc/ip/spi.md + * + * Rendered documentation is served from: ...
Minor nit, but we normally use `/** */` blocks for multi-line doc comments.
cheriot-rtos
github_2023
others
234
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,175 @@ +#pragma once +#include <cdefs.h> +#include <stdint.h> + +/** + * A Simple Driver for the Sonata's SPI. + * + * Documentation source can be found at: + * https://github.com/lowRISC/sonata-system/blob/1a59633d2515d4fe186a07d53e49ff95c18d9bbf/doc/ip/spi.md + * + * Rendered documentation is served from: ...
```suggestion /// be modified only whilst the SPI block is idle. ```
cheriot-rtos
github_2023
others
234
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,175 @@ +#pragma once +#include <cdefs.h> +#include <stdint.h> + +/** + * A Simple Driver for the Sonata's SPI. + * + * Documentation source can be found at: + * https://github.com/lowRISC/sonata-system/blob/1a59633d2515d4fe186a07d53e49ff95c18d9bbf/doc/ip/spi.md + * + * Rendered documentation is served from: ...
This took a few attempts for me to parse. Maybe 'Writes to this begin an SPI operation'?
cheriot-rtos
github_2023
others
234
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,175 @@ +#pragma once +#include <cdefs.h> +#include <stdint.h> + +/** + * A Simple Driver for the Sonata's SPI. + * + * Documentation source can be found at: + * https://github.com/lowRISC/sonata-system/blob/1a59633d2515d4fe186a07d53e49ff95c18d9bbf/doc/ip/spi.md + * + * Rendered documentation is served from: ...
```suggestion ConfigurationHalfClockPeriodMask = 0xffu << 0, ```
cheriot-rtos
github_2023
others
234
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,175 @@ +#pragma once +#include <cdefs.h> +#include <stdint.h> + +/** + * A Simple Driver for the Sonata's SPI. + * + * Documentation source can be found at: + * https://github.com/lowRISC/sonata-system/blob/1a59633d2515d4fe186a07d53e49ff95c18d9bbf/doc/ip/spi.md + * + * Rendered documentation is served from: ...
What is CPHA?
cheriot-rtos
github_2023
others
234
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,175 @@ +#pragma once +#include <cdefs.h> +#include <stdint.h> + +/** + * A Simple Driver for the Sonata's SPI. + * + * Documentation source can be found at: + * https://github.com/lowRISC/sonata-system/blob/1a59633d2515d4fe186a07d53e49ff95c18d9bbf/doc/ip/spi.md + * + * Rendered documentation is served from: ...
What is CPOL?
cheriot-rtos
github_2023
others
234
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,175 @@ +#pragma once +#include <cdefs.h> +#include <stdint.h> + +/** + * A Simple Driver for the Sonata's SPI. + * + * Documentation source can be found at: + * https://github.com/lowRISC/sonata-system/blob/1a59633d2515d4fe186a07d53e49ff95c18d9bbf/doc/ip/spi.md + * + * Rendered documentation is served from: ...
```suggestion ControlTransmitClear = 1 << 0, ``` The abbreviations tx and rx are ubiquitous in communication literature, but not everyone reading this code will have that background. Same comment for the next ones down.
cheriot-rtos
github_2023
others
234
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,227 @@ +#pragma once +#include <cdefs.h> +#include <debug.hh> +#include <stdint.h> + +/** + * A Simple Driver for the Sonata's SPI. + * + * Documentation source can be found at: + * https://github.com/lowRISC/sonata-system/blob/1a59633d2515d4fe186a07d53e49ff95c18d9bbf/doc/ip/spi.md + * + * Rendered documenta...
It looks as if we're missing the interrupt definitions in the board JSON?
cheriot-rtos
github_2023
others
234
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,227 @@ +#pragma once +#include <cdefs.h> +#include <debug.hh> +#include <stdint.h> + +/** + * A Simple Driver for the Sonata's SPI. + * + * Documentation source can be found at: + * https://github.com/lowRISC/sonata-system/blob/1a59633d2515d4fe186a07d53e49ff95c18d9bbf/doc/ip/spi.md + * + * Rendered documenta...
```suggestion * receives data. This register can be modified only whilst the SPI block ```
cheriot-rtos
github_2023
others
234
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,227 @@ +#pragma once +#include <cdefs.h> +#include <debug.hh> +#include <stdint.h> + +/** + * A Simple Driver for the Sonata's SPI. + * + * Documentation source can be found at: + * https://github.com/lowRISC/sonata-system/blob/1a59633d2515d4fe186a07d53e49ff95c18d9bbf/doc/ip/spi.md + * + * Rendered documenta...
It would be nice to add a `clang::flag_enum` attribute here.
cheriot-rtos
github_2023
others
234
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,227 @@ +#pragma once +#include <cdefs.h> +#include <debug.hh> +#include <stdint.h> + +/** + * A Simple Driver for the Sonata's SPI. + * + * Documentation source can be found at: + * https://github.com/lowRISC/sonata-system/blob/1a59633d2515d4fe186a07d53e49ff95c18d9bbf/doc/ip/spi.md + * + * Rendered documenta...
Same here (`clang::flag_enum`)
cheriot-rtos
github_2023
others
234
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,227 @@ +#pragma once +#include <cdefs.h> +#include <debug.hh> +#include <stdint.h> + +/** + * A Simple Driver for the Sonata's SPI. + * + * Documentation source can be found at: + * https://github.com/lowRISC/sonata-system/blob/1a59633d2515d4fe186a07d53e49ff95c18d9bbf/doc/ip/spi.md + * + * Rendered documenta...
```suggestion const uint16_t HalfClockPeriod) volatile ``` Can the doc comment explain the parameters too? This is the thing that pops up when you type this in an IDE, so it's nice if it can be self-contained.
cheriot-rtos
github_2023
others
234
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,227 @@ +#pragma once +#include <cdefs.h> +#include <debug.hh> +#include <stdint.h> + +/** + * A Simple Driver for the Sonata's SPI. + * + * Documentation source can be found at: + * https://github.com/lowRISC/sonata-system/blob/1a59633d2515d4fe186a07d53e49ff95c18d9bbf/doc/ip/spi.md + * + * Rendered documenta...
I'm not really a fan of `const` parameters that are value types. `const` on a parameter is a contract not to mutate the object, but when mutation is not observable in the caller then it's not helpful, and making this `const` means that we then have a new `LenTruncated` rather than just having a `length` parameter that...
cheriot-rtos
github_2023
others
234
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,227 @@ +#pragma once +#include <cdefs.h> +#include <debug.hh> +#include <stdint.h> + +/** + * A Simple Driver for the Sonata's SPI. + * + * Documentation source can be found at: + * https://github.com/lowRISC/sonata-system/blob/1a59633d2515d4fe186a07d53e49ff95c18d9bbf/doc/ip/spi.md + * + * Rendered documenta...
Same comment here.
cheriot-rtos
github_2023
others
220
CHERIoT-Platform
davidchisnall
@@ -2,6 +2,15 @@ #include <cdefs.h> #include <stdint.h> +enum class SonataJoystick : uint8_t
It would be nice to have a doc comment for this, especially to tell me if more than one direction can be returned at once (are diagonals represented as two directions? Can it be pressed and moved in a direction)?
cheriot-rtos
github_2023
others
220
CHERIoT-Platform
davidchisnall
@@ -58,4 +67,37 @@ struct SonataGPIO { output = output ^ led_bit(index); } + + /** + * The bit index of the first GPIO pin connected to a user switch. + */ + static constexpr uint32_t FirstSwitch = 5; + /** + * The bit index of the last GPIO pin connected to a user switch. + */ + static constexpr uint32_t La...
Please can you add doc comments on these too?
cheriot-rtos
github_2023
others
224
CHERIoT-Platform
rmn30
@@ -302,6 +314,16 @@ after_zero: // ca1, used for second return value zeroAllRegistersExcept ra, sp, gp, s0, s1, a0, a1 cret + + // If the stack is too small, we don't do the call, but to avoid leaking + // any other state we still go through the same return path as normal. We + // set the return registers to -E...
Why can't we use the macro here?
cheriot-rtos
github_2023
cpp
224
CHERIoT-Platform
rmn30
@@ -104,6 +107,7 @@ void __cheri_compartment("test_runner") run_tests() debug_log("Trying to print string: {}", std::string_view{testString, 13}); run_timed("All tests", []() { + run_timed("Stacks exhaustion in the switcher", test_stack);
I guess this was left in from testing as it also appears below? ```suggestion ```
cheriot-rtos
github_2023
cpp
224
CHERIoT-Platform
rmn30
@@ -72,8 +72,42 @@ namespace debug_log("Expected to invoke the handler? {}", handlerExpected); set_expected_behaviour(&threadStackTestFailed, handlerExpected); } + + __attribute__((used)) extern "C" int test_small_stack() + { + return test_stack_requirement(); + } } // namespace +__attribute__((used)) exter...
I guess this is 2 x spills here, 2 x spills in test_small_stack and 4 x switcher spills? Would be good to be explicit. Are the spills in `test_small_stack` dependent on anything like debug level? Concerned the magic number may be a bit brittle.
cheriot-rtos
github_2023
others
191
CHERIoT-Platform
davidchisnall
@@ -13,6 +13,10 @@ .p2align 2 .type start,@function start: + .rept 32 + .word 0x0000006f + .endr +
CHERIoT doesn't use the vectored interrupts (CHERIoT Ibex doesn't support the vectored mode). Interrupts are configured by the scheduler. If Sunburst is doing something different for the interrupt controller (not the CLINT + PLIC combination that we normally use with CHERIoT Ibex) then you'll need a custom driver.
cheriot-rtos
github_2023
others
191
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,32 @@ +{ + "devices": { + "clint": { + "start": 0x2000000, + "length": 0x10000 + }, + "uart": { + "start": 0x81000000, + "length": 0x1000
This doesn't seem to match the size of the UART device that you've added (should be 52 bytes?)
cheriot-rtos
github_2023
others
191
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,70 @@ +#pragma once +#pragma push_macro("CHERIOT_PLATFORM_CUSTOM_UART") +#define CHERIOT_PLATFORM_CUSTOM_UART +#include_next <platform-uart.hh> +#pragma pop_macro("CHERIOT_PLATFORM_CUSTOM_UART") + +/** + * OpenTitan UART + */ +template<unsigned DefaultBaudRate = 115'200> +class OpenTitanUart +{ + public: + u...
These should all have doc comments explaining what they do (which are for reading, which for writing, which bits are valid in each and what they mean, and so on).
cheriot-rtos
github_2023
others
191
CHERIoT-Platform
davidchisnall
@@ -0,0 +1,70 @@ +#pragma once +#pragma push_macro("CHERIOT_PLATFORM_CUSTOM_UART") +#define CHERIOT_PLATFORM_CUSTOM_UART +#include_next <platform-uart.hh> +#pragma pop_macro("CHERIOT_PLATFORM_CUSTOM_UART") + +/** + * OpenTitan UART + */ +template<unsigned DefaultBaudRate = 115'200> +class OpenTitanUart +{ + public: + u...
I don't understand this comment.
cheriot-rtos
github_2023
cpp
206
CHERIoT-Platform
davidchisnall
@@ -211,6 +262,7 @@ namespace */ void test_ticket_lock_ordering() { + counter = 0;
What is this?
cheriot-rtos
github_2023
cpp
212
CHERIoT-Platform
davidchisnall
@@ -108,3 +108,12 @@ int __cheri_libcall eventgroup_get(struct EventGroup *group, uint32_t *outBits); */ int __cheri_libcall eventgroup_destroy(struct SObjStruct *heapCapability, struct EventGroup *group); + +/** + * Destroy an event group without tacking the lock. + * + * Thi...
Cleanup is a noun, you mean clean up.
cheriot-rtos
github_2023
others
195
CHERIoT-Platform
davidchisnall
@@ -741,6 +741,17 @@ class KunyanEthernet while (transmitControl & 1) {} // Write the frame to the transmit buffer. auto transmitBuffer = transmit_buffer_pointer(); + // We must check the frame pointer and its length. Although it + // is supplied by the firewall which is trusted, the firewall + // does not ...
Please can you add brackets around the < expression?
cheriot-rtos
github_2023
others
199
CHERIoT-Platform
davidchisnall
@@ -9,6 +9,7 @@ DECLARE_AND_DEFINE_INTERRUPT_CAPABILITY(RevokerInterruptEntropy, InterruptName::EthernetReceiveInterrupt, true, false) +#define REVOKER_INTERRUPT_ENTROPY STATIC_SEALED_VALUE(Revoker...
Why a define for a thing that is used only once?
cheriot-rtos
github_2023
others
170
CHERIoT-Platform
davidchisnall
@@ -224,3 +224,119 @@ For Ubuntu, you can do: # apt install xmake ``` +Running on the Arty A7 +---------------------- + +We previously run the test suite in the simulator.
```suggestion We previously ran the test suite in the simulator. ```
cheriot-rtos
github_2023
others
170
CHERIoT-Platform
davidchisnall
@@ -224,3 +224,119 @@ For Ubuntu, you can do: # apt install xmake ``` +Running on the Arty A7 +---------------------- + +We previously run the test suite in the simulator. +Let us now run it on the Arty A7 FPGA development board. + +We will first build the CHERIoT [small and fast FPGA emulator](https://github.com/m...
```suggestion We first need to reconfigure the build, and rebuild. For this example, we'll show rebuilding the test suite, but the same set of steps should work for any CHERIoT RTOS project (try the examples!): ```
cheriot-rtos
github_2023
others
170
CHERIoT-Platform
davidchisnall
@@ -224,3 +224,119 @@ For Ubuntu, you can do: # apt install xmake ``` +Running on the Arty A7 +---------------------- + +We previously run the test suite in the simulator. +Let us now run it on the Arty A7 FPGA development board. + +We will first build the CHERIoT [small and fast FPGA emulator](https://github.com/m...
```suggestion This value may vary if you did not use the dev container and must be the directory containing a `bin` directory with your LLVM build in it. ```
cheriot-rtos
github_2023
others
170
CHERIoT-Platform
davidchisnall
@@ -224,3 +224,119 @@ For Ubuntu, you can do: # apt install xmake ``` +Running on the Arty A7 +---------------------- + +We previously run the test suite in the simulator. +Let us now run it on the Arty A7 FPGA development board. + +We will first build the CHERIoT [small and fast FPGA emulator](https://github.com/m...
```suggestion Make sure that hardware and software flow control are *off*. On macOS, the kernel silently ignores these if they are not supported but on Linux the kernel will refuse to send data unless the flow control is in the correct state. Unfortunately, the hardware flow control lines in the Arty A7's UART are n...
cheriot-rtos
github_2023
others
170
CHERIoT-Platform
davidchisnall
@@ -224,3 +224,119 @@ For Ubuntu, you can do: # apt install xmake ``` +Running on the Arty A7 +---------------------- + +We previously run the test suite in the simulator. +Let us now run it on the Arty A7 FPGA development board. + +We will first build the CHERIoT [small and fast FPGA emulator](https://github.com/m...
```suggestion The "Press CTRL-A Z for help on special keys" message tells you which meta key is configured on your system. Here, the meta key is `CTRL-A`. The meta key varies across systems (the default on macOS is `<ESC>`) and configurations and so we refer to it as `<META>`. ```
cheriot-rtos
github_2023
others
170
CHERIoT-Platform
davidchisnall
@@ -224,3 +224,119 @@ For Ubuntu, you can do: # apt install xmake ``` +Running on the Arty A7 +---------------------- + +We previously run the test suite in the simulator. +Let us now run it on the Arty A7 FPGA development board. + +We will first build the CHERIoT [small and fast FPGA emulator](https://github.com/m...
```suggestion Minicom may block after printing a one or a small number of dots. If it does, then it will resume if you press any key that would be sent over the serial link. Each dot represents 1 KiB of transmitted data. ```
cheriot-rtos
github_2023
cpp
169
CHERIoT-Platform
davidchisnall
@@ -57,6 +46,13 @@ namespace { while (true) { + // Do not attempt to lock if the destruct mode + // flag is set + if ((lockWord & Flag::LockedInDestructMode) != 0)
Possibly move this down. The common case should be that the lock is not held, so we want the fast path to be the CAS. If the lock is in destruction mode then we're happy to be in a slow path.
cheriot-rtos
github_2023
cpp
169
CHERIoT-Platform
davidchisnall
@@ -136,14 +132,33 @@ namespace lockWord.notify_all(); } } + + /** + * Set the destruction bit in the flag lock word and wake + * waiters. Assumes that the lock is hold by the caller.
```suggestion * waiters. Assumes that the lock is held by the caller.```
cheriot-rtos
github_2023
cpp
169
CHERIoT-Platform
davidchisnall
@@ -65,6 +65,81 @@ namespace } } + /** + * Test that destructing a lock automatically wakes up all waiters, + * failing them to acquire the lock. + */ + template<typename Lock> + void test_destruct_lock_wake_up(Lock &lock) + { + modified = false; + + Timeout t{1}; + lock.try_lock(&t); + + // Create an asy...
```suggestion // Try to acquire the lock in a background thread ```
cheriot-rtos
github_2023
cpp
169
CHERIoT-Platform
davidchisnall
@@ -65,6 +65,81 @@ namespace } } + /** + * Test that destructing a lock automatically wakes up all waiters, + * failing them to acquire the lock. + */ + template<typename Lock> + void test_destruct_lock_wake_up(Lock &lock) + { + modified = false; + + Timeout t{1}; + lock.try_lock(&t); + + // Create an asy...
```suggestion // When the lock is upgraded in destruction mode, `lock` will return failure. ```
cheriot-rtos
github_2023
cpp
169
CHERIoT-Platform
davidchisnall
@@ -65,6 +65,81 @@ namespace } } + /** + * Test that destructing a lock automatically wakes up all waiters, + * failing them to acquire the lock. + */ + template<typename Lock> + void test_destruct_lock_wake_up(Lock &lock) + { + modified = false; + + Timeout t{1}; + lock.try_lock(&t); + + // Create an asy...
```suggestion // Make sure that we don't prevent the thread pool making progress if this test fails. Timeout t{100}; TEST(lock.lock() == false, "Lock acquisition should not succeed!"); ```
cheriot-rtos
github_2023
cpp
169
CHERIoT-Platform
davidchisnall
@@ -65,6 +65,81 @@ namespace } } + /** + * Test that destructing a lock automatically wakes up all waiters, + * failing them to acquire the lock. + */ + template<typename Lock> + void test_destruct_lock_wake_up(Lock &lock) + { + modified = false; + + Timeout t{1}; + lock.try_lock(&t); + + // Create an asy...
We should have an assert that's triggered here if this is reached, because this is always API misuse.
cheriot-rtos
github_2023
cpp
169
CHERIoT-Platform
davidchisnall
@@ -63,6 +52,17 @@ namespace { return 0; } + + // Do not attempt to lock if the destruct mode + // flag is set. Note that the lock is in + // destruction mode iff it is already locked,
I don't think that's the reason: we are not setting the locked-for-destruction bit in our expected value, so the CAS will always fail if the lock-for-destruction bit is set. Being locked for destruction is unusual (it should happen exactly once for any lock) and so we don't bother to check for it unless we've failed t...
cheriot-rtos
github_2023
cpp
169
CHERIoT-Platform
davidchisnall
@@ -136,14 +136,33 @@ namespace lockWord.notify_all(); } } + + /** + * Set the destruction bit in the flag lock word and wake + * waiters. Assumes that the lock is held by the caller. + * + * Note: This does not check that the lock is owned by the
It might be nice to have an assert that does check. There's a `Debug::Assert` variant that takes a lambda as the argument and so will not compile the check unless we've turned on lock debugging.
cheriot-rtos
github_2023
cpp
169
CHERIoT-Platform
davidchisnall
@@ -90,8 +90,8 @@ int __cheri_libcall flaglock_trylock(Timeout *timeout, * that thread either releases the lock with `flaglock_unlock` or the timeout * expires. * - * Returns 0 on success, -ETIMEDOUT if the timeout expired, or -EINVAL if the - * arguments are invalid. + * Returns 0 on success, -ETIME...
```suggestion * arguments are invalid, or -ENOENT if the lock is set in destruction mode. * Note: if the object is deallocated while trying to acquire the lock, then this will fault. In many cases, this is called at a compartment boundary and so this is fine. If it is not acceptable, use `heap_claim_fast` to ens...
cheriot-rtos
github_2023
cpp
169
CHERIoT-Platform
davidchisnall
@@ -123,6 +123,11 @@ flaglock_priority_inheriting_lock(struct FlagLockState *lock) */ void __cheri_libcall flaglock_unlock(struct FlagLockState *lock); +/** + * Set a flag lock in destruction mode.
```suggestion * Set a flag lock in destruction mode. * Locks in destruction mode cannot be acquired by other threads. Any threads currently attempting to acquire the lock will wake and fail to acquire the lock. This should be called before deallocating an object that contains a lock. ```
cheriot-rtos
github_2023
others
169
CHERIoT-Platform
davidchisnall
@@ -21,10 +21,9 @@ __clang_ignored_warning_push("-Watomic-alignment"); /** * A simple flag log, wrapping an atomic word used with the `futex` calls. - * Threads blocked on this will be woken in priority order but this does not - * propagate priority and so can lead to priority inversion if a low-priority - * threa...
```suggestion * `IsPriorityInherited` is set, priority is inherited by waiters to avoid ```
cheriot-rtos
github_2023
others
169
CHERIoT-Platform
davidchisnall
@@ -85,6 +84,14 @@ class FlagLockGeneric { flaglock_unlock(&state); } + + /** + * Set the lock in destruction mode.
```suggestion * Set the lock in destruction mode. See the documentation of `flaglock_upgrade_for_destruction` for more information. ```
cheriot-rtos
github_2023
cpp
169
CHERIoT-Platform
davidchisnall
@@ -63,6 +52,19 @@ namespace { return 0; } + + // We are not setting the LockedInDestructMode + // bit in our expected value (`old`), so the + // CAS will always fail if the bit is set. + // Being locked for destruction is unusual (it + // should happen at most once for any lock) and + ...
```suggestion if ((old & Flag::LockedInDestructMode) != 0) ``` We don't need to do another atomic load, the CAS has just told us the value that was there. If it has the destruction bit set, we can abort.
cheriot-rtos
github_2023
cpp
169
CHERIoT-Platform
davidchisnall
@@ -136,14 +138,33 @@ namespace lockWord.notify_all(); } } + + /** + * Set the destruction bit in the flag lock word and wake + * waiters. Assumes that the lock is held by the caller. + * + * Note: This does not check that the lock is owned by the + * calling thread. + */ + void upgrade_for_de...
I think this is correct, but it might be worth a comment that explains the ordering. Importantly: - The `lock` path always adds the waiters bit with a CAS and retries from the top in case of failure. - We always do an atomic or to set the destruct mode bit. - We load to check the waiters bit *after* setting it. ...
cheriot-rtos
github_2023
cpp
169
CHERIoT-Platform
davidchisnall
@@ -65,6 +65,85 @@ namespace } } + /** + * Test that destructing a lock automatically wakes up all waiters, + * failing them to acquire the lock. + */ + template<typename Lock> + void test_destruct_lock_wake_up(Lock &lock) + { + modified = false; + + Timeout t{1}; + lock.try_lock(&t);
Doing `try_lock` and not checking the return is confusing. You probably want this to be `TEST(lock.try_lock(&t), "Failed to acquire uncontended lock");`