Edward O'Callaghan submitted this change.
5 is the latest approved patch-set.
No files were changed between the latest approved patch-set and the submitted one.
flashrom_tester: Simplify WriteProtectState
Remove the WriteProtectState 'stack' implementation and the push
function. This functionality allowed states to be stacked and then
automatically unrolled via RAII lifetimes. This was useful. However this
unrolling could make errors in a flashrom_tester run much harder to
understand, as the actual failure would be followed by multiple write
protect calls that could subsequently fail, potentially causing panicing
inside the panic handler and the process to be hard aborted and the
restore golden image function would not be run.
The new approach is to prefer code simplicity. Ideally this makes errors
easier to diagnose from logs. To that end the lifetime has been
simplified. The stack has been removed. The mutex has been removed.
This means tests may not be running in the same environment they were
previously. However if they continue to specify their requirements with
set_sw and set_hw there will be no difference and the errors will be
clear.
BUG=b:259494812
BRANCH=None
TEST=flashrom_tester --libflashrom host
Change-Id: I1c4251f69b42a327383b8a99fa933f411feb9568
Signed-off-by: Evan Benn <evanbenn@chromium.org>
Reviewed-on: https://review.coreboot.org/c/flashrom/+/69401
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Edward O'Callaghan <quasisec@chromium.org>
---
M util/flashrom_tester/src/tester.rs
1 file changed, 71 insertions(+), 187 deletions(-)
diff --git a/util/flashrom_tester/src/tester.rs b/util/flashrom_tester/src/tester.rs
index 323b536..c43c7d2 100644
--- a/util/flashrom_tester/src/tester.rs
+++ b/util/flashrom_tester/src/tester.rs
@@ -41,11 +41,9 @@
use serde_json::json;
use std::fs::File;
use std::io::Write;
-use std::mem::MaybeUninit;
use std::path::Path;
use std::path::PathBuf;
use std::sync::atomic::{AtomicBool, Ordering};
-use std::sync::Mutex;
// type-signature comes from the return type of lib.rs workers.
type TestError = Box<dyn std::error::Error>;
@@ -60,7 +58,7 @@
pub cmd: &'a dyn Flashrom,
layout: LayoutSizes,
- pub wp: WriteProtectState<'a, 'static>,
+ pub wp: WriteProtectState<'a>,
/// The path to a file containing the flash contents at test start.
original_flash_contents: PathBuf,
/// The path to a file containing flash-sized random data
@@ -174,64 +172,33 @@
}
}
+struct WriteProtect {
+ hw: bool,
+ sw: bool,
+}
+
/// RAII handle for setting write protect in either hardware or software.
///
/// Given an instance, the state of either write protect can be modified by calling
-/// `set` or `push`. When it goes out of scope, the write protects will be returned
+/// `set`. When it goes out of scope, the write protects will be returned
/// to the state they had then it was created.
-///
-/// The lifetime `'p` on this struct is the parent state it derives from; `'static`
-/// implies it is derived from hardware, while anything else is part of a stack
-/// created by `push`ing states. An initial state is always static, and the stack
-/// forms a lifetime chain `'static -> 'p -> 'p1 -> ... -> 'pn`.
-pub struct WriteProtectState<'a, 'p> {
- /// The parent state this derives from.
- ///
- /// If it's a root (gotten via `from_hardware`), then this is Hardware and the
- /// liveness flag will be reset on drop.
- initial: InitialState<'p>,
- // Tuples are (hardware, software)
- current: (bool, bool),
+pub struct WriteProtectState<'a> {
+ current: WriteProtect,
+ initial: WriteProtect,
cmd: &'a dyn Flashrom,
fc: FlashChip,
}
-enum InitialState<'p> {
- Hardware(bool, bool),
- Previous(&'p WriteProtectState<'p, 'p>),
-}
-
-impl InitialState<'_> {
- fn get_target(&self) -> (bool, bool) {
- match self {
- InitialState::Hardware(hw, sw) => (*hw, *sw),
- InitialState::Previous(s) => s.current,
- }
- }
-}
-
-impl<'a> WriteProtectState<'a, 'static> {
+impl<'a> WriteProtectState<'a> {
/// Initialize a state from the current state of the hardware.
- ///
- /// Panics if there is already a live state derived from hardware. In such a situation the
- /// new state must be derived from the live one, or the live one must be dropped first.
pub fn from_hardware(cmd: &'a dyn Flashrom, fc: FlashChip) -> Result<Self, FlashromError> {
- let mut lock = Self::get_liveness_lock()
- .lock()
- .expect("Somebody panicked during WriteProtectState init from hardware");
- if *lock {
- drop(lock); // Don't poison the lock
- panic!("Attempted to create a new WriteProtectState when one is already live");
- }
-
let hw = Self::get_hw(cmd)?;
let sw = Self::get_sw(cmd)?;
- info!("Initial hardware write protect: HW={} SW={}", hw, sw);
+ info!("Initial write protect state: HW={} SW={}", hw, sw);
- *lock = true;
Ok(WriteProtectState {
- initial: InitialState::Hardware(hw, sw),
- current: (hw, sw),
+ current: WriteProtect { hw, sw },
+ initial: WriteProtect { hw, sw },
cmd,
fc,
})
@@ -251,9 +218,7 @@
let b = cmd.wp_status(true)?;
Ok(b)
}
-}
-impl<'a, 'p> WriteProtectState<'a, 'p> {
/// Return true if the current programmer supports setting the hardware
/// write protect.
///
@@ -262,22 +227,23 @@
self.cmd.can_control_hw_wp()
}
- /// Set the software write protect.
- pub fn set_sw(&mut self, enable: bool) -> Result<&mut Self, FlashromError> {
- info!("request={}, current={}", enable, self.current.1);
- if self.current.1 != enable {
- self.cmd.wp_toggle(/* en= */ enable)?;
- self.current.1 = enable;
+ /// Set the software write protect and check that the state is as expected.
+ pub fn set_sw(&mut self, enable: bool) -> Result<&mut Self, String> {
+ info!("request={}, current={}", enable, self.current.sw);
+ if self.current.sw != enable {
+ self.cmd
+ .wp_toggle(/* en= */ enable)
+ .map_err(|e| e.to_string())?;
}
Ok(self)
}
/// Set the hardware write protect.
pub fn set_hw(&mut self, enable: bool) -> Result<&mut Self, String> {
- if self.current.0 != enable {
- if self.can_control_hw_wp() {
+ if self.can_control_hw_wp() {
+ if self.current.hw != enable {
super::utils::toggle_hw_wp(/* dis= */ !enable)?;
- self.current.0 = enable;
+ self.current.hw = enable;
} else if enable {
info!(
"Ignoring attempt to enable hardware WP with {:?} programmer",
@@ -288,151 +254,35 @@
Ok(self)
}
- /// Stack a new write protect state on top of the current one.
- ///
- /// This is useful if you need to temporarily make a change to write protection:
- ///
- /// ```no_run
- /// # fn main() -> Result<(), Box<dyn std::error::Error>> {
- /// # let cmd: flashrom::FlashromCmd = unimplemented!();
- /// let wp = flashrom_tester::tester::WriteProtectState::from_hardware(&cmd, flashrom::FlashChip::SERVO)?;
- /// {
- /// let mut wp = wp.push();
- /// wp.set_sw(false)?;
- /// // Do something with software write protect disabled
- /// }
- /// // Now software write protect returns to its original state, even if
- /// // set_sw() failed.
- /// # Ok(())
- /// # }
- /// ```
- ///
- /// This returns a new state which restores the original when it is dropped- the new state
- /// refers to the old, so the compiler enforces that states are disposed of in the reverse
- /// order of their creation and correctly restore the original state.
- pub fn push<'p1>(&'p1 self) -> WriteProtectState<'a, 'p1> {
- WriteProtectState {
- initial: InitialState::Previous(self),
- current: self.current,
- cmd: self.cmd,
- fc: self.fc,
- }
- }
-
- fn get_liveness_lock() -> &'static Mutex<bool> {
- static INIT: std::sync::Once = std::sync::Once::new();
- /// Value becomes true when there is a live WriteProtectState derived `from_hardware`,
- /// blocking duplicate initialization.
- ///
- /// This is required because hardware access is not synchronized; it's possible to leave the
- /// hardware in an unintended state by creating a state handle from it, modifying the state,
- /// creating another handle from the hardware then dropping the first handle- then on drop
- /// of the second handle it will restore the state to the modified one rather than the initial.
- ///
- /// This flag ensures that a duplicate root state cannot be created.
- ///
- /// This is a Mutex<bool> rather than AtomicBool because acquiring the flag needs to perform
- /// several operations that may themselves fail- acquisitions must be fully synchronized.
- static mut LIVE_FROM_HARDWARE: MaybeUninit<Mutex<bool>> = MaybeUninit::uninit();
-
- unsafe {
- INIT.call_once(|| {
- LIVE_FROM_HARDWARE.as_mut_ptr().write(Mutex::new(false));
- });
- &*LIVE_FROM_HARDWARE.as_ptr()
- }
- }
-
/// Reset the hardware to what it was when this state was created, reporting errors.
///
/// This behaves exactly like allowing a state to go out of scope, but it can return
/// errors from that process rather than panicking.
pub fn close(mut self) -> Result<(), String> {
- unsafe {
- let out = self.drop_internal();
- // We just ran drop, don't do it again
- std::mem::forget(self);
- out
- }
+ let out = self.drop_internal();
+ // We just ran drop, don't do it again
+ std::mem::forget(self);
+ out
}
- /// Internal Drop impl.
- ///
- /// This is unsafe because it effectively consumes self when clearing the
- /// liveness lock. Callers must be able to guarantee that self will be forgotten
- /// if the state was constructed from hardware in order to uphold the liveness
- /// invariant (that only a single state constructed from hardware exists at any
- /// time).
- unsafe fn drop_internal(&mut self) -> Result<(), String> {
- let lock = match self.initial {
- InitialState::Hardware(_, _) => Some(
- Self::get_liveness_lock()
- .lock()
- .expect("Somebody panicked during WriteProtectState drop from hardware"),
- ),
- _ => None,
- };
- let (hw, sw) = self.initial.get_target();
-
- fn enable_str(enable: bool) -> &'static str {
- if enable {
- "en"
- } else {
- "dis"
- }
- }
-
+ /// Sets both write protects to the state they had when this state was created.
+ fn drop_internal(&mut self) -> Result<(), String> {
// Toggle both protects back to their initial states.
// Software first because we can't change it once hardware is enabled.
- if sw != self.current.1 {
- // Is the hw wp currently enabled?
- if self.current.0 {
- super::utils::toggle_hw_wp(/* dis= */ true).map_err(|e| {
- format!(
- "Failed to {}able hardware write protect: {}",
- enable_str(false),
- e
- )
- })?;
- }
- self.cmd.wp_toggle(/* en= */ sw).map_err(|e| {
- format!(
- "Failed to {}able software write protect: {}",
- enable_str(sw),
- e
- )
- })?;
+ if self.set_sw(self.initial.sw).is_err() {
+ self.set_hw(false)?;
+ self.set_sw(self.initial.sw)?;
}
+ self.set_hw(self.initial.hw)?;
- assert!(
- self.cmd.can_control_hw_wp() || (!self.current.0 && !hw),
- "HW WP must be disabled if it cannot be controlled"
- );
- if hw != self.current.0 {
- super::utils::toggle_hw_wp(/* dis= */ !hw).map_err(|e| {
- format!(
- "Failed to {}able hardware write protect: {}",
- enable_str(hw),
- e
- )
- })?;
- }
-
- if let Some(mut lock) = lock {
- // Initial state was constructed via from_hardware, now we can clear the liveness
- // lock since reset is complete.
- *lock = false;
- }
Ok(())
}
}
-impl<'a, 'p> Drop for WriteProtectState<'a, 'p> {
- /// Sets both write protects to the state they had when this state was created.
- ///
- /// Panics on error because there is no mechanism to report errors in Drop.
+impl<'a> Drop for WriteProtectState<'a> {
fn drop(&mut self) {
- unsafe { self.drop_internal() }.expect("Error while dropping WriteProtectState")
+ self.drop_internal()
+ .expect("Error while dropping WriteProtectState")
}
}
To view, visit change 69401. To unsubscribe, or for help writing mail filters, visit settings.