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
svsm
github_2023
others
243
coconut-svsm
00xc
@@ -1262,6 +1541,12 @@ unsafe impl GlobalAlloc for SvsmAllocator { ret.map_or_else(|_| ptr::null_mut(), |addr| addr.as_mut_ptr::<u8>()) } + /// Deallocates memory based on the specified pointer and layout. + /// + /// # Safety + /// + /// This function assumes that the pointer and layout ...
Same thing here, no need to document this
svsm
github_2023
others
243
coconut-svsm
00xc
@@ -1074,22 +1304,34 @@ impl SlabCommon { } } +/// Represents a slab page for the [`SlabPageSlab`] allocator. #[derive(Debug)] struct SlabPageSlab { common: SlabCommon, } impl SlabPageSlab { + /// Creates a new [`SlabPageSlab`] with a default [`SlabCommon`]. const fn new() -> Self { ...
This would also be a candidate for removing `# Returns` in my opinion, the function signature is quite clear I would say.
svsm
github_2023
others
243
coconut-svsm
00xc
@@ -1324,13 +1611,23 @@ pub struct TestRootMem<'a>(LockGuard<'a, ()>); #[cfg(any(test, fuzzing))] impl TestRootMem<'_> { #[cfg(test_in_svsm)] + /// Sets up a test environment, returning a guard to ensure memory is + /// held for the test's duration. + /// + /// # Returns + /// + /// A guard tha...
These two are the same function with more or less the same idea behind them, so their description should probably be the same or similar enough. It's just that one version of the function requires more setup because it does not run inside the SVSM.
svsm
github_2023
others
243
coconut-svsm
00xc
@@ -77,6 +92,17 @@ pub fn init_memory_map( Ok(()) } +/// Checks if a physical address is considered valid based on the configured +/// memory regions.
Perhaps we should explain what "valid" means (i.e. contained within the configured memory regions).
svsm
github_2023
others
243
coconut-svsm
00xc
@@ -77,6 +92,17 @@ pub fn init_memory_map( Ok(()) } +/// Checks if a physical address is considered valid based on the configured +/// memory regions. +/// +/// # Arguments +/// +/// * `paddr` - The physical address to check for validity. +/// +/// # Returns +/// +/// Returns `true` if the provided physical add...
Perhaps too much boilerplate, I'd say the function signature is pretty clear. A few extra works would be enough in my opinion.
svsm
github_2023
others
243
coconut-svsm
00xc
@@ -93,9 +119,23 @@ pub fn valid_phys_address(paddr: PhysAddr) -> bool { .any(|region| region.contains(paddr)) } +/// The starting address of the ISA range. const ISA_RANGE_START: PhysAddr = PhysAddr::new(0xa0000); + +/// The ending address of the ISA range. const ISA_RANGE_END: PhysAddr = PhysAddr::new(0...
Same here, too much boilerplate in my opinion.
svsm
github_2023
others
258
coconut-svsm
joergroedel
@@ -120,6 +120,10 @@ svsm.bin: stage1/stage1 svsm-test.bin: stage1/stage1-test objcopy -O binary $< $@ +clippy:
`clippy` also needs to be added to the `.PHONY` target.
svsm
github_2023
others
255
coconut-svsm
roy-hopkins
@@ -353,7 +362,10 @@ impl IgvmBuilder { gpa_start: u64, ) -> Result<(), Box<dyn Error>> { let mut gpa = gpa_start; - let mut in_file = File::open(path).expect("Could not open input file"); + let mut in_file = File::open(path).map_err(|e| { + eprintln!("Could not open inpu...
Missing space between in `file{}`.
svsm
github_2023
others
257
coconut-svsm
00xc
@@ -14,6 +14,11 @@ else ifeq ($(V), 2) CARGO_ARGS += -vv endif +ifeq ($(OFFLINE), true) +CARGO_ARGS += --locked --offline +endif
Maybe we could use `ifdef` so it's more consistent with other options (e.g. `RELEASE`).
svsm
github_2023
others
245
coconut-svsm
00xc
@@ -1,40 +1,66 @@ -[package] -name = "svsm" -version = "0.1.0" -edition = "2021" +[workspace] +resolver = "2" +members = [ + # repo tooling + "igvmbuilder", + # binary targets + "kernel", + # fuzzing + "fuzz" +] -[[bin]] -name = "stage2" -path = "src/stage2.rs" -test = false -[[bin]] -name = "svsm...
These will be fixed with #248
svsm
github_2023
others
245
coconut-svsm
stefano-garzarella
@@ -60,7 +60,7 @@ impl CmdOptions { #[derive(Copy, Clone, PartialEq, Eq, PartialOrd, Ord, ValueEnum, Debug)] pub enum Hypervisor { /// Build an IGVM file compatible with QEMU - QEMU, + Qemu,
I'd suggest doing this change in another commit or PR.
svsm
github_2023
others
245
coconut-svsm
stefano-garzarella
@@ -178,7 +178,7 @@ impl SvsmFs { self.root = Some(root.clone()); } - #[cfg(all(any(test, fuzzing), not(test_in_svsm)))] + #[cfg(all(any(test, feature = "fuzzing-hooks"), not(test_in_svsm)))]
Not a strong opinion, but I would have made this change in a separate preparation commit.
svsm
github_2023
others
241
coconut-svsm
stefano-garzarella
@@ -31,10 +31,14 @@ C_BIT_POS ?= 51 STAGE1_OBJS = stage1/stage1.o stage1/reset.o IGVM_FILES = bin/coconut-qemu.igvm bin/coconut-hyperv.igvm IGVMBUILDER = "target/x86_64-unknown-linux-gnu/${TARGET_PATH}/igvmbuilder" +IGVMBIN = bin/igvmbld
Is there a particular reason for renaming it? I mean copying it to bin is okay IMO, but I'd leave it with the same name, so there's no confusion.
svsm
github_2023
others
241
coconut-svsm
stefano-garzarella
@@ -31,10 +31,14 @@ C_BIT_POS ?= 51 STAGE1_OBJS = stage1/stage1.o stage1/reset.o IGVM_FILES = bin/coconut-qemu.igvm bin/coconut-hyperv.igvm IGVMBUILDER = "target/x86_64-unknown-linux-gnu/${TARGET_PATH}/igvmbuilder" +IGVMBIN = bin/igvmbld all: stage1/kernel.elf svsm.bin igvm -igvm: $(IGVM_FILES) +igvm: $(IGVM_FI...
Should we move this line on the target we are adding in this PR (line 40)?
svsm
github_2023
others
241
coconut-svsm
stefano-garzarella
@@ -31,19 +31,26 @@ C_BIT_POS ?= 51 STAGE1_OBJS = stage1/stage1.o stage1/reset.o IGVM_FILES = bin/coconut-qemu.igvm bin/coconut-hyperv.igvm IGVMBUILDER = "target/x86_64-unknown-linux-gnu/${TARGET_PATH}/igvmbuilder" +IGVMBIN = bin/igvmbld all: stage1/kernel.elf svsm.bin igvm -igvm: $(IGVM_FILES) +igvm: $(IGVM_FI...
So we can remove this now, right?
svsm
github_2023
others
241
coconut-svsm
stefano-garzarella
@@ -31,19 +31,26 @@ C_BIT_POS ?= 51 STAGE1_OBJS = stage1/stage1.o stage1/reset.o IGVM_FILES = bin/coconut-qemu.igvm bin/coconut-hyperv.igvm IGVMBUILDER = "target/x86_64-unknown-linux-gnu/${TARGET_PATH}/igvmbuilder" +IGVMBIN = bin/igvmbld all: stage1/kernel.elf svsm.bin igvm -igvm: $(IGVM_FILES) +igvm: $(IGVM_FI...
Is this an intentional change?
svsm
github_2023
others
232
coconut-svsm
00xc
@@ -458,8 +458,13 @@ impl PerCpu { self.reset_ip = reset_ip; } - pub fn ghcb(&mut self) -> &'static mut GHCB { - unsafe { self.ghcb.as_mut().unwrap() } + /// # Safety + /// + /// This routine performs no checks on the validity of the current GHCB + /// and should only be used by ca...
Getting a pointer is not an unsafe operation in Rust. The unsafety is in using the pointer, like creating a reference (`&*ptr`) or dereferencing it (`*ptr`). For example, see slices' `as_mut_ptr()` method in the standard library, which is not marked as unsafe: https://doc.rust-lang.org/std/primitive.slice.html#meth...
svsm
github_2023
others
232
coconut-svsm
00xc
@@ -0,0 +1,38 @@ +// SPDX-License-Identifier: MIT OR Apache-2.0 +// +// Copyright (c) Microsoft Corporation +// +// Author: Jon Lange (jlange@microsoft.com) + +use crate::cpu::percpu::this_cpu_mut; +use crate::sev::ghcb::GHCB; + +use core::ops::{Deref, DerefMut}; + +#[derive(Debug)] +pub struct GHCBRef { + ghcb: *mu...
What is this line doing exactly? `cpu_ptr` is already a `&mut PerCpu`.
svsm
github_2023
others
232
coconut-svsm
joergroedel
@@ -607,12 +622,98 @@ impl PerCpu { } } -pub fn this_cpu() -> &'static PerCpu { - unsafe { SVSM_PERCPU_BASE.as_ptr::<PerCpu>().as_ref().unwrap() } +pub fn get_current_apic_id() -> u32 { + // The APIC ID never changes once a CPU has been constructed, so it is safe + // to reach into the raw CPU storage ...
These assertions need to be warning or error messages for now, at least until we fixed all cases that trigger them. Currently this breaks booting in debug builds.
svsm
github_2023
others
232
coconut-svsm
00xc
@@ -612,6 +612,11 @@ impl PerCpu { } } +pub fn get_current_apic_id() -> u32 { + let cpu = this_cpu(); + (*cpu).get_apic_id() +}
I don't see how this solves anything vs `this_cpu().get_apic_id()`. We are still going through an immutable borrow and the borrow is temporary in both cases. If we *really* want to avoid creating a reference we would need to do a `ptr::read()` from `SVSM_PERCPU_BASE` plus a certain offset.
svsm
github_2023
others
232
coconut-svsm
00xc
@@ -613,16 +624,97 @@ impl PerCpu { } pub fn get_current_apic_id() -> u32 { - let cpu = this_cpu(); - (*cpu).get_apic_id() + // The APIC ID never changes once a CPU has been constructed, so it is safe + // to reach into the raw CPU storage to extract the APIC ID without regard + // to borrow safety. ...
Instead of doing all of this accounting manually perhaps `CpuRef` should be something like a `RefCell<*mut PerCpu>` when `debug_assertions` is enabled.
svsm
github_2023
others
218
coconut-svsm
00xc
@@ -104,6 +104,7 @@ fn set_c_bit(paddr: PhysAddr) -> PhysAddr { bitflags! { #[derive(Copy, Clone, Debug)] pub struct PTEntryFlags: u64 { + const NONE = 0;
The `bitflag` docs explicitly discourage flags with no bits set. Instead we should use `PTEntryFlags::empty()`
svsm
github_2023
others
218
coconut-svsm
roy-hopkins
@@ -421,6 +428,49 @@ pub unsafe fn current_task_terminated() { TASKLIST.lock().terminate(task_node.clone()); } +fn task_pointer(taskptr: TaskPointer) -> *mut Task { + let mut guard = taskptr.task.lock_write(); + let task = guard.as_mut(); + + task as *mut Task +}
This provides mutable access to the task pointer without borrowing or holding a lock. I see it's only used prior to a task switch but the function should probably be marked `unsafe`, perhaps with a `// SAFETY` comment.
svsm
github_2023
others
218
coconut-svsm
roy-hopkins
@@ -421,6 +428,49 @@ pub unsafe fn current_task_terminated() { TASKLIST.lock().terminate(task_node.clone()); } +fn task_pointer(taskptr: TaskPointer) -> *mut Task { + let mut guard = taskptr.task.lock_write(); + let task = guard.as_mut(); + + task as *mut Task +} + +fn switch_to(prev: *mut Task, next: ...
`current` and `next` are `Arc<Task>` aren't they? If so, won't this hold a reference count over the task switch, so the task can only be dropped when we switch back to the current task?
svsm
github_2023
others
218
coconut-svsm
roy-hopkins
@@ -575,17 +573,7 @@ pub mod svsm_gdbstub { Some(t) => { let t = t.task.lock_read(); match t.state { - TaskState::RUNNING => { - if let Some(allocation) = t.allocation { - if this_...
The "Running" and "Stopped" strings here are intended to indicate a different state to the actual task state. Due to how the gdbstub works in SVSM, when the gdbstub state machine is in the 'idle' state then all tasks on the CPU that is processing the gdbstub are stopped. All other tasks on other CPUs are still running ...
svsm
github_2023
others
218
coconut-svsm
00xc
@@ -74,12 +47,11 @@ impl RunQueue { .unwrap_or(self.idle_task.get().unwrap().clone()) } - fn handle_task(&mut self, taskptr: TaskPointer) { - let task = taskptr.task.lock_read(); + fn handle_task(&mut self, task: TaskPointer) { if task.is_running() && !task.is_idle_task() { - ...
We don't need these `task.clone()` calls since `task` is already passed by value.
svsm
github_2023
others
218
coconut-svsm
roy-hopkins
@@ -446,20 +445,16 @@ pub mod svsm_gdbstub { let context = (self.ctx as *const TaskContext).as_ref().unwrap(); *regs = X86_64CoreRegs::from(context); } + } else if let Some(task) = TASKLIST.lock().get_task(tid.get() as u32) { + // The ...
The previous code deliberately assigned the task `Option` to a variable before checking if it was `Some` or `None` as otherwise the `TASKLIST` lock is held until the if scope is exited, causing a deadlock in `switch_to_task()`. This is what was causing the thread list to deadlock as GDB queries the register state of ea...
svsm
github_2023
others
218
coconut-svsm
roy-hopkins
@@ -105,9 +105,8 @@ pub mod svsm_gdbstub { ret_addr: ctx.frame.rip as u64, }; - if let Some(task_node) = this_cpu_mut().runqueue().lock_read().get_task(id) { - task_node.task.lock_write().rsp = &task_ctx as *const TaskContext as u64; - } + // FIXME: current_task()...
I've taken a look at the code and this change doesn't seem to break the switching of threads in gdb. `task.rsp` is only used for reading registers in the non-current thread whereas the disabled code sets `task.rsp` for the current thread only. I think it can safely be removed.
svsm
github_2023
others
218
coconut-svsm
roy-hopkins
@@ -550,7 +545,7 @@ pub mod svsm_gdbstub { let mut cursor = tl.list().front_mut(); while cursor.get().is_some() { - let this_task = cursor.get().unwrap().task.lock_read().id; + let this_task = cursor.get().unwrap().get_task_id(); ...
The idle task for each CPU is not added to `TASKLIST` which means this loop misses all of the idle tasks. Is there any way we can include the idle tasks here?
svsm
github_2023
others
218
coconut-svsm
roy-hopkins
@@ -572,21 +567,11 @@ pub mod svsm_gdbstub { // Get the current task from the stopped CPU so we can mark it as stopped let tl = TASKLIST.lock(); let str = match tl.get_task(tid.get() as u32) { - Some(t) => { - let t = t.task.lock_read(); - ...
Now the idle tasks appear in the task list it has highlighted one last issue where blocked tasks are marked as terminated: ```suggestion Some(task) => { if task.is_running() { "Running".as_bytes() } else if task.is_terminated() { ...
svsm
github_2023
others
71
coconut-svsm
cclaudio
@@ -250,7 +250,7 @@ pub fn parse_fw_meta_data() -> Result<SevFWMetaData, SvsmError> { let ptr = curr.as_ptr::<u16>(); let full_len = ptr.read() as usize; - let len = full_len - mem::size_of::<u16>() + mem::size_of::<Uuid>(); + let len = full_len - (mem::size_of::<u16>() + mem::size_of:...
If we have no control over `full_len`, we may want to use `usize::checked_sub()` to detect overflow.
svsm
github_2023
others
71
coconut-svsm
cclaudio
@@ -223,98 +186,205 @@ const SEV_META_DESC_TYPE_SECRETS: u32 = 2; const SEV_META_DESC_TYPE_CPUID: u32 = 3; const SEV_META_DESC_TYPE_CAA: u32 = 4; +#[derive(Debug, Clone, Copy)] +#[repr(C, packed)] +struct RawMetaHeader { + len: u16, + uuid: [u8; size_of::<Uuid>()], +} + +impl RawMetaHeader { + fn data_len(...
Maybe `s,//,///` ?
svsm
github_2023
others
71
coconut-svsm
cclaudio
@@ -223,98 +186,205 @@ const SEV_META_DESC_TYPE_SECRETS: u32 = 2; const SEV_META_DESC_TYPE_CPUID: u32 = 3; const SEV_META_DESC_TYPE_CAA: u32 = 4; +#[derive(Debug, Clone, Copy)] +#[repr(C, packed)] +struct RawMetaHeader { + len: u16, + uuid: [u8; size_of::<Uuid>()], +} + +impl RawMetaHeader { + fn data_len(...
IIRC, with `packed` the compiler can't do field reordering. So, I am wondering if the header should come before the data? If there is an unexpected issue with the data, I think it would still be safe to access the header (at least for debugging)
svsm
github_2023
others
213
coconut-svsm
00xc
@@ -9,39 +9,74 @@ extern crate alloc; use crate::cpu::vc::VcError; use crate::cpu::vc::VcErrorType; use crate::error::SvsmError; +use core::ops::{Index, IndexMut}; pub const MAX_INSN_SIZE: usize = 15; pub const MAX_INSN_FIELD_SIZE: usize = 3; -#[derive(Default, Debug, Copy, Clone)] -pub struct Instruction { - ...
I think the `Index<usize>` bound is not necessary, as it is always met by any array of size `N`. `Default` is necessary because it is only implemented for certain values of `N` due to some toolchain limitations AFAIK.
svsm
github_2023
others
210
coconut-svsm
joergroedel
@@ -18,6 +21,21 @@ matches_any() { return 1 } +# Check that the body for the given commit is not empty +nonempty_body() { + body=$(git show --no-patch --format="%b" "$1" | sed '/^ *$/d') + trailers=$(git show --no-patch --format="%(trailers:only)" "$1") + + # Remove trailers + for trailer in "${trailers[@]}"; do +...
Instead of removing the trailer, can we just count the number of lines for body and trailers and fail if they are equal?
svsm
github_2023
others
207
coconut-svsm
msft-jlange
@@ -20,6 +20,8 @@ pub struct KernelLaunchInfo { pub kernel_fs_end: u64, pub cpuid_page: u64, pub secrets_page: u64, + pub stage2_igvm_params_phys_addr: u64,
I think it would be cleaner if we avoid adding fields to the kernel launch info. That should be possible by changing `SvsmConfig::invalidate_boot_data()` to return a vector of regions to be invalidated instead of a bool. It could take the kernel launch info as its input and then it could add whatever regions are appr...
svsm
github_2023
others
202
coconut-svsm
joergroedel
@@ -66,6 +94,40 @@ pub fn verify_ghcb_version() { ); } +pub fn hypervisor_ghcb_features() -> GHCBHvFeatures { + *GHCB_HV_FEATURES +} + +pub fn init_hypervisor_ghcb_features() -> Result<(), GhcbMsrError> { + write_msr(SEV_GHCB, GHCBMsr::SNP_HV_FEATURES_REQ); + raw_vmgexit(); + let result = read_msr(S...
This check fails on current KVM because the HV does apparently not report the SEV_SNP_MULTI_VMPL and SEV_PAGE_STATE_CHANGE flags. Can the panic be temporarily removed? We have to fix up KVM first. Please also report the `features` and `required` fields in the error message and format them with `{:#x}`.
svsm
github_2023
c
201
coconut-svsm
roy-hopkins
@@ -130,17 +98,31 @@ DATA_OBJ *insert_data_object(DATA_OBJ *data_object) DATA_OBJ *allocate_data_object(uint64_t address, uint32_t size, uint32_t data_size) { + uint32_t allocation_size; DATA_OBJ *data_object; data_object = malloc(sizeof(DATA_OBJ)); data_object->address = address; data_obje...
This doesn't zero the padding. It is setting the start of the buffer to the first byte of `data_size`. ```suggestion memset(&((uint8_t *)data_object->data)[data_size], 0, allocation_size - data_size); ```
svsm
github_2023
cpp
201
coconut-svsm
roy-hopkins
@@ -0,0 +1,50 @@ +// SPDX-License-Identifier: MIT OR Apache-2.0 +// +// Copyright (c) Microsoft Corporation +// +// Author: Jon Lange (jlange@microsoft.com) +
Missing guard to prevent including more than once.
svsm
github_2023
others
201
coconut-svsm
roy-hopkins
@@ -48,7 +48,11 @@ pub struct IgvmParamBlockFwInfo { /// no firmware is launched after initialization is complete. pub size: u32, - _reserved: u32, + /// Indicates that the initial location of firmware is at the base of + /// memory and will not be loaded into the ROM range. + pub in_low_memory:...
I wonder if it makes more sense to invert this flag and name it something like `in_system_region` (see comment below regarding 'system')? This would then explain the change in behaviour when the firmware uses the ROM region. Or just rename it to `outside_system_region`?
svsm
github_2023
others
201
coconut-svsm
roy-hopkins
@@ -19,6 +19,44 @@ use crate::serial::SERIAL_PORT; use crate::utils::MemoryRegion; use alloc::vec::Vec; +fn check_ovmf_regions( + flash_regions: &[MemoryRegion<PhysAddr>], + kernel_region: &MemoryRegion<PhysAddr>, +) {
Further to our email/call discussion about the name of 'flash' in SVSM, I wonder if this function and parameters should be renamed. I'm not sure what would be a good name for the region from 3GB-4GB. ROM region? System region? ```suggestion fn check_system_region( fw_regions: &[MemoryRegion<PhysAddr>], ke...
svsm
github_2023
others
201
coconut-svsm
roy-hopkins
@@ -93,12 +131,30 @@ impl<'a> SvsmConfig<'a> { } } - pub fn get_fw_regions(&self) -> Result<Vec<MemoryRegion<PhysAddr>>, SvsmError> { + pub fn get_fw_regions( + &self, + kernel_region: &MemoryRegion<PhysAddr>, + ) -> Vec<MemoryRegion<PhysAddr>> { match self { ...
Renaming `in_low_memory` and `check_ovmf_regions()` as per my above comments would result in something like this, which I think conveys the intent more clearly. ```suggestion if igvm_params.fw_in_system_region() { check_system_region(&fw_regions, kernel_region); ...
svsm
github_2023
others
201
coconut-svsm
roy-hopkins
@@ -199,14 +200,26 @@ impl IgvmParams<'_> { } } - pub fn get_fw_regions(&self) -> Result<Vec<MemoryRegion<PhysAddr>>, SvsmError> { - if !self.should_launch_fw() { - Err(Firmware) - } else { - Ok(vec![MemoryRegion::new( - PhysAddr::new(self.igvm_param...
You could use `fw_in_low_memory()` here.
svsm
github_2023
c
201
coconut-svsm
roy-hopkins
@@ -948,25 +923,51 @@ int main(int argc, const char *argv[]) return 1; } - address = (stage2_data->address + stage2_data->size + PAGE_SIZE - 1) & - ~(PAGE_SIZE - 1); - if (address > 0x9F000) + stage2_top = (stage2_data->address + stage2_data->size + PAGE_SIZE - 1) & + ...
This `else` condition can never be executed as it's already in an `if (is_hyperv)`.
svsm
github_2023
others
199
coconut-svsm
msft-jlange
@@ -19,6 +19,13 @@ KERNEL_ELF = "target/x86_64-unknown-none/${TARGET_PATH}/svsm" TEST_KERNEL_ELF = target/x86_64-unknown-none/${TARGET_PATH}/svsm-test FS_FILE ?= none +FW_FILE ?= none +ifneq ($(FW_FILE), none) +BUILD_FW = --firmware ${FW_FILE} --fwtop 100000000
It seems a little unusual to specify a hex address without the 0x prefix, but it's probably not significant enough to merit updating the parsing logic to recognize the number base.
svsm
github_2023
c
199
coconut-svsm
msft-jlange
@@ -688,6 +674,65 @@ int generate_igvm_file(const char *filename) return 0; } +static int check_firmware_options() +{ + FILE *fp; + + if (!fw_filename) + { + // Firmware image is optional. + return 0; + } + + // Get the firmware file size so we can determine the top and bottom addr...
This is going to conflict with Hyper-V requirements. Is there any way to make this check, and the subsequent address validation, specific to QEMU configurations?
svsm
github_2023
c
199
coconut-svsm
msft-jlange
@@ -785,6 +830,39 @@ int parse_options(int argc, const char *argv[]) { is_hyperv = 1; } + else if (0 == strcmp(argv[0], "--firmware")) + { + if (argc == 1) + { + fprintf(stderr, "missing argument for --firmware\n"); + retur...
Should add logic to ensure that --firmware cannot be specified twice.
svsm
github_2023
c
199
coconut-svsm
msft-jlange
@@ -984,6 +1086,24 @@ int main(int argc, const char *argv[]) address += PAGE_SIZE; } + // If a firmware file has been specified then add it and set the relevant + // parameter block entries. + if (fw_filename) + { + igvm_parameter_block->firmware.size = fw_size; + igvm_paramet...
Perhaps gate this behind a -v or --verbose argument?
svsm
github_2023
others
195
coconut-svsm
roy-hopkins
@@ -94,29 +95,64 @@ pub fn init_page_table(launch_info: &KernelLaunchInfo, kernel_elf: &elf::Elf64Fi set_init_pgtable(pgtable); } -pub fn invalidate_stage2(config: &SvsmConfig) -> Result<(), SvsmError> { - let pstart = PhysAddr::null(); - let pend = pstart + (640 * 1024); - let mut paddr = pstart; - - ...
This calls `invalidate_boot_memory_region()` even if the filesystem range is empty. If the config specifies that a page state change is required then this will result in a call to `ghcb().page_state_change()`. Maybe this should only call `invalidate_boot_memory_region()` if `kernel_fs_size` is > 0?
svsm
github_2023
others
189
coconut-svsm
joergroedel
@@ -114,18 +124,5 @@ pub fn request_loop() { // Write back results params.write_back(vmsa); - - // Make VMSA runable again by setting EFER.SVME - vmsa.enable(); - - flush_tlb_global_sync(); - - // Check if mappings still valid - if update_mappings().is_ok() {
I don't see the call to update_mappings() moved, why is it safe to remove it? Also, I think the commit message should explain the reasoning behind this change.
svsm
github_2023
others
177
coconut-svsm
joergroedel
@@ -171,6 +186,7 @@ impl<T: BitmapAllocator + Debug> BitmapAllocator for BitmapAllocatorTree<T> { fn alloc_aligned(ba: &mut impl BitmapAllocator, entries: usize, align: usize) -> Option<usize> { // Iterate through the bitmap checking on each alignment boundary // for a free range of the requested size + a...
I am wondering whether this should be an assert or a `return None`?
svsm
github_2023
others
183
coconut-svsm
stefano-garzarella
@@ -219,13 +221,14 @@ impl VirtualMapping for VMFileMapping { write: bool, ) -> Result<VMPageFaultResolution, SvsmError> { let page_size = self.page_size(); + let page_size_bytes = usize::from(page_size); if write { if let Some(write_copy) = self.write_copy.as_mut() {...
What about adding `PageSize::mask()`? I mean something like this: ```diff diff --git a/src/mm/vm/mapping/file_mapping.rs b/src/mm/vm/mapping/file_mapping.rs index 51540aa..a687442 100644 --- a/src/mm/vm/mapping/file_mapping.rs +++ b/src/mm/vm/mapping/file_mapping.rs @@ -222,7 +222,6 @@ impl VirtualMapping for VM...
svsm
github_2023
others
182
coconut-svsm
stefano-garzarella
@@ -22,8 +22,22 @@ FS_FILE ?= none C_BIT_POS ?= 51 STAGE1_OBJS = stage1/stage1.o stage1/reset.o +IGVM_FILES = bin/coconut-qemu.igvm bin/coconut-hyperv.igvm
we have some binaries in the root (e.g. `svsm.bin`) others in `stage1/`, what about place those new binaries in `igvm/`? This is not a strong opinion, I'm fine also with `bin`, but I'd move all the binaries there, but we can do in another PR ;-)
svsm
github_2023
cpp
182
coconut-svsm
roy-hopkins
@@ -0,0 +1,86 @@ +// SPDX-License-Identifier: MIT OR Apache-2.0 +// +// Copyright (c) Microsoft Corporation +// +// Author: Jon Lange (jlange@microsoft.com) + +typedef enum { + IGVM_VHT_SUPPORTED_PLATFORM = 0x1, + IGVM_VHT_PARAMETER_AREA = 0x301, + IGVM_VHT_PAGE_DATA = 0x302, + IGVM_VHT_PARAMETER_INSERT = 0...
It would be nice if we could use the header file generated as part of this PR: https://github.com/microsoft/igvm/pull/3 Or, even better, depend on the C library. Something to consider when that PR gets merged. But maybe it's not important if this is going to be replace with a rust version soon.
svsm
github_2023
c
182
coconut-svsm
roy-hopkins
@@ -0,0 +1,892 @@ +// SPDX-License-Identifier: MIT OR Apache-2.0 +// +// Copyright (c) Microsoft Corporation +// +// Author: Jon Lange (jlange@microsoft.com) +// +// This module is provided in C as a bootstrapper to permit generation of IGVM +// files pending creation of a Rust version. The C version should be +// mai...
Shouldn't this be 0 for SNP, being the highest privilege level activated on the guest?
svsm
github_2023
others
178
coconut-svsm
joergroedel
@@ -663,17 +663,39 @@ impl MemoryRegion { self.write_page_info(i, pg); } - self.nr_pages[0] = self.page_count - meta_pages; - /* Mark all pages as allocated */ for i in meta_pages..self.page_count { let pg = PageInfo::Allocated(AllocatedInfo { order: 0 }); ...
This is a good optimization, thanks. I think there is one thing missing, which is the initializations of the PageInfos for the compound pages that are freed via `free_page_order()`. The allocator manages one PageInfo per PAGE_SIZE chunk of memory. This means that for pages with `order > 0` there are multiple `PageInfo...
svsm
github_2023
others
178
coconut-svsm
joergroedel
@@ -663,17 +665,40 @@ impl MemoryRegion { self.write_page_info(i, pg); } - self.nr_pages[0] = self.page_count - meta_pages; - /* Mark all pages as allocated */ for i in meta_pages..self.page_count { let pg = PageInfo::Allocated(AllocatedInfo { order: 0 }); ...
Thanks, looks good now.
svsm
github_2023
others
181
coconut-svsm
msft-jlange
@@ -473,8 +470,7 @@ pub extern "C" fn svsm_main() { if let Err(e) = copy_tables_to_fw(fw_meta) {
Your PR moves the location of the firmware to the IGVM file (which is good) but there is still a hardcoded assumption that the firmware wants CPUID/secrets information in a particular place (see map_and_parse_fw_meta). Shouldn't this be abstracted in the IGVM as well so that the SVSM is not making assumptions about OV...
svsm
github_2023
others
181
coconut-svsm
00xc
@@ -217,7 +217,7 @@ impl<'a> FwCfg<'a> { // This needs to be &mut self to prevent iterator invalidation, where the caller // could do fw_cfg.select() while iterating. Having a mutable reference prevents // other references. - pub fn iter_flash_regions(&mut self) -> impl Iterator<Item = MemoryRegion<Ph...
This change ignores the comment above the function. Not taking `&mut self` allows potential misbehavior. The comment should be updated or the change reverted.
svsm
github_2023
others
180
coconut-svsm
stefano-garzarella
@@ -334,6 +330,10 @@ pub extern "C" fn svsm_start(li: &KernelLaunchInfo, vb_addr: usize) { load_gdt(); early_idt_init(); + // Capture the debug serial port befure the launch info disappears from
s/befure/before
svsm
github_2023
others
169
coconut-svsm
stefano-garzarella
@@ -148,7 +148,12 @@ impl ValidBitmap { /// The number of u64's in the bitmap fn bitmap_len(&self) -> usize { let num_pages = self.region.len() / PAGE_SIZE; - num_pages.div_ceil(u64::BITS as usize) + let additional_u64 = if self.region.len() % PAGE_SIZE != 0 { + 1 + } ...
Not a strong opinion, but eventually we can avoid `additional_u64` and do something like this: ```suggestion if self.region.len() % PAGE_SIZE != 0 { num_pages + 1 } else { num_pages } ```
svsm
github_2023
others
176
coconut-svsm
roy-hopkins
@@ -51,6 +51,10 @@ startup_32: movl $STAGE2_START, %esp /* Write startup information to stage2 stack */ + xorl %eax, eax
The second eax should be `%eax`.
svsm
github_2023
others
172
coconut-svsm
00xc
@@ -102,3 +102,25 @@ pub fn writable_phys_addr(paddr: PhysAddr) -> bool { valid_phys_address(paddr) } + +#[cfg(test)] +#[cfg_attr(test_in_svsm, ignore = "Offline testing")]
Please remove this. Everything else looks good. ``` warning: `#[ignore]` only has an effect on functions --> src/mm/memory.rs:107:26 | 107 | #[cfg_attr(test_in_svsm, ignore = "Offline testing")] | ^^^^^^^^^^^^^^^^^^^^^^^^^^ | = note: `#[warn(unused_attributes)]` on ...
svsm
github_2023
others
168
coconut-svsm
00xc
@@ -449,31 +449,43 @@ pub extern "C" fn svsm_main() { start_secondary_cpus(&cpus); - let fw_meta = map_and_parse_fw_meta() - .unwrap_or_else(|e| panic!("Failed to parse FW SEV meta-data: {:#?}", e)); + let fw_metadata: Option<SevFWMetaData>; + if config.should_launch_fw() { + fw_metadata...
The `fw_metadata` can be constructed directly: ```suggestion let fw_metadata = if config.should_launch_fw() { Some( map_and_parse_fw_meta() .unwrap_or_else(|e| panic!("Failed to parse FW SEV meta-data: {:#?}", e)), ) } else { None }; ``` Or ...
svsm
github_2023
others
170
coconut-svsm
00xc
@@ -99,3 +99,23 @@ pub fn writable_phys_addr(paddr: PhysAddr) -> bool { valid_phys_address(paddr) } + +#[cfg(test)] +mod tests { + use super::*; + use crate::address::PhysAddr; + + #[test] + fn test_valid_phys_address() { + let start = PhysAddr::new(0x1000); + let end = PhysAddr::new(0...
Please remove the region again from `MEMORY_MAP` so this does not have a chance to interfere with something else when running inside the SVSM.
svsm
github_2023
others
170
coconut-svsm
00xc
@@ -222,3 +222,31 @@ impl<T: Sized + Copy> GuestPtr<T> { unsafe { GuestPtr::from_ptr(self.ptr.offset(count)) } } } + +mod tests { + + #[test] + fn test_read_u8_valid_address() {
This ends up using inline assembly, so Miri cannot handle it. ```suggestion #[test] #[cfg_attr(miri, ignore)] fn test_read_u8_valid_address() { ```
svsm
github_2023
others
170
coconut-svsm
00xc
@@ -222,3 +222,31 @@ impl<T: Sized + Copy> GuestPtr<T> { unsafe { GuestPtr::from_ptr(self.ptr.offset(count)) } } } + +mod tests { + + #[test] + fn test_read_u8_valid_address() { + use crate::mm::guestmem::*; + // Create a region to read from + let test_buffer: [u8; 6] = [0; 6];...
Nit: instead of using "as" casting, you can simply do: ```suggestion let test_address = VirtAddr::from(test_buffer.as_ptr()); let result = read_u8(test_address).unwrap(); ```
svsm
github_2023
others
170
coconut-svsm
00xc
@@ -222,3 +222,31 @@ impl<T: Sized + Copy> GuestPtr<T> { unsafe { GuestPtr::from_ptr(self.ptr.offset(count)) } } } + +mod tests { + + #[test] + fn test_read_u8_valid_address() { + use crate::mm::guestmem::*; + // Create a region to read from + let test_buffer: [u8; 6] = [0; 6];...
```suggestion #[test] #[cfg_attr(miri, ignore)] fn test_write_u8_valid_address() { ```
svsm
github_2023
others
170
coconut-svsm
00xc
@@ -222,3 +222,31 @@ impl<T: Sized + Copy> GuestPtr<T> { unsafe { GuestPtr::from_ptr(self.ptr.offset(count)) } } } + +mod tests { + + #[test] + fn test_read_u8_valid_address() { + use crate::mm::guestmem::*; + // Create a region to read from + let test_buffer: [u8; 6] = [0; 6];...
Nit: same here, instead of 'as' casting: ```suggestion let test_address = VirtAddr::from(test_buffer.as_mut_ptr()); let data_to_write = 0x42; write_u8(test_address, data_to_write).unwrap(); ```
svsm
github_2023
others
170
coconut-svsm
00xc
@@ -153,3 +153,67 @@ pub const SVSM_PERTASK_END: VirtAddr = SVSM_PERTASK_BASE.const_add(SIZE_LEVEL3); /// Kernel stack for a task pub const SVSM_PERTASK_STACK_BASE: VirtAddr = SVSM_PERTASK_BASE; + +#[cfg(test)] +mod tests { + use super::*; + + static KERNEL_MAPPING_TEST: ImmutAfterInitCell<KernelMapping> = Im...
This test fails inside the SVSM when calling `phys_to_virt()` in this panic: https://github.com/coconut-svsm/svsm/blob/be10d1b5ef8f4365434238e916623f14b59359d5/src/mm/address_space.rs#L43-L47 I suspect some of the other tests might also fail.
svsm
github_2023
others
143
coconut-svsm
00xc
@@ -157,5 +157,68 @@ pub const SVSM_PERTASK_BASE: VirtAddr = virt_from_idx(PGTABLE_LVL3_IDX_PERTASK); /// End address of task memory region pub const SVSM_PERTASK_END: VirtAddr = SVSM_PERTASK_BASE.const_add(SIZE_LEVEL3); -/// Kernel stack for a task -pub const SVSM_PERTASK_STACK_BASE: VirtAddr = SVSM_PERTASK_BASE; ...
I think this will be problematic when running tests inside the SVSM. If a test runs before `test_init_kernel_mapping_info()`, it will access an uninitialized `KERNEL_MAPPING` (via `phys_to_virt()` or `virt_to_phys()`), leading to undefined behavior. This does not happen when running tests regularly because `virt_to_...
svsm
github_2023
others
162
coconut-svsm
00xc
@@ -0,0 +1,18 @@ +#![no_main] + +use libfuzzer_sys::{fuzz_target, Corpus}; +use svsm::cpu::insn::{Instruction, MAX_INSN_SIZE}; + +fuzz_target!(|input: &[u8]| -> Corpus { + if input.len() < MAX_INSN_SIZE { + return Corpus::Reject; + } + + let mut data = [0u8; MAX_INSN_SIZE]; + data.copy_from_slice(&in...
You can merge the bounds check and the access into one. This might not look too useful, but since most harnesses are ran with debug assertions enabled this *might* avoid a needless bounds check when doing `&input[..MAX_INSN_SIZE]`. ```suggestion let Some(input) = input.get(..MAX_INSN_SIZE) else { retur...
svsm
github_2023
others
162
coconut-svsm
00xc
@@ -0,0 +1,18 @@ +#![no_main] + +use libfuzzer_sys::{fuzz_target, Corpus}; +use svsm::cpu::insn::{Instruction, MAX_INSN_SIZE}; + +fuzz_target!(|input: &[u8]| -> Corpus { + if input.len() < MAX_INSN_SIZE { + return Corpus::Reject; + } + + let mut data = [0u8; MAX_INSN_SIZE]; + data.copy_from_slice(&in...
The compiler might optimize this away depending on compilation flags. ```suggestion let _ = core::hint::black_box(insn.decode()); ```
svsm
github_2023
others
166
coconut-svsm
00xc
@@ -62,20 +62,25 @@ pub mod svsm_gdbstub { Ok(()) } - #[derive(PartialEq, Eq)] + #[derive(PartialEq, Eq, Debug)] enum ExceptionType { Debug, SwBreakpoint, PageFault, } + impl From<usize> for ExceptionType { + fn from(value: usize) -> Self { + ...
And now you can inline this function in the callers in my opinion, since it's only one line :)
svsm
github_2023
others
159
coconut-svsm
00xc
@@ -72,15 +72,12 @@ pub mod svsm_gdbstub { pub fn handle_debug_exception(ctx: &mut X86ExceptionContext, exception: usize) { let tp = match exception { BP_VECTOR => ExceptionType::SwBreakpoint, + VC_VECTOR => ExceptionType::Debug, _ => ExceptionType::PageFault, ...
While you are cleaning things up, could you implement `From<usize> for ExceptionType` and do this match case there?
svsm
github_2023
others
134
coconut-svsm
Freax13
@@ -152,11 +152,20 @@ impl ValidBitmap { bitmap_alloc_order(self.pbase, self.pend) } - fn migrate(&mut self, new_bitmap: *mut u64) { - let (count, _) = self.index(self.pend); + /// The number of u64's in the bitmap + fn bitmap_len(&self) -> usize { + let num_pages = self.pend.bits...
Is the check correct? I'd expect it to be `num_pages % u64::BITS as usize != 0`. Note that you may be able to use [`usize::div_ceil`](https://doc.rust-lang.org/std/primitive.usize.html#method.div_ceil) here (which was stabilized very recently).
svsm
github_2023
others
55
coconut-svsm
joergroedel
@@ -0,0 +1,60 @@ +// SPDX-License-Identifier: MIT OR Apache-2.0 +// +// Copyright (c) 2022-2023 SUSE LLC +// +// Author: Thomas Leroy <tleroy@suse.de>
When you move code to a new file, please also add the original authors to the file header.
svsm
github_2023
others
55
coconut-svsm
AdamCDunlap
@@ -39,18 +35,22 @@ pub const _HV_VECTOR: usize = 28; pub const VC_VECTOR: usize = 29; pub const _SX_VECTOR: usize = 30; +pub const SVM_EXIT_EXCP_BASE: usize = 0x40; +pub const X86_TRAP_DB: usize = 0x01; + #[repr(C, packed)] #[derive(Default, Debug)] pub struct X86ExceptionContext { pub regs: X86GeneralRegs...
This struct is constructed manually in assembly, and I don't believe this PR adds space for the insn member, meaning that when insn is written it overwrites things on the stack from before the interrupt happens. I think it'd be better to leave insn out of this context and just have it as a separate variable returned...
svsm
github_2023
others
55
coconut-svsm
AdamCDunlap
@@ -0,0 +1,136 @@ +// SPDX-License-Identifier: MIT OR Apache-2.0 +// +// Copyright (c) 2022-2023 SUSE LLC +// +// Authors: Joerg Roedel <jroedel@suse.de> + +use super::super::control_regs::read_cr2; +use super::super::extable::handle_exception_table; +use super::super::tss::IST_DF; +use super::super::vc::handle_vc_exce...
I believe this is copied from the old code, but still worth changing: This function is called manually from assembly, so it should have extern "C". Otherwise, it uses the rust ABI which may not have the argument be passed in %rdi.
svsm
github_2023
others
55
coconut-svsm
00xc
@@ -4,28 +4,266 @@ // // Author: Joerg Roedel <jroedel@suse.de> -use super::idt::X86ExceptionContext; -use crate::cpu::extable::handle_exception_table; +use super::idt::common::X86ExceptionContext; +use crate::cpu::cpuid::{cpuid_table_raw, CpuidLeaf}; +use crate::cpu::insn::{insn_fetch, Instruction}; +use crate::cp...
`SVM_EXIT_EXCP_BASE` is duplicated, it could be imported from `src/cpu/idt/common.rs`.
svsm
github_2023
others
55
coconut-svsm
00xc
@@ -4,28 +4,266 @@ // // Author: Joerg Roedel <jroedel@suse.de> -use super::idt::X86ExceptionContext; -use crate::cpu::extable::handle_exception_table; +use super::idt::common::X86ExceptionContext; +use crate::cpu::cpuid::{cpuid_table_raw, CpuidLeaf}; +use crate::cpu::insn::{insn_fetch, Instruction}; +use crate::cp...
`X86_TRAP_DB` is also duplicate from `src/cpu/idt/common.rs`.
svsm
github_2023
others
55
coconut-svsm
00xc
@@ -4,28 +4,266 @@ // // Author: Joerg Roedel <jroedel@suse.de> -use super::idt::X86ExceptionContext; -use crate::cpu::extable::handle_exception_table; +use super::idt::common::X86ExceptionContext; +use crate::cpu::cpuid::{cpuid_table_raw, CpuidLeaf}; +use crate::cpu::insn::{insn_fetch, Instruction}; +use crate::cp...
If we are going to panic on error then perhaps `panic!()` should be inside the `_ => ` branch of the match statement.
svsm
github_2023
others
55
coconut-svsm
00xc
@@ -4,28 +4,266 @@ // // Author: Joerg Roedel <jroedel@suse.de> -use super::idt::X86ExceptionContext; -use crate::cpu::extable::handle_exception_table; +use super::idt::common::X86ExceptionContext; +use crate::cpu::cpuid::{cpuid_table_raw, CpuidLeaf}; +use crate::cpu::insn::{insn_fetch, Instruction}; +use crate::cp...
There's a few places where we panic on `SvsmError`s containing this error and print the RIP and error code, so perhaps these variants should contain information about that error, so that we do not have to explicitly write out all that information. Something like: ```rust pub enum VcError { ... DecodeFaile...
svsm
github_2023
others
55
coconut-svsm
00xc
@@ -4,28 +4,266 @@ // // Author: Joerg Roedel <jroedel@suse.de> -use super::idt::X86ExceptionContext; -use crate::cpu::extable::handle_exception_table; +use super::idt::common::X86ExceptionContext; +use crate::cpu::cpuid::{cpuid_table_raw, CpuidLeaf}; +use crate::cpu::insn::{insn_fetch, Instruction}; +use crate::cp...
Same here, the error handling should be inside the match statement.
svsm
github_2023
others
55
coconut-svsm
00xc
@@ -4,28 +4,266 @@ // // Author: Joerg Roedel <jroedel@suse.de> -use super::idt::X86ExceptionContext; -use crate::cpu::extable::handle_exception_table; +use super::idt::common::X86ExceptionContext; +use crate::cpu::cpuid::{cpuid_table_raw, CpuidLeaf}; +use crate::cpu::insn::{insn_fetch, Instruction}; +use crate::cp...
```suggestion let Some(ret) = cpuid_table_raw(leaf.cpuid_fn, leaf.cpuid_subfn, 0, 0) else { return Err(SvsmError::Vc(VcError::UnknownCpuidLeaf)); }; ```
svsm
github_2023
others
55
coconut-svsm
00xc
@@ -4,28 +4,266 @@ // // Author: Joerg Roedel <jroedel@suse.de> -use super::idt::X86ExceptionContext; -use crate::cpu::extable::handle_exception_table; +use super::idt::common::X86ExceptionContext; +use crate::cpu::cpuid::{cpuid_table_raw, CpuidLeaf}; +use crate::cpu::insn::{insn_fetch, Instruction}; +use crate::cp...
```suggestion let (size, mask) = if insn.prefixes.nb_bytes > 0 { // inw instruction has a 0x66 operand-size prefix for word-sized operands (GHCBIOSize::Size16, u16::MAX as u64) } else { (GHCBIOSize::Size32, u32::MAX as u64) }; ``...
svsm
github_2023
others
55
coconut-svsm
00xc
@@ -4,28 +4,266 @@ // // Author: Joerg Roedel <jroedel@suse.de> -use super::idt::X86ExceptionContext; -use crate::cpu::extable::handle_exception_table; +use super::idt::common::X86ExceptionContext; +use crate::cpu::cpuid::{cpuid_table_raw, CpuidLeaf}; +use crate::cpu::insn::{insn_fetch, Instruction}; +use crate::cp...
```suggestion let size = if insn.prefixes.nb_bytes > 0 { // outw instruction has a 0x66 operand-size prefix for word-sized operands. GHCBIOSize::Size16 } else { GHCBIOSize::Size32 }; ```
svsm
github_2023
others
55
coconut-svsm
00xc
@@ -4,28 +4,266 @@ // // Author: Joerg Roedel <jroedel@suse.de> -use super::idt::X86ExceptionContext; -use crate::cpu::extable::handle_exception_table; +use super::idt::common::X86ExceptionContext; +use crate::cpu::cpuid::{cpuid_table_raw, CpuidLeaf}; +use crate::cpu::insn::{insn_fetch, Instruction}; +use crate::cp...
```suggestion // TODO: the instruction fetch will likely to be handled differently when // #VC exception will be raised from CPL > 0. ```
svsm
github_2023
others
55
coconut-svsm
00xc
@@ -0,0 +1,122 @@ +// SPDX-License-Identifier: MIT OR Apache-2.0 +// +// Copyright (c) 2022-2023 SUSE LLC +// +// Author: Thomas Leroy <tleroy@suse.de> + +extern crate alloc; + +use crate::cpu::vc::VcError; +use crate::error::SvsmError; +use core::ptr; + +const MAX_INSN_SIZE: usize = 15; +const MAX_INSN_FIELD_SIZE: usi...
Let's remove the double negative ```suggestion /// The caller should validate that `rip` is set to a valid address. ```
svsm
github_2023
others
55
coconut-svsm
00xc
@@ -4,28 +4,266 @@ // // Author: Joerg Roedel <jroedel@suse.de> -use super::idt::X86ExceptionContext; -use crate::cpu::extable::handle_exception_table; +use super::idt::common::X86ExceptionContext; +use crate::cpu::cpuid::{cpuid_table_raw, CpuidLeaf}; +use crate::cpu::insn::{insn_fetch, Instruction}; +use crate::cp...
Please add a `// SAFETY:` comment here. In fact I'm not sure this is safe, as RIP could be set close to a boundary to an unmaped page. `insn_fetch` will read `MAX_INSN_SIZE` bytes, which may or may not be in bounds.
svsm
github_2023
others
55
coconut-svsm
00xc
@@ -0,0 +1,122 @@ +// SPDX-License-Identifier: MIT OR Apache-2.0 +// +// Copyright (c) 2022-2023 SUSE LLC +// +// Author: Thomas Leroy <tleroy@suse.de> + +extern crate alloc; + +use crate::cpu::vc::VcError; +use crate::error::SvsmError; +use core::ptr; + +const MAX_INSN_SIZE: usize = 15; +const MAX_INSN_FIELD_SIZE: usi...
```suggestion pub unsafe fn insn_fetch(rip: *const u8) -> [u8; MAX_INSN_SIZE] { rip.cast::<[u8; MAX_INSN_SIZE]>.read() } ```
svsm
github_2023
others
55
coconut-svsm
00xc
@@ -0,0 +1,114 @@ +// SPDX-License-Identifier: MIT OR Apache-2.0 +// +// Copyright (c) 2022-2023 SUSE LLC +// +// Author: Thomas Leroy <tleroy@suse.de> + +extern crate alloc; + +use crate::cpu::vc::VcError; +use crate::error::SvsmError; + +const MAX_INSN_SIZE: usize = 15; +const MAX_INSN_FIELD_SIZE: usize = 3; + +#[der...
There are several arrays with a field to track length, so how about the following to reduce code duplication: ```rust #[derive(Debug, Copy, Clone, Default)] pub(super) struct InsBuffer<const N: usize> where [u8; N]: Default, { buf: [u8; N], len: usize, } #[derive(Debug, Copy, Clone, Default)] ...
svsm
github_2023
others
55
coconut-svsm
00xc
@@ -0,0 +1,114 @@ +// SPDX-License-Identifier: MIT OR Apache-2.0 +// +// Copyright (c) 2022-2023 SUSE LLC +// +// Author: Thomas Leroy <tleroy@suse.de> + +extern crate alloc; + +use crate::cpu::vc::VcError; +use crate::error::SvsmError; + +const MAX_INSN_SIZE: usize = 15; +const MAX_INSN_FIELD_SIZE: usize = 3; + +#[der...
```suggestion opcode: InstructionField::default(), // we'll copy content later ```
svsm
github_2023
others
55
coconut-svsm
00xc
@@ -0,0 +1,114 @@ +// SPDX-License-Identifier: MIT OR Apache-2.0 +// +// Copyright (c) 2022-2023 SUSE LLC +// +// Author: Thomas Leroy <tleroy@suse.de> + +extern crate alloc; + +use crate::cpu::vc::VcError; +use crate::error::SvsmError; + +const MAX_INSN_SIZE: usize = 15; +const MAX_INSN_FIELD_SIZE: usize = 3; + +#[der...
```suggestion /// # Safety /// /// The caller should validate that `rip` is set to a valid address /// and that the next [`MAX_INSN_SIZE`] bytes are within valid memory. ```
svsm
github_2023
others
124
coconut-svsm
00xc
@@ -19,6 +19,9 @@ all: svsm.bin test: cargo test --target=x86_64-unknown-linux-gnu +doc: + cargo doc --open
We probably also want `--all-features`. Perhaps even `--document-private-items`.
svsm
github_2023
others
124
coconut-svsm
00xc
@@ -19,6 +19,9 @@ all: svsm.bin test: cargo test --target=x86_64-unknown-linux-gnu +doc: + cargo doc --open --document-private-items
`--all-features` still missing
svsm
github_2023
others
109
coconut-svsm
00xc
@@ -46,27 +45,20 @@ impl PerCpuInfo { // PERCPU areas virtual addresses into shared memory pub static PERCPU_AREAS: PerCpuAreas = PerCpuAreas::new(); -// We use an UnsafeCell to allow for a static with interior -// mutability. Normally, we would need to guarantee synchronization -// on the backing datatype, but thi...
`RWLock` introduces unnecessary overhead in my opinion, so I'd say this change is not needed, as the comment explains. Safety is already documented and limited to private functions.
svsm
github_2023
others
109
coconut-svsm
joergroedel
@@ -5,62 +5,62 @@ // Author: Joerg Roedel <jroedel@suse.de> use core::cell::UnsafeCell; -use core::fmt::Debug; use core::ops::{Deref, DerefMut}; use core::sync::atomic::{AtomicU64, Ordering}; #[derive(Debug)] -pub struct ReadLockGuard<'a, T: Debug> { +pub struct ReadLockGuard<'a, T> { rwlock: &'a AtomicU6...
Hmm, I don't buy the justification for this. In my thinking, the purpose of `RWLock` is to add `Sync` to any type by: 1. Serializing writes wrt. other writes and readers (there is no reader active when write-lock is aquired) 2. Allowing multiple readers at the same time by using the fact that immutable values are a...
svsm
github_2023
others
109
coconut-svsm
joergroedel
@@ -5,62 +5,62 @@ // Author: Joerg Roedel <jroedel@suse.de> use core::cell::UnsafeCell; -use core::fmt::Debug; use core::ops::{Deref, DerefMut}; use core::sync::atomic::{AtomicU64, Ordering}; #[derive(Debug)] -pub struct ReadLockGuard<'a, T: Debug> { +pub struct ReadLockGuard<'a, T> {
There was some reasoning behind adding those bounds. @00xc might know more, probably because it will make it easier to add a Debug trait to structs containing a lock.
svsm
github_2023
others
109
coconut-svsm
joergroedel
@@ -129,20 +125,17 @@ impl SvsmFs { } } -static mut FS_ROOT: SvsmFs = SvsmFs::new(); +static FS_ROOT: RWLock<SvsmFs> = RWLock::new(SvsmFs::new());
Is there another reason for the lock besides removing the `mut`? This change has a significant impact on the scalability of the FS code, so if there is no other reason I'd rather live the the `mut` than introducing a lock.