| Age | Commit message (Collapse) | Author |
|---|
|
To prevent user programs from performing any operations related to
system memory structures (such as the GDT, TSS, etc), we will always set
CR4.UMIP if possible to improve the security of Hyra.
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
Some older CPUs may not even support the TSC so it would be a good idea
to check before we start assuming its existence.
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
Introduce the USER_TSC kernel config option to control whether or not
the 'rdtsc' instruction should be accessible in a user context.
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
This commit introduces a vendor field within the 'cpu_info' structure.
This allows kernel code to detect the CPU vendor by comparing it with
the CPU_VENDOR_* defines that have also been added.
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
- Extend 'nswitch' to 64-bits as it increments fast
- Ensure writes to 'nswitch' are atomic
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
Sometimes we may rely on a specific kind of timer. While we are able to
grab scheduler specific timers, general purpose timers, etc. One might
need to for example, check whether their general purpose timer timer is
monotonic or not.
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
This commit greatly improves the handling of how inter-processor
interrupts are allocated and managed. Previously Hyra used handlers at
fixed interrupt vectors for IPIs, however, that is very rigid and not
very scalable at all. With the new IPI framework, one may allocate IPIs
by calling 'md_ipi_alloc()' and initializing the handler field.
To send an IPI to a specific processor, one can use cpu_ipi_send() and
pass the specific CPU structure as well as the IPI identifier.
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
A backtrace usually occurs during critcal system events like panics.
When the system in a error state, we cannot trust the correctness of the
stack. This commit aims to improve checks done in a731d4e by adding a
max frame depth that limits how many frames can be walked.
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
Improve handling within the stacktrace logic used during system panics
and critical events.
- Perform checks on 'rbp' *before* logging
- Ensure RBP is 8-byte aligned
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
Give each core a bit that indicates whether the core is online or
halted. We also provide a cpu_halt() which marks the core as offline.
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
Each core on the CPU gets one idle thread so that it is always up to
something. To ensure these threads are not bouncing wildly between cores
and are always ready for their respective processor to take when they
have no threads to grab. We must call proc_pin() on each for every
processor.
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
It is a good idea to have a completely different stack during syscalls.
This commit implements a TSS IST entry for syscalls.
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
Certain operations dealing with context switches are better off being
implemented as machine dependent routines which allows us to be flexible
with what we can do, as well as the extra benefit of requiring less
workarounds to keep things MI.
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
Keep track of how many times an interrupt fires, this can be used for
interrupt statistics later on.
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
The CPU family ID and model number are sometimes useful in order to
detect the presence of certain CPU features as well as detecting CPU
specific quirks.
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
Avoid using kprintf() during a system panic event as it relies on
internal locking that may hang the system preventing diagnostics from
being logged
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
To make debugging userland program crashes easier, this commit
introduces coredumps that store the last known process state into a
temporary /tmp/core.X file (where X is the PID of the faulting process).
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
Allow the user to configure if SMEP would be enabled for their build of
the Hyra AMD64 port
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
Some CPUs support Supervisor Memory Execution Protection that prevent ring 3
code from being executed in a ring 0 context. Enable this on CPUs that
support it.
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
Keep track of per-cpu features by using a 'feat' field which contains
the results of various CPUIDs.
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
- Rename regdump() to trap_fatal()
- Use panic() rather than kprintf()
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
Having the same amount of idle threads per core allow the CPU to always
have a thread assigned to one of its cores even when there are currently
no other threads for it to take.
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
Before the Hyra kernel starts up, the Limine bootloader creates a
mapping of the first 4 GiB of memory (identity mapped for base revisions
of zero and mapped to the HDDM on higher base revisions). However, it
usually will create a huge mapping over this which would be problematic
during remaps of device MMIO base addresses as they typically reside at
the 4 GiB region of RAM and as of this commit, huge pages are not a
supported feature of the Hyra AMD64 pmap layer.
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
The function md_td_kick() is only used for user threads and there is no
need reference any kernel segments. Furthermore, the previous code would
introduce a bug where rpl would always be 0.
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
Allow the intr_hand structure to carry driver specific and interrupt
related data so that it can be passed to an interrupt handler when it
is invoked.
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
- Use pmap_update_tbl() to ensure that the TLB is flushed when caching
attributes are updated.
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
Add an `alloc' paramater to pmap_update_tbl() to ensure that functions
like pmap_map() and pmap_unmap() have control over, whether or not to
allocate new page table entries. This is useful as unmapping memory does
not require new page table entries to be allocated and doing so anyways
may result in undefined behaviour such as hangs.
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
- Add prot var to keep track of flags to map with
- OR prot flags with PROT_USER if rpl is 3
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
Add support for managing the page table entry dirty bit. This will allow
vm(9) to know when to page-out data to backing store, among other
things.
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
Signed-off-by: Ian Moffett <ian@osmora.org>
|
|
Signed-off-by: Ian Moffett <ian@osmora.org>
|
