In order to bring consistency in-tree, migrate all arch code to the new
prefix <zephyr/...>. Note that the conversion has been scripted, refer
to zephyrproject-rtos#45388 for more details.
Signed-off-by: Gerard Marull-Paretas <gerard.marull@nordicsemi.no>
If generic section is not present at boot, the thread stack
may not be in physical memory. Unconditionally page in the stack
instead of relying on page fault to speed up a little bit
on starting the thread.
Also, this prevents a double fault during thread setup when
setting up stack permission in z_x86_userspace_enter().
Signed-off-by: Daniel Leung <daniel.leung@intel.com>
This marks code and data within x86/ia32 so they are going to
reside in boot and pinned regions. This is a step to enable
demand paging for whole kernel.
Signed-off-by: Daniel Leung <daniel.leung@intel.com>
The x86_64 SysV ABI requires 16 byte alignment for the stack pointer
during execution of normal code. That means that on entry to an
ABI-compatible C function (which is reached via a CALL instruction
that pushes the return address) the RSP register must be MISaligned by
exactly 8 bytes. The kernel mode thread setup got this right, but we
missed the equivalent condition in userspace entry.
The end result was a misaligned stack, which is surprisingly robust
for most use. But recent toolchains have starting doing some more
elaborate vectorization, and the resulting SSE instructions started
failing in userspace on the misaliged loads.
Note that there's a comment about optimization: we're doing the stack
alignment in the "wrong place" and are needlessly wasting bytes in
some cases. We should see the raw stack boundaries where we are
setting up RSP values. Add a FIXME to this effect, but don't touch
anything as this patch is a targeted bugfix.
Also fix a somewhat embarassing 32-bit-ism that would have truncated
the address of a userspace stack that we tried to put above 4G.
Fixes#31018
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
We provide an option for low-memory systems to use a single set
of page tables for all threads. This is only supported if
KPTI and SMP are disabled. This configuration saves a considerable
amount of RAM, especially if multiple memory domains are used,
at a cost of context switching overhead.
Some caching techniques are used to reduce the amount of context
switch updates; the page tables aren't updated if switching to
a supervisor thread, and the page table configuration of the last
user thread switched in is cached.
Signed-off-by: Andrew Boie <andrew.p.boie@intel.com>
- z_x86_userspace_enter() for both 32-bit and 64-bit now
call into C code to clear the stack buffer and set the
US bits in the page tables for the memory range.
- Page tables are now associated with memory domains,
instead of having separate page tables per thread.
A spinlock protects write access to these page tables,
and read/write access to the list of active page
tables.
- arch_mem_domain_init() implemented, allocating and
copying page tables from the boot page tables.
- struct arch_mem_domain defined for x86. It has
a page table link and also a list node for iterating
over them.
Signed-off-by: Andrew Boie <andrew.p.boie@intel.com>
This will be needed when we support memory un-mapping, or
the same user mode page tables on multiple CPUs. Neither
are implemented yet.
Signed-off-by: Andrew Boie <andrew.p.boie@intel.com>
These days all threads are always a member of a memory domain,
remove this NULL check as it won't ever be false.
Signed-off-by: Andrew Boie <andrew.p.boie@intel.com>
This function iterates over the thread's memory domain
and updates page tables based on it. We need to be holding
z_mem_domain_lock while this happens.
Signed-off-by: Andrew Boie <andrew.p.boie@intel.com>
We no longer plan to support a split address space with
the kernel in high memory and per-process address spaces.
Because of this, we can simplify some things. System RAM
is now always identity mapped at boot.
We no longer require any virtual-to-physical translation
for page tables, and can remove the dual-mapping logic
from the page table generation script since we won't need
to transition the instruction point off of physical
addresses.
CONFIG_KERNEL_VM_BASE and CONFIG_KERNEL_VM_LIMIT
have been removed. The kernel's address space always
starts at CONFIG_SRAM_BASE_ADDRESS, of a fixed size
specified by CONFIG_KERNEL_VM_SIZE.
Driver MMIOs and other uses of k_mem_map() are still
virtually mapped, and the later introduction of demand
paging will result in only a subset of system RAM being
a fixed identity mapping instead of all of it.
Signed-off-by: Andrew Boie <andrew.p.boie@intel.com>
The x86 paging code has been rewritten to support another paging mode
and non-identity virtual mappings.
- Paging code now uses an array of paging level characteristics and
walks tables using for loops. This is opposed to having different
functions for every paging level and lots of #ifdefs. The code is
now more concise and adding new paging modes should be trivial.
- We now support 32-bit, PAE, and IA-32e page tables.
- The page tables created by gen_mmu.py are now installed at early
boot. There are no longer separate "flat" page tables. These tables
are mutable at any time.
- The x86_mmu code now has a private header. Many definitions that did
not need to be in public scope have been moved out of mmustructs.h
and either placed in the C file or in the private header.
- Improvements to dumping page table information, with the physical
mapping and flags all shown
- arch_mem_map() implemented
- x86 userspace/memory domain code ported to use the new
infrastructure.
- add logic for physical -> virtual instruction pointer transition,
including cleaning up identity mappings after this takes place.
Signed-off-by: Andrew Boie <andrew.p.boie@intel.com>
The core kernel computes the initial stack pointer
for a thread, properly aligning it and subtracting out
any random offsets or thread-local storage areas.
arch_new_thread() no longer needs to make any calculations,
an initial stack frame may be placed at the bounds of
the new 'stack_ptr' parameter passed in. This parameter
replaces 'stack_size'.
thread->stack_info is now set before arch_new_thread()
is invoked, z_new_thread_init() has been removed.
The values populated may need to be adjusted on arches
which carve-out MPU guard space from the actual stack
buffer.
thread->stack_info now has a new member 'delta' which
indicates any offset applied for TLS or random offset.
It's used so the calculations don't need to be repeated
if the thread later drops to user mode.
CONFIG_INIT_STACKS logic is now performed inside
z_setup_new_thread(), before arch_new_thread() is called.
thread->stack_info is now defined as the canonical
user-accessible area within the stack object, including
random offsets and TLS. It will never include any
carved-out memory for MPU guards and must be updated at
runtime if guards are removed.
Available stack space is now optimized. Some arches may
need to significantly round up the buffer size to account
for page-level granularity or MPU power-of-two requirements.
This space is now accounted for and used by virtue of
the Z_THREAD_STACK_SIZE_ADJUST() call in z_setup_new_thread.
Signed-off-by: Andrew Boie <andrew.p.boie@intel.com>
This operation is formally defined as rounding down a potential
stack pointer value to meet CPU and ABI requirments.
This was previously defined ad-hoc as STACK_ROUND_DOWN().
A new architecture constant ARCH_STACK_PTR_ALIGN is added.
Z_STACK_PTR_ALIGN() is defined in terms of it. This used to
be inconsistently specified as STACK_ALIGN or STACK_PTR_ALIGN;
in the latter case, STACK_ALIGN meant something else, typically
a required alignment for the base of a stack buffer.
STACK_ROUND_UP() only used in practice by Risc-V, delete
elsewhere.
Signed-off-by: Andrew Boie <andrew.p.boie@intel.com>
Implement a set of per-cpu trampoline stacks which all
interrupts and exceptions will initially land on, and also
as an intermediate stack for privilege changes as we need
some stack space to swap page tables.
Set up the special trampoline page which contains all the
trampoline stacks, TSS, and GDT. This page needs to be
present in the user page tables or interrupts don't work.
CPU exceptions, with KPTI turned on, are treated as interrupts
and not traps so that we have IRQs locked on exception entry.
Add some additional macros for defining IDT entries.
Add special handling of locore text/rodata sections when
creating user mode page tables on x86-64.
Restore qemu_x86_64 to use KPTI, and remove restrictions on
enabling user mode on x86-64.
Signed-off-by: Andrew Boie <andrew.p.boie@intel.com>
This is causing problems, as if we create a thread in
a system call we will *not* be using the kernel page
tables if CONFIG_KPTI=n.
Just don't fiddle with this page's permissions; we don't
need it as a guard area anyway since we have a stack
guard placed immediately before it, and this page
is unused if user mode isn't active.
Signed-off-by: Andrew Boie <andrew.p.boie@intel.com>
These are now common code, all are related to user mode
threads. The rat's nest of ifdefs in ia32's arch_new_thread
has been greatly simplified, there is now just one hook
if user mode is turned on.
Signed-off-by: Andrew Boie <andrew.p.boie@intel.com>
Gerard Marull-Paretas