Both #DB itself, as all other IST users (NMI, #MC) now clear DR7 on
entry. Combined with not allowing breakpoints on entry/noinstr/NOKPROBE
text and no single step (EFLAGS.TF) inside the #DB handler should guarantee
no nested #DB.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lkml.kernel.org/r/20200529213321.303027161@infradead.org
- Untangle the somewhat incestous way of how VMALLOC_START is used all across the
kernel, but is, on x86, defined deep inside one of the lowest level page table headers.
It doesn't help that vmalloc.h only includes a single asm header:
#include <asm/page.h> /* pgprot_t */
So there was no existing cross-arch way to decouple address layout
definitions from page.h details. I used this:
#ifndef VMALLOC_START
# include <asm/vmalloc.h>
#endif
This way every architecture that wants to simplify page.h can do so.
- Also on x86 we had a couple of LDT related inline functions that used
the late-stage address space layout positions - but these could be
uninlined without real trouble - the end result is cleaner this way as
well.
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: linux-kernel@vger.kernel.org
Cc: linux-mm@kvack.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
There are three problems with the current layout of the doublefault
stack and TSS. First, the TSS is only cacheline-aligned, which is
not enough -- if the hardware portion of the TSS (struct x86_hw_tss)
crosses a page boundary, horrible things happen [0]. Second, the
stack and TSS are global, so simultaneous double faults on different
CPUs will cause massive corruption. Third, the whole mechanism
won't work if user CR3 is loaded, resulting in a triple fault [1].
Let the doublefault stack and TSS share a page (which prevents the
TSS from spanning a page boundary), make it percpu, and move it into
cpu_entry_area. Teach the stack dump code about the doublefault
stack.
[0] Real hardware will read past the end of the page onto the next
*physical* page if a task switch happens. Virtual machines may
have any number of bugs, and I would consider it reasonable for
a VM to summarily kill the guest if it tries to task-switch to
a page-spanning TSS.
[1] Real hardware triple faults. At least some VMs seem to hang.
I'm not sure what's going on.
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
When two recent commits that increased the size of the 'struct cpu_entry_area'
were merged in -tip, the 32-bit defconfig build started failing on the following
build time assert:
./include/linux/compiler.h:391:38: error: call to ‘__compiletime_assert_189’ declared with attribute error: BUILD_BUG_ON failed: CPU_ENTRY_AREA_PAGES * PAGE_SIZE < CPU_ENTRY_AREA_MAP_SIZE
arch/x86/mm/cpu_entry_area.c:189:2: note: in expansion of macro ‘BUILD_BUG_ON’
In function ‘setup_cpu_entry_area_ptes’,
Which corresponds to the following build time assert:
BUILD_BUG_ON(CPU_ENTRY_AREA_PAGES * PAGE_SIZE < CPU_ENTRY_AREA_MAP_SIZE);
The purpose of this assert is to sanity check the fixed-value definition of
CPU_ENTRY_AREA_PAGES arch/x86/include/asm/pgtable_32_types.h:
#define CPU_ENTRY_AREA_PAGES (NR_CPUS * 41)
The '41' is supposed to match sizeof(struct cpu_entry_area)/PAGE_SIZE, which value
we didn't want to define in such a low level header, because it would cause
dependency hell.
Every time the size of cpu_entry_area is changed, we have to adjust CPU_ENTRY_AREA_PAGES
accordingly - and this assert is checking that constraint.
But the assert is both imprecise and buggy, primarily because it doesn't
include the single readonly IDT page that is mapped at CPU_ENTRY_AREA_BASE
(which begins at a PMD boundary).
This bug was hidden by the fact that by accident CPU_ENTRY_AREA_PAGES is defined
too large upstream (v5.4-rc8):
#define CPU_ENTRY_AREA_PAGES (NR_CPUS * 40)
While 'struct cpu_entry_area' is 155648 bytes, or 38 pages. So we had two extra
pages, which hid the bug.
The following commit (not yet upstream) increased the size to 40 pages:
x86/iopl: ("Restrict iopl() permission scope")
... but increased CPU_ENTRY_AREA_PAGES only 41 - i.e. shortening the gap
to just 1 extra page.
Then another not-yet-upstream commit changed the size again:
880a98c339: ("x86/cpu_entry_area: Add guard page for entry stack on 32bit")
Which increased the cpu_entry_area size from 38 to 39 pages, but
didn't change CPU_ENTRY_AREA_PAGES (kept it at 40). This worked
fine, because we still had a page left from the accidental 'reserve'.
But when these two commits were merged into the same tree, the
combined size of cpu_entry_area grew from 38 to 40 pages, while
CPU_ENTRY_AREA_PAGES finally caught up to 40 as well.
Which is fine in terms of functionality, but the assert broke:
BUILD_BUG_ON(CPU_ENTRY_AREA_PAGES * PAGE_SIZE < CPU_ENTRY_AREA_MAP_SIZE);
because CPU_ENTRY_AREA_MAP_SIZE is the total size of the area,
which is 1 page larger due to the IDT page.
To fix all this, change the assert to two precise asserts:
BUILD_BUG_ON((CPU_ENTRY_AREA_PAGES+1)*PAGE_SIZE != CPU_ENTRY_AREA_MAP_SIZE);
BUILD_BUG_ON(CPU_ENTRY_AREA_TOTAL_SIZE != CPU_ENTRY_AREA_MAP_SIZE);
This takes the IDT page into account, and also connects the size-based
define of CPU_ENTRY_AREA_TOTAL_SIZE with the address-subtraction based
define of CPU_ENTRY_AREA_MAP_SIZE.
Also clean up some of the names which made it rather confusing:
- 'CPU_ENTRY_AREA_TOT_SIZE' wasn't actually the 'total' size of
the cpu-entry-area, but the per-cpu array size, so rename this
to CPU_ENTRY_AREA_ARRAY_SIZE.
- Introduce CPU_ENTRY_AREA_TOTAL_SIZE that _is_ the total mapping
size, with the IDT included.
- Add comments where '+1' denotes the IDT mapping - it wasn't
obvious and took me about 3 hours to decode...
Finally, because this particular commit is actually applied after
this patch:
880a98c339: ("x86/cpu_entry_area: Add guard page for entry stack on 32bit")
Fix the CPU_ENTRY_AREA_PAGES value from 40 pages to the correct 39 pages.
All future commits that change cpu_entry_area will have to adjust
this value precisely.
As a side note, we should probably attempt to remove CPU_ENTRY_AREA_PAGES
and derive its value directly from the structure, without causing
header hell - but that is an adventure for another day! :-)
Fixes: 880a98c339: ("x86/cpu_entry_area: Add guard page for entry stack on 32bit")
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: stable@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The entry stack in the cpu entry area is protected against overflow by the
readonly GDT on 64-bit, but on 32-bit the GDT needs to be writeable and
therefore does not trigger a fault on stack overflow.
Add a guard page.
Fixes: c482feefe1 ("x86/entry/64: Make cpu_entry_area.tss read-only")
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: stable@kernel.org
The debug IST stack is actually two separate debug stacks to handle #DB
recursion. This is required because the CPU starts always at top of stack
on exception entry, which means on #DB recursion the second #DB would
overwrite the stack of the first.
The low level entry code therefore adjusts the top of stack on entry so a
secondary #DB starts from a different stack page. But the stack pages are
adjacent without a guard page between them.
Split the debug stack into 3 stacks which are separated by guard pages. The
3rd stack is never mapped into the cpu_entry_area and is only there to
catch triple #DB nesting:
--- top of DB_stack <- Initial stack
--- end of DB_stack
guard page
--- top of DB1_stack <- Top of stack after entering first #DB
--- end of DB1_stack
guard page
--- top of DB2_stack <- Top of stack after entering second #DB
--- end of DB2_stack
guard page
If DB2 would not act as the final guard hole, a second #DB would point the
top of #DB stack to the stack below #DB1 which would be valid and not catch
the not so desired triple nesting.
The backing store does not allocate any memory for DB2 and its guard page
as it is not going to be mapped into the cpu_entry_area.
- Adjust the low level entry code so it adjusts top of #DB with the offset
between the stacks instead of exception stack size.
- Make the dumpstack code aware of the new stacks.
- Adjust the in_debug_stack() implementation and move it into the NMI code
where it belongs. As this is NMI hotpath code, it just checks the full
area between top of DB_stack and bottom of DB1_stack without checking
for the guard page. That's correct because the NMI cannot hit a
stackpointer pointing to the guard page between DB and DB1 stack. Even
if it would, then the NMI operation still is unaffected, but the resume
of the debug exception on the topmost DB stack will crash by touching
the guard page.
[ bp: Make exception_stack_names static const char * const ]
Suggested-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Borislav Petkov <bp@suse.de>
Reviewed-by: Sean Christopherson <sean.j.christopherson@intel.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Baoquan He <bhe@redhat.com>
Cc: "Chang S. Bae" <chang.seok.bae@intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Dominik Brodowski <linux@dominikbrodowski.net>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Joerg Roedel <jroedel@suse.de>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Juergen Gross <jgross@suse.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Cc: linux-doc@vger.kernel.org
Cc: Masahiro Yamada <yamada.masahiro@socionext.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Qian Cai <cai@lca.pw>
Cc: Sean Christopherson <sean.j.christopherson@intel.com>
Cc: x86-ml <x86@kernel.org>
Link: https://lkml.kernel.org/r/20190414160145.439944544@linutronix.de
All usage sites which expected that the exception stacks in the CPU entry
area are mapped linearly are fixed up. Enable guard pages between the
IST stacks.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Borislav Petkov <bp@suse.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Sean Christopherson <sean.j.christopherson@intel.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: x86-ml <x86@kernel.org>
Link: https://lkml.kernel.org/r/20190414160145.349862042@linutronix.de
The entry order of the TSS.IST array and the order of the stack
storage/mapping are not required to be the same.
With the upcoming split of the debug stack this is going to fall apart as
the number of TSS.IST array entries stays the same while the actual stacks
are increasing.
Make them separate so that code like dumpstack can just utilize the mapping
order. The IST index is solely required for the actual TSS.IST array
initialization.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Borislav Petkov <bp@suse.de>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Baoquan He <bhe@redhat.com>
Cc: "Chang S. Bae" <chang.seok.bae@intel.com>
Cc: Dominik Brodowski <linux@dominikbrodowski.net>
Cc: Dou Liyang <douly.fnst@cn.fujitsu.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jann Horn <jannh@google.com>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Cc: Nicolai Stange <nstange@suse.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Qian Cai <cai@lca.pw>
Cc: Sean Christopherson <sean.j.christopherson@intel.com>
Cc: x86-ml <x86@kernel.org>
Link: https://lkml.kernel.org/r/20190414160145.241588113@linutronix.de
Store a pointer to the per cpu entry area exception stack mappings to allow
fast retrieval.
Required for converting various places from using the shadow IST array to
directly doing address calculations on the actual mapping address.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Borislav Petkov <bp@suse.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sean Christopherson <sean.j.christopherson@intel.com>
Cc: x86-ml <x86@kernel.org>
Link: https://lkml.kernel.org/r/20190414160144.680960459@linutronix.de
At the moment everything assumes a full linear mapping of the various
exception stacks. Adding guard pages to the cpu entry area mapping of the
exception stacks will break that assumption.
As a preparatory step convert both the real storage and the effective
mapping in the cpu entry area from character arrays to structures.
To ensure that both arrays have the same ordering and the same size of the
individual stacks fill the members with a macro. The guard size is the only
difference between the two resulting structures. For now both have guard
size 0 until the preparation of all usage sites is done.
Provide a couple of helper macros which are used in the following
conversions.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Borislav Petkov <bp@suse.de>
Reviewed-by: Sean Christopherson <sean.j.christopherson@intel.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: "Chang S. Bae" <chang.seok.bae@intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Dominik Brodowski <linux@dominikbrodowski.net>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: x86-ml <x86@kernel.org>
Link: https://lkml.kernel.org/r/20190414160144.506807893@linutronix.de
The SYSCALL64 trampoline has a couple of nice properties:
- The usual sequence of SWAPGS followed by two GS-relative accesses to
set up RSP is somewhat slow because the GS-relative accesses need
to wait for SWAPGS to finish. The trampoline approach allows
RIP-relative accesses to set up RSP, which avoids the stall.
- The trampoline avoids any percpu access before CR3 is set up,
which means that no percpu memory needs to be mapped in the user
page tables. This prevents using Meltdown to read any percpu memory
outside the cpu_entry_area and prevents using timing leaks
to directly locate the percpu areas.
The downsides of using a trampoline may outweigh the upsides, however.
It adds an extra non-contiguous I$ cache line to system calls, and it
forces an indirect jump to transfer control back to the normal kernel
text after CR3 is set up. The latter is because x86 lacks a 64-bit
direct jump instruction that could jump from the trampoline to the entry
text. With retpolines enabled, the indirect jump is extremely slow.
Change the code to map the percpu TSS into the user page tables to allow
the non-trampoline SYSCALL64 path to work under PTI. This does not add a
new direct information leak, since the TSS is readable by Meltdown from the
cpu_entry_area alias regardless. It does allow a timing attack to locate
the percpu area, but KASLR is more or less a lost cause against local
attack on CPUs vulnerable to Meltdown regardless. As far as I'm concerned,
on current hardware, KASLR is only useful to mitigate remote attacks that
try to attack the kernel without first gaining RCE against a vulnerable
user process.
On Skylake, with CONFIG_RETPOLINE=y and KPTI on, this reduces syscall
overhead from ~237ns to ~228ns.
There is a possible alternative approach: Move the trampoline within 2G of
the entry text and make a separate copy for each CPU. This would allow a
direct jump to rejoin the normal entry path. There are pro's and con's for
this approach:
+ It avoids a pipeline stall
- It executes from an extra page and read from another extra page during
the syscall. The latter is because it needs to use a relative
addressing mode to find sp1 -- it's the same *cacheline*, but accessed
using an alias, so it's an extra TLB entry.
- Slightly more memory. This would be one page per CPU for a simple
implementation and 64-ish bytes per CPU or one page per node for a more
complex implementation.
- More code complexity.
The current approach is chosen for simplicity and because the alternative
does not provide a significant benefit, which makes it worth.
[ tglx: Added the alternative discussion to the changelog ]
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Borislav Petkov <bp@suse.de>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Joerg Roedel <joro@8bytes.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: https://lkml.kernel.org/r/8c7c6e483612c3e4e10ca89495dc160b1aa66878.1536015544.git.luto@kernel.org
The Intel PEBS/BTS debug store is a design trainwreck as it expects virtual
addresses which must be visible in any execution context.
So it is required to make these mappings visible to user space when kernel
page table isolation is active.
Provide enough room for the buffer mappings in the cpu_entry_area so the
buffers are available in the user space visible page tables.
At the point where the kernel side entry area is populated there is no
buffer available yet, but the kernel PMD must be populated. To achieve this
set the entries for these buffers to non present.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: David Laight <David.Laight@aculab.com>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: Eduardo Valentin <eduval@amazon.com>
Cc: Greg KH <gregkh@linuxfoundation.org>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Juergen Gross <jgross@suse.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Will Deacon <will.deacon@arm.com>
Cc: aliguori@amazon.com
Cc: daniel.gruss@iaik.tugraz.at
Cc: hughd@google.com
Cc: keescook@google.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Put the cpu_entry_area into a separate P4D entry. The fixmap gets too big
and 0-day already hit a case where the fixmap PTEs were cleared by
cleanup_highmap().
Aside of that the fixmap API is a pain as it's all backwards.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Juergen Gross <jgross@suse.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Separate the cpu_entry_area code out of cpu/common.c and the fixmap.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Juergen Gross <jgross@suse.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
