| Message ID | 20221205201210.463781-2-ardb@kernel.org |
|---|---|
| State | Accepted |
| Commit | ff7a167961d1b97e0e205f245f806e564d3505e7 |
| Headers | show |
| Series | arm64: efi: Robustify EFI runtime wrapper code | expand |
Hi Ard, One drive-by comment below... On Mon, Dec 05, 2022 at 09:12:09PM +0100, Ard Biesheuvel wrote: > With the introduction of PRMT in the ACPI subsystem, the EFI rts > workqueue is no longer the only caller of efi_call_virt_pointer() in the > kernel. This means the EFI runtime services lock is no longer sufficient > to manage concurrent calls into firmware, but also that firmware calls > may occur that are not marshalled via the workqueue mechanism, but > originate directly from the caller context. > > For added robustness, and to ensure that the runtime services have 8 KiB > of stack space available as per the EFI spec, introduce a spinlock > protected EFI runtime stack of 8 KiB, where the spinlock also ensures > serialization between the EFI rts workqueue (which itself serializes EFI > runtime calls) and other callers of efi_call_virt_pointer(). > > While at it, use the stack pivot to avoid reloading the shadow call > stack pointer from the ordinary stack, as doing so could produce a > gadget to defeat it. > > Signed-off-by: Ard Biesheuvel <ardb@kernel.org> > --- > arch/arm64/include/asm/efi.h | 3 +++ > arch/arm64/kernel/efi-rt-wrapper.S | 13 +++++++++- > arch/arm64/kernel/efi.c | 25 ++++++++++++++++++++ We'll need to teach the stack unwinder about this, or if we take an exception from the EFI stack, the backtrace will terminate as soon as it hits a frame record on the EFI stack. In arch/arm64/kernel/stacktrace.c's arch_stack_walk(), that'll need to be added to the array of stack bounds. Ideally we'd only add that when a thread is making an EFI call. Thanks, Mark. > 3 files changed, 40 insertions(+), 1 deletion(-) > > diff --git a/arch/arm64/include/asm/efi.h b/arch/arm64/include/asm/efi.h > index 7c12e01c2b312e7b..1c408ec3c8b3a883 100644 > --- a/arch/arm64/include/asm/efi.h > +++ b/arch/arm64/include/asm/efi.h > @@ -25,6 +25,7 @@ int efi_set_mapping_permissions(struct mm_struct *mm, efi_memory_desc_t *md); > ({ \ > efi_virtmap_load(); \ > __efi_fpsimd_begin(); \ > + spin_lock(&efi_rt_lock); \ > }) > > #undef arch_efi_call_virt > @@ -33,10 +34,12 @@ int efi_set_mapping_permissions(struct mm_struct *mm, efi_memory_desc_t *md); > > #define arch_efi_call_virt_teardown() \ > ({ \ > + spin_unlock(&efi_rt_lock); \ > __efi_fpsimd_end(); \ > efi_virtmap_unload(); \ > }) > > +extern spinlock_t efi_rt_lock; > efi_status_t __efi_rt_asm_wrapper(void *, const char *, ...); > > #define ARCH_EFI_IRQ_FLAGS_MASK (PSR_D_BIT | PSR_A_BIT | PSR_I_BIT | PSR_F_BIT) > diff --git a/arch/arm64/kernel/efi-rt-wrapper.S b/arch/arm64/kernel/efi-rt-wrapper.S > index 75691a2641c1c0f8..b2786b968fee68dd 100644 > --- a/arch/arm64/kernel/efi-rt-wrapper.S > +++ b/arch/arm64/kernel/efi-rt-wrapper.S > @@ -16,6 +16,12 @@ SYM_FUNC_START(__efi_rt_asm_wrapper) > */ > stp x1, x18, [sp, #16] > > + ldr_l x16, efi_rt_stack_top > + mov sp, x16 > +#ifdef CONFIG_SHADOW_CALL_STACK > + str x18, [sp, #-16]! > +#endif > + > /* > * We are lucky enough that no EFI runtime services take more than > * 5 arguments, so all are passed in registers rather than via the > @@ -29,6 +35,7 @@ SYM_FUNC_START(__efi_rt_asm_wrapper) > mov x4, x6 > blr x8 > > + mov sp, x29 > ldp x1, x2, [sp, #16] > cmp x2, x18 > ldp x29, x30, [sp], #32 > @@ -42,6 +49,10 @@ SYM_FUNC_START(__efi_rt_asm_wrapper) > * called with preemption disabled and a separate shadow stack is used > * for interrupts. > */ > - mov x18, x2 > +#ifdef CONFIG_SHADOW_CALL_STACK > + ldr_l x18, efi_rt_stack_top > + ldr x18, [x18, #-16] > +#endif > + > b efi_handle_corrupted_x18 // tail call > SYM_FUNC_END(__efi_rt_asm_wrapper) > diff --git a/arch/arm64/kernel/efi.c b/arch/arm64/kernel/efi.c > index a908a37f03678b6b..8cb2e005f8aca589 100644 > --- a/arch/arm64/kernel/efi.c > +++ b/arch/arm64/kernel/efi.c > @@ -144,3 +144,28 @@ asmlinkage efi_status_t efi_handle_corrupted_x18(efi_status_t s, const char *f) > pr_err_ratelimited(FW_BUG "register x18 corrupted by EFI %s\n", f); > return s; > } > + > +DEFINE_SPINLOCK(efi_rt_lock); > + > +asmlinkage u64 *efi_rt_stack_top __ro_after_init; > + > +/* required by the EFI spec */ > +static_assert(THREAD_SIZE >= SZ_8K); > + > +int __init arm64_efi_rt_init(void) > +{ > + void *p = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN, > + VMALLOC_START, VMALLOC_END, GFP_KERNEL, > + PAGE_KERNEL, 0, NUMA_NO_NODE, > + __builtin_return_address(0)); > + > + if (!p) { > + pr_warn("Failed to allocate EFI runtime stack\n"); > + clear_bit(EFI_RUNTIME_SERVICES, &efi.flags); > + return -ENOMEM; > + } > + > + efi_rt_stack_top = p + THREAD_SIZE; > + return 0; > +} > +core_initcall(arm64_efi_rt_init); > -- > 2.35.1 >
On Fri, 9 Dec 2022 at 11:51, Mark Rutland <mark.rutland@arm.com> wrote: > > Hi Ard, > > One drive-by comment below... > > On Mon, Dec 05, 2022 at 09:12:09PM +0100, Ard Biesheuvel wrote: > > With the introduction of PRMT in the ACPI subsystem, the EFI rts > > workqueue is no longer the only caller of efi_call_virt_pointer() in the > > kernel. This means the EFI runtime services lock is no longer sufficient > > to manage concurrent calls into firmware, but also that firmware calls > > may occur that are not marshalled via the workqueue mechanism, but > > originate directly from the caller context. > > > > For added robustness, and to ensure that the runtime services have 8 KiB > > of stack space available as per the EFI spec, introduce a spinlock > > protected EFI runtime stack of 8 KiB, where the spinlock also ensures > > serialization between the EFI rts workqueue (which itself serializes EFI > > runtime calls) and other callers of efi_call_virt_pointer(). > > > > While at it, use the stack pivot to avoid reloading the shadow call > > stack pointer from the ordinary stack, as doing so could produce a > > gadget to defeat it. > > > > Signed-off-by: Ard Biesheuvel <ardb@kernel.org> > > --- > > arch/arm64/include/asm/efi.h | 3 +++ > > arch/arm64/kernel/efi-rt-wrapper.S | 13 +++++++++- > > arch/arm64/kernel/efi.c | 25 ++++++++++++++++++++ > > We'll need to teach the stack unwinder about this, or if we take an exception > from the EFI stack, the backtrace will terminate as soon as it hits a frame > record on the EFI stack. > > In arch/arm64/kernel/stacktrace.c's arch_stack_walk(), that'll need to be added > to the array of stack bounds. Ideally we'd only add that when a thread is > making an EFI call. > Thanks, I'll look into that.
On Wed, Jan 04, 2023 at 02:56:19PM +0100, Ard Biesheuvel wrote: > On Wed, 4 Jan 2023 at 11:40, Lee Jones <lee@kernel.org> wrote: > > > > On Mon, 05 Dec 2022, Ard Biesheuvel wrote: > > > > > With the introduction of PRMT in the ACPI subsystem, the EFI rts > > > workqueue is no longer the only caller of efi_call_virt_pointer() in the > > > kernel. This means the EFI runtime services lock is no longer sufficient > > > to manage concurrent calls into firmware, but also that firmware calls > > > may occur that are not marshalled via the workqueue mechanism, but > > > originate directly from the caller context. > > > > > > For added robustness, and to ensure that the runtime services have 8 KiB > > > of stack space available as per the EFI spec, introduce a spinlock > > > protected EFI runtime stack of 8 KiB, where the spinlock also ensures > > > serialization between the EFI rts workqueue (which itself serializes EFI > > > runtime calls) and other callers of efi_call_virt_pointer(). > > > > > > While at it, use the stack pivot to avoid reloading the shadow call > > > stack pointer from the ordinary stack, as doing so could produce a > > > gadget to defeat it. > > > > > > Signed-off-by: Ard Biesheuvel <ardb@kernel.org> > > > --- > > > arch/arm64/include/asm/efi.h | 3 +++ > > > arch/arm64/kernel/efi-rt-wrapper.S | 13 +++++++++- > > > arch/arm64/kernel/efi.c | 25 ++++++++++++++++++++ > > > 3 files changed, 40 insertions(+), 1 deletion(-) > > > > Could we have this in Stable please? > > > > Upstream commit: ff7a167961d1b ("arm64: efi: Execute runtime services from a dedicated stack") > > > > Ard, do we need Patch 2 as well, or can this be applied on its own? > > > > Thanks for the reminder. > > Only patch #1 is needed. It should be applied to v5.10 and later. Hold on, why did this go into mainline when I had an outstanding comment w.r.t. the stack unwinder?
On Wed, 4 Jan 2023 at 17:13, Mark Rutland <mark.rutland@arm.com> wrote: > > On Wed, Jan 04, 2023 at 02:56:19PM +0100, Ard Biesheuvel wrote: > > On Wed, 4 Jan 2023 at 11:40, Lee Jones <lee@kernel.org> wrote: > > > > > > On Mon, 05 Dec 2022, Ard Biesheuvel wrote: > > > > > > > With the introduction of PRMT in the ACPI subsystem, the EFI rts > > > > workqueue is no longer the only caller of efi_call_virt_pointer() in the > > > > kernel. This means the EFI runtime services lock is no longer sufficient > > > > to manage concurrent calls into firmware, but also that firmware calls > > > > may occur that are not marshalled via the workqueue mechanism, but > > > > originate directly from the caller context. > > > > > > > > For added robustness, and to ensure that the runtime services have 8 KiB > > > > of stack space available as per the EFI spec, introduce a spinlock > > > > protected EFI runtime stack of 8 KiB, where the spinlock also ensures > > > > serialization between the EFI rts workqueue (which itself serializes EFI > > > > runtime calls) and other callers of efi_call_virt_pointer(). > > > > > > > > While at it, use the stack pivot to avoid reloading the shadow call > > > > stack pointer from the ordinary stack, as doing so could produce a > > > > gadget to defeat it. > > > > > > > > Signed-off-by: Ard Biesheuvel <ardb@kernel.org> > > > > --- > > > > arch/arm64/include/asm/efi.h | 3 +++ > > > > arch/arm64/kernel/efi-rt-wrapper.S | 13 +++++++++- > > > > arch/arm64/kernel/efi.c | 25 ++++++++++++++++++++ > > > > 3 files changed, 40 insertions(+), 1 deletion(-) > > > > > > Could we have this in Stable please? > > > > > > Upstream commit: ff7a167961d1b ("arm64: efi: Execute runtime services from a dedicated stack") > > > > > > Ard, do we need Patch 2 as well, or can this be applied on its own? > > > > > > > Thanks for the reminder. > > > > Only patch #1 is needed. It should be applied to v5.10 and later. > > Hold on, why did this go into mainline when I had an outstanding comment w.r.t. > the stack unwinder? > > From your last reply to me there I was expecting a respin with that fixed. > Apologies for the confusion. I have a patch for this queued up, but AIUI, that cannot be merged all the way back to v5.10, so these need to remain separate changes in any case. https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git/commit/?id=c2530a04a73e6b75ed71ed14d09d7b42d6300013
On Wed, Jan 04, 2023 at 05:15:34PM +0100, Ard Biesheuvel wrote: > On Wed, 4 Jan 2023 at 17:13, Mark Rutland <mark.rutland@arm.com> wrote: > > > > On Wed, Jan 04, 2023 at 02:56:19PM +0100, Ard Biesheuvel wrote: > > > On Wed, 4 Jan 2023 at 11:40, Lee Jones <lee@kernel.org> wrote: > > > > > > > > On Mon, 05 Dec 2022, Ard Biesheuvel wrote: > > > > > > > > > With the introduction of PRMT in the ACPI subsystem, the EFI rts > > > > > workqueue is no longer the only caller of efi_call_virt_pointer() in the > > > > > kernel. This means the EFI runtime services lock is no longer sufficient > > > > > to manage concurrent calls into firmware, but also that firmware calls > > > > > may occur that are not marshalled via the workqueue mechanism, but > > > > > originate directly from the caller context. > > > > > > > > > > For added robustness, and to ensure that the runtime services have 8 KiB > > > > > of stack space available as per the EFI spec, introduce a spinlock > > > > > protected EFI runtime stack of 8 KiB, where the spinlock also ensures > > > > > serialization between the EFI rts workqueue (which itself serializes EFI > > > > > runtime calls) and other callers of efi_call_virt_pointer(). > > > > > > > > > > While at it, use the stack pivot to avoid reloading the shadow call > > > > > stack pointer from the ordinary stack, as doing so could produce a > > > > > gadget to defeat it. > > > > > > > > > > Signed-off-by: Ard Biesheuvel <ardb@kernel.org> > > > > > --- > > > > > arch/arm64/include/asm/efi.h | 3 +++ > > > > > arch/arm64/kernel/efi-rt-wrapper.S | 13 +++++++++- > > > > > arch/arm64/kernel/efi.c | 25 ++++++++++++++++++++ > > > > > 3 files changed, 40 insertions(+), 1 deletion(-) > > > > > > > > Could we have this in Stable please? > > > > > > > > Upstream commit: ff7a167961d1b ("arm64: efi: Execute runtime services from a dedicated stack") > > > > > > > > Ard, do we need Patch 2 as well, or can this be applied on its own? > > > > > > > > > > Thanks for the reminder. > > > > > > Only patch #1 is needed. It should be applied to v5.10 and later. > > > > Hold on, why did this go into mainline when I had an outstanding comment w.r.t. > > the stack unwinder? > > > > From your last reply to me there I was expecting a respin with that fixed. > > > > Apologies for the confusion. > > I have a patch for this queued up, but AIUI, that cannot be merged all > the way back to v5.10, so these need to remain separate changes in any > case. > > https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git/commit/?id=c2530a04a73e6b75ed71ed14d09d7b42d6300013 Ah, ok, thanks for the pointer! I'm a little uneasy here, still. By backporting this we're also backporting the new breakage of the stack unwinder, and the minimal change for backports would be to add the lock and not the new stack (which was added for additinoal robustness, not to fix the bug the lock fixes). I do appreciate that the additional stack is likely more useful than the occasional diagnostic output from the kernel, but it does seem like this has traded off one bug for another, and I'm just a little annoyed because I pointed that out before the first pull request was made. I do know that this isn't malicious, and I'm not trying to start a fight, but now we have to consider whether we want/need to backport a stack unwinder fix to account for this, and we hadn't had that discussion before. Thanks, Mark.
On Wed, 4 Jan 2023 at 17:30, Mark Rutland <mark.rutland@arm.com> wrote: > > On Wed, Jan 04, 2023 at 05:15:34PM +0100, Ard Biesheuvel wrote: > > On Wed, 4 Jan 2023 at 17:13, Mark Rutland <mark.rutland@arm.com> wrote: > > > > > > On Wed, Jan 04, 2023 at 02:56:19PM +0100, Ard Biesheuvel wrote: > > > > On Wed, 4 Jan 2023 at 11:40, Lee Jones <lee@kernel.org> wrote: > > > > > > > > > > On Mon, 05 Dec 2022, Ard Biesheuvel wrote: > > > > > > > > > > > With the introduction of PRMT in the ACPI subsystem, the EFI rts > > > > > > workqueue is no longer the only caller of efi_call_virt_pointer() in the > > > > > > kernel. This means the EFI runtime services lock is no longer sufficient > > > > > > to manage concurrent calls into firmware, but also that firmware calls > > > > > > may occur that are not marshalled via the workqueue mechanism, but > > > > > > originate directly from the caller context. > > > > > > > > > > > > For added robustness, and to ensure that the runtime services have 8 KiB > > > > > > of stack space available as per the EFI spec, introduce a spinlock > > > > > > protected EFI runtime stack of 8 KiB, where the spinlock also ensures > > > > > > serialization between the EFI rts workqueue (which itself serializes EFI > > > > > > runtime calls) and other callers of efi_call_virt_pointer(). > > > > > > > > > > > > While at it, use the stack pivot to avoid reloading the shadow call > > > > > > stack pointer from the ordinary stack, as doing so could produce a > > > > > > gadget to defeat it. > > > > > > > > > > > > Signed-off-by: Ard Biesheuvel <ardb@kernel.org> > > > > > > --- > > > > > > arch/arm64/include/asm/efi.h | 3 +++ > > > > > > arch/arm64/kernel/efi-rt-wrapper.S | 13 +++++++++- > > > > > > arch/arm64/kernel/efi.c | 25 ++++++++++++++++++++ > > > > > > 3 files changed, 40 insertions(+), 1 deletion(-) > > > > > > > > > > Could we have this in Stable please? > > > > > > > > > > Upstream commit: ff7a167961d1b ("arm64: efi: Execute runtime services from a dedicated stack") > > > > > > > > > > Ard, do we need Patch 2 as well, or can this be applied on its own? > > > > > > > > > > > > > Thanks for the reminder. > > > > > > > > Only patch #1 is needed. It should be applied to v5.10 and later. > > > > > > Hold on, why did this go into mainline when I had an outstanding comment w.r.t. > > > the stack unwinder? > > > > > > From your last reply to me there I was expecting a respin with that fixed. > > > > > > > Apologies for the confusion. > > > > I have a patch for this queued up, but AIUI, that cannot be merged all > > the way back to v5.10, so these need to remain separate changes in any > > case. > > > > https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git/commit/?id=c2530a04a73e6b75ed71ed14d09d7b42d6300013 > > Ah, ok, thanks for the pointer! > > I'm a little uneasy here, still. > > By backporting this we're also backporting the new breakage of the stack > unwinder, and the minimal change for backports would be to add the lock and not > the new stack (which was added for additinoal robustness, not to fix the bug > the lock fixes). > > I do appreciate that the additional stack is likely more useful than the > occasional diagnostic output from the kernel, but it does seem like this has > traded off one bug for another, and I'm just a little annoyed because I pointed > that out before the first pull request was made. > > I do know that this isn't malicious, and I'm not trying to start a fight, but > now we have to consider whether we want/need to backport a stack unwinder fix > to account for this, and we hadn't had that discussion before. > In that case, let's drop these backports for the time being, and collaborate on a solution that works for all of us. Greg, could you please drop these again? Thanks.
On Wed, Jan 04, 2023 at 05:32:18PM +0100, Ard Biesheuvel wrote: > On Wed, 4 Jan 2023 at 17:30, Mark Rutland <mark.rutland@arm.com> wrote: > > > > On Wed, Jan 04, 2023 at 05:15:34PM +0100, Ard Biesheuvel wrote: > > > On Wed, 4 Jan 2023 at 17:13, Mark Rutland <mark.rutland@arm.com> wrote: > > > > > > > > On Wed, Jan 04, 2023 at 02:56:19PM +0100, Ard Biesheuvel wrote: > > > > > On Wed, 4 Jan 2023 at 11:40, Lee Jones <lee@kernel.org> wrote: > > > > > > > > > > > > On Mon, 05 Dec 2022, Ard Biesheuvel wrote: > > > > > > > > > > > > > With the introduction of PRMT in the ACPI subsystem, the EFI rts > > > > > > > workqueue is no longer the only caller of efi_call_virt_pointer() in the > > > > > > > kernel. This means the EFI runtime services lock is no longer sufficient > > > > > > > to manage concurrent calls into firmware, but also that firmware calls > > > > > > > may occur that are not marshalled via the workqueue mechanism, but > > > > > > > originate directly from the caller context. > > > > > > > > > > > > > > For added robustness, and to ensure that the runtime services have 8 KiB > > > > > > > of stack space available as per the EFI spec, introduce a spinlock > > > > > > > protected EFI runtime stack of 8 KiB, where the spinlock also ensures > > > > > > > serialization between the EFI rts workqueue (which itself serializes EFI > > > > > > > runtime calls) and other callers of efi_call_virt_pointer(). > > > > > > > > > > > > > > While at it, use the stack pivot to avoid reloading the shadow call > > > > > > > stack pointer from the ordinary stack, as doing so could produce a > > > > > > > gadget to defeat it. > > > > > > > > > > > > > > Signed-off-by: Ard Biesheuvel <ardb@kernel.org> > > > > > > > --- > > > > > > > arch/arm64/include/asm/efi.h | 3 +++ > > > > > > > arch/arm64/kernel/efi-rt-wrapper.S | 13 +++++++++- > > > > > > > arch/arm64/kernel/efi.c | 25 ++++++++++++++++++++ > > > > > > > 3 files changed, 40 insertions(+), 1 deletion(-) > > > > > > > > > > > > Could we have this in Stable please? > > > > > > > > > > > > Upstream commit: ff7a167961d1b ("arm64: efi: Execute runtime services from a dedicated stack") > > > > > > > > > > > > Ard, do we need Patch 2 as well, or can this be applied on its own? > > > > > > > > > > > > > > > > Thanks for the reminder. > > > > > > > > > > Only patch #1 is needed. It should be applied to v5.10 and later. > > > > > > > > Hold on, why did this go into mainline when I had an outstanding comment w.r.t. > > > > the stack unwinder? > > > > > > > > From your last reply to me there I was expecting a respin with that fixed. > > > > > > > > > > Apologies for the confusion. > > > > > > I have a patch for this queued up, but AIUI, that cannot be merged all > > > the way back to v5.10, so these need to remain separate changes in any > > > case. > > > > > > https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git/commit/?id=c2530a04a73e6b75ed71ed14d09d7b42d6300013 > > > > Ah, ok, thanks for the pointer! > > > > I'm a little uneasy here, still. > > > > By backporting this we're also backporting the new breakage of the stack > > unwinder, and the minimal change for backports would be to add the lock and not > > the new stack (which was added for additinoal robustness, not to fix the bug > > the lock fixes). > > > > I do appreciate that the additional stack is likely more useful than the > > occasional diagnostic output from the kernel, but it does seem like this has > > traded off one bug for another, and I'm just a little annoyed because I pointed > > that out before the first pull request was made. > > > > I do know that this isn't malicious, and I'm not trying to start a fight, but > > now we have to consider whether we want/need to backport a stack unwinder fix > > to account for this, and we hadn't had that discussion before. > > > > In that case, let's drop these backports for the time being, and > collaborate on a solution that works for all of us. > > Greg, could you please drop these again? Thanks. Dropped now from all queues, thanks. greg k-h
On Wed, Jan 04, 2023 at 05:32:18PM +0100, Ard Biesheuvel wrote: > On Wed, 4 Jan 2023 at 17:30, Mark Rutland <mark.rutland@arm.com> wrote: > > > > On Wed, Jan 04, 2023 at 05:15:34PM +0100, Ard Biesheuvel wrote: > > > On Wed, 4 Jan 2023 at 17:13, Mark Rutland <mark.rutland@arm.com> wrote: > > > > > > > > On Wed, Jan 04, 2023 at 02:56:19PM +0100, Ard Biesheuvel wrote: > > > > > On Wed, 4 Jan 2023 at 11:40, Lee Jones <lee@kernel.org> wrote: > > > > > > > > > > > > On Mon, 05 Dec 2022, Ard Biesheuvel wrote: > > > > > > > > > > > > > With the introduction of PRMT in the ACPI subsystem, the EFI rts > > > > > > > workqueue is no longer the only caller of efi_call_virt_pointer() in the > > > > > > > kernel. This means the EFI runtime services lock is no longer sufficient > > > > > > > to manage concurrent calls into firmware, but also that firmware calls > > > > > > > may occur that are not marshalled via the workqueue mechanism, but > > > > > > > originate directly from the caller context. > > > > > > > > > > > > > > For added robustness, and to ensure that the runtime services have 8 KiB > > > > > > > of stack space available as per the EFI spec, introduce a spinlock > > > > > > > protected EFI runtime stack of 8 KiB, where the spinlock also ensures > > > > > > > serialization between the EFI rts workqueue (which itself serializes EFI > > > > > > > runtime calls) and other callers of efi_call_virt_pointer(). > > > > > > > > > > > > > > While at it, use the stack pivot to avoid reloading the shadow call > > > > > > > stack pointer from the ordinary stack, as doing so could produce a > > > > > > > gadget to defeat it. > > > > > > > > > > > > > > Signed-off-by: Ard Biesheuvel <ardb@kernel.org> > > > > > > > --- > > > > > > > arch/arm64/include/asm/efi.h | 3 +++ > > > > > > > arch/arm64/kernel/efi-rt-wrapper.S | 13 +++++++++- > > > > > > > arch/arm64/kernel/efi.c | 25 ++++++++++++++++++++ > > > > > > > 3 files changed, 40 insertions(+), 1 deletion(-) > > > > > > > > > > > > Could we have this in Stable please? > > > > > > > > > > > > Upstream commit: ff7a167961d1b ("arm64: efi: Execute runtime services from a dedicated stack") > > > > > > > > > > > > Ard, do we need Patch 2 as well, or can this be applied on its own? > > > > > > > > > > > > > > > > Thanks for the reminder. > > > > > > > > > > Only patch #1 is needed. It should be applied to v5.10 and later. > > > > > > > > Hold on, why did this go into mainline when I had an outstanding comment w.r.t. > > > > the stack unwinder? > > > > > > > > From your last reply to me there I was expecting a respin with that fixed. > > > > > > > > > > Apologies for the confusion. > > > > > > I have a patch for this queued up, but AIUI, that cannot be merged all > > > the way back to v5.10, so these need to remain separate changes in any > > > case. > > > > > > https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git/commit/?id=c2530a04a73e6b75ed71ed14d09d7b42d6300013 > > > > Ah, ok, thanks for the pointer! > > > > I'm a little uneasy here, still. > > > > By backporting this we're also backporting the new breakage of the stack > > unwinder, and the minimal change for backports would be to add the lock and not > > the new stack (which was added for additinoal robustness, not to fix the bug > > the lock fixes). > > > > I do appreciate that the additional stack is likely more useful than the > > occasional diagnostic output from the kernel, but it does seem like this has > > traded off one bug for another, and I'm just a little annoyed because I pointed > > that out before the first pull request was made. > > > > I do know that this isn't malicious, and I'm not trying to start a fight, but > > now we have to consider whether we want/need to backport a stack unwinder fix > > to account for this, and we hadn't had that discussion before. > > In that case, let's drop these backports for the time being, and > collaborate on a solution that works for all of us. Thanks! IIUC our options here are: 1) Create a cut-down patch for stable that just adds the new lock but leaves out the new stack. I may be missing a reason why that's insufficient or painful. 2) Backport this *but* also backport the follow-up fixes from your other series: https://lore.kernel.org/r/20230104174433.1259428-1-ardb@kernel.org Above you mentioned something about v5.10, was that just to say that some manual backporting was required, or that there was a structural problem that would require more invasive changes / prerequisites? 3) Something else? My preference would be (1), but if we are encountering issue with stack size on stable kernels, then I'd be happy to help with manual backporting effort for (2), as long as we backported all the relevant bits in one go. Does that make sense, and does that sound reasonable to you? Thanks, Mark.
On Thu, 5 Jan 2023 at 13:56, Mark Rutland <mark.rutland@arm.com> wrote: > > On Wed, Jan 04, 2023 at 05:32:18PM +0100, Ard Biesheuvel wrote: > > On Wed, 4 Jan 2023 at 17:30, Mark Rutland <mark.rutland@arm.com> wrote: > > > > > > On Wed, Jan 04, 2023 at 05:15:34PM +0100, Ard Biesheuvel wrote: > > > > On Wed, 4 Jan 2023 at 17:13, Mark Rutland <mark.rutland@arm.com> wrote: > > > > > > > > > > On Wed, Jan 04, 2023 at 02:56:19PM +0100, Ard Biesheuvel wrote: > > > > > > On Wed, 4 Jan 2023 at 11:40, Lee Jones <lee@kernel.org> wrote: > > > > > > > > > > > > > > On Mon, 05 Dec 2022, Ard Biesheuvel wrote: > > > > > > > > > > > > > > > With the introduction of PRMT in the ACPI subsystem, the EFI rts > > > > > > > > workqueue is no longer the only caller of efi_call_virt_pointer() in the > > > > > > > > kernel. This means the EFI runtime services lock is no longer sufficient > > > > > > > > to manage concurrent calls into firmware, but also that firmware calls > > > > > > > > may occur that are not marshalled via the workqueue mechanism, but > > > > > > > > originate directly from the caller context. > > > > > > > > > > > > > > > > For added robustness, and to ensure that the runtime services have 8 KiB > > > > > > > > of stack space available as per the EFI spec, introduce a spinlock > > > > > > > > protected EFI runtime stack of 8 KiB, where the spinlock also ensures > > > > > > > > serialization between the EFI rts workqueue (which itself serializes EFI > > > > > > > > runtime calls) and other callers of efi_call_virt_pointer(). > > > > > > > > > > > > > > > > While at it, use the stack pivot to avoid reloading the shadow call > > > > > > > > stack pointer from the ordinary stack, as doing so could produce a > > > > > > > > gadget to defeat it. > > > > > > > > > > > > > > > > Signed-off-by: Ard Biesheuvel <ardb@kernel.org> > > > > > > > > --- > > > > > > > > arch/arm64/include/asm/efi.h | 3 +++ > > > > > > > > arch/arm64/kernel/efi-rt-wrapper.S | 13 +++++++++- > > > > > > > > arch/arm64/kernel/efi.c | 25 ++++++++++++++++++++ > > > > > > > > 3 files changed, 40 insertions(+), 1 deletion(-) > > > > > > > > > > > > > > Could we have this in Stable please? > > > > > > > > > > > > > > Upstream commit: ff7a167961d1b ("arm64: efi: Execute runtime services from a dedicated stack") > > > > > > > > > > > > > > Ard, do we need Patch 2 as well, or can this be applied on its own? > > > > > > > > > > > > > > > > > > > Thanks for the reminder. > > > > > > > > > > > > Only patch #1 is needed. It should be applied to v5.10 and later. > > > > > > > > > > Hold on, why did this go into mainline when I had an outstanding comment w.r.t. > > > > > the stack unwinder? > > > > > > > > > > From your last reply to me there I was expecting a respin with that fixed. > > > > > > > > > > > > > Apologies for the confusion. > > > > > > > > I have a patch for this queued up, but AIUI, that cannot be merged all > > > > the way back to v5.10, so these need to remain separate changes in any > > > > case. > > > > > > > > https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git/commit/?id=c2530a04a73e6b75ed71ed14d09d7b42d6300013 > > > > > > Ah, ok, thanks for the pointer! > > > > > > I'm a little uneasy here, still. > > > > > > By backporting this we're also backporting the new breakage of the stack > > > unwinder, and the minimal change for backports would be to add the lock and not > > > the new stack (which was added for additinoal robustness, not to fix the bug > > > the lock fixes). > > > > > > I do appreciate that the additional stack is likely more useful than the > > > occasional diagnostic output from the kernel, but it does seem like this has > > > traded off one bug for another, and I'm just a little annoyed because I pointed > > > that out before the first pull request was made. > > > > > > I do know that this isn't malicious, and I'm not trying to start a fight, but > > > now we have to consider whether we want/need to backport a stack unwinder fix > > > to account for this, and we hadn't had that discussion before. > > > > In that case, let's drop these backports for the time being, and > > collaborate on a solution that works for all of us. > > Thanks! > > IIUC our options here are: > > 1) Create a cut-down patch for stable that just adds the new lock but leaves > out the new stack. > The lock by itself does nothing useful - it is only needed because there is now a single stack that is shared by all callers of EFI runtime code. The existing ones are serialized already, but ACPI may invoke efi_call_virt_pointer() as well, which is why the additional spinlock is required for completeness. However, that ACPI feature is relatively recent (and I am not aware of any arm64 systems that actually implement it in their firmware) > I may be missing a reason why that's insufficient or painful. > The reason for the backport is that it also allows us to stash the shadow call stack pointer elsewhere, as storing it on the normal stack as we do currently defeats the purpose. > 2) Backport this *but* also backport the follow-up fixes from your other series: > https://lore.kernel.org/r/20230104174433.1259428-1-ardb@kernel.org > > Above you mentioned something about v5.10, was that just to say that some > manual backporting was required, or that there was a structural problem that > would require more invasive changes / prerequisites? > This is when the shadow call stack was introduced. Apologies for not making this clearer in the commit log. > 3) Something else? > > My preference would be (1), but if we are encountering issue with stack size on > stable kernels, then I'd be happy to help with manual backporting effort for > (2), as long as we backported all the relevant bits in one go. > > Does that make sense, and does that sound reasonable to you? > What I had in mind (but did not communicate clearly) is to backport the patch that introduces the new stack to v5.10 and later, and to backport the patch that adds the stacktrace declaration as well, but separately (and probably not all the way back to v5.10)
On Thu, 05 Jan 2023, Greg KH wrote: > On Wed, Jan 04, 2023 at 05:32:18PM +0100, Ard Biesheuvel wrote: > > On Wed, 4 Jan 2023 at 17:30, Mark Rutland <mark.rutland@arm.com> wrote: > > > > > > On Wed, Jan 04, 2023 at 05:15:34PM +0100, Ard Biesheuvel wrote: > > > > On Wed, 4 Jan 2023 at 17:13, Mark Rutland <mark.rutland@arm.com> wrote: > > > > > > > > > > On Wed, Jan 04, 2023 at 02:56:19PM +0100, Ard Biesheuvel wrote: > > > > > > On Wed, 4 Jan 2023 at 11:40, Lee Jones <lee@kernel.org> wrote: > > > > > > > > > > > > > > On Mon, 05 Dec 2022, Ard Biesheuvel wrote: > > > > > > > > > > > > > > > With the introduction of PRMT in the ACPI subsystem, the EFI rts > > > > > > > > workqueue is no longer the only caller of efi_call_virt_pointer() in the > > > > > > > > kernel. This means the EFI runtime services lock is no longer sufficient > > > > > > > > to manage concurrent calls into firmware, but also that firmware calls > > > > > > > > may occur that are not marshalled via the workqueue mechanism, but > > > > > > > > originate directly from the caller context. > > > > > > > > > > > > > > > > For added robustness, and to ensure that the runtime services have 8 KiB > > > > > > > > of stack space available as per the EFI spec, introduce a spinlock > > > > > > > > protected EFI runtime stack of 8 KiB, where the spinlock also ensures > > > > > > > > serialization between the EFI rts workqueue (which itself serializes EFI > > > > > > > > runtime calls) and other callers of efi_call_virt_pointer(). > > > > > > > > > > > > > > > > While at it, use the stack pivot to avoid reloading the shadow call > > > > > > > > stack pointer from the ordinary stack, as doing so could produce a > > > > > > > > gadget to defeat it. > > > > > > > > > > > > > > > > Signed-off-by: Ard Biesheuvel <ardb@kernel.org> > > > > > > > > --- > > > > > > > > arch/arm64/include/asm/efi.h | 3 +++ > > > > > > > > arch/arm64/kernel/efi-rt-wrapper.S | 13 +++++++++- > > > > > > > > arch/arm64/kernel/efi.c | 25 ++++++++++++++++++++ > > > > > > > > 3 files changed, 40 insertions(+), 1 deletion(-) > > > > > > > > > > > > > > Could we have this in Stable please? > > > > > > > > > > > > > > Upstream commit: ff7a167961d1b ("arm64: efi: Execute runtime services from a dedicated stack") > > > > > > > > > > > > > > Ard, do we need Patch 2 as well, or can this be applied on its own? > > > > > > > > > > > > > > > > > > > Thanks for the reminder. > > > > > > > > > > > > Only patch #1 is needed. It should be applied to v5.10 and later. > > > > > > > > > > Hold on, why did this go into mainline when I had an outstanding comment w.r.t. > > > > > the stack unwinder? > > > > > > > > > > From your last reply to me there I was expecting a respin with that fixed. > > > > > > > > > > > > > Apologies for the confusion. > > > > > > > > I have a patch for this queued up, but AIUI, that cannot be merged all > > > > the way back to v5.10, so these need to remain separate changes in any > > > > case. > > > > > > > > https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git/commit/?id=c2530a04a73e6b75ed71ed14d09d7b42d6300013 > > > > > > Ah, ok, thanks for the pointer! > > > > > > I'm a little uneasy here, still. > > > > > > By backporting this we're also backporting the new breakage of the stack > > > unwinder, and the minimal change for backports would be to add the lock and not > > > the new stack (which was added for additinoal robustness, not to fix the bug > > > the lock fixes). > > > > > > I do appreciate that the additional stack is likely more useful than the > > > occasional diagnostic output from the kernel, but it does seem like this has > > > traded off one bug for another, and I'm just a little annoyed because I pointed > > > that out before the first pull request was made. > > > > > > I do know that this isn't malicious, and I'm not trying to start a fight, but > > > now we have to consider whether we want/need to backport a stack unwinder fix > > > to account for this, and we hadn't had that discussion before. > > > > > > > In that case, let's drop these backports for the time being, and > > collaborate on a solution that works for all of us. > > > > Greg, could you please drop these again? Thanks. > > Dropped now from all queues, thanks. Now in Mainline as: 18bba1843fc7f efi: rt-wrapper: Add missing include ff7a167961d1b arm64: efi: Execute runtime services from a dedicated stack Would you be kind enough to re-collect them please? Thank you.
On Tue, Jan 17, 2023 at 04:56:19PM +0000, Lee Jones wrote: > On Thu, 05 Jan 2023, Greg KH wrote: > > > On Wed, Jan 04, 2023 at 05:32:18PM +0100, Ard Biesheuvel wrote: > > > On Wed, 4 Jan 2023 at 17:30, Mark Rutland <mark.rutland@arm.com> wrote: > > > > > > > > On Wed, Jan 04, 2023 at 05:15:34PM +0100, Ard Biesheuvel wrote: > > > > > On Wed, 4 Jan 2023 at 17:13, Mark Rutland <mark.rutland@arm.com> wrote: > > > > > > > > > > > > On Wed, Jan 04, 2023 at 02:56:19PM +0100, Ard Biesheuvel wrote: > > > > > > > On Wed, 4 Jan 2023 at 11:40, Lee Jones <lee@kernel.org> wrote: > > > > > > > > > > > > > > > > On Mon, 05 Dec 2022, Ard Biesheuvel wrote: > > > > > > > > > > > > > > > > > With the introduction of PRMT in the ACPI subsystem, the EFI rts > > > > > > > > > workqueue is no longer the only caller of efi_call_virt_pointer() in the > > > > > > > > > kernel. This means the EFI runtime services lock is no longer sufficient > > > > > > > > > to manage concurrent calls into firmware, but also that firmware calls > > > > > > > > > may occur that are not marshalled via the workqueue mechanism, but > > > > > > > > > originate directly from the caller context. > > > > > > > > > > > > > > > > > > For added robustness, and to ensure that the runtime services have 8 KiB > > > > > > > > > of stack space available as per the EFI spec, introduce a spinlock > > > > > > > > > protected EFI runtime stack of 8 KiB, where the spinlock also ensures > > > > > > > > > serialization between the EFI rts workqueue (which itself serializes EFI > > > > > > > > > runtime calls) and other callers of efi_call_virt_pointer(). > > > > > > > > > > > > > > > > > > While at it, use the stack pivot to avoid reloading the shadow call > > > > > > > > > stack pointer from the ordinary stack, as doing so could produce a > > > > > > > > > gadget to defeat it. > > > > > > > > > > > > > > > > > > Signed-off-by: Ard Biesheuvel <ardb@kernel.org> > > > > > > > > > --- > > > > > > > > > arch/arm64/include/asm/efi.h | 3 +++ > > > > > > > > > arch/arm64/kernel/efi-rt-wrapper.S | 13 +++++++++- > > > > > > > > > arch/arm64/kernel/efi.c | 25 ++++++++++++++++++++ > > > > > > > > > 3 files changed, 40 insertions(+), 1 deletion(-) > > > > > > > > > > > > > > > > Could we have this in Stable please? > > > > > > > > > > > > > > > > Upstream commit: ff7a167961d1b ("arm64: efi: Execute runtime services from a dedicated stack") > > > > > > > > > > > > > > > > Ard, do we need Patch 2 as well, or can this be applied on its own? > > > > > > > > > > > > > > > > > > > > > > Thanks for the reminder. > > > > > > > > > > > > > > Only patch #1 is needed. It should be applied to v5.10 and later. > > > > > > > > > > > > Hold on, why did this go into mainline when I had an outstanding comment w.r.t. > > > > > > the stack unwinder? > > > > > > > > > > > > From your last reply to me there I was expecting a respin with that fixed. > > > > > > > > > > > > > > > > Apologies for the confusion. > > > > > > > > > > I have a patch for this queued up, but AIUI, that cannot be merged all > > > > > the way back to v5.10, so these need to remain separate changes in any > > > > > case. > > > > > > > > > > https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git/commit/?id=c2530a04a73e6b75ed71ed14d09d7b42d6300013 > > > > > > > > Ah, ok, thanks for the pointer! > > > > > > > > I'm a little uneasy here, still. > > > > > > > > By backporting this we're also backporting the new breakage of the stack > > > > unwinder, and the minimal change for backports would be to add the lock and not > > > > the new stack (which was added for additinoal robustness, not to fix the bug > > > > the lock fixes). > > > > > > > > I do appreciate that the additional stack is likely more useful than the > > > > occasional diagnostic output from the kernel, but it does seem like this has > > > > traded off one bug for another, and I'm just a little annoyed because I pointed > > > > that out before the first pull request was made. > > > > > > > > I do know that this isn't malicious, and I'm not trying to start a fight, but > > > > now we have to consider whether we want/need to backport a stack unwinder fix > > > > to account for this, and we hadn't had that discussion before. > > > > > > > > > > In that case, let's drop these backports for the time being, and > > > collaborate on a solution that works for all of us. > > > > > > Greg, could you please drop these again? Thanks. > > > > Dropped now from all queues, thanks. > > Now in Mainline as: > > 18bba1843fc7f efi: rt-wrapper: Add missing include > ff7a167961d1b arm64: efi: Execute runtime services from a dedicated stack > > Would you be kind enough to re-collect them please? Now queued up for 5.10.y and newer. greg k-h
diff --git a/arch/arm64/include/asm/efi.h b/arch/arm64/include/asm/efi.h index 7c12e01c2b312e7b..1c408ec3c8b3a883 100644 --- a/arch/arm64/include/asm/efi.h +++ b/arch/arm64/include/asm/efi.h @@ -25,6 +25,7 @@ int efi_set_mapping_permissions(struct mm_struct *mm, efi_memory_desc_t *md); ({ \ efi_virtmap_load(); \ __efi_fpsimd_begin(); \ + spin_lock(&efi_rt_lock); \ }) #undef arch_efi_call_virt @@ -33,10 +34,12 @@ int efi_set_mapping_permissions(struct mm_struct *mm, efi_memory_desc_t *md); #define arch_efi_call_virt_teardown() \ ({ \ + spin_unlock(&efi_rt_lock); \ __efi_fpsimd_end(); \ efi_virtmap_unload(); \ }) +extern spinlock_t efi_rt_lock; efi_status_t __efi_rt_asm_wrapper(void *, const char *, ...); #define ARCH_EFI_IRQ_FLAGS_MASK (PSR_D_BIT | PSR_A_BIT | PSR_I_BIT | PSR_F_BIT) diff --git a/arch/arm64/kernel/efi-rt-wrapper.S b/arch/arm64/kernel/efi-rt-wrapper.S index 75691a2641c1c0f8..b2786b968fee68dd 100644 --- a/arch/arm64/kernel/efi-rt-wrapper.S +++ b/arch/arm64/kernel/efi-rt-wrapper.S @@ -16,6 +16,12 @@ SYM_FUNC_START(__efi_rt_asm_wrapper) */ stp x1, x18, [sp, #16] + ldr_l x16, efi_rt_stack_top + mov sp, x16 +#ifdef CONFIG_SHADOW_CALL_STACK + str x18, [sp, #-16]! +#endif + /* * We are lucky enough that no EFI runtime services take more than * 5 arguments, so all are passed in registers rather than via the @@ -29,6 +35,7 @@ SYM_FUNC_START(__efi_rt_asm_wrapper) mov x4, x6 blr x8 + mov sp, x29 ldp x1, x2, [sp, #16] cmp x2, x18 ldp x29, x30, [sp], #32 @@ -42,6 +49,10 @@ SYM_FUNC_START(__efi_rt_asm_wrapper) * called with preemption disabled and a separate shadow stack is used * for interrupts. */ - mov x18, x2 +#ifdef CONFIG_SHADOW_CALL_STACK + ldr_l x18, efi_rt_stack_top + ldr x18, [x18, #-16] +#endif + b efi_handle_corrupted_x18 // tail call SYM_FUNC_END(__efi_rt_asm_wrapper) diff --git a/arch/arm64/kernel/efi.c b/arch/arm64/kernel/efi.c index a908a37f03678b6b..8cb2e005f8aca589 100644 --- a/arch/arm64/kernel/efi.c +++ b/arch/arm64/kernel/efi.c @@ -144,3 +144,28 @@ asmlinkage efi_status_t efi_handle_corrupted_x18(efi_status_t s, const char *f) pr_err_ratelimited(FW_BUG "register x18 corrupted by EFI %s\n", f); return s; } + +DEFINE_SPINLOCK(efi_rt_lock); + +asmlinkage u64 *efi_rt_stack_top __ro_after_init; + +/* required by the EFI spec */ +static_assert(THREAD_SIZE >= SZ_8K); + +int __init arm64_efi_rt_init(void) +{ + void *p = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN, + VMALLOC_START, VMALLOC_END, GFP_KERNEL, + PAGE_KERNEL, 0, NUMA_NO_NODE, + __builtin_return_address(0)); + + if (!p) { + pr_warn("Failed to allocate EFI runtime stack\n"); + clear_bit(EFI_RUNTIME_SERVICES, &efi.flags); + return -ENOMEM; + } + + efi_rt_stack_top = p + THREAD_SIZE; + return 0; +} +core_initcall(arm64_efi_rt_init);
With the introduction of PRMT in the ACPI subsystem, the EFI rts workqueue is no longer the only caller of efi_call_virt_pointer() in the kernel. This means the EFI runtime services lock is no longer sufficient to manage concurrent calls into firmware, but also that firmware calls may occur that are not marshalled via the workqueue mechanism, but originate directly from the caller context. For added robustness, and to ensure that the runtime services have 8 KiB of stack space available as per the EFI spec, introduce a spinlock protected EFI runtime stack of 8 KiB, where the spinlock also ensures serialization between the EFI rts workqueue (which itself serializes EFI runtime calls) and other callers of efi_call_virt_pointer(). While at it, use the stack pivot to avoid reloading the shadow call stack pointer from the ordinary stack, as doing so could produce a gadget to defeat it. Signed-off-by: Ard Biesheuvel <ardb@kernel.org> --- arch/arm64/include/asm/efi.h | 3 +++ arch/arm64/kernel/efi-rt-wrapper.S | 13 +++++++++- arch/arm64/kernel/efi.c | 25 ++++++++++++++++++++ 3 files changed, 40 insertions(+), 1 deletion(-)