]> git.ipfire.org Git - thirdparty/kernel/stable-queue.git/commitdiff
4.19-stable patches
authorGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Mon, 29 Apr 2019 13:36:17 +0000 (15:36 +0200)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Mon, 29 Apr 2019 13:36:17 +0000 (15:36 +0200)
added patches:
x86-fpu-don-t-export-__kernel_fpu_-begin-end.patch

queue-4.19/series
queue-4.19/x86-fpu-don-t-export-__kernel_fpu_-begin-end.patch [new file with mode: 0644]

index b33883c7ca7f696bb2a116ccd5046d8ec17513a0..9e6c750bf8ac51cc827ed2596de7b6c48eeeca33 100644 (file)
@@ -80,3 +80,4 @@ fix-aio_poll-races.patch
 x86-retpolines-raise-limit-for-generating-indirect-calls-from-switch-case.patch
 x86-retpolines-disable-switch-jump-tables-when-retpolines-are-enabled.patch
 mm-fix-warning-in-insert_pfn.patch
+x86-fpu-don-t-export-__kernel_fpu_-begin-end.patch
diff --git a/queue-4.19/x86-fpu-don-t-export-__kernel_fpu_-begin-end.patch b/queue-4.19/x86-fpu-don-t-export-__kernel_fpu_-begin-end.patch
new file mode 100644 (file)
index 0000000..b04027a
--- /dev/null
@@ -0,0 +1,130 @@
+From 12209993e98c5fa1855c467f22a24e3d5b8be205 Mon Sep 17 00:00:00 2001
+From: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
+Date: Thu, 29 Nov 2018 16:02:10 +0100
+Subject: x86/fpu: Don't export __kernel_fpu_{begin,end}()
+MIME-Version: 1.0
+Content-Type: text/plain; charset=UTF-8
+Content-Transfer-Encoding: 8bit
+
+From: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
+
+commit 12209993e98c5fa1855c467f22a24e3d5b8be205 upstream.
+
+There is one user of __kernel_fpu_begin() and before invoking it,
+it invokes preempt_disable(). So it could invoke kernel_fpu_begin()
+right away. The 32bit version of arch_efi_call_virt_setup() and
+arch_efi_call_virt_teardown() does this already.
+
+The comment above *kernel_fpu*() claims that before invoking
+__kernel_fpu_begin() preemption should be disabled and that KVM is a
+good example of doing it. Well, KVM doesn't do that since commit
+
+  f775b13eedee2 ("x86,kvm: move qemu/guest FPU switching out to vcpu_run")
+
+so it is not an example anymore.
+
+With EFI gone as the last user of __kernel_fpu_{begin|end}(), both can
+be made static and not exported anymore.
+
+Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
+Signed-off-by: Borislav Petkov <bp@suse.de>
+Reviewed-by: Rik van Riel <riel@surriel.com>
+Cc: "H. Peter Anvin" <hpa@zytor.com>
+Cc: "Jason A. Donenfeld" <Jason@zx2c4.com>
+Cc: Andy Lutomirski <luto@kernel.org>
+Cc: Ard Biesheuvel <ard.biesheuvel@linaro.org>
+Cc: Dave Hansen <dave.hansen@linux.intel.com>
+Cc: Ingo Molnar <mingo@redhat.com>
+Cc: Nicolai Stange <nstange@suse.de>
+Cc: Paolo Bonzini <pbonzini@redhat.com>
+Cc: Radim Krčmář <rkrcmar@redhat.com>
+Cc: Thomas Gleixner <tglx@linutronix.de>
+Cc: kvm ML <kvm@vger.kernel.org>
+Cc: linux-efi <linux-efi@vger.kernel.org>
+Cc: x86-ml <x86@kernel.org>
+Link: https://lkml.kernel.org/r/20181129150210.2k4mawt37ow6c2vq@linutronix.de
+Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
+
+---
+ arch/x86/include/asm/efi.h     |    6 ++----
+ arch/x86/include/asm/fpu/api.h |   15 +++++----------
+ arch/x86/kernel/fpu/core.c     |    6 ++----
+ 3 files changed, 9 insertions(+), 18 deletions(-)
+
+--- a/arch/x86/include/asm/efi.h
++++ b/arch/x86/include/asm/efi.h
+@@ -82,8 +82,7 @@ struct efi_scratch {
+ #define arch_efi_call_virt_setup()                                    \
+ ({                                                                    \
+       efi_sync_low_kernel_mappings();                                 \
+-      preempt_disable();                                              \
+-      __kernel_fpu_begin();                                           \
++      kernel_fpu_begin();                                             \
+       firmware_restrict_branch_speculation_start();                   \
+                                                                       \
+       if (!efi_enabled(EFI_OLD_MEMMAP))                               \
+@@ -99,8 +98,7 @@ struct efi_scratch {
+               efi_switch_mm(efi_scratch.prev_mm);                     \
+                                                                       \
+       firmware_restrict_branch_speculation_end();                     \
+-      __kernel_fpu_end();                                             \
+-      preempt_enable();                                               \
++      kernel_fpu_end();                                               \
+ })
+ extern void __iomem *__init efi_ioremap(unsigned long addr, unsigned long size,
+--- a/arch/x86/include/asm/fpu/api.h
++++ b/arch/x86/include/asm/fpu/api.h
+@@ -12,17 +12,12 @@
+ #define _ASM_X86_FPU_API_H
+ /*
+- * Careful: __kernel_fpu_begin/end() must be called with preempt disabled
+- * and they don't touch the preempt state on their own.
+- * If you enable preemption after __kernel_fpu_begin(), preempt notifier
+- * should call the __kernel_fpu_end() to prevent the kernel/user FPU
+- * state from getting corrupted. KVM for example uses this model.
+- *
+- * All other cases use kernel_fpu_begin/end() which disable preemption
+- * during kernel FPU usage.
++ * Use kernel_fpu_begin/end() if you intend to use FPU in kernel context. It
++ * disables preemption so be careful if you intend to use it for long periods
++ * of time.
++ * If you intend to use the FPU in softirq you need to check first with
++ * irq_fpu_usable() if it is possible.
+  */
+-extern void __kernel_fpu_begin(void);
+-extern void __kernel_fpu_end(void);
+ extern void kernel_fpu_begin(void);
+ extern void kernel_fpu_end(void);
+ extern bool irq_fpu_usable(void);
+--- a/arch/x86/kernel/fpu/core.c
++++ b/arch/x86/kernel/fpu/core.c
+@@ -93,7 +93,7 @@ bool irq_fpu_usable(void)
+ }
+ EXPORT_SYMBOL(irq_fpu_usable);
+-void __kernel_fpu_begin(void)
++static void __kernel_fpu_begin(void)
+ {
+       struct fpu *fpu = &current->thread.fpu;
+@@ -111,9 +111,8 @@ void __kernel_fpu_begin(void)
+               __cpu_invalidate_fpregs_state();
+       }
+ }
+-EXPORT_SYMBOL(__kernel_fpu_begin);
+-void __kernel_fpu_end(void)
++static void __kernel_fpu_end(void)
+ {
+       struct fpu *fpu = &current->thread.fpu;
+@@ -122,7 +121,6 @@ void __kernel_fpu_end(void)
+       kernel_fpu_enable();
+ }
+-EXPORT_SYMBOL(__kernel_fpu_end);
+ void kernel_fpu_begin(void)
+ {