Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
Merge branch 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip
Pull x86 fixes from Ingo Molnar:
"This fixes 3 FPU handling related bugs, an EFI boot crash and a
runtime warning.
The EFI fix arrived late but I didn't want to delay it to after v4.5
because the effects are pretty bad for the systems that are affected
by it"
[ Actually, I don't think the EFI fix really matters yet, because we
haven't switched to the separate EFI page tables in mainline yet ]
* 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86/efi: Fix boot crash by always mapping boot service regions into new EFI page tables
x86/fpu: Fix eager-FPU handling on legacy FPU machines
x86/delay: Avoid preemptible context checks in delay_mwaitx()
x86/fpu: Revert ("x86/fpu: Disable AVX when eagerfpu is off")
x86/fpu: Fix 'no387' regression
+79
-37
+4
-5
arch/x86/include/asm/fpu/xstate.h
+4
-5
arch/x86/include/asm/fpu/xstate.h
···
20
20
21
21
/* Supported features which support lazy state saving */
22
22
#define XFEATURE_MASK_LAZY (XFEATURE_MASK_FP | \
23
-
XFEATURE_MASK_SSE)
24
-
25
-
/* Supported features which require eager state saving */
26
-
#define XFEATURE_MASK_EAGER (XFEATURE_MASK_BNDREGS | \
27
-
XFEATURE_MASK_BNDCSR | \
23
+
XFEATURE_MASK_SSE | \
28
24
XFEATURE_MASK_YMM | \
29
25
XFEATURE_MASK_OPMASK | \
30
26
XFEATURE_MASK_ZMM_Hi256 | \
31
27
XFEATURE_MASK_Hi16_ZMM)
28
+
29
+
/* Supported features which require eager state saving */
30
+
#define XFEATURE_MASK_EAGER (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR)
32
31
33
32
/* All currently supported features */
34
33
#define XCNTXT_MASK (XFEATURE_MASK_LAZY | XFEATURE_MASK_EAGER)
+3
-1
arch/x86/kernel/fpu/core.c
+3
-1
arch/x86/kernel/fpu/core.c
+9
-13
arch/x86/kernel/fpu/init.c
+9
-13
arch/x86/kernel/fpu/init.c
···
78
78
cr0 &= ~(X86_CR0_TS | X86_CR0_EM);
79
79
write_cr0(cr0);
80
80
81
-
asm volatile("fninit ; fnstsw %0 ; fnstcw %1"
82
-
: "+m" (fsw), "+m" (fcw));
81
+
if (!test_bit(X86_FEATURE_FPU, (unsigned long *)cpu_caps_cleared)) {
82
+
asm volatile("fninit ; fnstsw %0 ; fnstcw %1"
83
+
: "+m" (fsw), "+m" (fcw));
83
84
84
-
if (fsw == 0 && (fcw & 0x103f) == 0x003f)
85
-
set_cpu_cap(c, X86_FEATURE_FPU);
86
-
else
87
-
clear_cpu_cap(c, X86_FEATURE_FPU);
85
+
if (fsw == 0 && (fcw & 0x103f) == 0x003f)
86
+
set_cpu_cap(c, X86_FEATURE_FPU);
87
+
else
88
+
clear_cpu_cap(c, X86_FEATURE_FPU);
89
+
}
88
90
89
91
#ifndef CONFIG_MATH_EMULATION
90
92
if (!cpu_has_fpu) {
···
134
132
* Set up the legacy init FPU context. (xstate init might overwrite this
135
133
* with a more modern format, if the CPU supports it.)
136
134
*/
137
-
fpstate_init_fxstate(&init_fpstate.fxsave);
135
+
fpstate_init(&init_fpstate);
138
136
139
137
fpu__init_system_mxcsr();
140
138
}
···
302
300
static void __init fpu__clear_eager_fpu_features(void)
303
301
{
304
302
setup_clear_cpu_cap(X86_FEATURE_MPX);
305
-
setup_clear_cpu_cap(X86_FEATURE_AVX);
306
-
setup_clear_cpu_cap(X86_FEATURE_AVX2);
307
-
setup_clear_cpu_cap(X86_FEATURE_AVX512F);
308
-
setup_clear_cpu_cap(X86_FEATURE_AVX512PF);
309
-
setup_clear_cpu_cap(X86_FEATURE_AVX512ER);
310
-
setup_clear_cpu_cap(X86_FEATURE_AVX512CD);
311
303
}
312
304
313
305
/*
+1
-1
arch/x86/lib/delay.c
+1
-1
arch/x86/lib/delay.c
···
102
102
* Use cpu_tss as a cacheline-aligned, seldomly
103
103
* accessed per-cpu variable as the monitor target.
104
104
*/
105
-
__monitorx(this_cpu_ptr(&cpu_tss), 0, 0);
105
+
__monitorx(raw_cpu_ptr(&cpu_tss), 0, 0);
106
106
107
107
/*
108
108
* AMD, like Intel, supports the EAX hint and EAX=0xf
+62
-17
arch/x86/platform/efi/quirks.c
+62
-17
arch/x86/platform/efi/quirks.c
···
131
131
EXPORT_SYMBOL_GPL(efi_query_variable_store);
132
132
133
133
/*
134
+
* Helper function for efi_reserve_boot_services() to figure out if we
135
+
* can free regions in efi_free_boot_services().
136
+
*
137
+
* Use this function to ensure we do not free regions owned by somebody
138
+
* else. We must only reserve (and then free) regions:
139
+
*
140
+
* - Not within any part of the kernel
141
+
* - Not the BIOS reserved area (E820_RESERVED, E820_NVS, etc)
142
+
*/
143
+
static bool can_free_region(u64 start, u64 size)
144
+
{
145
+
if (start + size > __pa_symbol(_text) && start <= __pa_symbol(_end))
146
+
return false;
147
+
148
+
if (!e820_all_mapped(start, start+size, E820_RAM))
149
+
return false;
150
+
151
+
return true;
152
+
}
153
+
154
+
/*
134
155
* The UEFI specification makes it clear that the operating system is free to do
135
156
* whatever it wants with boot services code after ExitBootServices() has been
136
157
* called. Ignoring this recommendation a significant bunch of EFI implementations
···
168
147
efi_memory_desc_t *md = p;
169
148
u64 start = md->phys_addr;
170
149
u64 size = md->num_pages << EFI_PAGE_SHIFT;
150
+
bool already_reserved;
171
151
172
152
if (md->type != EFI_BOOT_SERVICES_CODE &&
173
153
md->type != EFI_BOOT_SERVICES_DATA)
174
154
continue;
175
-
/* Only reserve where possible:
176
-
* - Not within any already allocated areas
177
-
* - Not over any memory area (really needed, if above?)
178
-
* - Not within any part of the kernel
179
-
* - Not the bios reserved area
180
-
*/
181
-
if ((start + size > __pa_symbol(_text)
182
-
&& start <= __pa_symbol(_end)) ||
183
-
!e820_all_mapped(start, start+size, E820_RAM) ||
184
-
memblock_is_region_reserved(start, size)) {
185
-
/* Could not reserve, skip it */
186
-
md->num_pages = 0;
187
-
memblock_dbg("Could not reserve boot range [0x%010llx-0x%010llx]\n",
188
-
start, start+size-1);
189
-
} else
155
+
156
+
already_reserved = memblock_is_region_reserved(start, size);
157
+
158
+
/*
159
+
* Because the following memblock_reserve() is paired
160
+
* with free_bootmem_late() for this region in
161
+
* efi_free_boot_services(), we must be extremely
162
+
* careful not to reserve, and subsequently free,
163
+
* critical regions of memory (like the kernel image) or
164
+
* those regions that somebody else has already
165
+
* reserved.
166
+
*
167
+
* A good example of a critical region that must not be
168
+
* freed is page zero (first 4Kb of memory), which may
169
+
* contain boot services code/data but is marked
170
+
* E820_RESERVED by trim_bios_range().
171
+
*/
172
+
if (!already_reserved) {
190
173
memblock_reserve(start, size);
174
+
175
+
/*
176
+
* If we are the first to reserve the region, no
177
+
* one else cares about it. We own it and can
178
+
* free it later.
179
+
*/
180
+
if (can_free_region(start, size))
181
+
continue;
182
+
}
183
+
184
+
/*
185
+
* We don't own the region. We must not free it.
186
+
*
187
+
* Setting this bit for a boot services region really
188
+
* doesn't make sense as far as the firmware is
189
+
* concerned, but it does provide us with a way to tag
190
+
* those regions that must not be paired with
191
+
* free_bootmem_late().
192
+
*/
193
+
md->attribute |= EFI_MEMORY_RUNTIME;
191
194
}
192
195
}
193
196
···
228
183
md->type != EFI_BOOT_SERVICES_DATA)
229
184
continue;
230
185
231
-
/* Could not reserve boot area */
232
-
if (!size)
186
+
/* Do not free, someone else owns it: */
187
+
if (md->attribute & EFI_MEMORY_RUNTIME)
233
188
continue;
234
189
235
190
free_bootmem_late(start, size);