The XSAVE init code initializes all enabled and supported components with XRSTOR(S) to init state. Then it XSAVEs the state of the components back into init_fpstate which is used in several places to fill in the init state of components. This works correctly with XSAVE, but not with XSAVEOPT and XSAVES because those use the init optimization and skip writing state of components which are in init state. So init_fpstate.xsave still contains all zeroes after this operation. There are two ways to solve that: 1) Use XSAVE unconditionally, but that requires to reshuffle the buffer when XSAVES is enabled because XSAVES uses compacted format. 2) Save the components which are known to have a non-zero init state by other means. Looking deeper, #2 is the right thing to do because all components the kernel supports have all-zeroes init state except the legacy features (FP, SSE). Those cannot be hard coded because the states are not identical on all CPUs, but they can be saved with FXSAVE which avoids all conditionals. Use FXSAVE to save the legacy FP/SSE components in init_fpstate along with a BUILD_BUG_ON() which reminds developers to validate that a newly added component has all zeroes init state. As a bonus remove the now unused copy_xregs_to_kernel_booting() crutch. The XSAVE and reshuffle method can still be implemented in the unlikely case that components are added which have a non-zero init state and no other means to save them. For now, FXSAVE is just simple and good enough. [ bp: Fix a typo or two in the text. ]
Unified · Split
| @@ -204,6 +204,14 @@ static inline void copy_fxregs_to_kernel(struct fpu *fpu) · 203 unmodified lines | |||
| 204 | asm volatile("fxsaveq %[fx]" : [fx] "=m" (fpu->state.fxsave)); | 204 | asm volatile("fxsaveq %[fx]" : [fx] "=m" (fpu->state.fxsave)); |
| 205 | } | 205 | } |
| 206 | 206 | ||
| 207 | static inline void fxsave(struct fxregs_state *fx) | ||
| 208 | { | ||
| 209 | if (IS_ENABLED(CONFIG_X86_32)) | ||
| 210 | asm volatile( "fxsave %[fx]" : [fx] "=m" (*fx)); | ||
| 211 | else | ||
| 212 | asm volatile("fxsaveq %[fx]" : [fx] "=m" (*fx)); | ||
| 213 | } | ||
| 214 | |||
| 207 | /* These macros all use (%edi)/(%rdi) as the single memory argument. */ | 215 | /* These macros all use (%edi)/(%rdi) as the single memory argument. */ |
| 208 | #define XSAVE ".byte " REX_PREFIX "0x0f,0xae,0x27" | 216 | #define XSAVE ".byte " REX_PREFIX "0x0f,0xae,0x27" |
| 209 | #define XSAVEOPT ".byte " REX_PREFIX "0x0f,0xae,0x37" | 217 | #define XSAVEOPT ".byte " REX_PREFIX "0x0f,0xae,0x37" |
| @@ -268,28 +276,6 @@ static inline void copy_fxregs_to_kernel(struct fpu *fpu) · 58 unmodified lines | |||
| 268 | : "D" (st), "m" (*st), "a" (lmask), "d" (hmask) \ | 276 | : "D" (st), "m" (*st), "a" (lmask), "d" (hmask) \ |
| 269 | : "memory") | 277 | : "memory") |
| 270 | 278 | ||
| 271 | /* | ||
| 272 | * This function is called only during boot time when x86 caps are not set | ||
| 273 | * up and alternative can not be used yet. | ||
| 274 | */ | ||
| 275 | static inline void copy_xregs_to_kernel_booting(struct xregs_state *xstate) | ||
| 276 | { | ||
| 277 | u64 mask = xfeatures_mask_all; | ||
| 278 | u32 lmask = mask; | ||
| 279 | u32 hmask = mask >> 32; | ||
| 280 | int err; | ||
| 281 | |||
| 282 | WARN_ON(system_state != SYSTEM_BOOTING); | ||
| 283 | |||
| 284 | if (boot_cpu_has(X86_FEATURE_XSAVES)) | ||
| 285 | XSTATE_OP(XSAVES, xstate, lmask, hmask, err); | ||
| 286 | else | ||
| 287 | XSTATE_OP(XSAVE, xstate, lmask, hmask, err); | ||
| 288 | |||
| 289 | /* We should never fault when copying to a kernel buffer: */ | ||
| 290 | WARN_ON_FPU(err); | ||
| 291 | } | ||
| 292 | |||
| 293 | /* | 279 | /* |
| 294 | * This function is called only during boot time when x86 caps are not set | 280 | * This function is called only during boot time when x86 caps are not set |
| 295 | * up and alternative can not be used yet. | 281 | * up and alternative can not be used yet. |
| @@ -440,6 +440,25 @@ static void __init print_xstate_offset_size(void) · 439 unmodified lines | |||
| 440 | } | 440 | } |
| 441 | } | 441 | } |
| 442 | 442 | ||
| 443 | /* | ||
| 444 | * All supported features have either init state all zeros or are | ||
| 445 | * handled in setup_init_fpu() individually. This is an explicit | ||
| 446 | * feature list and does not use XFEATURE_MASK*SUPPORTED to catch | ||
| 447 | * newly added supported features at build time and make people | ||
| 448 | * actually look at the init state for the new feature. | ||
| 449 | */ | ||
| 450 | #define XFEATURES_INIT_FPSTATE_HANDLED \ | ||
| 451 | (XFEATURE_MASK_FP | \ | ||
| 452 | XFEATURE_MASK_SSE | \ | ||
| 453 | XFEATURE_MASK_YMM | \ | ||
| 454 | XFEATURE_MASK_OPMASK | \ | ||
| 455 | XFEATURE_MASK_ZMM_Hi256 | \ | ||
| 456 | XFEATURE_MASK_Hi16_ZMM | \ | ||
| 457 | XFEATURE_MASK_PKRU | \ | ||
| 458 | XFEATURE_MASK_BNDREGS | \ | ||
| 459 | XFEATURE_MASK_BNDCSR | \ | ||
| 460 | XFEATURE_MASK_PASID) | ||
| 461 | |||
| 443 | /* | 462 | /* |
| 444 | * setup the xstate image representing the init state | 463 | * setup the xstate image representing the init state |
| 445 | */ | 464 | */ |
| @@ -447,6 +466,10 @@ static void __init setup_init_fpu_buf(void) · 1 unmodified line | |||
| 447 | { | 466 | { |
| 448 | static int on_boot_cpu __initdata = 1; | 467 | static int on_boot_cpu __initdata = 1; |
| 449 | 468 | ||
| 469 | BUILD_BUG_ON((XFEATURE_MASK_USER_SUPPORTED | | ||
| 470 | XFEATURE_MASK_SUPERVISOR_SUPPORTED) != | ||
| 471 | XFEATURES_INIT_FPSTATE_HANDLED); | ||
| 472 | |||
| 450 | WARN_ON_FPU(!on_boot_cpu); | 473 | WARN_ON_FPU(!on_boot_cpu); |
| 451 | on_boot_cpu = 0; | 474 | on_boot_cpu = 0; |
| 452 | 475 | ||
| @@ -466,10 +489,22 @@ static void __init setup_init_fpu_buf(void) · 13 unmodified lines | |||
| 466 | copy_kernel_to_xregs_booting(&init_fpstate.xsave); | 489 | copy_kernel_to_xregs_booting(&init_fpstate.xsave); |
| 467 | 490 | ||
| 468 | /* | 491 | /* |
| 469 | * Dump the init state again. This is to identify the init state | 492 | * All components are now in init state. Read the state back so |
| 470 | * of any feature which is not represented by all zero's. | 493 | * that init_fpstate contains all non-zero init state. This only |
| 494 | * works with XSAVE, but not with XSAVEOPT and XSAVES because | ||
| 495 | * those use the init optimization which skips writing data for | ||
| 496 | * components in init state. | ||
| 497 | * | ||
| 498 | * XSAVE could be used, but that would require to reshuffle the | ||
| 499 | * data when XSAVES is available because XSAVES uses xstate | ||
| 500 | * compaction. But doing so is a pointless exercise because most | ||
| 501 | * components have an all zeros init state except for the legacy | ||
| 502 | * ones (FP and SSE). Those can be saved with FXSAVE into the | ||
| 503 | * legacy area. Adding new features requires to ensure that init | ||
| 504 | * state is all zeroes or if not to add the necessary handling | ||
| 505 | * here. | ||
| 471 | */ | 506 | */ |
| 472 | copy_xregs_to_kernel_booting(&init_fpstate.xsave); | 507 | fxsave(&init_fpstate.fxsave); |
| 473 | } | 508 | } |
| 474 | 509 | ||
| 475 | static int xfeature_uncompacted_offset(int xfeature_nr) | 510 | static int xfeature_uncompacted_offset(int xfeature_nr) |