FPU: Fix KABI issue for copy_user_generic_string opt patch

Signed-off-by: yuehongwu <yuehongwu@tencent.com>
Reviewed-by: caelli <caelli@tencent.com>
Signed-off-by: Jianping Liu <frankjpliu@tencent.com>

arch/x86/include/asm/fpu/internal.h

@@ -642,9 +642,10 @@ static inline void switch_kernel_fpu_prepare(struct task_struct *prev, int cpu)
 }
 
 /* Internal helper for switch_kernel_fpu_finish() and signal frame setup */
+extern unsigned long get_fpu_registers_pos(struct fpu *fpu, unsigned int neg_off);
 static inline void fpregs_restore_kernelregs(struct fpu *kfpu)
 {
-	kernel_fpu_states_restore(NULL, &kfpu->kernel_state, sizeof(kfpu->kernel_state));
+	kernel_fpu_states_restore(NULL, (void *)get_fpu_registers_pos(kfpu, MAX_FPU_CTX_SIZE), MAX_FPU_CTX_SIZE);
 }
 
 /*

arch/x86/include/asm/fpu/types.h

@@ -300,9 +300,6 @@ struct fpu {
 	 */
 	unsigned long			avx512_timestamp;
 
-#ifdef CONFIG_USING_FPU_IN_KERNEL_NONATOMIC
-	union fpregs_state              kernel_state;
-#endif
 	/*
 	 * @state:
 	 *

arch/x86/include/asm/uaccess_64.h

@@ -40,6 +40,9 @@ copy_user_sse2_opt_string(void *to, const void *from, unsigned len);
 
 #define copy_user_large_memory_generic_string  copy_user_sse2_opt_string
 
+#define MAX_FPU_CTX_SIZE                64
+#define KERNEL_FPU_NONATOMIC_SIZE       (2*(MAX_FPU_CTX_SIZE))
+
 #endif //CONFIG_USING_SSE2_FOR_LARGE_MEMORY_COPY
 
 #ifdef CONFIG_USING_AVX2_FOR_LARGE_MEMORY_COPY
@@ -54,6 +57,9 @@ __must_check unsigned long
 copy_user_avx2_pf64_nt_string(void *to, const void *from, unsigned len);
 
 #define copy_user_large_memory_generic_stirng  copy_user_avx2_pf64_nt_string
+#define MAX_FPU_CTX_SIZE		256
+#define KERNEL_FPU_NONATOMIC_SIZE	(2*(MAX_FPU_CTX_SIZE))
+
 #endif //NO DEFINE CONFIG_USING_SSE2_FOR_LARGE_MEMORY_COPY
 #endif //CONFIG_USING_AVX2_FOR_LAGRE_MEMORY_COPY
 unsigned long get_nt_block_copy_mini_len(void);

arch/x86/kernel/fpu/core.c

@@ -137,6 +137,25 @@ void kernel_fpu_end(void)
 EXPORT_SYMBOL_GPL(kernel_fpu_end);
 
 #ifdef CONFIG_USING_FPU_IN_KERNEL_NONATOMIC
+extern unsigned int fpu_kernel_nonatomic_xstate_size;
+unsigned int get_fpustate_free_space(struct fpu *fpu)
+{
+	if ((fpu_kernel_xstate_size + fpu_kernel_nonatomic_xstate_size) > sizeof(fpu->state))
+		return 0;
+
+	return fpu_kernel_nonatomic_xstate_size;
+}
+
+unsigned long get_fpu_registers_pos(struct fpu *fpu, unsigned int off)
+{
+	unsigned long addr = 0;
+	if (fpu && (fpu_kernel_nonatomic_xstate_size > off)) {
+		addr = (unsigned long)&fpu->state.__padding[0];
+		addr += fpu_kernel_xstate_size + off;
+	}
+	return addr;
+}
+
 /*
  * We can call kernel_fpu_begin_nonatomic in non-atomic task context.
  */
@@ -148,6 +167,9 @@ int kernel_fpu_begin_nonatomic_mask(unsigned int kfpu_mask)
 	if (test_thread_flag(TIF_USING_FPU_NONATOMIC))
 		goto nested_err;
 
+	if (KERNEL_FPU_NONATOMIC_SIZE > get_fpustate_free_space(&current->thread.fpu))
+		goto err;
+
 	/*
 	 * This means we call kernel_fpu_begin_nonatomic after kernel_fpu_begin,
 	 * but before kernel_fpu_end.
@@ -209,7 +231,7 @@ EXPORT_SYMBOL_GPL(kernel_fpu_end_nonatomic);
 
 void save_fpregs_to_fpkernelstate(struct fpu *kfpu)
 {
-	kernel_fpu_states_save(&kfpu->kernel_state, NULL, sizeof(kfpu->kernel_state));
+	kernel_fpu_states_save((void *)get_fpu_registers_pos(kfpu, MAX_FPU_CTX_SIZE), NULL, MAX_FPU_CTX_SIZE);
 }
 #endif //CONFIG_USING_FPU_IN_KERNEL_NONATOMIC
 

arch/x86/kernel/fpu/init.c

@@ -138,6 +138,10 @@ static void __init fpu__init_system_generic(void)
  */
 unsigned int fpu_kernel_xstate_size;
 EXPORT_SYMBOL_GPL(fpu_kernel_xstate_size);
+#ifdef CONFIG_USING_FPU_IN_KERNEL_NONATOMIC
+unsigned int fpu_kernel_nonatomic_xstate_size;
+EXPORT_SYMBOL_GPL(fpu_kernel_nonatomic_xstate_size);
+#endif
 
 /*
  * Enforce that 'MEMBER' is the last field of 'TYPE'.
@@ -168,6 +172,9 @@ static void __init fpu__init_task_struct_size(void)
 	 */
 	task_size += fpu_kernel_xstate_size;
 
+#ifdef CONFIG_USING_FPU_IN_KERNEL_NONATOMIC
+	task_size += fpu_kernel_nonatomic_xstate_size;
+#endif
 	/*
 	 * We dynamically size 'struct fpu', so we require that
 	 * it be at the end of 'thread_struct' and that

arch/x86/kernel/fpu/xstate.c

@@ -700,6 +700,9 @@ static int __init init_xstate_size(void)
 	/* Recompute the context size for enabled features: */
 	unsigned int possible_xstate_size;
 	unsigned int xsave_size;
+#ifdef CONFIG_USING_FPU_IN_KERNEL_NONATOMIC
+	extern unsigned int fpu_kernel_nonatomic_xstate_size;
+#endif
 
 	xsave_size = get_xsave_size();
 
@@ -723,6 +726,9 @@ static int __init init_xstate_size(void)
 	 * User space is always in standard format.
 	 */
 	fpu_user_xstate_size = xsave_size;
+#ifdef CONFIG_USING_FPU_IN_KERNEL_NONATOMIC
+	fpu_kernel_nonatomic_xstate_size = KERNEL_FPU_NONATOMIC_SIZE;
+#endif
 	return 0;
 }