MFC 274817,274878,276801,276840,278976:

Improve support for XSAVE with debuggers.
- Dump an NT_X86_XSTATE note if XSAVE is in use. This note is designed
  to match what Linux does in that 1) it dumps the entire XSAVE area
  including the fxsave state, and 2) it stashes a copy of the current
  xsave mask in the unused padding between the fxsave state and the
  xstate header at the same location used by Linux.
- Teach readelf() to recognize NT_X86_XSTATE notes.
- Change PT_GET/SETXSTATE to take the entire XSAVE state instead of
  only the extra portion. This avoids having to always make two
  ptrace() calls to get or set the full XSAVE state.
- Add a PT_GET_XSTATE_INFO which returns the length of the current
  XSTATE save area (so the size of the buffer needed for PT_GETXSTATE)
  and the current XSAVE mask (%xcr0).

3 files changed, 72 insertions, 16 deletions

diff --git a/sys/amd64/amd64/elf_machdep.c b/sys/amd64/amd64/elf_machdep.c
index fdc4d56..23fa39b 100644
--- a/sys/amd64/amd64/elf_machdep.c
+++ b/sys/amd64/amd64/elf_machdep.c
@@ -44,6 +44,7 @@ __FBSDID("$FreeBSD$");
 #include <vm/vm_param.h>
 
 #include <machine/elf.h>
+#include <machine/fpu.h>
 #include <machine/md_var.h>
 
 struct sysentvec elf64_freebsd_sysvec = {
@@ -133,12 +134,27 @@ SYSINIT(kelf64, SI_SUB_EXEC, SI_ORDER_ANY,
 	&kfreebsd_brand_info);
 
 void
-elf64_dump_thread(struct thread *td __unused, void *dst __unused,
-    size_t *off __unused)
+elf64_dump_thread(struct thread *td, void *dst, size_t *off)
 {
+	void *buf;
+	size_t len;
+
+	len = 0;
+	if (use_xsave) {
+		if (dst != NULL) {
+			fpugetregs(td);
+			len += elf64_populate_note(NT_X86_XSTATE,
+			    get_pcb_user_save_td(td), dst,
+			    cpu_max_ext_state_size, &buf);
+			*(uint64_t *)((char *)buf + X86_XSTATE_XCR0_OFFSET) =
+			    xsave_mask;
+		} else
+			len += elf64_populate_note(NT_X86_XSTATE, NULL, NULL,
+			    cpu_max_ext_state_size, NULL);
+	}
+	*off = len;
 }
 
-
 /* Process one elf relocation with addend. */
 static int
 elf_reloc_internal(linker_file_t lf, Elf_Addr relocbase, const void *data,
diff --git a/sys/amd64/amd64/fpu.c b/sys/amd64/amd64/fpu.c
index b0f25ba..f30c073 100644
--- a/sys/amd64/amd64/fpu.c
+++ b/sys/amd64/amd64/fpu.c
@@ -127,6 +127,13 @@ CTASSERT(sizeof(struct savefpu_ymm) == 832);
  */
 CTASSERT(sizeof(struct pcb) % XSAVE_AREA_ALIGN == 0);
 
+/*
+ * Ensure the copy of XCR0 saved in a core is contained in the padding
+ * area.
+ */
+CTASSERT(X86_XSTATE_XCR0_OFFSET >= offsetof(struct savefpu, sv_pad) &&
+    X86_XSTATE_XCR0_OFFSET + sizeof(uint64_t) <= sizeof(struct savefpu));
+
 static	void	fpu_clean_state(void);
 
 SYSCTL_INT(_hw, HW_FLOATINGPT, floatingpoint, CTLFLAG_RD,
diff --git a/sys/amd64/amd64/ptrace_machdep.c b/sys/amd64/amd64/ptrace_machdep.c
index 9fa1917..4cd5bf2 100644
--- a/sys/amd64/amd64/ptrace_machdep.c
+++ b/sys/amd64/amd64/ptrace_machdep.c
@@ -42,6 +42,7 @@ __FBSDID("$FreeBSD$");
 static int
 cpu_ptrace_xstate(struct thread *td, int req, void *addr, int data)
 {
+	struct ptrace_xstate_info info;
 	char *savefpu;
 	int error;
 
@@ -49,14 +50,14 @@ cpu_ptrace_xstate(struct thread *td, int req, void *addr, int data)
 		return (EOPNOTSUPP);
 
 	switch (req) {
-	case PT_GETXSTATE:
+	case PT_GETXSTATE_OLD:
 		fpugetregs(td);
 		savefpu = (char *)(get_pcb_user_save_td(td) + 1);
 		error = copyout(savefpu, addr,
 		    cpu_max_ext_state_size - sizeof(struct savefpu));
 		break;
 
-	case PT_SETXSTATE:
+	case PT_SETXSTATE_OLD:
 		if (data > cpu_max_ext_state_size - sizeof(struct savefpu)) {
 			error = EINVAL;
 			break;
@@ -70,6 +71,37 @@ cpu_ptrace_xstate(struct thread *td, int req, void *addr, int data)
 		free(savefpu, M_TEMP);
 		break;
 
+	case PT_GETXSTATE_INFO:
+		if (data != sizeof(info)) {
+			error  = EINVAL;
+			break;
+		}
+		info.xsave_len = cpu_max_ext_state_size;
+		info.xsave_mask = xsave_mask;
+		error = copyout(&info, addr, data);
+		break;
+
+	case PT_GETXSTATE:
+		fpugetregs(td);
+		savefpu = (char *)(get_pcb_user_save_td(td));
+		error = copyout(savefpu, addr, cpu_max_ext_state_size);
+		break;
+
+	case PT_SETXSTATE:
+		if (data < sizeof(struct savefpu) ||
+		    data > cpu_max_ext_state_size) {
+			error = EINVAL;
+			break;
+		}
+		savefpu = malloc(data, M_TEMP, M_WAITOK);
+		error = copyin(addr, savefpu, data);
+		if (error == 0)
+			error = fpusetregs(td, (struct savefpu *)savefpu,
+			    savefpu + sizeof(struct savefpu), data -
+			    sizeof(struct savefpu));
+		free(savefpu, M_TEMP);
+		break;
+
 	default:
 		error = EINVAL;
 		break;
@@ -81,8 +113,6 @@ cpu_ptrace_xstate(struct thread *td, int req, void *addr, int data)
 #ifdef COMPAT_FREEBSD32
 #define PT_I386_GETXMMREGS	(PT_FIRSTMACH + 0)
 #define PT_I386_SETXMMREGS	(PT_FIRSTMACH + 1)
-#define PT_I386_GETXSTATE	(PT_FIRSTMACH + 2)
-#define PT_I386_SETXSTATE	(PT_FIRSTMACH + 3)
 
 static int
 cpu32_ptrace(struct thread *td, int req, void *addr, int data)
@@ -104,12 +134,12 @@ cpu32_ptrace(struct thread *td, int req, void *addr, int data)
 		fpstate->sv_env.en_mxcsr &= cpu_mxcsr_mask;
 		break;
 
-	case PT_I386_GETXSTATE:
-		error = cpu_ptrace_xstate(td, PT_GETXSTATE, addr, data);
-		break;
-
-	case PT_I386_SETXSTATE:
-		error = cpu_ptrace_xstate(td, PT_SETXSTATE, addr, data);
+	case PT_GETXSTATE_OLD:
+	case PT_SETXSTATE_OLD:
+	case PT_GETXSTATE_INFO:
+	case PT_GETXSTATE:
+	case PT_SETXSTATE:
+		error = cpu_ptrace_xstate(td, req, addr, data);
 		break;
 
 	default:
@@ -131,13 +161,16 @@ cpu_ptrace(struct thread *td, int req, void *addr, int data)
 		return (cpu32_ptrace(td, req, addr, data));
 #endif
 
-	/* Support old values of PT_GETXSTATE and PT_SETXSTATE. */
+	/* Support old values of PT_GETXSTATE_OLD and PT_SETXSTATE_OLD. */
 	if (req == PT_FIRSTMACH + 0)
-		req = PT_GETXSTATE;
+		req = PT_GETXSTATE_OLD;
 	if (req == PT_FIRSTMACH + 1)
-		req = PT_SETXSTATE;
+		req = PT_SETXSTATE_OLD;
 
 	switch (req) {
+	case PT_GETXSTATE_OLD:
+	case PT_SETXSTATE_OLD:
+	case PT_GETXSTATE_INFO:
 	case PT_GETXSTATE:
 	case PT_SETXSTATE:
 		error = cpu_ptrace_xstate(td, req, addr, data);