linuxppc-dev.lists.ozlabs.org archive mirror
 help / color / mirror / Atom feed
From: Cyril Bur <cyrilbur@gmail.com>
To: linuxppc-dev@ozlabs.org
Cc: naveen.n.rao@linux.vnet.ibm.com
Subject: [PATCH v5 5/9] powerpc: Restore FPU/VEC/VSX if previously used
Date: Tue, 23 Feb 2016 14:38:18 +1100	[thread overview]
Message-ID: <1456198702-29657-6-git-send-email-cyrilbur@gmail.com> (raw)
In-Reply-To: <1456198702-29657-1-git-send-email-cyrilbur@gmail.com>

Currently the FPU, VEC and VSX facilities are lazily loaded. This is not a
problem unless a process is using these facilities.

Modern versions of GCC are very good at automatically vectorising code, new
and modernised workloads make use of floating point and vector facilities,
even the kernel makes use of vectorised memcpy.

All this combined greatly increases the cost of a syscall since the kernel
uses the facilities sometimes even in syscall fast-path making it
increasingly common for a thread to take an *_unavailable exception soon
after a syscall, not to mention potentially taking all three.

The obvious overcompensation to this problem is to simply always load all
the facilities on every exit to userspace. Loading up all FPU, VEC and VSX
registers every time can be expensive and if a workload does avoid using
them, it should not be forced to incur this penalty.

An 8bit counter is used to detect if the registers have been used in the
past and the registers are always loaded until the value wraps to back to
zero.

Several versions of the assembly in entry_64.S. 1. Always calling C, 2.
Performing a common case check and then calling C and 3. A complex check in
asm. After some benchmarking it was determined that avoiding C in the
common case is a performance benefit. The full check in asm greatly
complicated that codepath for a negligible performance gain and the
trade-off was deemed not worth it.

Signed-off-by: Cyril Bur <cyrilbur@gmail.com>
---
 arch/powerpc/include/asm/processor.h |  2 +
 arch/powerpc/kernel/asm-offsets.c    |  2 +
 arch/powerpc/kernel/entry_64.S       | 21 +++++++--
 arch/powerpc/kernel/fpu.S            |  4 ++
 arch/powerpc/kernel/process.c        | 88 +++++++++++++++++++++++++++++++-----
 arch/powerpc/kernel/vector.S         |  4 ++
 6 files changed, 107 insertions(+), 14 deletions(-)

diff --git a/arch/powerpc/include/asm/processor.h b/arch/powerpc/include/asm/processor.h
index ac23308..dcab21f 100644
--- a/arch/powerpc/include/asm/processor.h
+++ b/arch/powerpc/include/asm/processor.h
@@ -236,11 +236,13 @@ struct thread_struct {
 #endif
 	struct arch_hw_breakpoint hw_brk; /* info on the hardware breakpoint */
 	unsigned long	trap_nr;	/* last trap # on this thread */
+	u8 load_fp;
 #ifdef CONFIG_ALTIVEC
 	struct thread_vr_state vr_state;
 	struct thread_vr_state *vr_save_area;
 	unsigned long	vrsave;
 	int		used_vr;	/* set if process has used altivec */
+	u8 load_vec;
 #endif /* CONFIG_ALTIVEC */
 #ifdef CONFIG_VSX
 	/* VSR status */
diff --git a/arch/powerpc/kernel/asm-offsets.c b/arch/powerpc/kernel/asm-offsets.c
index 07cebc3..10d5eab 100644
--- a/arch/powerpc/kernel/asm-offsets.c
+++ b/arch/powerpc/kernel/asm-offsets.c
@@ -95,12 +95,14 @@ int main(void)
 	DEFINE(THREAD_FPSTATE, offsetof(struct thread_struct, fp_state));
 	DEFINE(THREAD_FPSAVEAREA, offsetof(struct thread_struct, fp_save_area));
 	DEFINE(FPSTATE_FPSCR, offsetof(struct thread_fp_state, fpscr));
+	DEFINE(THREAD_LOAD_FP, offsetof(struct thread_struct, load_fp));
 #ifdef CONFIG_ALTIVEC
 	DEFINE(THREAD_VRSTATE, offsetof(struct thread_struct, vr_state));
 	DEFINE(THREAD_VRSAVEAREA, offsetof(struct thread_struct, vr_save_area));
 	DEFINE(THREAD_VRSAVE, offsetof(struct thread_struct, vrsave));
 	DEFINE(THREAD_USED_VR, offsetof(struct thread_struct, used_vr));
 	DEFINE(VRSTATE_VSCR, offsetof(struct thread_vr_state, vscr));
+	DEFINE(THREAD_LOAD_VEC, offsetof(struct thread_struct, load_vec));
 #endif /* CONFIG_ALTIVEC */
 #ifdef CONFIG_VSX
 	DEFINE(THREAD_USED_VSR, offsetof(struct thread_struct, used_vsr));
diff --git a/arch/powerpc/kernel/entry_64.S b/arch/powerpc/kernel/entry_64.S
index 0d525ce..038e0a1 100644
--- a/arch/powerpc/kernel/entry_64.S
+++ b/arch/powerpc/kernel/entry_64.S
@@ -210,7 +210,20 @@ system_call:			/* label this so stack traces look sane */
 	li	r11,-MAX_ERRNO
 	andi.	r0,r9,(_TIF_SYSCALL_DOTRACE|_TIF_SINGLESTEP|_TIF_USER_WORK_MASK|_TIF_PERSYSCALL_MASK)
 	bne-	syscall_exit_work
-	cmpld	r3,r11
+
+	andi.	r0,r8,MSR_FP
+	beq 2f
+#ifdef CONFIG_ALTIVEC
+	andis.	r0,r8,MSR_VEC@h
+	bne	3f
+#endif
+2:	addi    r3,r1,STACK_FRAME_OVERHEAD
+	bl	restore_math
+	ld	r8,_MSR(r1)
+	ld	r3,RESULT(r1)
+	li	r11,-MAX_ERRNO
+
+3:	cmpld	r3,r11
 	ld	r5,_CCR(r1)
 	bge-	syscall_error
 .Lsyscall_error_cont:
@@ -602,8 +615,8 @@ _GLOBAL(ret_from_except_lite)
 
 	/* Check current_thread_info()->flags */
 	andi.	r0,r4,_TIF_USER_WORK_MASK
-#ifdef CONFIG_PPC_BOOK3E
 	bne	1f
+#ifdef CONFIG_PPC_BOOK3E
 	/*
 	 * Check to see if the dbcr0 register is set up to debug.
 	 * Use the internal debug mode bit to do this.
@@ -618,7 +631,9 @@ _GLOBAL(ret_from_except_lite)
 	mtspr	SPRN_DBSR,r10
 	b	restore
 #else
-	beq	restore
+	addi	r3,r1,STACK_FRAME_OVERHEAD
+	bl	restore_math
+	b	restore
 #endif
 1:	andi.	r0,r4,_TIF_NEED_RESCHED
 	beq	2f
diff --git a/arch/powerpc/kernel/fpu.S b/arch/powerpc/kernel/fpu.S
index 2117eac..b063524 100644
--- a/arch/powerpc/kernel/fpu.S
+++ b/arch/powerpc/kernel/fpu.S
@@ -130,6 +130,10 @@ END_FTR_SECTION_IFSET(CPU_FTR_VSX)
 	or	r12,r12,r4
 	std	r12,_MSR(r1)
 #endif
+	/* Don't care if r4 overflows, this is desired behaviour */
+	lbz	r4,THREAD_LOAD_FP(r5)
+	addi	r4,r4,1
+	stb	r4,THREAD_LOAD_FP(r5)
 	addi	r10,r5,THREAD_FPSTATE
 	lfd	fr0,FPSTATE_FPSCR(r10)
 	MTFSF_L(fr0)
diff --git a/arch/powerpc/kernel/process.c b/arch/powerpc/kernel/process.c
index e0c3d2d..55c1eb0 100644
--- a/arch/powerpc/kernel/process.c
+++ b/arch/powerpc/kernel/process.c
@@ -187,9 +187,22 @@ void enable_kernel_fp(void)
 	}
 }
 EXPORT_SYMBOL(enable_kernel_fp);
+
+static int restore_fp(struct task_struct *tsk) {
+	if (tsk->thread.load_fp) {
+		load_fp_state(&current->thread.fp_state);
+		current->thread.load_fp++;
+		return 1;
+	}
+	return 0;
+}
+#else
+static int restore_fp(struct task_struct *tsk) { return 0; }
 #endif /* CONFIG_PPC_FPU */
 
 #ifdef CONFIG_ALTIVEC
+#define loadvec(thr) ((thr).load_vec)
+
 void giveup_altivec(struct task_struct *tsk)
 {
 	check_if_tm_restore_required(tsk);
@@ -229,6 +242,21 @@ void flush_altivec_to_thread(struct task_struct *tsk)
 	}
 }
 EXPORT_SYMBOL_GPL(flush_altivec_to_thread);
+
+static int restore_altivec(struct task_struct *tsk)
+{
+	if (cpu_has_feature(CPU_FTR_ALTIVEC) && tsk->thread.load_vec) {
+		load_vr_state(&tsk->thread.vr_state);
+		tsk->thread.used_vr = 1;
+		tsk->thread.load_vec++;
+
+		return 1;
+	}
+	return 0;
+}
+#else
+#define loadvec(thr) 0
+static inline int restore_altivec(struct task_struct *tsk) { return 0; }
 #endif /* CONFIG_ALTIVEC */
 
 #ifdef CONFIG_VSX
@@ -275,6 +303,18 @@ void flush_vsx_to_thread(struct task_struct *tsk)
 	}
 }
 EXPORT_SYMBOL_GPL(flush_vsx_to_thread);
+
+static int restore_vsx(struct task_struct *tsk)
+{
+	if (cpu_has_feature(CPU_FTR_VSX)) {
+		tsk->thread.used_vsr = 1;
+		return 1;
+	}
+
+	return 0;
+}
+#else
+static inline int restore_vsx(struct task_struct *tsk) { return 0; }
 #endif /* CONFIG_VSX */
 
 #ifdef CONFIG_SPE
@@ -374,6 +414,36 @@ void giveup_all(struct task_struct *tsk)
 }
 EXPORT_SYMBOL(giveup_all);
 
+void restore_math(struct pt_regs *regs)
+{
+	unsigned long msr;
+
+	if (!current->thread.load_fp && !loadvec(current->thread))
+		return;
+
+	msr = regs->msr;
+	msr_check_and_set(msr_all_available);
+
+	/*
+	 * Only reload if the bit is not set in the user MSR, the bit BEING set
+	 * indicates that the registers are hot
+	 */
+	if ((!(msr & MSR_FP)) && restore_fp(current))
+		msr |= MSR_FP | current->thread.fpexc_mode;
+
+	if ((!(msr & MSR_VEC)) && restore_altivec(current))
+		msr |= MSR_VEC;
+
+	if ((msr & (MSR_FP | MSR_VEC)) == (MSR_FP | MSR_VEC) &&
+			restore_vsx(current)) {
+		msr |= MSR_VSX;
+	}
+
+	msr_check_and_clear(msr_all_available);
+
+	regs->msr = msr;
+}
+
 void flush_all_to_thread(struct task_struct *tsk)
 {
 	if (tsk->thread.regs) {
@@ -832,17 +902,9 @@ void restore_tm_state(struct pt_regs *regs)
 
 	msr_diff = current->thread.ckpt_regs.msr & ~regs->msr;
 	msr_diff &= MSR_FP | MSR_VEC | MSR_VSX;
-	if (msr_diff & MSR_FP) {
-		msr_check_and_set(MSR_FP);
-		load_fp_state(&current->thread.fp_state);
-		msr_check_and_clear(MSR_FP);
-		regs->msr |= current->thread.fpexc_mode;
-	}
-	if (msr_diff & MSR_VEC) {
-		msr_check_and_set(MSR_VEC);
-		load_vr_state(&current->thread.vr_state);
-		msr_check_and_clear(MSR_VEC);
-	}
+
+	restore_math(regs);
+
 	regs->msr |= msr_diff;
 }
 
@@ -1006,6 +1068,10 @@ struct task_struct *__switch_to(struct task_struct *prev,
 		batch = this_cpu_ptr(&ppc64_tlb_batch);
 		batch->active = 1;
 	}
+
+	if (current_thread_info()->task->thread.regs)
+		restore_math(current_thread_info()->task->thread.regs);
+
 #endif /* CONFIG_PPC_BOOK3S_64 */
 
 	return last;
diff --git a/arch/powerpc/kernel/vector.S b/arch/powerpc/kernel/vector.S
index 162d0f7..038cff8 100644
--- a/arch/powerpc/kernel/vector.S
+++ b/arch/powerpc/kernel/vector.S
@@ -91,6 +91,10 @@ _GLOBAL(load_up_altivec)
 	oris	r12,r12,MSR_VEC@h
 	std	r12,_MSR(r1)
 #endif
+	/* Don't care if r4 overflows, this is desired behaviour */
+	lbz	r4,THREAD_LOAD_VEC(r5)
+	addi	r4,r4,1
+	stb	r4,THREAD_LOAD_VEC(r5)
 	addi	r6,r5,THREAD_VRSTATE
 	li	r4,1
 	li	r10,VRSTATE_VSCR
-- 
2.7.1

  parent reply	other threads:[~2016-02-23  3:40 UTC|newest]

Thread overview: 16+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2016-02-23  3:38 [PATCH v5 0/9] FP/VEC/VSX switching optimisations Cyril Bur
2016-02-23  3:38 ` [PATCH v5 1/9] selftests/powerpc: Test the preservation of FPU and VMX regs across syscall Cyril Bur
2016-02-24 14:27   ` Naveen N. Rao
2016-02-24 23:44     ` Cyril Bur
2016-02-25  6:22       ` Naveen N. Rao
2016-02-25 22:18         ` Cyril Bur
2016-02-29  5:53           ` Naveen N. Rao
2016-02-25  6:22   ` Cyril Bur
2016-02-23  3:38 ` [PATCH v5 2/9] selftests/powerpc: Test preservation of FPU and VMX regs across preemption Cyril Bur
2016-02-23  3:38 ` [PATCH v5 3/9] selftests/powerpc: Test FPU and VMX regs in signal ucontext Cyril Bur
2016-02-23  3:38 ` [PATCH v5 4/9] powerpc: Explicitly disable math features when copying thread Cyril Bur
2016-02-23  3:38 ` Cyril Bur [this message]
2016-02-23  3:38 ` [PATCH v5 6/9] powerpc: Prepare for splitting giveup_{fpu, altivec, vsx} in two Cyril Bur
2016-02-23  3:38 ` [PATCH v5 7/9] powerpc: Add the ability to save FPU without giving it up Cyril Bur
2016-02-23  3:38 ` [PATCH v5 8/9] powerpc: Add the ability to save Altivec " Cyril Bur
2016-02-23  3:38 ` [PATCH v5 9/9] powerpc: Add the ability to save VSX " Cyril Bur

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=1456198702-29657-6-git-send-email-cyrilbur@gmail.com \
    --to=cyrilbur@gmail.com \
    --cc=linuxppc-dev@ozlabs.org \
    --cc=naveen.n.rao@linux.vnet.ibm.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body. 
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).