[PATCH] powerpc: Don't export cvt_fd & _df when CONFIG_PPC_FPU is not set

[PATCH] powerpc: Don't export cvt_fd & _df when CONFIG_PPC_FPU is not set

[Benjamin Herrenschmidt]

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
---
 arch/powerpc/kernel/ppc_ksyms.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/arch/powerpc/kernel/ppc_ksyms.c b/arch/powerpc/kernel/ppc_ksyms.c
index bc9f39d..3b4dcc8 100644
--- a/arch/powerpc/kernel/ppc_ksyms.c
+++ b/arch/powerpc/kernel/ppc_ksyms.c
@@ -101,7 +101,7 @@ EXPORT_SYMBOL(pci_dram_offset);
 EXPORT_SYMBOL(start_thread);
 EXPORT_SYMBOL(kernel_thread);
 
-#ifndef CONFIG_BOOKE
+#ifdef CONFIG_PPC_FPU
 EXPORT_SYMBOL_GPL(cvt_df);
 EXPORT_SYMBOL_GPL(cvt_fd);
 #endif

Re: [PATCH] powerpc: Don't export cvt_fd & _df when CONFIG_PPC_FPU is not set

[Benjamin Herrenschmidt]

On Mon, 2010-05-31 at 11:50 +1000, Benjamin Herrenschmidt wrote:
> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
> ---
>  arch/powerpc/kernel/ppc_ksyms.c |    2 +-
>  1 files changed, 1 insertions(+), 1 deletions(-)

Alex, this is just a temporary fix to remove the build breakage for 40x
etc... but please, update KVM to just build-in its own.

Don't we have .o's that are always built into modules nowadays ?

Cheers,
Ben.

Re: [PATCH] powerpc: Don't export cvt_fd & _df when CONFIG_PPC_FPU is not set

[Alexander Graf]

On 31.05.2010, at 04:00, Benjamin Herrenschmidt wrote:

> On Mon, 2010-05-31 at 11:50 +1000, Benjamin Herrenschmidt wrote:
>> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
>> ---
>> arch/powerpc/kernel/ppc_ksyms.c |    2 +-
>> 1 files changed, 1 insertions(+), 1 deletions(-)
> 
> Alex, this is just a temporary fix to remove the build breakage for 40x
> etc... but please, update KVM to just build-in its own.

What's the bad part in reusing the existing code?

> Don't we have .o's that are always built into modules nowadays ?

I don't think I understand what you mean.


Alex

Re: [PATCH] powerpc: Don't export cvt_fd & _df when CONFIG_PPC_FPU is not set

[Andreas Schwab]

Alexander Graf <agraf@suse.de> writes:

> On 31.05.2010, at 04:00, Benjamin Herrenschmidt wrote:
>
>> On Mon, 2010-05-31 at 11:50 +1000, Benjamin Herrenschmidt wrote:
>>> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
>>> ---
>>> arch/powerpc/kernel/ppc_ksyms.c |    2 +-
>>> 1 files changed, 1 insertions(+), 1 deletions(-)
>> 
>> Alex, this is just a temporary fix to remove the build breakage for 40x
>> etc... but please, update KVM to just build-in its own.
>
> What's the bad part in reusing the existing code?

The bad part is that KVM is wasting a ridiculous amount of stack space
by allocating multiple instances of struct task_struct on it.  I have a
patch removing all of it, but I haven't tested it yet.

Andreas.

-- 
Andreas Schwab, schwab@redhat.com
GPG Key fingerprint = D4E8 DBE3 3813 BB5D FA84  5EC7 45C6 250E 6F00 984E
"And now for something completely different."

Re: [PATCH] powerpc: Don't export cvt_fd & _df when CONFIG_PPC_FPU is not set

[Alexander Graf]

Andreas Schwab wrote:
> Alexander Graf <agraf@suse.de> writes:
>
>   
>> On 31.05.2010, at 04:00, Benjamin Herrenschmidt wrote:
>>
>>     
>>> On Mon, 2010-05-31 at 11:50 +1000, Benjamin Herrenschmidt wrote:
>>>       
>>>> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
>>>> ---
>>>> arch/powerpc/kernel/ppc_ksyms.c |    2 +-
>>>> 1 files changed, 1 insertions(+), 1 deletions(-)
>>>>         
>>> Alex, this is just a temporary fix to remove the build breakage for 40x
>>> etc... but please, update KVM to just build-in its own.
>>>       
>> What's the bad part in reusing the existing code?
>>     
>
> The bad part is that KVM is wasting a ridiculous amount of stack space
> by allocating multiple instances of struct task_struct on it.  I have a
> patch removing all of it, but I haven't tested it yet.
>   

Mind to send it over so I can take a look at it :)? Getting rid of the
task_struct structs lying in the stack is certainly a good idea.

Alex

Re: [PATCH] powerpc: Don't export cvt_fd & _df when CONFIG_PPC_FPU is not set

[Andreas Schwab]

Alexander Graf <agraf@suse.de> writes:

> Mind to send it over so I can take a look at it :)? Getting rid of the
> task_struct structs lying in the stack is certainly a good idea.

Still untested, because rc1 does not boot.

Andreas.

---
 arch/powerpc/include/asm/kvm_fpu.h       |   29 +++++----
 arch/powerpc/kvm/book3s.c                |   49 ++++++++--------
 arch/powerpc/kvm/book3s_paired_singles.c |   94 ++++++++++++-------------------
 arch/powerpc/kvm/fpu.S                   |   18 +++++
 4 files changed, 98 insertions(+), 92 deletions(-)

--- linux-2.6.35-rc1.orig/arch/powerpc/include/asm/kvm_fpu.h
+++ linux-2.6.35-rc1/arch/powerpc/include/asm/kvm_fpu.h
@@ -22,24 +22,24 @@
 
 #include <linux/types.h>
 
-extern void fps_fres(struct thread_struct *t, u32 *dst, u32 *src1);
-extern void fps_frsqrte(struct thread_struct *t, u32 *dst, u32 *src1);
-extern void fps_fsqrts(struct thread_struct *t, u32 *dst, u32 *src1);
-
-extern void fps_fadds(struct thread_struct *t, u32 *dst, u32 *src1, u32 *src2);
-extern void fps_fdivs(struct thread_struct *t, u32 *dst, u32 *src1, u32 *src2);
-extern void fps_fmuls(struct thread_struct *t, u32 *dst, u32 *src1, u32 *src2);
-extern void fps_fsubs(struct thread_struct *t, u32 *dst, u32 *src1, u32 *src2);
+extern void fps_fres(u64 *fpscr, u32 *dst, u32 *src1);
+extern void fps_frsqrte(u64 *fpscr, u32 *dst, u32 *src1);
+extern void fps_fsqrts(u64 *fpscr, u32 *dst, u32 *src1);
+
+extern void fps_fadds(u64 *fpscr, u32 *dst, u32 *src1, u32 *src2);
+extern void fps_fdivs(u64 *fpscr, u32 *dst, u32 *src1, u32 *src2);
+extern void fps_fmuls(u64 *fpscr, u32 *dst, u32 *src1, u32 *src2);
+extern void fps_fsubs(u64 *fpscr, u32 *dst, u32 *src1, u32 *src2);
 
-extern void fps_fmadds(struct thread_struct *t, u32 *dst, u32 *src1, u32 *src2,
+extern void fps_fmadds(u64 *fpscr, u32 *dst, u32 *src1, u32 *src2,
 		       u32 *src3);
-extern void fps_fmsubs(struct thread_struct *t, u32 *dst, u32 *src1, u32 *src2,
+extern void fps_fmsubs(u64 *fpscr, u32 *dst, u32 *src1, u32 *src2,
 		       u32 *src3);
-extern void fps_fnmadds(struct thread_struct *t, u32 *dst, u32 *src1, u32 *src2,
+extern void fps_fnmadds(u64 *fpscr, u32 *dst, u32 *src1, u32 *src2,
 		        u32 *src3);
-extern void fps_fnmsubs(struct thread_struct *t, u32 *dst, u32 *src1, u32 *src2,
+extern void fps_fnmsubs(u64 *fpscr, u32 *dst, u32 *src1, u32 *src2,
 		        u32 *src3);
-extern void fps_fsel(struct thread_struct *t, u32 *dst, u32 *src1, u32 *src2,
+extern void fps_fsel(u64 *fpscr, u32 *dst, u32 *src1, u32 *src2,
 		     u32 *src3);
 
 #define FPD_ONE_IN(name) extern void fpd_ ## name(u64 *fpscr, u32 *cr, \
@@ -82,4 +82,7 @@ FPD_THREE_IN(fmadd)
 FPD_THREE_IN(fnmsub)
 FPD_THREE_IN(fnmadd)
 
+extern void kvm_cvt_fd(u32 *from, u64 *to, u64 *fpscr);
+extern void kvm_cvt_df(u64 *from, u32 *to, u64 *fpscr);
+
 #endif
--- linux-2.6.35-rc1.orig/arch/powerpc/kvm/book3s.c
+++ linux-2.6.35-rc1/arch/powerpc/kvm/book3s.c
@@ -1309,12 +1309,17 @@ extern int __kvmppc_vcpu_entry(struct kv
 int __kvmppc_vcpu_run(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
 {
 	int ret;
-	struct thread_struct ext_bkp;
+	double fpr[32][TS_FPRWIDTH];
+	unsigned int fpscr;
+	int fpexc_mode;
 #ifdef CONFIG_ALTIVEC
-	bool save_vec = current->thread.used_vr;
+	vector128 vr[32];
+	vector128 vscr;
+	unsigned long uninitialized_var(vrsave);
+	int used_vr;
 #endif
 #ifdef CONFIG_VSX
-	bool save_vsx = current->thread.used_vsr;
+	int used_vsr;
 #endif
 	ulong ext_msr;
 
@@ -1327,27 +1332,27 @@ int __kvmppc_vcpu_run(struct kvm_run *kv
 	/* Save FPU state in stack */
 	if (current->thread.regs->msr & MSR_FP)
 		giveup_fpu(current);
-	memcpy(ext_bkp.fpr, current->thread.fpr, sizeof(current->thread.fpr));
-	ext_bkp.fpscr = current->thread.fpscr;
-	ext_bkp.fpexc_mode = current->thread.fpexc_mode;
+	memcpy(fpr, current->thread.fpr, sizeof(current->thread.fpr));
+	fpscr = current->thread.fpscr.val;
+	fpexc_mode = current->thread.fpexc_mode;
 
 #ifdef CONFIG_ALTIVEC
 	/* Save Altivec state in stack */
-	if (save_vec) {
+	used_vr = current->thread.used_vr;
+	if (used_vr) {
 		if (current->thread.regs->msr & MSR_VEC)
 			giveup_altivec(current);
-		memcpy(ext_bkp.vr, current->thread.vr, sizeof(ext_bkp.vr));
-		ext_bkp.vscr = current->thread.vscr;
-		ext_bkp.vrsave = current->thread.vrsave;
+		memcpy(vr, current->thread.vr, sizeof(current->thread.vr));
+		vscr = current->thread.vscr;
+		vrsave = current->thread.vrsave;
 	}
-	ext_bkp.used_vr = current->thread.used_vr;
 #endif
 
 #ifdef CONFIG_VSX
 	/* Save VSX state in stack */
-	if (save_vsx && (current->thread.regs->msr & MSR_VSX))
+	used_vsr = current->thread.used_vsr;
+	if (used_vsr && (current->thread.regs->msr & MSR_VSX))
 			__giveup_vsx(current);
-	ext_bkp.used_vsr = current->thread.used_vsr;
 #endif
 
 	/* Remember the MSR with disabled extensions */
@@ -1372,22 +1377,22 @@ int __kvmppc_vcpu_run(struct kvm_run *kv
 	kvmppc_giveup_ext(vcpu, MSR_VSX);
 
 	/* Restore FPU state from stack */
-	memcpy(current->thread.fpr, ext_bkp.fpr, sizeof(ext_bkp.fpr));
-	current->thread.fpscr = ext_bkp.fpscr;
-	current->thread.fpexc_mode = ext_bkp.fpexc_mode;
+	memcpy(current->thread.fpr, fpr, sizeof(current->thread.fpr));
+	current->thread.fpscr.val = fpscr;
+	current->thread.fpexc_mode = fpexc_mode;
 
 #ifdef CONFIG_ALTIVEC
 	/* Restore Altivec state from stack */
-	if (save_vec && current->thread.used_vr) {
-		memcpy(current->thread.vr, ext_bkp.vr, sizeof(ext_bkp.vr));
-		current->thread.vscr = ext_bkp.vscr;
-		current->thread.vrsave= ext_bkp.vrsave;
+	if (used_vr && current->thread.used_vr) {
+		memcpy(current->thread.vr, vr, sizeof(current->thread.vr));
+		current->thread.vscr = vscr;
+		current->thread.vrsave = vrsave;
 	}
-	current->thread.used_vr = ext_bkp.used_vr;
+	current->thread.used_vr = used_vr;
 #endif
 
 #ifdef CONFIG_VSX
-	current->thread.used_vsr = ext_bkp.used_vsr;
+	current->thread.used_vsr = used_vsr;
 #endif
 
 	return ret;
--- linux-2.6.35-rc1.orig/arch/powerpc/kvm/book3s_paired_singles.c
+++ linux-2.6.35-rc1/arch/powerpc/kvm/book3s_paired_singles.c
@@ -159,10 +159,7 @@
 
 static inline void kvmppc_sync_qpr(struct kvm_vcpu *vcpu, int rt)
 {
-	struct thread_struct t;
-
-	t.fpscr.val = vcpu->arch.fpscr;
-	cvt_df((double*)&vcpu->arch.fpr[rt], (float*)&vcpu->arch.qpr[rt], &t);
+	kvm_cvt_df(&vcpu->arch.fpr[rt], &vcpu->arch.qpr[rt], &vcpu->arch.fpscr);
 }
 
 static void kvmppc_inject_pf(struct kvm_vcpu *vcpu, ulong eaddr, bool is_store)
@@ -183,7 +180,6 @@ static int kvmppc_emulate_fpr_load(struc
 				   int rs, ulong addr, int ls_type)
 {
 	int emulated = EMULATE_FAIL;
-	struct thread_struct t;
 	int r;
 	char tmp[8];
 	int len = sizeof(u32);
@@ -191,8 +187,6 @@ static int kvmppc_emulate_fpr_load(struc
 	if (ls_type == FPU_LS_DOUBLE)
 		len = sizeof(u64);
 
-	t.fpscr.val = vcpu->arch.fpscr;
-
 	/* read from memory */
 	r = kvmppc_ld(vcpu, &addr, len, tmp, true);
 	vcpu->arch.paddr_accessed = addr;
@@ -210,7 +204,7 @@ static int kvmppc_emulate_fpr_load(struc
 	/* put in registers */
 	switch (ls_type) {
 	case FPU_LS_SINGLE:
-		cvt_fd((float*)tmp, (double*)&vcpu->arch.fpr[rs], &t);
+		kvm_cvt_fd((u32*)tmp, &vcpu->arch.fpr[rs], &vcpu->arch.fpscr);
 		vcpu->arch.qpr[rs] = *((u32*)tmp);
 		break;
 	case FPU_LS_DOUBLE:
@@ -229,17 +223,14 @@ static int kvmppc_emulate_fpr_store(stru
 				    int rs, ulong addr, int ls_type)
 {
 	int emulated = EMULATE_FAIL;
-	struct thread_struct t;
 	int r;
 	char tmp[8];
 	u64 val;
 	int len;
 
-	t.fpscr.val = vcpu->arch.fpscr;
-
 	switch (ls_type) {
 	case FPU_LS_SINGLE:
-		cvt_df((double*)&vcpu->arch.fpr[rs], (float*)tmp, &t);
+		kvm_cvt_df(&vcpu->arch.fpr[rs], (u32*)tmp, &vcpu->arch.fpscr);
 		val = *((u32*)tmp);
 		len = sizeof(u32);
 		break;
@@ -278,13 +269,10 @@ static int kvmppc_emulate_psq_load(struc
 				   int rs, ulong addr, bool w, int i)
 {
 	int emulated = EMULATE_FAIL;
-	struct thread_struct t;
 	int r;
 	float one = 1.0;
 	u32 tmp[2];
 
-	t.fpscr.val = vcpu->arch.fpscr;
-
 	/* read from memory */
 	if (w) {
 		r = kvmppc_ld(vcpu, &addr, sizeof(u32), tmp, true);
@@ -308,7 +296,7 @@ static int kvmppc_emulate_psq_load(struc
 	emulated = EMULATE_DONE;
 
 	/* put in registers */
-	cvt_fd((float*)&tmp[0], (double*)&vcpu->arch.fpr[rs], &t);
+	kvm_cvt_fd(&tmp[0], &vcpu->arch.fpr[rs], &vcpu->arch.fpscr);
 	vcpu->arch.qpr[rs] = tmp[1];
 
 	dprintk(KERN_INFO "KVM: PSQ_LD [0x%x, 0x%x] at 0x%lx (%d)\n", tmp[0],
@@ -322,14 +310,11 @@ static int kvmppc_emulate_psq_store(stru
 				    int rs, ulong addr, bool w, int i)
 {
 	int emulated = EMULATE_FAIL;
-	struct thread_struct t;
 	int r;
 	u32 tmp[2];
 	int len = w ? sizeof(u32) : sizeof(u64);
 
-	t.fpscr.val = vcpu->arch.fpscr;
-
-	cvt_df((double*)&vcpu->arch.fpr[rs], (float*)&tmp[0], &t);
+	kvm_cvt_df(&vcpu->arch.fpr[rs], &tmp[0], &vcpu->arch.fpscr);
 	tmp[1] = vcpu->arch.qpr[rs];
 
 	r = kvmppc_st(vcpu, &addr, len, tmp, true);
@@ -517,7 +502,7 @@ static int get_d_signext(u32 inst)
 static int kvmppc_ps_three_in(struct kvm_vcpu *vcpu, bool rc,
 				      int reg_out, int reg_in1, int reg_in2,
 				      int reg_in3, int scalar,
-				      void (*func)(struct thread_struct *t,
+				      void (*func)(u64 *fpscr,
 						 u32 *dst, u32 *src1,
 						 u32 *src2, u32 *src3))
 {
@@ -526,27 +511,25 @@ static int kvmppc_ps_three_in(struct kvm
 	u32 ps0_out;
 	u32 ps0_in1, ps0_in2, ps0_in3;
 	u32 ps1_in1, ps1_in2, ps1_in3;
-	struct thread_struct t;
-	t.fpscr.val = vcpu->arch.fpscr;
 
 	/* RC */
 	WARN_ON(rc);
 
 	/* PS0 */
-	cvt_df((double*)&fpr[reg_in1], (float*)&ps0_in1, &t);
-	cvt_df((double*)&fpr[reg_in2], (float*)&ps0_in2, &t);
-	cvt_df((double*)&fpr[reg_in3], (float*)&ps0_in3, &t);
+	kvm_cvt_df(&fpr[reg_in1], &ps0_in1, &vcpu->arch.fpscr);
+	kvm_cvt_df(&fpr[reg_in2], &ps0_in2, &vcpu->arch.fpscr);
+	kvm_cvt_df(&fpr[reg_in3], &ps0_in3, &vcpu->arch.fpscr);
 
 	if (scalar & SCALAR_LOW)
 		ps0_in2 = qpr[reg_in2];
 
-	func(&t, &ps0_out, &ps0_in1, &ps0_in2, &ps0_in3);
+	func(&vcpu->arch.fpscr, &ps0_out, &ps0_in1, &ps0_in2, &ps0_in3);
 
 	dprintk(KERN_INFO "PS3 ps0 -> f(0x%x, 0x%x, 0x%x) = 0x%x\n",
 			  ps0_in1, ps0_in2, ps0_in3, ps0_out);
 
 	if (!(scalar & SCALAR_NO_PS0))
-		cvt_fd((float*)&ps0_out, (double*)&fpr[reg_out], &t);
+		kvm_cvt_fd(&ps0_out, &fpr[reg_out], &vcpu->arch.fpscr);
 
 	/* PS1 */
 	ps1_in1 = qpr[reg_in1];
@@ -557,7 +540,7 @@ static int kvmppc_ps_three_in(struct kvm
 		ps1_in2 = ps0_in2;
 
 	if (!(scalar & SCALAR_NO_PS1))
-		func(&t, &qpr[reg_out], &ps1_in1, &ps1_in2, &ps1_in3);
+		func(&vcpu->arch.fpscr, &qpr[reg_out], &ps1_in1, &ps1_in2, &ps1_in3);
 
 	dprintk(KERN_INFO "PS3 ps1 -> f(0x%x, 0x%x, 0x%x) = 0x%x\n",
 			  ps1_in1, ps1_in2, ps1_in3, qpr[reg_out]);
@@ -568,7 +551,7 @@ static int kvmppc_ps_three_in(struct kvm
 static int kvmppc_ps_two_in(struct kvm_vcpu *vcpu, bool rc,
 				    int reg_out, int reg_in1, int reg_in2,
 				    int scalar,
-				    void (*func)(struct thread_struct *t,
+				    void (*func)(u64 *fpscr,
 						 u32 *dst, u32 *src1,
 						 u32 *src2))
 {
@@ -578,27 +561,25 @@ static int kvmppc_ps_two_in(struct kvm_v
 	u32 ps0_in1, ps0_in2;
 	u32 ps1_out;
 	u32 ps1_in1, ps1_in2;
-	struct thread_struct t;
-	t.fpscr.val = vcpu->arch.fpscr;
 
 	/* RC */
 	WARN_ON(rc);
 
 	/* PS0 */
-	cvt_df((double*)&fpr[reg_in1], (float*)&ps0_in1, &t);
+	kvm_cvt_df(&fpr[reg_in1], &ps0_in1, &vcpu->arch.fpscr);
 
 	if (scalar & SCALAR_LOW)
 		ps0_in2 = qpr[reg_in2];
 	else
-		cvt_df((double*)&fpr[reg_in2], (float*)&ps0_in2, &t);
+		kvm_cvt_df(&fpr[reg_in2], &ps0_in2, &vcpu->arch.fpscr);
 
-	func(&t, &ps0_out, &ps0_in1, &ps0_in2);
+	func(&vcpu->arch.fpscr, &ps0_out, &ps0_in1, &ps0_in2);
 
 	if (!(scalar & SCALAR_NO_PS0)) {
 		dprintk(KERN_INFO "PS2 ps0 -> f(0x%x, 0x%x) = 0x%x\n",
 				  ps0_in1, ps0_in2, ps0_out);
 
-		cvt_fd((float*)&ps0_out, (double*)&fpr[reg_out], &t);
+		kvm_cvt_fd(&ps0_out, &fpr[reg_out], &vcpu->arch.fpscr);
 	}
 
 	/* PS1 */
@@ -608,7 +589,7 @@ static int kvmppc_ps_two_in(struct kvm_v
 	if (scalar & SCALAR_HIGH)
 		ps1_in2 = ps0_in2;
 
-	func(&t, &ps1_out, &ps1_in1, &ps1_in2);
+	func(&vcpu->arch.fpscr, &ps1_out, &ps1_in1, &ps1_in2);
 
 	if (!(scalar & SCALAR_NO_PS1)) {
 		qpr[reg_out] = ps1_out;
@@ -622,31 +603,29 @@ static int kvmppc_ps_two_in(struct kvm_v
 
 static int kvmppc_ps_one_in(struct kvm_vcpu *vcpu, bool rc,
 				    int reg_out, int reg_in,
-				    void (*func)(struct thread_struct *t,
+				    void (*func)(u64 *t,
 						 u32 *dst, u32 *src1))
 {
 	u32 *qpr = vcpu->arch.qpr;
 	u64 *fpr = vcpu->arch.fpr;
 	u32 ps0_out, ps0_in;
 	u32 ps1_in;
-	struct thread_struct t;
-	t.fpscr.val = vcpu->arch.fpscr;
 
 	/* RC */
 	WARN_ON(rc);
 
 	/* PS0 */
-	cvt_df((double*)&fpr[reg_in], (float*)&ps0_in, &t);
-	func(&t, &ps0_out, &ps0_in);
+	kvm_cvt_df(&fpr[reg_in], &ps0_in, &vcpu->arch.fpscr);
+	func(&vcpu->arch.fpscr, &ps0_out, &ps0_in);
 
 	dprintk(KERN_INFO "PS1 ps0 -> f(0x%x) = 0x%x\n",
 			  ps0_in, ps0_out);
 
-	cvt_fd((float*)&ps0_out, (double*)&fpr[reg_out], &t);
+	kvm_cvt_fd(&ps0_out, &fpr[reg_out], &vcpu->arch.fpscr);
 
 	/* PS1 */
 	ps1_in = qpr[reg_in];
-	func(&t, &qpr[reg_out], &ps1_in);
+	func(&vcpu->arch.fpscr, &qpr[reg_out], &ps1_in);
 
 	dprintk(KERN_INFO "PS1 ps1 -> f(0x%x) = 0x%x\n",
 			  ps1_in, qpr[reg_out]);
@@ -672,13 +651,10 @@ int kvmppc_emulate_paired_single(struct
 
 	bool rcomp = (inst & 1) ? true : false;
 	u32 cr = kvmppc_get_cr(vcpu);
-	struct thread_struct t;
 #ifdef DEBUG
 	int i;
 #endif
 
-	t.fpscr.val = vcpu->arch.fpscr;
-
 	if (!kvmppc_inst_is_paired_single(vcpu, inst))
 		return EMULATE_FAIL;
 
@@ -695,7 +671,7 @@ int kvmppc_emulate_paired_single(struct
 #ifdef DEBUG
 	for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++) {
 		u32 f;
-		cvt_df((double*)&vcpu->arch.fpr[i], (float*)&f, &t);
+		kvm_cvt_df(&vcpu->arch.fpr[i], &f, &vcpu->arch.fpscr);
 		dprintk(KERN_INFO "FPR[%d] = 0x%x / 0x%llx    QPR[%d] = 0x%x\n",
 			i, f, vcpu->arch.fpr[i], i, vcpu->arch.qpr[i]);
 	}
@@ -819,8 +795,9 @@ int kvmppc_emulate_paired_single(struct
 			WARN_ON(rcomp);
 			vcpu->arch.fpr[ax_rd] = vcpu->arch.fpr[ax_ra];
 			/* vcpu->arch.qpr[ax_rd] = vcpu->arch.fpr[ax_rb]; */
-			cvt_df((double*)&vcpu->arch.fpr[ax_rb],
-			       (float*)&vcpu->arch.qpr[ax_rd], &t);
+			kvm_cvt_df(&vcpu->arch.fpr[ax_rb],
+				   &vcpu->arch.qpr[ax_rd],
+				   &vcpu->arch.fpscr);
 			break;
 		case OP_4X_PS_MERGE01:
 			WARN_ON(rcomp);
@@ -830,17 +807,20 @@ int kvmppc_emulate_paired_single(struct
 		case OP_4X_PS_MERGE10:
 			WARN_ON(rcomp);
 			/* vcpu->arch.fpr[ax_rd] = vcpu->arch.qpr[ax_ra]; */
-			cvt_fd((float*)&vcpu->arch.qpr[ax_ra],
-			       (double*)&vcpu->arch.fpr[ax_rd], &t);
+			kvm_cvt_fd(&vcpu->arch.qpr[ax_ra],
+				   &vcpu->arch.fpr[ax_rd],
+				   &vcpu->arch.fpscr);
 			/* vcpu->arch.qpr[ax_rd] = vcpu->arch.fpr[ax_rb]; */
-			cvt_df((double*)&vcpu->arch.fpr[ax_rb],
-			       (float*)&vcpu->arch.qpr[ax_rd], &t);
+			kvm_cvt_df(&vcpu->arch.fpr[ax_rb],
+				   &vcpu->arch.qpr[ax_rd],
+				   &vcpu->arch.fpscr);
 			break;
 		case OP_4X_PS_MERGE11:
 			WARN_ON(rcomp);
 			/* vcpu->arch.fpr[ax_rd] = vcpu->arch.qpr[ax_ra]; */
-			cvt_fd((float*)&vcpu->arch.qpr[ax_ra],
-			       (double*)&vcpu->arch.fpr[ax_rd], &t);
+			kvm_cvt_fd(&vcpu->arch.qpr[ax_ra],
+				   &vcpu->arch.fpr[ax_rd],
+				   &vcpu->arch.fpscr);
 			vcpu->arch.qpr[ax_rd] = vcpu->arch.qpr[ax_rb];
 			break;
 		}
@@ -1275,7 +1255,7 @@ int kvmppc_emulate_paired_single(struct
 #ifdef DEBUG
 	for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++) {
 		u32 f;
-		cvt_df((double*)&vcpu->arch.fpr[i], (float*)&f, &t);
+		kvm_cvt_df(&vcpu->arch.fpr[i], &f, &vcpu->arch.fpscr);
 		dprintk(KERN_INFO "FPR[%d] = 0x%x\n", i, f);
 	}
 #endif
--- linux-2.6.35-rc1.orig/arch/powerpc/kvm/fpu.S
+++ linux-2.6.35-rc1/arch/powerpc/kvm/fpu.S
@@ -271,3 +271,21 @@ FPD_THREE_IN(fmsub)
 FPD_THREE_IN(fmadd)
 FPD_THREE_IN(fnmsub)
 FPD_THREE_IN(fnmadd)
+
+_GLOBAL(kvm_cvt_fd)
+	lfd	0,0(r5)			/* load up fpscr value */
+	MTFSF_L(0)
+	lfs	0,0(r3)
+	stfd	0,0(r4)
+	mffs	0
+	stfd	0,0(r5)			/* save new fpscr value */
+	blr
+
+_GLOBAL(kvm_cvt_df)
+	lfd	0,0(r5)			/* load up fpscr value */
+	MTFSF_L(0)
+	lfd	0,0(r3)
+	stfs	0,0(r4)
+	mffs	0
+	stfd	0,0(r5)			/* save new fpscr value */
+	blr

-- 
Andreas Schwab, schwab@linux-m68k.org
GPG Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: [PATCH] powerpc: Don't export cvt_fd & _df when CONFIG_PPC_FPU is not set

[Alexander Graf]

Andreas Schwab wrote:
> Alexander Graf <agraf@suse.de> writes:
>
>   
>> Mind to send it over so I can take a look at it :)? Getting rid of the
>> task_struct structs lying in the stack is certainly a good idea.
>>     
>
> Still untested, because rc1 does not boot.
>   

Normal guests work great with the patch and from a quick look over it,
it seems sane. I couldn't test the paired single implementation yet,
because my userspace app for that is in a flux right now.

Either way, thanks a lot for the patch. Mind to wrap it up and submit
it? I'll fix breakages later on. Please CC kvm-ppc@vger.kernel.org and
kvm@vger.kernel.org, so Avi or Marcelo can pick it up on the KVM tree.

Alex

Re: [PATCH] powerpc: Don't export cvt_fd & _df when CONFIG_PPC_FPU is not set

[Benjamin Herrenschmidt]

On Mon, 2010-05-31 at 20:06 +0200, Andreas Schwab wrote:
> Alexander Graf <agraf@suse.de> writes:
> 
> > Mind to send it over so I can take a look at it :)? Getting rid of the
> > task_struct structs lying in the stack is certainly a good idea.
> 
> Still untested, because rc1 does not boot.

Ouch ! What machine ?

Cheers,
Ben.

Re: [PATCH] powerpc: Don't export cvt_fd & _df when CONFIG_PPC_FPU is not set

[Andreas Schwab]

Benjamin Herrenschmidt <benh@kernel.crashing.org> writes:

> On Mon, 2010-05-31 at 20:06 +0200, Andreas Schwab wrote:
>> Alexander Graf <agraf@suse.de> writes:
>> 
>> > Mind to send it over so I can take a look at it :)? Getting rid of the
>> > task_struct structs lying in the stack is certainly a good idea.
>> 
>> Still untested, because rc1 does not boot.
>
> Ouch ! What machine ?

It's crashing in macio_pci_add_devices.

Andreas.

-- 
Andreas Schwab, schwab@redhat.com
GPG Key fingerprint = D4E8 DBE3 3813 BB5D FA84  5EC7 45C6 250E 6F00 984E
"And now for something completely different."

Re: [PATCH] powerpc: Don't export cvt_fd & _df when CONFIG_PPC_FPU is not set

[Benjamin Herrenschmidt]

On Tue, 2010-06-01 at 11:36 +0200, Andreas Schwab wrote:
> Benjamin Herrenschmidt <benh@kernel.crashing.org> writes:
> 
> > On Mon, 2010-05-31 at 20:06 +0200, Andreas Schwab wrote:
> >> Alexander Graf <agraf@suse.de> writes:
> >> 
> >> > Mind to send it over so I can take a look at it :)? Getting rid of the
> >> > task_struct structs lying in the stack is certainly a good idea.
> >> 
> >> Still untested, because rc1 does not boot.
> >
> > Ouch ! What machine ?
> 
> It's crashing in macio_pci_add_devices.

Ok, so a G5 ? I'll have a look tomorrow.

Cheers,
Ben.

[PATCH] macio: don't dereference pointer before null check

[Andreas Schwab]

Signed-off-by: Andreas Schwab <schwab@linux-m68k.org>
---
 drivers/macintosh/macio_asic.c |    7 ++++---
 1 files changed, 4 insertions(+), 3 deletions(-)

diff --git a/drivers/macintosh/macio_asic.c b/drivers/macintosh/macio_asic.c
index 9714780..40390ac 100644
--- a/drivers/macintosh/macio_asic.c
+++ b/drivers/macintosh/macio_asic.c
@@ -492,8 +492,8 @@ static void macio_pci_add_devices(struct macio_chip *chip)
 	}
 
 	/* Add media bay devices if any */
-	pnode = mbdev->ofdev.dev.of_node;
-	if (mbdev)
+	if (mbdev) {
+		pnode = mbdev->ofdev.dev.of_node;
 		for (np = NULL; (np = of_get_next_child(pnode, np)) != NULL;) {
 			if (macio_skip_device(np))
 				continue;
@@ -502,10 +502,11 @@ static void macio_pci_add_devices(struct macio_chip *chip)
 						 mbdev,  root_res) == NULL)
 				of_node_put(np);
 		}
+	}
 
 	/* Add serial ports if any */
-	pnode = sdev->ofdev.dev.of_node;
 	if (sdev) {
+		pnode = sdev->ofdev.dev.of_node;
 		for (np = NULL; (np = of_get_next_child(pnode, np)) != NULL;) {
 			if (macio_skip_device(np))
 				continue;
-- 
1.7.1

-- 
Andreas Schwab, schwab@linux-m68k.org
GPG Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: [PATCH] powerpc: Don't export cvt_fd & _df when CONFIG_PPC_FPU is not set

[Andreas Schwab]

Andreas Schwab <schwab@redhat.com> writes:

> Benjamin Herrenschmidt <benh@kernel.crashing.org> writes:
>
>> On Mon, 2010-05-31 at 20:06 +0200, Andreas Schwab wrote:
>>> Alexander Graf <agraf@suse.de> writes:
>>> 
>>> > Mind to send it over so I can take a look at it :)? Getting rid of the
>>> > task_struct structs lying in the stack is certainly a good idea.
>>> 
>>> Still untested, because rc1 does not boot.
>>
>> Ouch ! What machine ?
>
> It's crashing in macio_pci_add_devices.

And snd-aoa is broken, too.

Andreas.

-- 
Andreas Schwab, schwab@linux-m68k.org
GPG Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: [PATCH] powerpc: Don't export cvt_fd & _df when CONFIG_PPC_FPU is not set

[Benjamin Herrenschmidt]

> > It's crashing in macio_pci_add_devices.
> 
> And snd-aoa is broken, too.

Right, all macio_driver's are. I have a fix, testing now.

Cheers,
Ben.

Re: [PATCH] powerpc: Don't export cvt_fd & _df when CONFIG_PPC_FPU is not set

[Benjamin Herrenschmidt]

On Mon, 2010-05-31 at 11:33 +0200, Alexander Graf wrote:
> > Alex, this is just a temporary fix to remove the build breakage for
> 40x
> > etc... but please, update KVM to just build-in its own.
> 
> What's the bad part in reusing the existing code?

I'm not saying we should dup the code, but I'm sure we can find a better
way for that sort of stuff, along with other base support bits like you
normally find in libgcc, to be linked directly with all modules.

calling out of modules symbols sucks for perf too anyways.

> > Don't we have .o's that are always built into modules nowadays ?
> 
> I don't think I understand what you mean.

Ben.