misc riscv cleanups

misc riscv cleanups

[Christoph Hellwig]

Hi all,

this series has a couple small cleanups and micro-optimizations
that resulted from the nommu work.  Pushing them out ASAP to avoid
conflicts the tlbflush work from Atish.


_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv

[PATCH 1/6] riscv: refactor the IPI code

[Christoph Hellwig]

This prepares for adding native non-SBI IPI code.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 arch/riscv/kernel/smp.c | 55 +++++++++++++++++++++++------------------
 1 file changed, 31 insertions(+), 24 deletions(-)

diff --git a/arch/riscv/kernel/smp.c b/arch/riscv/kernel/smp.c
index 5a9834503a2f..8cd730239613 100644
--- a/arch/riscv/kernel/smp.c
+++ b/arch/riscv/kernel/smp.c
@@ -78,13 +78,38 @@ static void ipi_stop(void)
 		wait_for_interrupt();
 }
 
+static void send_ipi_mask(const struct cpumask *mask, enum ipi_message_type op)
+{
+	int cpuid, hartid;
+	struct cpumask hartid_mask;
+
+	cpumask_clear(&hartid_mask);
+	mb();
+	for_each_cpu(cpuid, mask) {
+		set_bit(op, &ipi_data[cpuid].bits);
+		hartid = cpuid_to_hartid_map(cpuid);
+		cpumask_set_cpu(hartid, &hartid_mask);
+	}
+	mb();
+	sbi_send_ipi(cpumask_bits(&hartid_mask));
+}
+
+static void send_ipi_single(int cpu, enum ipi_message_type op)
+{
+	send_ipi_mask(cpumask_of(cpu), op);
+}
+
+static inline void clear_ipi(void)
+{
+	csr_clear(CSR_SIP, SIE_SSIE);
+}
+
 void riscv_software_interrupt(void)
 {
 	unsigned long *pending_ipis = &ipi_data[smp_processor_id()].bits;
 	unsigned long *stats = ipi_data[smp_processor_id()].stats;
 
-	/* Clear pending IPI */
-	csr_clear(CSR_SIP, SIE_SSIE);
+	clear_ipi();
 
 	while (true) {
 		unsigned long ops;
@@ -118,23 +143,6 @@ void riscv_software_interrupt(void)
 	}
 }
 
-static void
-send_ipi_message(const struct cpumask *to_whom, enum ipi_message_type operation)
-{
-	int cpuid, hartid;
-	struct cpumask hartid_mask;
-
-	cpumask_clear(&hartid_mask);
-	mb();
-	for_each_cpu(cpuid, to_whom) {
-		set_bit(operation, &ipi_data[cpuid].bits);
-		hartid = cpuid_to_hartid_map(cpuid);
-		cpumask_set_cpu(hartid, &hartid_mask);
-	}
-	mb();
-	sbi_send_ipi(cpumask_bits(&hartid_mask));
-}
-
 static const char * const ipi_names[] = {
 	[IPI_RESCHEDULE]	= "Rescheduling interrupts",
 	[IPI_CALL_FUNC]		= "Function call interrupts",
@@ -156,12 +164,12 @@ void show_ipi_stats(struct seq_file *p, int prec)
 
 void arch_send_call_function_ipi_mask(struct cpumask *mask)
 {
-	send_ipi_message(mask, IPI_CALL_FUNC);
+	send_ipi_mask(mask, IPI_CALL_FUNC);
 }
 
 void arch_send_call_function_single_ipi(int cpu)
 {
-	send_ipi_message(cpumask_of(cpu), IPI_CALL_FUNC);
+	send_ipi_single(cpu, IPI_CALL_FUNC);
 }
 
 void smp_send_stop(void)
@@ -176,7 +184,7 @@ void smp_send_stop(void)
 
 		if (system_state <= SYSTEM_RUNNING)
 			pr_crit("SMP: stopping secondary CPUs\n");
-		send_ipi_message(&mask, IPI_CPU_STOP);
+		send_ipi_mask(&mask, IPI_CPU_STOP);
 	}
 
 	/* Wait up to one second for other CPUs to stop */
@@ -191,6 +199,5 @@ void smp_send_stop(void)
 
 void smp_send_reschedule(int cpu)
 {
-	send_ipi_message(cpumask_of(cpu), IPI_RESCHEDULE);
+	send_ipi_single(cpu, IPI_RESCHEDULE);
 }
-
-- 
2.20.1


_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv

[PATCH 2/6] riscv: cleanup send_ipi_mask

[Christoph Hellwig]

Use the special barriers for atomic bitops to make the intention
a little more clear, and use riscv_cpuid_to_hartid_mask instead of
open coding it.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 arch/riscv/kernel/smp.c | 16 +++++++---------
 1 file changed, 7 insertions(+), 9 deletions(-)

diff --git a/arch/riscv/kernel/smp.c b/arch/riscv/kernel/smp.c
index 8cd730239613..2e21669aa068 100644
--- a/arch/riscv/kernel/smp.c
+++ b/arch/riscv/kernel/smp.c
@@ -80,17 +80,15 @@ static void ipi_stop(void)
 
 static void send_ipi_mask(const struct cpumask *mask, enum ipi_message_type op)
 {
-	int cpuid, hartid;
 	struct cpumask hartid_mask;
+	int cpu;
 
-	cpumask_clear(&hartid_mask);
-	mb();
-	for_each_cpu(cpuid, mask) {
-		set_bit(op, &ipi_data[cpuid].bits);
-		hartid = cpuid_to_hartid_map(cpuid);
-		cpumask_set_cpu(hartid, &hartid_mask);
-	}
-	mb();
+	smp_mb__before_atomic();
+	for_each_cpu(cpu, mask)
+		set_bit(op, &ipi_data[cpu].bits);
+	smp_mb__after_atomic();
+
+	riscv_cpuid_to_hartid_mask(mask, &hartid_mask);
 	sbi_send_ipi(cpumask_bits(&hartid_mask));
 }
 
-- 
2.20.1


_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv

[PATCH 3/6] riscv: optimize send_ipi_single

[Christoph Hellwig]

Don't go through send_ipi_mask, but just set the op bit an then pass a
simple generate hartid mask directly to sbi_send_ipi.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 arch/riscv/kernel/smp.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/arch/riscv/kernel/smp.c b/arch/riscv/kernel/smp.c
index 2e21669aa068..a3715d621f60 100644
--- a/arch/riscv/kernel/smp.c
+++ b/arch/riscv/kernel/smp.c
@@ -94,7 +94,13 @@ static void send_ipi_mask(const struct cpumask *mask, enum ipi_message_type op)
 
 static void send_ipi_single(int cpu, enum ipi_message_type op)
 {
-	send_ipi_mask(cpumask_of(cpu), op);
+	int hartid = cpuid_to_hartid_map(cpu);
+
+	smp_mb__before_atomic();
+	set_bit(op, &ipi_data[cpu].bits);
+	smp_mb__after_atomic();
+
+	sbi_send_ipi(cpumask_bits(cpumask_of(hartid)));
 }
 
 static inline void clear_ipi(void)
-- 
2.20.1


_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv

[PATCH 4/6] riscv: cleanup riscv_cpuid_to_hartid_mask

[Christoph Hellwig]

Move the initial clearing of the mask from the callers to
riscv_cpuid_to_hartid_mask, and remove the unused !CONFIG_SMP stub.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 arch/riscv/include/asm/smp.h      | 6 ------
 arch/riscv/include/asm/tlbflush.h | 1 -
 arch/riscv/kernel/smp.c           | 1 +
 arch/riscv/mm/cacheflush.c        | 1 -
 4 files changed, 1 insertion(+), 8 deletions(-)

diff --git a/arch/riscv/include/asm/smp.h b/arch/riscv/include/asm/smp.h
index c6ed4d691def..a83451d73a4e 100644
--- a/arch/riscv/include/asm/smp.h
+++ b/arch/riscv/include/asm/smp.h
@@ -61,11 +61,5 @@ static inline unsigned long cpuid_to_hartid_map(int cpu)
 	return boot_cpu_hartid;
 }
 
-static inline void riscv_cpuid_to_hartid_mask(const struct cpumask *in,
-					      struct cpumask *out)
-{
-	cpumask_set_cpu(cpuid_to_hartid_map(0), out);
-}
-
 #endif /* CONFIG_SMP */
 #endif /* _ASM_RISCV_SMP_H */
diff --git a/arch/riscv/include/asm/tlbflush.h b/arch/riscv/include/asm/tlbflush.h
index 4d9bbe8438bf..df31fe2ed09c 100644
--- a/arch/riscv/include/asm/tlbflush.h
+++ b/arch/riscv/include/asm/tlbflush.h
@@ -47,7 +47,6 @@ static inline void remote_sfence_vma(struct cpumask *cmask, unsigned long start,
 {
 	struct cpumask hmask;
 
-	cpumask_clear(&hmask);
 	riscv_cpuid_to_hartid_mask(cmask, &hmask);
 	sbi_remote_sfence_vma(hmask.bits, start, size);
 }
diff --git a/arch/riscv/kernel/smp.c b/arch/riscv/kernel/smp.c
index a3715d621f60..3836760d7aaf 100644
--- a/arch/riscv/kernel/smp.c
+++ b/arch/riscv/kernel/smp.c
@@ -56,6 +56,7 @@ void riscv_cpuid_to_hartid_mask(const struct cpumask *in, struct cpumask *out)
 {
 	int cpu;
 
+	cpumask_clear(out);
 	for_each_cpu(cpu, in)
 		cpumask_set_cpu(cpuid_to_hartid_map(cpu), out);
 }
diff --git a/arch/riscv/mm/cacheflush.c b/arch/riscv/mm/cacheflush.c
index 9ebcff8ba263..3f15938dec89 100644
--- a/arch/riscv/mm/cacheflush.c
+++ b/arch/riscv/mm/cacheflush.c
@@ -47,7 +47,6 @@ void flush_icache_mm(struct mm_struct *mm, bool local)
 	cpumask_andnot(&others, mm_cpumask(mm), cpumask_of(cpu));
 	local |= cpumask_empty(&others);
 	if (mm != current->active_mm || !local) {
-		cpumask_clear(&hmask);
 		riscv_cpuid_to_hartid_mask(&others, &hmask);
 		sbi_remote_fence_i(hmask.bits);
 	} else {
-- 
2.20.1


_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv

[PATCH 5/6] riscv: don't use the rdtime(h) pseudo-instructions

[Christoph Hellwig]

If we just use the CSRs that these map to directly the code is simpler
and doesn't require extra inline assembly code.  Also fix up the top-level
comment in timer-riscv.c to not talk about the cycle count or mention
details of the clocksource interface, of which this file is just a
consumer.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 arch/riscv/include/asm/timex.h    | 44 +++++++++++++++----------------
 drivers/clocksource/timer-riscv.c | 17 +++---------
 2 files changed, 25 insertions(+), 36 deletions(-)

diff --git a/arch/riscv/include/asm/timex.h b/arch/riscv/include/asm/timex.h
index 6a703ec9d796..c7ef131b9e4c 100644
--- a/arch/riscv/include/asm/timex.h
+++ b/arch/riscv/include/asm/timex.h
@@ -6,43 +6,41 @@
 #ifndef _ASM_RISCV_TIMEX_H
 #define _ASM_RISCV_TIMEX_H
 
-#include <asm/param.h>
+#include <asm/csr.h>
 
 typedef unsigned long cycles_t;
 
-static inline cycles_t get_cycles_inline(void)
+static inline cycles_t get_cycles(void)
 {
-	cycles_t n;
-
-	__asm__ __volatile__ (
-		"rdtime %0"
-		: "=r" (n));
-	return n;
+	return csr_read(CSR_TIME);
 }
-#define get_cycles get_cycles_inline
+#define get_cycles get_cycles
 
 #ifdef CONFIG_64BIT
-static inline uint64_t get_cycles64(void)
+static inline u64 get_cycles64(void)
+{
+	return get_cycles();
+}
+#else /* CONFIG_64BIT */
+static inline u32 get_cycles_hi(void)
 {
-        return get_cycles();
+	return csr_read(CSR_TIMEH);
 }
-#else
-static inline uint64_t get_cycles64(void)
+
+static inline u64 get_cycles64(void)
 {
-	u32 lo, hi, tmp;
-	__asm__ __volatile__ (
-		"1:\n"
-		"rdtimeh %0\n"
-		"rdtime %1\n"
-		"rdtimeh %2\n"
-		"bne %0, %2, 1b"
-		: "=&r" (hi), "=&r" (lo), "=&r" (tmp));
+	u32 hi, lo;
+
+	do {
+		hi = get_cycles_hi();
+		lo = get_cycles();
+	} while (hi != get_cycles_hi());
+
 	return ((u64)hi << 32) | lo;
 }
-#endif
+#endif /* CONFIG_64BIT */
 
 #define ARCH_HAS_READ_CURRENT_TIMER
-
 static inline int read_current_timer(unsigned long *timer_val)
 {
 	*timer_val = get_cycles();
diff --git a/drivers/clocksource/timer-riscv.c b/drivers/clocksource/timer-riscv.c
index 09e031176bc6..470c7ef02ea4 100644
--- a/drivers/clocksource/timer-riscv.c
+++ b/drivers/clocksource/timer-riscv.c
@@ -2,6 +2,10 @@
 /*
  * Copyright (C) 2012 Regents of the University of California
  * Copyright (C) 2017 SiFive
+ *
+ * All RISC-V systems have a timer attached to every hart.  These timers can be
+ * read from the "time" and "timeh" CSRs, and can use the SBI to setup
+ * events.
  */
 #include <linux/clocksource.h>
 #include <linux/clockchips.h>
@@ -12,19 +16,6 @@
 #include <asm/smp.h>
 #include <asm/sbi.h>
 
-/*
- * All RISC-V systems have a timer attached to every hart.  These timers can be
- * read by the 'rdcycle' pseudo instruction, and can use the SBI to setup
- * events.  In order to abstract the architecture-specific timer reading and
- * setting functions away from the clock event insertion code, we provide
- * function pointers to the clockevent subsystem that perform two basic
- * operations: rdtime() reads the timer on the current CPU, and
- * next_event(delta) sets the next timer event to 'delta' cycles in the future.
- * As the timers are inherently a per-cpu resource, these callbacks perform
- * operations on the current hart.  There is guaranteed to be exactly one timer
- * per hart on all RISC-V systems.
- */
-
 static int riscv_clock_next_event(unsigned long delta,
 		struct clock_event_device *ce)
 {
-- 
2.20.1


_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv

[PATCH 6/6] riscv: move the TLB flush logic out of line

[Christoph Hellwig]

The TLB flush logic is going to become more complex.  Start moving
it out of line.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 arch/riscv/include/asm/tlbflush.h | 37 ++++++-------------------------
 arch/riscv/mm/Makefile            |  3 +++
 arch/riscv/mm/tlbflush.c          | 35 +++++++++++++++++++++++++++++
 3 files changed, 45 insertions(+), 30 deletions(-)
 create mode 100644 arch/riscv/mm/tlbflush.c

diff --git a/arch/riscv/include/asm/tlbflush.h b/arch/riscv/include/asm/tlbflush.h
index df31fe2ed09c..075a784c66c5 100644
--- a/arch/riscv/include/asm/tlbflush.h
+++ b/arch/riscv/include/asm/tlbflush.h
@@ -25,8 +25,13 @@ static inline void local_flush_tlb_page(unsigned long addr)
 	__asm__ __volatile__ ("sfence.vma %0" : : "r" (addr) : "memory");
 }
 
-#ifndef CONFIG_SMP
-
+#ifdef CONFIG_SMP
+void flush_tlb_all(void);
+void flush_tlb_mm(struct mm_struct *mm);
+void flush_tlb_page(struct vm_area_struct *vma, unsigned long addr);
+void flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
+		unsigned long end);
+#else /* CONFIG_SMP */
 #define flush_tlb_all() local_flush_tlb_all()
 #define flush_tlb_page(vma, addr) local_flush_tlb_page(addr)
 
@@ -37,34 +42,6 @@ static inline void flush_tlb_range(struct vm_area_struct *vma,
 }
 
 #define flush_tlb_mm(mm) flush_tlb_all()
-
-#else /* CONFIG_SMP */
-
-#include <asm/sbi.h>
-
-static inline void remote_sfence_vma(struct cpumask *cmask, unsigned long start,
-				     unsigned long size)
-{
-	struct cpumask hmask;
-
-	riscv_cpuid_to_hartid_mask(cmask, &hmask);
-	sbi_remote_sfence_vma(hmask.bits, start, size);
-}
-
-#define flush_tlb_all() sbi_remote_sfence_vma(NULL, 0, -1)
-
-#define flush_tlb_range(vma, start, end) \
-	remote_sfence_vma(mm_cpumask((vma)->vm_mm), start, (end) - (start))
-
-static inline void flush_tlb_page(struct vm_area_struct *vma,
-				  unsigned long addr)
-{
-	flush_tlb_range(vma, addr, addr + PAGE_SIZE);
-}
-
-#define flush_tlb_mm(mm)				\
-	remote_sfence_vma(mm_cpumask(mm), 0, -1)
-
 #endif /* CONFIG_SMP */
 
 /* Flush a range of kernel pages */
diff --git a/arch/riscv/mm/Makefile b/arch/riscv/mm/Makefile
index 74055e1d6f21..9d9a17335686 100644
--- a/arch/riscv/mm/Makefile
+++ b/arch/riscv/mm/Makefile
@@ -13,4 +13,7 @@ obj-y += cacheflush.o
 obj-y += context.o
 obj-y += sifive_l2_cache.o
 
+ifeq ($(CONFIG_MMU),y)
+obj-$(CONFIG_SMP) += tlbflush.o
+endif
 obj-$(CONFIG_HUGETLB_PAGE) += hugetlbpage.o
diff --git a/arch/riscv/mm/tlbflush.c b/arch/riscv/mm/tlbflush.c
new file mode 100644
index 000000000000..df93b26f1b9d
--- /dev/null
+++ b/arch/riscv/mm/tlbflush.c
@@ -0,0 +1,35 @@
+// SPDX-License-Identifier: GPL-2.0
+
+#include <linux/mm.h>
+#include <linux/smp.h>
+#include <asm/sbi.h>
+
+void flush_tlb_all(void)
+{
+	sbi_remote_sfence_vma(NULL, 0, -1);
+}
+
+static void __sbi_tlb_flush_range(struct cpumask *cmask, unsigned long start,
+		unsigned long size)
+{
+	struct cpumask hmask;
+
+	riscv_cpuid_to_hartid_mask(cmask, &hmask);
+	sbi_remote_sfence_vma(hmask.bits, start, size);
+}
+
+void flush_tlb_mm(struct mm_struct *mm)
+{
+	__sbi_tlb_flush_range(mm_cpumask(mm), 0, -1);
+}
+
+void flush_tlb_page(struct vm_area_struct *vma, unsigned long addr)
+{
+	__sbi_tlb_flush_range(mm_cpumask(vma->vm_mm), addr, PAGE_SIZE);
+}
+
+void flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
+		unsigned long end)
+{
+	__sbi_tlb_flush_range(mm_cpumask(vma->vm_mm), start, end - start);
+}
-- 
2.20.1


_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv

Re: [PATCH 4/6] riscv: cleanup riscv_cpuid_to_hartid_mask

[Atish Patra]

On Wed, 2019-08-21 at 23:58 +0900, Christoph Hellwig wrote:
> Move the initial clearing of the mask from the callers to
> riscv_cpuid_to_hartid_mask, and remove the unused !CONFIG_SMP stub.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> ---
>  arch/riscv/include/asm/smp.h      | 6 ------
>  arch/riscv/include/asm/tlbflush.h | 1 -
>  arch/riscv/kernel/smp.c           | 1 +
>  arch/riscv/mm/cacheflush.c        | 1 -
>  4 files changed, 1 insertion(+), 8 deletions(-)
> 
> diff --git a/arch/riscv/include/asm/smp.h
> b/arch/riscv/include/asm/smp.h
> index c6ed4d691def..a83451d73a4e 100644
> --- a/arch/riscv/include/asm/smp.h
> +++ b/arch/riscv/include/asm/smp.h
> @@ -61,11 +61,5 @@ static inline unsigned long
> cpuid_to_hartid_map(int cpu)
>  	return boot_cpu_hartid;
>  }
>  
> -static inline void riscv_cpuid_to_hartid_mask(const struct cpumask
> *in,
> -					      struct cpumask *out)
> -{
> -	cpumask_set_cpu(cpuid_to_hartid_map(0), out);
> -}
> -
>  #endif /* CONFIG_SMP */
>  #endif /* _ASM_RISCV_SMP_H */
> diff --git a/arch/riscv/include/asm/tlbflush.h
> b/arch/riscv/include/asm/tlbflush.h
> index 4d9bbe8438bf..df31fe2ed09c 100644
> --- a/arch/riscv/include/asm/tlbflush.h
> +++ b/arch/riscv/include/asm/tlbflush.h
> @@ -47,7 +47,6 @@ static inline void remote_sfence_vma(struct cpumask
> *cmask, unsigned long start,
>  {
>  	struct cpumask hmask;
>  
> -	cpumask_clear(&hmask);
>  	riscv_cpuid_to_hartid_mask(cmask, &hmask);
>  	sbi_remote_sfence_vma(hmask.bits, start, size);
>  }
> diff --git a/arch/riscv/kernel/smp.c b/arch/riscv/kernel/smp.c
> index a3715d621f60..3836760d7aaf 100644
> --- a/arch/riscv/kernel/smp.c
> +++ b/arch/riscv/kernel/smp.c
> @@ -56,6 +56,7 @@ void riscv_cpuid_to_hartid_mask(const struct
> cpumask *in, struct cpumask *out)
>  {
>  	int cpu;
>  
> +	cpumask_clear(out);
>  	for_each_cpu(cpu, in)
>  		cpumask_set_cpu(cpuid_to_hartid_map(cpu), out);
>  }
> diff --git a/arch/riscv/mm/cacheflush.c b/arch/riscv/mm/cacheflush.c
> index 9ebcff8ba263..3f15938dec89 100644
> --- a/arch/riscv/mm/cacheflush.c
> +++ b/arch/riscv/mm/cacheflush.c
> @@ -47,7 +47,6 @@ void flush_icache_mm(struct mm_struct *mm, bool
> local)
>  	cpumask_andnot(&others, mm_cpumask(mm), cpumask_of(cpu));
>  	local |= cpumask_empty(&others);
>  	if (mm != current->active_mm || !local) {
> -		cpumask_clear(&hmask);
>  		riscv_cpuid_to_hartid_mask(&others, &hmask);
>  		sbi_remote_fence_i(hmask.bits);
>  	} else {

Reviewed-by: Atish Patra <atish.patra@wdc.com>

-- 
Regards,
Atish
_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv

Re: [PATCH 6/6] riscv: move the TLB flush logic out of line

[Atish Patra]

On Wed, 2019-08-21 at 23:58 +0900, Christoph Hellwig wrote:
> The TLB flush logic is going to become more complex.  Start moving
> it out of line.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> ---
>  arch/riscv/include/asm/tlbflush.h | 37 ++++++-----------------------
> --
>  arch/riscv/mm/Makefile            |  3 +++
>  arch/riscv/mm/tlbflush.c          | 35 +++++++++++++++++++++++++++++
>  3 files changed, 45 insertions(+), 30 deletions(-)
>  create mode 100644 arch/riscv/mm/tlbflush.c
> 
> diff --git a/arch/riscv/include/asm/tlbflush.h
> b/arch/riscv/include/asm/tlbflush.h
> index df31fe2ed09c..075a784c66c5 100644
> --- a/arch/riscv/include/asm/tlbflush.h
> +++ b/arch/riscv/include/asm/tlbflush.h
> @@ -25,8 +25,13 @@ static inline void local_flush_tlb_page(unsigned
> long addr)
>  	__asm__ __volatile__ ("sfence.vma %0" : : "r" (addr) :
> "memory");
>  }
>  
> -#ifndef CONFIG_SMP
> -
> +#ifdef CONFIG_SMP
> +void flush_tlb_all(void);
> +void flush_tlb_mm(struct mm_struct *mm);
> +void flush_tlb_page(struct vm_area_struct *vma, unsigned long addr);
> +void flush_tlb_range(struct vm_area_struct *vma, unsigned long
> start,
> +		unsigned long end);
> +#else /* CONFIG_SMP */
>  #define flush_tlb_all() local_flush_tlb_all()
>  #define flush_tlb_page(vma, addr) local_flush_tlb_page(addr)
>  
> @@ -37,34 +42,6 @@ static inline void flush_tlb_range(struct
> vm_area_struct *vma,
>  }
>  
>  #define flush_tlb_mm(mm) flush_tlb_all()
> -
> -#else /* CONFIG_SMP */
> -
> -#include <asm/sbi.h>
> -
> -static inline void remote_sfence_vma(struct cpumask *cmask, unsigned
> long start,
> -				     unsigned long size)
> -{
> -	struct cpumask hmask;
> -
> -	riscv_cpuid_to_hartid_mask(cmask, &hmask);
> -	sbi_remote_sfence_vma(hmask.bits, start, size);
> -}
> -
> -#define flush_tlb_all() sbi_remote_sfence_vma(NULL, 0, -1)
> -
> -#define flush_tlb_range(vma, start, end) \
> -	remote_sfence_vma(mm_cpumask((vma)->vm_mm), start, (end) -
> (start))
> -
> -static inline void flush_tlb_page(struct vm_area_struct *vma,
> -				  unsigned long addr)
> -{
> -	flush_tlb_range(vma, addr, addr + PAGE_SIZE);
> -}
> -
> -#define flush_tlb_mm(mm)				\
> -	remote_sfence_vma(mm_cpumask(mm), 0, -1)
> -
>  #endif /* CONFIG_SMP */
>  
>  /* Flush a range of kernel pages */
> diff --git a/arch/riscv/mm/Makefile b/arch/riscv/mm/Makefile
> index 74055e1d6f21..9d9a17335686 100644
> --- a/arch/riscv/mm/Makefile
> +++ b/arch/riscv/mm/Makefile
> @@ -13,4 +13,7 @@ obj-y += cacheflush.o
>  obj-y += context.o
>  obj-y += sifive_l2_cache.o
>  
> +ifeq ($(CONFIG_MMU),y)
> +obj-$(CONFIG_SMP) += tlbflush.o
> +endif
>  obj-$(CONFIG_HUGETLB_PAGE) += hugetlbpage.o
> diff --git a/arch/riscv/mm/tlbflush.c b/arch/riscv/mm/tlbflush.c
> new file mode 100644
> index 000000000000..df93b26f1b9d
> --- /dev/null
> +++ b/arch/riscv/mm/tlbflush.c
> @@ -0,0 +1,35 @@
> +// SPDX-License-Identifier: GPL-2.0
> +
> +#include <linux/mm.h>
> +#include <linux/smp.h>
> +#include <asm/sbi.h>
> +
> +void flush_tlb_all(void)
> +{
> +	sbi_remote_sfence_vma(NULL, 0, -1);
> +}
> +
> +static void __sbi_tlb_flush_range(struct cpumask *cmask, unsigned
> long start,
> +		unsigned long size)
> +{
> +	struct cpumask hmask;
> +
> +	riscv_cpuid_to_hartid_mask(cmask, &hmask);
> +	sbi_remote_sfence_vma(hmask.bits, start, size);
> +}
> +
> +void flush_tlb_mm(struct mm_struct *mm)
> +{
> +	__sbi_tlb_flush_range(mm_cpumask(mm), 0, -1);
> +}
> +
> +void flush_tlb_page(struct vm_area_struct *vma, unsigned long addr)
> +{
> +	__sbi_tlb_flush_range(mm_cpumask(vma->vm_mm), addr, PAGE_SIZE);
> +}
> +
> +void flush_tlb_range(struct vm_area_struct *vma, unsigned long
> start,
> +		unsigned long end)
> +{
> +	__sbi_tlb_flush_range(mm_cpumask(vma->vm_mm), start, end -
> start);
> +}

Reviewed-by: Atish Patra <atish.patra@wdc.com>

-- 
Regards,
Atish
_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv

Re: [PATCH 2/6] riscv: cleanup send_ipi_mask

[Atish Patra]

On Wed, 2019-08-21 at 23:58 +0900, Christoph Hellwig wrote:
> Use the special barriers for atomic bitops to make the intention
> a little more clear, and use riscv_cpuid_to_hartid_mask instead of
> open coding it.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> ---
>  arch/riscv/kernel/smp.c | 16 +++++++---------
>  1 file changed, 7 insertions(+), 9 deletions(-)
> 
> diff --git a/arch/riscv/kernel/smp.c b/arch/riscv/kernel/smp.c
> index 8cd730239613..2e21669aa068 100644
> --- a/arch/riscv/kernel/smp.c
> +++ b/arch/riscv/kernel/smp.c
> @@ -80,17 +80,15 @@ static void ipi_stop(void)
>  
>  static void send_ipi_mask(const struct cpumask *mask, enum
> ipi_message_type op)
>  {
> -	int cpuid, hartid;
>  	struct cpumask hartid_mask;
> +	int cpu;
>  
> -	cpumask_clear(&hartid_mask);
> -	mb();
> -	for_each_cpu(cpuid, mask) {
> -		set_bit(op, &ipi_data[cpuid].bits);
> -		hartid = cpuid_to_hartid_map(cpuid);
> -		cpumask_set_cpu(hartid, &hartid_mask);
> -	}
> -	mb();
> +	smp_mb__before_atomic();
> +	for_each_cpu(cpu, mask)
> +		set_bit(op, &ipi_data[cpu].bits);
> +	smp_mb__after_atomic();
> +
> +	riscv_cpuid_to_hartid_mask(mask, &hartid_mask);

Isn't that less optimized than previous one ?

This will iterate all the cpus set in mask twice during every ipi sent.
For now, we won't see any different. As we have more number of cpus in
RISC-V (hopefully one day ;) ;)), this may affect the performance.

>  	sbi_send_ipi(cpumask_bits(&hartid_mask));
>  }
>  

-- 
Regards,
Atish
_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv

Re: [PATCH 3/6] riscv: optimize send_ipi_single

[Atish Patra]

On Wed, 2019-08-21 at 23:58 +0900, Christoph Hellwig wrote:
> Don't go through send_ipi_mask, but just set the op bit an then pass
> a
> simple generate hartid mask directly to sbi_send_ipi.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> ---
>  arch/riscv/kernel/smp.c | 8 +++++++-
>  1 file changed, 7 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/riscv/kernel/smp.c b/arch/riscv/kernel/smp.c
> index 2e21669aa068..a3715d621f60 100644
> --- a/arch/riscv/kernel/smp.c
> +++ b/arch/riscv/kernel/smp.c
> @@ -94,7 +94,13 @@ static void send_ipi_mask(const struct cpumask
> *mask, enum ipi_message_type op)
>  
>  static void send_ipi_single(int cpu, enum ipi_message_type op)
>  {
> -	send_ipi_mask(cpumask_of(cpu), op);
> +	int hartid = cpuid_to_hartid_map(cpu);
> +
> +	smp_mb__before_atomic();
> +	set_bit(op, &ipi_data[cpu].bits);
> +	smp_mb__after_atomic();
> +
> +	sbi_send_ipi(cpumask_bits(cpumask_of(hartid)));

The only cost save I see is you don't have to cpumask_clear anymore.
Is there any other cost save ? If not is it worth duplicating the code
?

May be I am being too pedantic here.. :) :)

>  }
>  
>  static inline void clear_ipi(void)

-- 
Regards,
Atish
_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv

Re: [PATCH 5/6] riscv: don't use the rdtime(h) pseudo-instructions

[Atish Patra]

On Wed, 2019-08-21 at 23:58 +0900, Christoph Hellwig wrote:
> If we just use the CSRs that these map to directly the code is
> simpler
> and doesn't require extra inline assembly code.  Also fix up the top-
> level
> comment in timer-riscv.c to not talk about the cycle count or mention
> details of the clocksource interface, of which this file is just a
> consumer.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> ---
>  arch/riscv/include/asm/timex.h    | 44 +++++++++++++++------------
> ----
>  drivers/clocksource/timer-riscv.c | 17 +++---------
>  2 files changed, 25 insertions(+), 36 deletions(-)
> 
> diff --git a/arch/riscv/include/asm/timex.h
> b/arch/riscv/include/asm/timex.h
> index 6a703ec9d796..c7ef131b9e4c 100644
> --- a/arch/riscv/include/asm/timex.h
> +++ b/arch/riscv/include/asm/timex.h
> @@ -6,43 +6,41 @@
>  #ifndef _ASM_RISCV_TIMEX_H
>  #define _ASM_RISCV_TIMEX_H
>  
> -#include <asm/param.h>
> +#include <asm/csr.h>
>  
>  typedef unsigned long cycles_t;
>  
> -static inline cycles_t get_cycles_inline(void)
> +static inline cycles_t get_cycles(void)
>  {
> -	cycles_t n;
> -
> -	__asm__ __volatile__ (
> -		"rdtime %0"
> -		: "=r" (n));
> -	return n;
> +	return csr_read(CSR_TIME);

Does this work correctly in QEMU ? I was looking at the qemu code and
it looks like it returns cpu_get_host_ticks which seems wrong to me.

https://github.com/qemu/qemu/blob/master/target/riscv/csr.c#L213


>  }
> -#define get_cycles get_cycles_inline
> +#define get_cycles get_cycles
>  
>  #ifdef CONFIG_64BIT
> -static inline uint64_t get_cycles64(void)
> +static inline u64 get_cycles64(void)
> +{
> +	return get_cycles();
> +}
> +#else /* CONFIG_64BIT */
> +static inline u32 get_cycles_hi(void)
>  {
> -        return get_cycles();
> +	return csr_read(CSR_TIMEH);
>  }
> -#else
> -static inline uint64_t get_cycles64(void)
> +
> +static inline u64 get_cycles64(void)
>  {
> -	u32 lo, hi, tmp;
> -	__asm__ __volatile__ (
> -		"1:\n"
> -		"rdtimeh %0\n"
> -		"rdtime %1\n"
> -		"rdtimeh %2\n"
> -		"bne %0, %2, 1b"
> -		: "=&r" (hi), "=&r" (lo), "=&r" (tmp));
> +	u32 hi, lo;
> +
> +	do {
> +		hi = get_cycles_hi();
> +		lo = get_cycles();
> +	} while (hi != get_cycles_hi());
> +
>  	return ((u64)hi << 32) | lo;
>  }
> -#endif
> +#endif /* CONFIG_64BIT */
>  
>  #define ARCH_HAS_READ_CURRENT_TIMER
> -
>  static inline int read_current_timer(unsigned long *timer_val)
>  {
>  	*timer_val = get_cycles();
> diff --git a/drivers/clocksource/timer-riscv.c
> b/drivers/clocksource/timer-riscv.c
> index 09e031176bc6..470c7ef02ea4 100644
> --- a/drivers/clocksource/timer-riscv.c
> +++ b/drivers/clocksource/timer-riscv.c
> @@ -2,6 +2,10 @@
>  /*
>   * Copyright (C) 2012 Regents of the University of California
>   * Copyright (C) 2017 SiFive
> + *
> + * All RISC-V systems have a timer attached to every hart.  These
> timers can be
> + * read from the "time" and "timeh" CSRs, and can use the SBI to
> setup
> + * events.
>   */
>  #include <linux/clocksource.h>
>  #include <linux/clockchips.h>
> @@ -12,19 +16,6 @@
>  #include <asm/smp.h>
>  #include <asm/sbi.h>
>  
> -/*
> - * All RISC-V systems have a timer attached to every hart.  These
> timers can be
> - * read by the 'rdcycle' pseudo instruction, and can use the SBI to
> setup
> - * events.  In order to abstract the architecture-specific timer
> reading and
> - * setting functions away from the clock event insertion code, we
> provide
> - * function pointers to the clockevent subsystem that perform two
> basic
> - * operations: rdtime() reads the timer on the current CPU, and
> - * next_event(delta) sets the next timer event to 'delta' cycles in
> the future.
> - * As the timers are inherently a per-cpu resource, these callbacks
> perform
> - * operations on the current hart.  There is guaranteed to be
> exactly one timer
> - * per hart on all RISC-V systems.
> - */
> -
>  static int riscv_clock_next_event(unsigned long delta,
>  		struct clock_event_device *ce)
>  {

-- 
Regards,
Atish
_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv

Re: [PATCH 5/6] riscv: don't use the rdtime(h) pseudo-instructions

[Atish Patra]

On Fri, 2019-08-23 at 17:37 -0700, Atish Patra wrote:
> On Wed, 2019-08-21 at 23:58 +0900, Christoph Hellwig wrote:
> > If we just use the CSRs that these map to directly the code is
> > simpler
> > and doesn't require extra inline assembly code.  Also fix up the
> > top-
> > level
> > comment in timer-riscv.c to not talk about the cycle count or
> > mention
> > details of the clocksource interface, of which this file is just a
> > consumer.
> > 
> > Signed-off-by: Christoph Hellwig <hch@lst.de>
> > ---
> >  arch/riscv/include/asm/timex.h    | 44 +++++++++++++++------------
> > ----
> >  drivers/clocksource/timer-riscv.c | 17 +++---------
> >  2 files changed, 25 insertions(+), 36 deletions(-)
> > 
> > diff --git a/arch/riscv/include/asm/timex.h
> > b/arch/riscv/include/asm/timex.h
> > index 6a703ec9d796..c7ef131b9e4c 100644
> > --- a/arch/riscv/include/asm/timex.h
> > +++ b/arch/riscv/include/asm/timex.h
> > @@ -6,43 +6,41 @@
> >  #ifndef _ASM_RISCV_TIMEX_H
> >  #define _ASM_RISCV_TIMEX_H
> >  
> > -#include <asm/param.h>
> > +#include <asm/csr.h>
> >  
> >  typedef unsigned long cycles_t;
> >  
> > -static inline cycles_t get_cycles_inline(void)
> > +static inline cycles_t get_cycles(void)
> >  {
> > -	cycles_t n;
> > -
> > -	__asm__ __volatile__ (
> > -		"rdtime %0"
> > -		: "=r" (n));
> > -	return n;
> > +	return csr_read(CSR_TIME);
> 
> Does this work correctly in QEMU ? I was looking at the qemu code and
> it looks like it returns cpu_get_host_ticks which seems wrong to me.
> 
> https://github.com/qemu/qemu/blob/master/target/riscv/csr.c#L213
> 
> 

Nevermind. I missed the CONFIG_USER_ONLY and got confused.
csr_read will also trap and get the correct value.

Regards,
Atish
> >  }
> > -#define get_cycles get_cycles_inline
> > +#define get_cycles get_cycles
> >  
> >  #ifdef CONFIG_64BIT
> > -static inline uint64_t get_cycles64(void)
> > +static inline u64 get_cycles64(void)
> > +{
> > +	return get_cycles();
> > +}
> > +#else /* CONFIG_64BIT */
> > +static inline u32 get_cycles_hi(void)
> >  {
> > -        return get_cycles();
> > +	return csr_read(CSR_TIMEH);
> >  }
> > -#else
> > -static inline uint64_t get_cycles64(void)
> > +
> > +static inline u64 get_cycles64(void)
> >  {
> > -	u32 lo, hi, tmp;
> > -	__asm__ __volatile__ (
> > -		"1:\n"
> > -		"rdtimeh %0\n"
> > -		"rdtime %1\n"
> > -		"rdtimeh %2\n"
> > -		"bne %0, %2, 1b"
> > -		: "=&r" (hi), "=&r" (lo), "=&r" (tmp));
> > +	u32 hi, lo;
> > +
> > +	do {
> > +		hi = get_cycles_hi();
> > +		lo = get_cycles();
> > +	} while (hi != get_cycles_hi());
> > +
> >  	return ((u64)hi << 32) | lo;
> >  }
> > -#endif
> > +#endif /* CONFIG_64BIT */
> >  
> >  #define ARCH_HAS_READ_CURRENT_TIMER
> > -
> >  static inline int read_current_timer(unsigned long *timer_val)
> >  {
> >  	*timer_val = get_cycles();
> > diff --git a/drivers/clocksource/timer-riscv.c
> > b/drivers/clocksource/timer-riscv.c
> > index 09e031176bc6..470c7ef02ea4 100644
> > --- a/drivers/clocksource/timer-riscv.c
> > +++ b/drivers/clocksource/timer-riscv.c
> > @@ -2,6 +2,10 @@
> >  /*
> >   * Copyright (C) 2012 Regents of the University of California
> >   * Copyright (C) 2017 SiFive
> > + *
> > + * All RISC-V systems have a timer attached to every hart.  These
> > timers can be
> > + * read from the "time" and "timeh" CSRs, and can use the SBI to
> > setup
> > + * events.
> >   */
> >  #include <linux/clocksource.h>
> >  #include <linux/clockchips.h>
> > @@ -12,19 +16,6 @@
> >  #include <asm/smp.h>
> >  #include <asm/sbi.h>
> >  
> > -/*
> > - * All RISC-V systems have a timer attached to every hart.  These
> > timers can be
> > - * read by the 'rdcycle' pseudo instruction, and can use the SBI
> > to
> > setup
> > - * events.  In order to abstract the architecture-specific timer
> > reading and
> > - * setting functions away from the clock event insertion code, we
> > provide
> > - * function pointers to the clockevent subsystem that perform two
> > basic
> > - * operations: rdtime() reads the timer on the current CPU, and
> > - * next_event(delta) sets the next timer event to 'delta' cycles
> > in
> > the future.
> > - * As the timers are inherently a per-cpu resource, these
> > callbacks
> > perform
> > - * operations on the current hart.  There is guaranteed to be
> > exactly one timer
> > - * per hart on all RISC-V systems.
> > - */
> > -
> >  static int riscv_clock_next_event(unsigned long delta,
> >  		struct clock_event_device *ce)
> >  {

-- 
Regards,
Atish
_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv

Re: [PATCH 1/6] riscv: refactor the IPI code

[Atish Patra]

On Wed, 2019-08-21 at 23:58 +0900, Christoph Hellwig wrote:
> This prepares for adding native non-SBI IPI code.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> ---
>  arch/riscv/kernel/smp.c | 55 +++++++++++++++++++++++--------------
> ----
>  1 file changed, 31 insertions(+), 24 deletions(-)
> 
> diff --git a/arch/riscv/kernel/smp.c b/arch/riscv/kernel/smp.c
> index 5a9834503a2f..8cd730239613 100644
> --- a/arch/riscv/kernel/smp.c
> +++ b/arch/riscv/kernel/smp.c
> @@ -78,13 +78,38 @@ static void ipi_stop(void)
>  		wait_for_interrupt();
>  }
>  
> +static void send_ipi_mask(const struct cpumask *mask, enum
> ipi_message_type op)
> +{
> +	int cpuid, hartid;
> +	struct cpumask hartid_mask;
> +
> +	cpumask_clear(&hartid_mask);
> +	mb();
> +	for_each_cpu(cpuid, mask) {
> +		set_bit(op, &ipi_data[cpuid].bits);
> +		hartid = cpuid_to_hartid_map(cpuid);
> +		cpumask_set_cpu(hartid, &hartid_mask);
> +	}
> +	mb();
> +	sbi_send_ipi(cpumask_bits(&hartid_mask));
> +}
> +
> +static void send_ipi_single(int cpu, enum ipi_message_type op)
> +{
> +	send_ipi_mask(cpumask_of(cpu), op);
> +}
> +
> +static inline void clear_ipi(void)
> +{
> +	csr_clear(CSR_SIP, SIE_SSIE);
> +}
> +
>  void riscv_software_interrupt(void)
>  {
>  	unsigned long *pending_ipis =
> &ipi_data[smp_processor_id()].bits;
>  	unsigned long *stats = ipi_data[smp_processor_id()].stats;
>  
> -	/* Clear pending IPI */
> -	csr_clear(CSR_SIP, SIE_SSIE);
> +	clear_ipi();
>  
>  	while (true) {
>  		unsigned long ops;
> @@ -118,23 +143,6 @@ void riscv_software_interrupt(void)
>  	}
>  }
>  
> -static void
> -send_ipi_message(const struct cpumask *to_whom, enum
> ipi_message_type operation)
> -{
> -	int cpuid, hartid;
> -	struct cpumask hartid_mask;
> -
> -	cpumask_clear(&hartid_mask);
> -	mb();
> -	for_each_cpu(cpuid, to_whom) {
> -		set_bit(operation, &ipi_data[cpuid].bits);
> -		hartid = cpuid_to_hartid_map(cpuid);
> -		cpumask_set_cpu(hartid, &hartid_mask);
> -	}
> -	mb();
> -	sbi_send_ipi(cpumask_bits(&hartid_mask));
> -}
> -
>  static const char * const ipi_names[] = {
>  	[IPI_RESCHEDULE]	= "Rescheduling interrupts",
>  	[IPI_CALL_FUNC]		= "Function call interrupts",
> @@ -156,12 +164,12 @@ void show_ipi_stats(struct seq_file *p, int
> prec)
>  
>  void arch_send_call_function_ipi_mask(struct cpumask *mask)
>  {
> -	send_ipi_message(mask, IPI_CALL_FUNC);
> +	send_ipi_mask(mask, IPI_CALL_FUNC);
>  }
>  
>  void arch_send_call_function_single_ipi(int cpu)
>  {
> -	send_ipi_message(cpumask_of(cpu), IPI_CALL_FUNC);
> +	send_ipi_single(cpu, IPI_CALL_FUNC);
>  }
>  
>  void smp_send_stop(void)
> @@ -176,7 +184,7 @@ void smp_send_stop(void)
>  
>  		if (system_state <= SYSTEM_RUNNING)
>  			pr_crit("SMP: stopping secondary CPUs\n");
> -		send_ipi_message(&mask, IPI_CPU_STOP);
> +		send_ipi_mask(&mask, IPI_CPU_STOP);
>  	}
>  
>  	/* Wait up to one second for other CPUs to stop */
> @@ -191,6 +199,5 @@ void smp_send_stop(void)
>  
>  void smp_send_reschedule(int cpu)
>  {
> -	send_ipi_message(cpumask_of(cpu), IPI_RESCHEDULE);
> +	send_ipi_single(cpu, IPI_RESCHEDULE);
>  }
> -

Reviewed-by: Atish Patra <atish.patra@wdc.com>

-- 
Regards,
Atish
_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv

Re: [PATCH 2/6] riscv: cleanup send_ipi_mask

[hch]

On Sat, Aug 24, 2019 at 12:11:15AM +0000, Atish Patra wrote:
> Isn't that less optimized than previous one ?
> 
> This will iterate all the cpus set in mask twice during every ipi sent.
> For now, we won't see any different. As we have more number of cpus in
> RISC-V (hopefully one day ;) ;)), this may affect the performance.

By then we are hopefully done with using the SBI IPI code :) The native
IPI code this refactor is preparing for won't need the hartid
translation for example.  The point of this patch isn't really to
micro-optimize, but to make the code clear and obvious.

_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv

Re: [PATCH 3/6] riscv: optimize send_ipi_single

[hch]

On Sat, Aug 24, 2019 at 12:26:26AM +0000, Atish Patra wrote:
> >  static void send_ipi_single(int cpu, enum ipi_message_type op)
> >  {
> > -	send_ipi_mask(cpumask_of(cpu), op);
> > +	int hartid = cpuid_to_hartid_map(cpu);
> > +
> > +	smp_mb__before_atomic();
> > +	set_bit(op, &ipi_data[cpu].bits);
> > +	smp_mb__after_atomic();
> > +
> > +	sbi_send_ipi(cpumask_bits(cpumask_of(hartid)));
> 
> The only cost save I see is you don't have to cpumask_clear anymore.
> Is there any other cost save ? If not is it worth duplicating the code
> ?
> 
> May be I am being too pedantic here.. :) :)

It avoids the additional potentially huge cpumask, and generally makes
the code a lot more obvious.  This might not really be needed, but
helps with sharing the code nicely with the native IPI path.

_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv

Re: [PATCH 5/6] riscv: don't use the rdtime(h) pseudo-instructions

[hch]

On Sat, Aug 24, 2019 at 12:37:02AM +0000, Atish Patra wrote:
> > -static inline cycles_t get_cycles_inline(void)
> > +static inline cycles_t get_cycles(void)
> >  {
> > -	cycles_t n;
> > -
> > -	__asm__ __volatile__ (
> > -		"rdtime %0"
> > -		: "=r" (n));
> > -	return n;
> > +	return csr_read(CSR_TIME);
> 
> Does this work correctly in QEMU ? I was looking at the qemu code and
> it looks like it returns cpu_get_host_ticks which seems wrong to me.

It better should.  rdtime is just a pseudo-instruction that the
assembler translates to a CSR read.  (in other words something totally
pointless, no idea why it even is in the spec).

_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv

Re: [PATCH 2/6] riscv: cleanup send_ipi_mask

[Atish Patra]

On Mon, 2019-08-26 at 13:28 +0200, hch@lst.de wrote:
> On Sat, Aug 24, 2019 at 12:11:15AM +0000, Atish Patra wrote:
> > Isn't that less optimized than previous one ?
> > 
> > This will iterate all the cpus set in mask twice during every ipi
> > sent.
> > For now, we won't see any different. As we have more number of cpus
> > in
> > RISC-V (hopefully one day ;) ;)), this may affect the performance.
> 
> By then we are hopefully done with using the SBI IPI code :) The
> native
> IPI code this refactor is preparing for won't need the hartid
> translation for example.  The point of this patch isn't really to
> micro-optimize, but to make the code clear and obvious.

ok. Sounds good to me.

Reviewed-by: Atish Patra <atish.patra@wdc.com>

-- 
Regards,
Atish
_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv

Re: [PATCH 3/6] riscv: optimize send_ipi_single

[Atish Patra]

On Mon, 2019-08-26 at 13:29 +0200, hch@lst.de wrote:
> On Sat, Aug 24, 2019 at 12:26:26AM +0000, Atish Patra wrote:
> > >  static void send_ipi_single(int cpu, enum ipi_message_type op)
> > >  {
> > > -	send_ipi_mask(cpumask_of(cpu), op);
> > > +	int hartid = cpuid_to_hartid_map(cpu);
> > > +
> > > +	smp_mb__before_atomic();
> > > +	set_bit(op, &ipi_data[cpu].bits);
> > > +	smp_mb__after_atomic();
> > > +
> > > +	sbi_send_ipi(cpumask_bits(cpumask_of(hartid)));
> > 
> > The only cost save I see is you don't have to cpumask_clear
> > anymore.
> > Is there any other cost save ? If not is it worth duplicating the
> > code
> > ?
> > 
> > May be I am being too pedantic here.. :) :)
> 
> It avoids the additional potentially huge cpumask, and generally
> makes
> the code a lot more obvious.  This might not really be needed, but
> helps with sharing the code nicely with the native IPI path.

ok.

Reviewed-by: Atish Patra <atish.patra@wdc.com>

-- 
Regards,
Atish
_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv

Re: [PATCH 5/6] riscv: don't use the rdtime(h) pseudo-instructions

[Atish Patra]

On Mon, 2019-08-26 at 13:30 +0200, hch@lst.de wrote:
> On Sat, Aug 24, 2019 at 12:37:02AM +0000, Atish Patra wrote:
> > > -static inline cycles_t get_cycles_inline(void)
> > > +static inline cycles_t get_cycles(void)
> > >  {
> > > -	cycles_t n;
> > > -
> > > -	__asm__ __volatile__ (
> > > -		"rdtime %0"
> > > -		: "=r" (n));
> > > -	return n;
> > > +	return csr_read(CSR_TIME);
> > 
> > Does this work correctly in QEMU ? I was looking at the qemu code
> > and
> > it looks like it returns cpu_get_host_ticks which seems wrong to
> > me.
> 
> It better should.  rdtime is just a pseudo-instruction that the
> assembler translates to a CSR read.  (in other words something
> totally
> pointless, no idea why it even is in the spec).

Yes. I did not see the usermode macro carefully in qemu code.

Reviewed-by: Atish Patra <atish.patra@wdc.com>

-- 
Regards,
Atish
_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv

Re: [PATCH 1/6] riscv: refactor the IPI code

[Paul Walmsley]

On Wed, 21 Aug 2019, Christoph Hellwig wrote:

> This prepares for adding native non-SBI IPI code.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>

Thanks - dropped the previous version from the v5.4-rc1 queue, and queued 
this one in its place, with Atish's Reviewed-by:.


- Paul

_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv

Re: [PATCH 2/6] riscv: cleanup send_ipi_mask

[Paul Walmsley]

On Wed, 21 Aug 2019, Christoph Hellwig wrote:

> Use the special barriers for atomic bitops to make the intention
> a little more clear, and use riscv_cpuid_to_hartid_mask instead of
> open coding it.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>

Thanks, queued for v5.4-rc1 with Atish's Reviewed-by:.


- Paul

_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv

Re: [PATCH 3/6] riscv: optimize send_ipi_single

[Paul Walmsley]

On Wed, 21 Aug 2019, Christoph Hellwig wrote:

> Don't go through send_ipi_mask, but just set the op bit an then pass a
> simple generate hartid mask directly to sbi_send_ipi.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>

Thanks, I fixed some minor issues in the patch description and queued 
the following for v5.4-rc1.


- Paul

From: Christoph Hellwig <hch@lst.de>
Date: Wed, 21 Aug 2019 23:58:34 +0900
Subject: [PATCH] riscv: optimize send_ipi_single

Don't go through send_ipi_mask, but just set the op bit and then pass
a simple generated hartid mask directly to sbi_send_ipi.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Atish Patra <atish.patra@wdc.com>
[paul.walmsley@sifive.com: minor patch description fixes]
Signed-off-by: Paul Walmsley <paul.walmsley@sifive.com>
---
 arch/riscv/kernel/smp.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/arch/riscv/kernel/smp.c b/arch/riscv/kernel/smp.c
index 2e21669aa068..a3715d621f60 100644
--- a/arch/riscv/kernel/smp.c
+++ b/arch/riscv/kernel/smp.c
@@ -94,7 +94,13 @@ static void send_ipi_mask(const struct cpumask *mask, enum ipi_message_type op)
 
 static void send_ipi_single(int cpu, enum ipi_message_type op)
 {
-	send_ipi_mask(cpumask_of(cpu), op);
+	int hartid = cpuid_to_hartid_map(cpu);
+
+	smp_mb__before_atomic();
+	set_bit(op, &ipi_data[cpu].bits);
+	smp_mb__after_atomic();
+
+	sbi_send_ipi(cpumask_bits(cpumask_of(hartid)));
 }
 
 static inline void clear_ipi(void)
-- 
2.23.0


_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv

Re: [PATCH 4/6] riscv: cleanup riscv_cpuid_to_hartid_mask

[Paul Walmsley]

On Wed, 21 Aug 2019, Christoph Hellwig wrote:

> Move the initial clearing of the mask from the callers to
> riscv_cpuid_to_hartid_mask, and remove the unused !CONFIG_SMP stub.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>

Thanks, queued for v5.4-rc1 with Atish's Reviewed-by:.


- Paul

_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv

Re: [PATCH 5/6] riscv: don't use the rdtime(h) pseudo-instructions

[Paul Walmsley]

On Wed, 21 Aug 2019, Christoph Hellwig wrote:

> If we just use the CSRs that these map to directly the code is simpler
> and doesn't require extra inline assembly code.  Also fix up the top-level
> comment in timer-riscv.c to not talk about the cycle count or mention
> details of the clocksource interface, of which this file is just a
> consumer.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>

Thanks, queued for v5.4-rc1 with Atish's Reviewed-by:.


- Paul

_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv

Re: [PATCH 6/6] riscv: move the TLB flush logic out of line

[Paul Walmsley]

On Wed, 21 Aug 2019, Christoph Hellwig wrote:

> The TLB flush logic is going to become more complex.  Start moving
> it out of line.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>

Thanks.  "checkpatch.pl --strict" reported several whitespace issues with 
this patch.  I fixed those up here and queued the following for v5.4-rc1.


- Paul

From: Christoph Hellwig <hch@lst.de>
Date: Wed, 21 Aug 2019 23:58:37 +0900
Subject: [PATCH] riscv: move the TLB flush logic out of line

The TLB flush logic is going to become more complex.  Start moving
it out of line.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Atish Patra <atish.patra@wdc.com>
[paul.walmsley@sifive.com: fixed checkpatch whitespace warnings]
Signed-off-by: Paul Walmsley <paul.walmsley@sifive.com>
---
 arch/riscv/include/asm/tlbflush.h | 37 ++++++-------------------------
 arch/riscv/mm/Makefile            |  3 +++
 arch/riscv/mm/tlbflush.c          | 35 +++++++++++++++++++++++++++++
 3 files changed, 45 insertions(+), 30 deletions(-)
 create mode 100644 arch/riscv/mm/tlbflush.c

diff --git a/arch/riscv/include/asm/tlbflush.h b/arch/riscv/include/asm/tlbflush.h
index df31fe2ed09c..37ae4e367ad2 100644
--- a/arch/riscv/include/asm/tlbflush.h
+++ b/arch/riscv/include/asm/tlbflush.h
@@ -25,8 +25,13 @@ static inline void local_flush_tlb_page(unsigned long addr)
 	__asm__ __volatile__ ("sfence.vma %0" : : "r" (addr) : "memory");
 }
 
-#ifndef CONFIG_SMP
-
+#ifdef CONFIG_SMP
+void flush_tlb_all(void);
+void flush_tlb_mm(struct mm_struct *mm);
+void flush_tlb_page(struct vm_area_struct *vma, unsigned long addr);
+void flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
+		     unsigned long end);
+#else /* CONFIG_SMP */
 #define flush_tlb_all() local_flush_tlb_all()
 #define flush_tlb_page(vma, addr) local_flush_tlb_page(addr)
 
@@ -37,34 +42,6 @@ static inline void flush_tlb_range(struct vm_area_struct *vma,
 }
 
 #define flush_tlb_mm(mm) flush_tlb_all()
-
-#else /* CONFIG_SMP */
-
-#include <asm/sbi.h>
-
-static inline void remote_sfence_vma(struct cpumask *cmask, unsigned long start,
-				     unsigned long size)
-{
-	struct cpumask hmask;
-
-	riscv_cpuid_to_hartid_mask(cmask, &hmask);
-	sbi_remote_sfence_vma(hmask.bits, start, size);
-}
-
-#define flush_tlb_all() sbi_remote_sfence_vma(NULL, 0, -1)
-
-#define flush_tlb_range(vma, start, end) \
-	remote_sfence_vma(mm_cpumask((vma)->vm_mm), start, (end) - (start))
-
-static inline void flush_tlb_page(struct vm_area_struct *vma,
-				  unsigned long addr)
-{
-	flush_tlb_range(vma, addr, addr + PAGE_SIZE);
-}
-
-#define flush_tlb_mm(mm)				\
-	remote_sfence_vma(mm_cpumask(mm), 0, -1)
-
 #endif /* CONFIG_SMP */
 
 /* Flush a range of kernel pages */
diff --git a/arch/riscv/mm/Makefile b/arch/riscv/mm/Makefile
index 74055e1d6f21..9d9a17335686 100644
--- a/arch/riscv/mm/Makefile
+++ b/arch/riscv/mm/Makefile
@@ -13,4 +13,7 @@ obj-y += cacheflush.o
 obj-y += context.o
 obj-y += sifive_l2_cache.o
 
+ifeq ($(CONFIG_MMU),y)
+obj-$(CONFIG_SMP) += tlbflush.o
+endif
 obj-$(CONFIG_HUGETLB_PAGE) += hugetlbpage.o
diff --git a/arch/riscv/mm/tlbflush.c b/arch/riscv/mm/tlbflush.c
new file mode 100644
index 000000000000..24cd33d2c48f
--- /dev/null
+++ b/arch/riscv/mm/tlbflush.c
@@ -0,0 +1,35 @@
+// SPDX-License-Identifier: GPL-2.0
+
+#include <linux/mm.h>
+#include <linux/smp.h>
+#include <asm/sbi.h>
+
+void flush_tlb_all(void)
+{
+	sbi_remote_sfence_vma(NULL, 0, -1);
+}
+
+static void __sbi_tlb_flush_range(struct cpumask *cmask, unsigned long start,
+				  unsigned long size)
+{
+	struct cpumask hmask;
+
+	riscv_cpuid_to_hartid_mask(cmask, &hmask);
+	sbi_remote_sfence_vma(hmask.bits, start, size);
+}
+
+void flush_tlb_mm(struct mm_struct *mm)
+{
+	__sbi_tlb_flush_range(mm_cpumask(mm), 0, -1);
+}
+
+void flush_tlb_page(struct vm_area_struct *vma, unsigned long addr)
+{
+	__sbi_tlb_flush_range(mm_cpumask(vma->vm_mm), addr, PAGE_SIZE);
+}
+
+void flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
+		     unsigned long end)
+{
+	__sbi_tlb_flush_range(mm_cpumask(vma->vm_mm), start, end - start);
+}
-- 
2.23.0


_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv