[PATCH V3 0/5] MIPS: Loongson: Add Loongson-3A R2 support

[PATCH V3 0/5] MIPS: Loongson: Add Loongson-3A R2 support

[Huacai Chen]

This patchset is is prepared for the next 4.6 release for Linux/MIPS.
It adds Loongson-3A R2 (Loongson-3A2000) support and fixes a potential
bug related to FTLB.

Loongson-3 CPU family:

Code-name       Brand-name       PRId
Loongson-3A R1  Loongson-3A1000  0x6305
Loongson-3A R2  Loongson-3A2000  0x6308
Loongson-3B R1  Loongson-3B1000  0x6306
Loongson-3B R2  Loongson-3B1500  0x6307

Features of R2 revision of Loongson-3A:
1, Primary cache includes I-Cache, D-Cache and V-Cache (Victim Cache).
2, I-Cache, D-Cache and V-Cache are 16-way set-associative, linesize is 64 Bytes.
3, 64 entries of VTLB (classic TLB), 1024 entries of FTLB (8-way set-associative).
4, Support DSP/DSPv2 instructions, UserLocal register and Read-Inhibit/Execute-Inhibit.

V1 -> V2:
1, Probe MIPS_CPU_PREFETCH by PRId.
2, Use PRID_REV_MASK instead of hardcode.
3, Update commit messages to avoid confusion.

V2 -> V3:
1, Remove the 4th patch since it is a bugfix not only for Loongson.
2, Split the 5th patch and remove the generic part since that is not only for Loongson.

Huacai Chen(5):
 MIPS: Loongson: Add Loongson-3A R2 basic support.
 MIPS: Loongson: Invalidate special TLBs when needed.
 MIPS: Loongson-3: Fast TLB refill handler.
 MIPS: Loongson-3: Use cpu_has_coherent_cache feature
 MIPS: Loongson-3: Introduce CONFIG_LOONGSON3_ENHANCEMENT.

Signed-off-by: Huacai Chen <chenhc@lemote.com>
---
 arch/mips/Kconfig                                  |  19 ++++
 arch/mips/include/asm/cacheops.h                   |   6 +
 arch/mips/include/asm/cpu-features.h               |   3 +
 arch/mips/include/asm/cpu-info.h                   |   1 +
 arch/mips/include/asm/cpu.h                        |   5 +-
 arch/mips/include/asm/hazards.h                    |   7 +-
 arch/mips/include/asm/io.h                         |  10 +-
 arch/mips/include/asm/irqflags.h                   |   5 +
 .../asm/mach-loongson64/cpu-feature-overrides.h    |  13 +--
 .../asm/mach-loongson64/kernel-entry-init.h        |  18 ++-
 arch/mips/include/asm/mipsregs.h                   |   8 ++
 arch/mips/include/asm/pgtable-bits.h               |   8 +-
 arch/mips/include/asm/pgtable.h                    |   4 +-
 arch/mips/include/asm/uasm.h                       |   3 +-
 arch/mips/include/uapi/asm/inst.h                  |  10 ++
 arch/mips/kernel/cpu-probe.c                       |  40 ++++++-
 arch/mips/kernel/idle.c                            |   5 +
 arch/mips/kernel/traps.c                           |   3 +-
 arch/mips/loongson64/common/env.c                  |   7 +-
 arch/mips/loongson64/loongson-3/smp.c              | 106 +++++++++++++++--
 arch/mips/mm/c-r4k.c                               |  29 +++++
 arch/mips/mm/page.c                                |   9 ++
 arch/mips/mm/tlb-r4k.c                             |  27 +++--
 arch/mips/mm/tlbex.c                               | 126 ++++++++++++++++++++-
 arch/mips/mm/uasm-mips.c                           |   2 +
 arch/mips/mm/uasm.c                                |   3 +
 drivers/platform/mips/cpu_hwmon.c                  |   4 +-
 27 files changed, 422 insertions(+), 59 deletions(-)
--
2.7.0

[PATCH] MIPS: tlbex: Fix bugs in tlbchange handler

[Huacai Chen]

If a tlb miss triggered when EXL=1, tlb refill exception is treated as
tlb invalid exception, so tlbp may fails. In this situation, CP0_Index
register doesn't contain a valid value. This may not be a problem for
VTLB since it is fully-associative. However, FTLB is set-associative so
not every tlb entry is valid for a specific address. Thus, we should
use tlbwr instead of tlbwi when tlbp fails.

There is a similar case for huge page, so build_huge_tlb_write_entry()
is also modified. If wmode != tlb_random, that means the caller is tlb
invalid handler, we should select tlbr/tlbi depend on the tlbp result.

Cc: stable@kernel.org
Signed-off-by: Huacai Chen <chenhc@lemote.com>
---
 arch/mips/mm/tlbex.c | 31 ++++++++++++++++++++++++++++++-
 1 file changed, 30 insertions(+), 1 deletion(-)

diff --git a/arch/mips/mm/tlbex.c b/arch/mips/mm/tlbex.c
index 5a04b6f..1ab9557 100644
--- a/arch/mips/mm/tlbex.c
+++ b/arch/mips/mm/tlbex.c
@@ -173,7 +173,10 @@ enum label_id {
 	label_large_segbits_fault,
 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
 	label_tlb_huge_update,
+	label_tail_huge_miss,
+	label_tail_huge_done,
 #endif
+	label_tail_miss,
 };
 
 UASM_L_LA(_second_part)
@@ -192,7 +195,10 @@ UASM_L_LA(_r3000_write_probe_fail)
 UASM_L_LA(_large_segbits_fault)
 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
 UASM_L_LA(_tlb_huge_update)
+UASM_L_LA(_tail_huge_miss)
+UASM_L_LA(_tail_huge_done)
 #endif
+UASM_L_LA(_tail_miss)
 
 static int hazard_instance;
 
@@ -701,8 +707,24 @@ static void build_huge_tlb_write_entry(u32 **p, struct uasm_label **l,
 	uasm_i_ori(p, tmp, tmp, PM_HUGE_MASK & 0xffff);
 	uasm_i_mtc0(p, tmp, C0_PAGEMASK);
 
-	build_tlb_write_entry(p, l, r, wmode);
+	if (wmode == tlb_random) { /* Caller is TLB Refill Handler */
+		build_tlb_write_entry(p, l, r, wmode);
+		build_restore_pagemask(p, r, tmp, label_leave, restore_scratch);
+		return;
+	}
+
+	/* Caller is TLB Load/Store/Modify Handler */
+	uasm_i_mfc0(p, tmp, C0_INDEX);
+	uasm_il_bltz(p, r, tmp, label_tail_huge_miss);
+	uasm_i_nop(p);
+	build_tlb_write_entry(p, l, r, tlb_indexed);
+	uasm_il_b(p, r, label_tail_huge_done);
+	uasm_i_nop(p);
+
+	uasm_l_tail_huge_miss(l, *p);
+	build_tlb_write_entry(p, l, r, tlb_random);
 
+	uasm_l_tail_huge_done(l, *p);
 	build_restore_pagemask(p, r, tmp, label_leave, restore_scratch);
 }
 
@@ -1913,7 +1935,14 @@ build_r4000_tlbchange_handler_tail(u32 **p, struct uasm_label **l,
 	uasm_i_ori(p, ptr, ptr, sizeof(pte_t));
 	uasm_i_xori(p, ptr, ptr, sizeof(pte_t));
 	build_update_entries(p, tmp, ptr);
+	uasm_i_mfc0(p, ptr, C0_INDEX);
+	uasm_il_bltz(p, r, ptr, label_tail_miss);
+	uasm_i_nop(p);
 	build_tlb_write_entry(p, l, r, tlb_indexed);
+	uasm_il_b(p, r, label_leave);
+	uasm_i_nop(p);
+	uasm_l_tail_miss(l, *p);
+	build_tlb_write_entry(p, l, r, tlb_random);
 	uasm_l_leave(l, *p);
 	build_restore_work_registers(p);
 	uasm_i_eret(p); /* return from trap */
-- 
2.7.0

Re: [PATCH] MIPS: tlbex: Fix bugs in tlbchange handler

[Maciej W. Rozycki]

On Wed, 24 Feb 2016, Huacai Chen wrote:

> If a tlb miss triggered when EXL=1, tlb refill exception is treated as
> tlb invalid exception, so tlbp may fails. In this situation, CP0_Index
> register doesn't contain a valid value. This may not be a problem for
> VTLB since it is fully-associative. However, FTLB is set-associative so
> not every tlb entry is valid for a specific address. Thus, we should
> use tlbwr instead of tlbwi when tlbp fails.

 Can you please explain exactly why this change is needed?  You're 
changing pretty generic code which has worked since forever.  So why is a 
change suddenly needed?  Our kernel entry/exit code has been written such 
that no mapped memory is accessed with EXL=1 so no TLB exception is 
expected to ever happen in these circumstances.  So what's your scenario 
you mean to address?  Your patch description does not explain it.

> @@ -1913,7 +1935,14 @@ build_r4000_tlbchange_handler_tail(u32 **p, struct uasm_label **l,
>  	uasm_i_ori(p, ptr, ptr, sizeof(pte_t));
>  	uasm_i_xori(p, ptr, ptr, sizeof(pte_t));
>  	build_update_entries(p, tmp, ptr);
> +	uasm_i_mfc0(p, ptr, C0_INDEX);
> +	uasm_il_bltz(p, r, ptr, label_tail_miss);
> +	uasm_i_nop(p);
>  	build_tlb_write_entry(p, l, r, tlb_indexed);
> +	uasm_il_b(p, r, label_leave);
> +	uasm_i_nop(p);
> +	uasm_l_tail_miss(l, *p);
> +	build_tlb_write_entry(p, l, r, tlb_random);
>  	uasm_l_leave(l, *p);
>  	build_restore_work_registers(p);
>  	uasm_i_eret(p); /* return from trap */

 Specifically you're causing a performance hit here, which is a fast path, 
for everyone.  If you have a scenario that needs this change, then please 
make it conditional on the circumstances and keep the handler unchanged in 
all the other cases.

  Maciej

Re: [PATCH] MIPS: tlbex: Fix bugs in tlbchange handler

[Maciej W. Rozycki]

On Wed, 24 Feb 2016, Huacai Chen wrote:

> If a tlb miss triggered when EXL=1, tlb refill exception is treated as
> tlb invalid exception, so tlbp may fails. In this situation, CP0_Index
> register doesn't contain a valid value. This may not be a problem for
> VTLB since it is fully-associative. However, FTLB is set-associative so
> not every tlb entry is valid for a specific address. Thus, we should
> use tlbwr instead of tlbwi when tlbp fails.

 Can you please explain exactly why this change is needed?  You're 
changing pretty generic code which has worked since forever.  So why is a 
change suddenly needed?  Our kernel entry/exit code has been written such 
that no mapped memory is accessed with EXL=1 so no TLB exception is 
expected to ever happen in these circumstances.  So what's your scenario 
you mean to address?  Your patch description does not explain it.

> @@ -1913,7 +1935,14 @@ build_r4000_tlbchange_handler_tail(u32 **p, struct uasm_label **l,
>  	uasm_i_ori(p, ptr, ptr, sizeof(pte_t));
>  	uasm_i_xori(p, ptr, ptr, sizeof(pte_t));
>  	build_update_entries(p, tmp, ptr);
> +	uasm_i_mfc0(p, ptr, C0_INDEX);
> +	uasm_il_bltz(p, r, ptr, label_tail_miss);
> +	uasm_i_nop(p);
>  	build_tlb_write_entry(p, l, r, tlb_indexed);
> +	uasm_il_b(p, r, label_leave);
> +	uasm_i_nop(p);
> +	uasm_l_tail_miss(l, *p);
> +	build_tlb_write_entry(p, l, r, tlb_random);
>  	uasm_l_leave(l, *p);
>  	build_restore_work_registers(p);
>  	uasm_i_eret(p); /* return from trap */

 Specifically you're causing a performance hit here, which is a fast path, 
for everyone.  If you have a scenario that needs this change, then please 
make it conditional on the circumstances and keep the handler unchanged in 
all the other cases.

  Maciej

Re: [PATCH] MIPS: tlbex: Fix bugs in tlbchange handler

[David Daney]

On 02/24/2016 04:40 PM, Maciej W. Rozycki wrote:
> On Wed, 24 Feb 2016, Huacai Chen wrote:
>
>> If a tlb miss triggered when EXL=1, tlb refill exception is treated as
>> tlb invalid exception, so tlbp may fails. In this situation, CP0_Index
>> register doesn't contain a valid value. This may not be a problem for
>> VTLB since it is fully-associative. However, FTLB is set-associative so
>> not every tlb entry is valid for a specific address. Thus, we should
>> use tlbwr instead of tlbwi when tlbp fails.
>
>   Can you please explain exactly why this change is needed?  You're
> changing pretty generic code which has worked since forever.  So why is a
> change suddenly needed?  Our kernel entry/exit code has been written such
> that no mapped memory is accessed with EXL=1 so no TLB exception is
> expected to ever happen in these circumstances.  So what's your scenario
> you mean to address?  Your patch description does not explain it.
>

I asked this exact same question back on Jan. 26, when the patch was 
previously posted.

No answer was given, all we got was the same thing again with no 
explanation.

David Daney



>> @@ -1913,7 +1935,14 @@ build_r4000_tlbchange_handler_tail(u32 **p, struct uasm_label **l,
>>   	uasm_i_ori(p, ptr, ptr, sizeof(pte_t));
>>   	uasm_i_xori(p, ptr, ptr, sizeof(pte_t));
>>   	build_update_entries(p, tmp, ptr);
>> +	uasm_i_mfc0(p, ptr, C0_INDEX);
>> +	uasm_il_bltz(p, r, ptr, label_tail_miss);
>> +	uasm_i_nop(p);
>>   	build_tlb_write_entry(p, l, r, tlb_indexed);
>> +	uasm_il_b(p, r, label_leave);
>> +	uasm_i_nop(p);
>> +	uasm_l_tail_miss(l, *p);
>> +	build_tlb_write_entry(p, l, r, tlb_random);
>>   	uasm_l_leave(l, *p);
>>   	build_restore_work_registers(p);
>>   	uasm_i_eret(p); /* return from trap */
>
>   Specifically you're causing a performance hit here, which is a fast path,
> for everyone.  If you have a scenario that needs this change, then please
> make it conditional on the circumstances and keep the handler unchanged in
> all the other cases.
>
>    Maciej
>
>
>

Re: [PATCH] MIPS: tlbex: Fix bugs in tlbchange handler

[Huacai Chen]

Hi, David and Maciej,

I *do* have answered your quesition, which you can see here:
https://patchwork.linux-mips.org/patch/12240/

When unaligned access triggered, do_ade() will access user address
with EXL=1, and that may trigger tlb refill.

Huacai

On Thu, Feb 25, 2016 at 9:06 AM, David Daney <ddaney.cavm@gmail.com> wrote:
> On 02/24/2016 04:40 PM, Maciej W. Rozycki wrote:
>>
>> On Wed, 24 Feb 2016, Huacai Chen wrote:
>>
>>> If a tlb miss triggered when EXL=1, tlb refill exception is treated as
>>> tlb invalid exception, so tlbp may fails. In this situation, CP0_Index
>>> register doesn't contain a valid value. This may not be a problem for
>>> VTLB since it is fully-associative. However, FTLB is set-associative so
>>> not every tlb entry is valid for a specific address. Thus, we should
>>> use tlbwr instead of tlbwi when tlbp fails.
>>
>>
>>   Can you please explain exactly why this change is needed?  You're
>> changing pretty generic code which has worked since forever.  So why is a
>> change suddenly needed?  Our kernel entry/exit code has been written such
>> that no mapped memory is accessed with EXL=1 so no TLB exception is
>> expected to ever happen in these circumstances.  So what's your scenario
>> you mean to address?  Your patch description does not explain it.
>>
>
> I asked this exact same question back on Jan. 26, when the patch was
> previously posted.
>
> No answer was given, all we got was the same thing again with no
> explanation.
>
> David Daney
>
>
>
>
>>> @@ -1913,7 +1935,14 @@ build_r4000_tlbchange_handler_tail(u32 **p, struct
>>> uasm_label **l,
>>>         uasm_i_ori(p, ptr, ptr, sizeof(pte_t));
>>>         uasm_i_xori(p, ptr, ptr, sizeof(pte_t));
>>>         build_update_entries(p, tmp, ptr);
>>> +       uasm_i_mfc0(p, ptr, C0_INDEX);
>>> +       uasm_il_bltz(p, r, ptr, label_tail_miss);
>>> +       uasm_i_nop(p);
>>>         build_tlb_write_entry(p, l, r, tlb_indexed);
>>> +       uasm_il_b(p, r, label_leave);
>>> +       uasm_i_nop(p);
>>> +       uasm_l_tail_miss(l, *p);
>>> +       build_tlb_write_entry(p, l, r, tlb_random);
>>>         uasm_l_leave(l, *p);
>>>         build_restore_work_registers(p);
>>>         uasm_i_eret(p); /* return from trap */
>>
>>
>>   Specifically you're causing a performance hit here, which is a fast
>> path,
>> for everyone.  If you have a scenario that needs this change, then please
>> make it conditional on the circumstances and keep the handler unchanged in
>> all the other cases.
>>
>>    Maciej
>>
>>
>>
>
>

Re: [PATCH] MIPS: tlbex: Fix bugs in tlbchange handler

[Maciej W. Rozycki]

On Thu, 25 Feb 2016, Huacai Chen wrote:

> I *do* have answered your quesition, which you can see here:
> https://patchwork.linux-mips.org/patch/12240/
> 
> When unaligned access triggered, do_ade() will access user address
> with EXL=1, and that may trigger tlb refill.

 Thank you and I can see your original answer as well, which would need to 
go into the patch description itself, however it is too vague for me I'm 
afraid.

 Unaligned accesses have been likewise handled since forever and no one 
else has hit an issue with `do_ade' making user accesses with EXL=1, as 
this handler is only called once KMODE has cleared EXL.  So please explain 
us the scenario in which it doesn't happen and you still have EXL=1 in 
`do_ade'.  What the conditions for this to be the case?

 Without a further explanation it looks like a bug elsewhere to me, which 
needs to be treated accordingly there, rather than by complicating TLB 
handlers.

  Maciej

Re: [PATCH] MIPS: tlbex: Fix bugs in tlbchange handler

[Huacai Chen]

OK, I know, let me have a think. This patch fix a "bug" on Loongson,
but maybe we don't have get the root cause.

Huacai

On Thu, Feb 25, 2016 at 7:02 PM, Maciej W. Rozycki <macro@imgtec.com> wrote:
> On Thu, 25 Feb 2016, Huacai Chen wrote:
>
>> I *do* have answered your quesition, which you can see here:
>> https://patchwork.linux-mips.org/patch/12240/
>>
>> When unaligned access triggered, do_ade() will access user address
>> with EXL=1, and that may trigger tlb refill.
>
>  Thank you and I can see your original answer as well, which would need to
> go into the patch description itself, however it is too vague for me I'm
> afraid.
>
>  Unaligned accesses have been likewise handled since forever and no one
> else has hit an issue with `do_ade' making user accesses with EXL=1, as
> this handler is only called once KMODE has cleared EXL.  So please explain
> us the scenario in which it doesn't happen and you still have EXL=1 in
> `do_ade'.  What the conditions for this to be the case?
>
>  Without a further explanation it looks like a bug elsewhere to me, which
> needs to be treated accordingly there, rather than by complicating TLB
> handlers.
>
>   Maciej
>

[PATCH] MIPS: Introduce cpu_has_coherent_cache feature

[Huacai Chen]

If a platform maintains cache coherency by hardware fully:
 1) It's icache is coherent with dcache.
 2) It's dcaches don't alias (maybe depend on PAGE_SIZE).
 3) It maintains cache coherency across cores (and for DMA).

So we introduce a MIPS_CPU_CACHE_COHERENT bit, and a cpu feature named
cpu_has_coherent_cache to modify MIPS's cache flushing functions.

Signed-off-by: Huacai Chen <chenhc@lemote.com>
---
 arch/mips/include/asm/cpu-features.h |  3 +++
 arch/mips/include/asm/cpu.h          |  1 +
 arch/mips/mm/c-r4k.c                 | 21 +++++++++++++++++++++
 3 files changed, 25 insertions(+)

diff --git a/arch/mips/include/asm/cpu-features.h b/arch/mips/include/asm/cpu-features.h
index eeec8c8..0b7f8a5 100644
--- a/arch/mips/include/asm/cpu-features.h
+++ b/arch/mips/include/asm/cpu-features.h
@@ -145,6 +145,9 @@
 #ifndef cpu_has_xpa
 #define cpu_has_xpa		(cpu_data[0].options & MIPS_CPU_XPA)
 #endif
+#ifndef cpu_has_coherent_cache
+#define cpu_has_coherent_cache	(cpu_data[0].options & MIPS_CPU_CACHE_COHERENT)
+#endif
 #ifndef cpu_has_vtag_icache
 #define cpu_has_vtag_icache	(cpu_data[0].icache.flags & MIPS_CACHE_VTAG)
 #endif
diff --git a/arch/mips/include/asm/cpu.h b/arch/mips/include/asm/cpu.h
index 7bea0f3..541ba9e 100644
--- a/arch/mips/include/asm/cpu.h
+++ b/arch/mips/include/asm/cpu.h
@@ -390,6 +390,7 @@ enum cpu_type_enum {
 #define MIPS_CPU_FTLB		0x20000000000ull /* CPU has Fixed-page-size TLB */
 #define MIPS_CPU_NAN_LEGACY	0x40000000000ull /* Legacy NaN implemented */
 #define MIPS_CPU_NAN_2008	0x80000000000ull /* 2008 NaN implemented */
+#define MIPS_CPU_CACHE_COHERENT	0x100000000000ull /* CPU maintains cache coherency by hardware */
 
 /*
  * CPU ASE encodings
diff --git a/arch/mips/mm/c-r4k.c b/arch/mips/mm/c-r4k.c
index caac3d7..04a38d8 100644
--- a/arch/mips/mm/c-r4k.c
+++ b/arch/mips/mm/c-r4k.c
@@ -428,6 +428,9 @@ static void r4k_blast_scache_setup(void)
 
 static inline void local_r4k___flush_cache_all(void * args)
 {
+	if (cpu_has_coherent_cache)
+		return;
+
 	switch (current_cpu_type()) {
 	case CPU_LOONGSON2:
 	case CPU_LOONGSON3:
@@ -456,6 +459,9 @@ static inline void local_r4k___flush_cache_all(void * args)
 
 static void r4k___flush_cache_all(void)
 {
+	if (cpu_has_coherent_cache)
+		return;
+
 	r4k_on_each_cpu(local_r4k___flush_cache_all, NULL);
 }
 
@@ -502,6 +508,9 @@ static void r4k_flush_cache_range(struct vm_area_struct *vma,
 {
 	int exec = vma->vm_flags & VM_EXEC;
 
+	if (cpu_has_coherent_cache)
+		return;
+
 	if (cpu_has_dc_aliases || (exec && !cpu_has_ic_fills_f_dc))
 		r4k_on_each_cpu(local_r4k_flush_cache_range, vma);
 }
@@ -626,6 +635,9 @@ static void r4k_flush_cache_page(struct vm_area_struct *vma,
 {
 	struct flush_cache_page_args args;
 
+	if (cpu_has_coherent_cache)
+		return;
+
 	args.vma = vma;
 	args.addr = addr;
 	args.pfn = pfn;
@@ -635,11 +647,17 @@ static void r4k_flush_cache_page(struct vm_area_struct *vma,
 
 static inline void local_r4k_flush_data_cache_page(void * addr)
 {
+	if (cpu_has_coherent_cache)
+		return;
+
 	r4k_blast_dcache_page((unsigned long) addr);
 }
 
 static void r4k_flush_data_cache_page(unsigned long addr)
 {
+	if (cpu_has_coherent_cache)
+		return;
+
 	if (in_atomic())
 		local_r4k_flush_data_cache_page((void *)addr);
 	else
@@ -824,6 +842,9 @@ static void local_r4k_flush_cache_sigtramp(void * arg)
 
 static void r4k_flush_cache_sigtramp(unsigned long addr)
 {
+	if (cpu_has_coherent_cache)
+		return;
+
 	r4k_on_each_cpu(local_r4k_flush_cache_sigtramp, (void *) addr);
 }
 
-- 
2.7.0

Re: [PATCH] MIPS: Introduce cpu_has_coherent_cache feature

[Maciej W. Rozycki]

On Wed, 24 Feb 2016, Huacai Chen wrote:

> If a platform maintains cache coherency by hardware fully:
>  1) It's icache is coherent with dcache.
>  2) It's dcaches don't alias (maybe depend on PAGE_SIZE).
>  3) It maintains cache coherency across cores (and for DMA).
> 
> So we introduce a MIPS_CPU_CACHE_COHERENT bit, and a cpu feature named
> cpu_has_coherent_cache to modify MIPS's cache flushing functions.
[...]
> diff --git a/arch/mips/mm/c-r4k.c b/arch/mips/mm/c-r4k.c
> index caac3d7..04a38d8 100644
> --- a/arch/mips/mm/c-r4k.c
> +++ b/arch/mips/mm/c-r4k.c
> @@ -428,6 +428,9 @@ static void r4k_blast_scache_setup(void)
>  
>  static inline void local_r4k___flush_cache_all(void * args)
>  {
> +	if (cpu_has_coherent_cache)
> +		return;
> +
[etc.]

 Have you considered setting the relevant handlers to `cache_noop' in 
`r4k_cache_init' instead?  It seems more natural to me and avoids the 
performance hit where `cpu_has_coherent_cache' is variable, which at this 
time means everywhere.

 Also you don't set the MIPS_CPU_CACHE_COHERENT bit anywhere within you 
patch set -- is there a follow-up change you're going to submit?

  Maciej

Re: [PATCH] MIPS: Introduce cpu_has_coherent_cache feature

[Maciej W. Rozycki]

On Wed, 24 Feb 2016, Huacai Chen wrote:

> If a platform maintains cache coherency by hardware fully:
>  1) It's icache is coherent with dcache.
>  2) It's dcaches don't alias (maybe depend on PAGE_SIZE).
>  3) It maintains cache coherency across cores (and for DMA).
> 
> So we introduce a MIPS_CPU_CACHE_COHERENT bit, and a cpu feature named
> cpu_has_coherent_cache to modify MIPS's cache flushing functions.
[...]
> diff --git a/arch/mips/mm/c-r4k.c b/arch/mips/mm/c-r4k.c
> index caac3d7..04a38d8 100644
> --- a/arch/mips/mm/c-r4k.c
> +++ b/arch/mips/mm/c-r4k.c
> @@ -428,6 +428,9 @@ static void r4k_blast_scache_setup(void)
>  
>  static inline void local_r4k___flush_cache_all(void * args)
>  {
> +	if (cpu_has_coherent_cache)
> +		return;
> +
[etc.]

 Have you considered setting the relevant handlers to `cache_noop' in 
`r4k_cache_init' instead?  It seems more natural to me and avoids the 
performance hit where `cpu_has_coherent_cache' is variable, which at this 
time means everywhere.

 Also you don't set the MIPS_CPU_CACHE_COHERENT bit anywhere within you 
patch set -- is there a follow-up change you're going to submit?

  Maciej

Re: [PATCH] MIPS: Introduce cpu_has_coherent_cache feature

[Huacai Chen]

Hi, Maciej

Joshua Kinard suggest me to split the patch, which you can see here:
https://patchwork.linux-mips.org/patch/12324/
If it is better to not split, please see my V2 patchset.

Huacai

On Thu, Feb 25, 2016 at 8:59 AM, Maciej W. Rozycki <macro@imgtec.com> wrote:
> On Wed, 24 Feb 2016, Huacai Chen wrote:
>
>> If a platform maintains cache coherency by hardware fully:
>>  1) It's icache is coherent with dcache.
>>  2) It's dcaches don't alias (maybe depend on PAGE_SIZE).
>>  3) It maintains cache coherency across cores (and for DMA).
>>
>> So we introduce a MIPS_CPU_CACHE_COHERENT bit, and a cpu feature named
>> cpu_has_coherent_cache to modify MIPS's cache flushing functions.
> [...]
>> diff --git a/arch/mips/mm/c-r4k.c b/arch/mips/mm/c-r4k.c
>> index caac3d7..04a38d8 100644
>> --- a/arch/mips/mm/c-r4k.c
>> +++ b/arch/mips/mm/c-r4k.c
>> @@ -428,6 +428,9 @@ static void r4k_blast_scache_setup(void)
>>
>>  static inline void local_r4k___flush_cache_all(void * args)
>>  {
>> +     if (cpu_has_coherent_cache)
>> +             return;
>> +
> [etc.]
>
>  Have you considered setting the relevant handlers to `cache_noop' in
> `r4k_cache_init' instead?  It seems more natural to me and avoids the
> performance hit where `cpu_has_coherent_cache' is variable, which at this
> time means everywhere.
>
>  Also you don't set the MIPS_CPU_CACHE_COHERENT bit anywhere within you
> patch set -- is there a follow-up change you're going to submit?
>
>   Maciej
>

Re: [PATCH] MIPS: Introduce cpu_has_coherent_cache feature

[Maciej W. Rozycki]

On Thu, 25 Feb 2016, Huacai Chen wrote:

> Joshua Kinard suggest me to split the patch, which you can see here:
> https://patchwork.linux-mips.org/patch/12324/
> If it is better to not split, please see my V2 patchset.

 Splitting is the correct action, however please answer my questions as 
well.

  Maciej

Re: [PATCH] MIPS: Introduce cpu_has_coherent_cache feature

[Huacai Chen]

OK, thanks. I will consider setting the relevant handlers to
cache_noop in r4k_cache_init().

On Thu, Feb 25, 2016 at 7:16 PM, Maciej W. Rozycki <macro@imgtec.com> wrote:
> On Thu, 25 Feb 2016, Huacai Chen wrote:
>
>> Joshua Kinard suggest me to split the patch, which you can see here:
>> https://patchwork.linux-mips.org/patch/12324/
>> If it is better to not split, please see my V2 patchset.
>
>  Splitting is the correct action, however please answer my questions as
> well.
>
>   Maciej
>

Re: [PATCH] MIPS: Introduce cpu_has_coherent_cache feature

[Huacai Chen]

Hi, Maciej,

Not all handlers have the same prototype as cache_noop(), so what should I do?

Huacai

On Thu, Feb 25, 2016 at 8:53 PM, Huacai Chen <chenhc@lemote.com> wrote:
> OK, thanks. I will consider setting the relevant handlers to
> cache_noop in r4k_cache_init().
>
> On Thu, Feb 25, 2016 at 7:16 PM, Maciej W. Rozycki <macro@imgtec.com> wrote:
>> On Thu, 25 Feb 2016, Huacai Chen wrote:
>>
>>> Joshua Kinard suggest me to split the patch, which you can see here:
>>> https://patchwork.linux-mips.org/patch/12324/
>>> If it is better to not split, please see my V2 patchset.
>>
>>  Splitting is the correct action, however please answer my questions as
>> well.
>>
>>   Maciej
>>

Re: [PATCH] MIPS: Introduce cpu_has_coherent_cache feature

[Maciej W. Rozycki]

On Fri, 26 Feb 2016, Huacai Chen wrote:

> Not all handlers have the same prototype as cache_noop(), so what should I do?

 Just add compatible empty handlers as required.  With recent GCC versions 
it may not even add binary code thanks to function epilogue merging.

  Maciej

[PATCH V3 1/5] MIPS: Loongson: Add Loongson-3A R2 basic support

[Huacai Chen]

Loongson-3 CPU family:

Code-name       Brand-name       PRId
Loongson-3A R1  Loongson-3A1000  0x6305
Loongson-3A R2  Loongson-3A2000  0x6308
Loongson-3B R1  Loongson-3B1000  0x6306
Loongson-3B R2  Loongson-3B1500  0x6307

Features of R2 revision of Loongson-3A:
1, Primary cache includes I-Cache, D-Cache and V-Cache (Victim Cache).
2, I-Cache, D-Cache and V-Cache are 16-way set-associative, linesize is 64 Bytes.
3, 64 entries of VTLB (classic TLB), 1024 entries of FTLB (8-way set-associative).
4, Support DSP/DSPv2 instructions, UserLocal register and Read-Inhibit/Execute-Inhibit.

Signed-off-by: Huacai Chen <chenhc@lemote.com>
---
 arch/mips/Kconfig                                  |   1 +
 arch/mips/include/asm/cacheops.h                   |   6 ++
 arch/mips/include/asm/cpu-info.h                   |   1 +
 arch/mips/include/asm/cpu.h                        |   4 +-
 .../asm/mach-loongson64/cpu-feature-overrides.h    |  12 +--
 .../asm/mach-loongson64/kernel-entry-init.h        |   6 +-
 arch/mips/include/asm/mipsregs.h                   |   2 +
 arch/mips/include/asm/pgtable-bits.h               |   8 +-
 arch/mips/include/asm/pgtable.h                    |   4 +-
 arch/mips/kernel/cpu-probe.c                       |  38 +++++++-
 arch/mips/kernel/idle.c                            |   5 +
 arch/mips/kernel/traps.c                           |   3 +-
 arch/mips/loongson64/common/env.c                  |   7 +-
 arch/mips/loongson64/loongson-3/smp.c              | 106 +++++++++++++++++++--
 arch/mips/mm/c-r4k.c                               |  27 ++++++
 arch/mips/mm/tlbex.c                               |   2 +-
 drivers/platform/mips/cpu_hwmon.c                  |   4 +-
 17 files changed, 201 insertions(+), 35 deletions(-)

diff --git a/arch/mips/Kconfig b/arch/mips/Kconfig
index c5bf89cb..38c0b11 100644
--- a/arch/mips/Kconfig
+++ b/arch/mips/Kconfig
@@ -1344,6 +1344,7 @@ config CPU_LOONGSON3
 	select CPU_SUPPORTS_HUGEPAGES
 	select WEAK_ORDERING
 	select WEAK_REORDERING_BEYOND_LLSC
+	select MIPS_PGD_C0_CONTEXT
 	select ARCH_REQUIRE_GPIOLIB
 	help
 		The Loongson 3 processor implements the MIPS64R2 instruction
diff --git a/arch/mips/include/asm/cacheops.h b/arch/mips/include/asm/cacheops.h
index c3212ff..8031fbc 100644
--- a/arch/mips/include/asm/cacheops.h
+++ b/arch/mips/include/asm/cacheops.h
@@ -21,6 +21,7 @@
 #define Cache_I				0x00
 #define Cache_D				0x01
 #define Cache_T				0x02
+#define Cache_V				0x02 /* Loongson-3 */
 #define Cache_S				0x03
 
 #define Index_Writeback_Inv		0x00
@@ -107,4 +108,9 @@
  */
 #define Hit_Invalidate_I_Loongson2	(Cache_I | 0x00)
 
+/*
+ * Loongson3-specific cacheops
+ */
+#define Index_Writeback_Inv_V		(Cache_V | Index_Writeback_Inv)
+
 #endif	/* __ASM_CACHEOPS_H */
diff --git a/arch/mips/include/asm/cpu-info.h b/arch/mips/include/asm/cpu-info.h
index e7dc785..6fd7b8bd 100644
--- a/arch/mips/include/asm/cpu-info.h
+++ b/arch/mips/include/asm/cpu-info.h
@@ -60,6 +60,7 @@ struct cpuinfo_mips {
 	int			tlbsizeftlbways;
 	struct cache_desc	icache; /* Primary I-cache */
 	struct cache_desc	dcache; /* Primary D or combined I/D cache */
+	struct cache_desc	vcache; /* Victim cache, between pcache and scache */
 	struct cache_desc	scache; /* Secondary cache */
 	struct cache_desc	tcache; /* Tertiary/split secondary cache */
 	int			srsets; /* Shadow register sets */
diff --git a/arch/mips/include/asm/cpu.h b/arch/mips/include/asm/cpu.h
index 541ba9e..858d036 100644
--- a/arch/mips/include/asm/cpu.h
+++ b/arch/mips/include/asm/cpu.h
@@ -42,6 +42,7 @@
 #define PRID_COMP_LEXRA		0x0b0000
 #define PRID_COMP_NETLOGIC	0x0c0000
 #define PRID_COMP_CAVIUM	0x0d0000
+#define PRID_COMP_LOONGSON	0x140000
 #define PRID_COMP_INGENIC_D0	0xd00000	/* JZ4740, JZ4750 */
 #define PRID_COMP_INGENIC_D1	0xd10000	/* JZ4770, JZ4775 */
 #define PRID_COMP_INGENIC_E1	0xe10000	/* JZ4780 */
@@ -239,9 +240,10 @@
 #define PRID_REV_LOONGSON1B	0x0020
 #define PRID_REV_LOONGSON2E	0x0002
 #define PRID_REV_LOONGSON2F	0x0003
-#define PRID_REV_LOONGSON3A	0x0005
+#define PRID_REV_LOONGSON3A_R1	0x0005
 #define PRID_REV_LOONGSON3B_R1	0x0006
 #define PRID_REV_LOONGSON3B_R2	0x0007
+#define PRID_REV_LOONGSON3A_R2	0x0008
 
 /*
  * Older processors used to encode processor version and revision in two
diff --git a/arch/mips/include/asm/mach-loongson64/cpu-feature-overrides.h b/arch/mips/include/asm/mach-loongson64/cpu-feature-overrides.h
index 98963c2..c3406db 100644
--- a/arch/mips/include/asm/mach-loongson64/cpu-feature-overrides.h
+++ b/arch/mips/include/asm/mach-loongson64/cpu-feature-overrides.h
@@ -16,11 +16,6 @@
 #ifndef __ASM_MACH_LOONGSON64_CPU_FEATURE_OVERRIDES_H
 #define __ASM_MACH_LOONGSON64_CPU_FEATURE_OVERRIDES_H
 
-#define cpu_dcache_line_size()	32
-#define cpu_icache_line_size()	32
-#define cpu_scache_line_size()	32
-
-
 #define cpu_has_32fpr		1
 #define cpu_has_3k_cache	0
 #define cpu_has_4k_cache	1
@@ -31,8 +26,6 @@
 #define cpu_has_counter		1
 #define cpu_has_dc_aliases	(PAGE_SIZE < 0x4000)
 #define cpu_has_divec		0
-#define cpu_has_dsp		0
-#define cpu_has_dsp2		0
 #define cpu_has_ejtag		0
 #define cpu_has_ic_fills_f_dc	0
 #define cpu_has_inclusive_pcaches	1
@@ -40,15 +33,11 @@
 #define cpu_has_mcheck		0
 #define cpu_has_mdmx		0
 #define cpu_has_mips16		0
-#define cpu_has_mips32r2	0
 #define cpu_has_mips3d		0
-#define cpu_has_mips64r2	0
 #define cpu_has_mipsmt		0
-#define cpu_has_prefetch	0
 #define cpu_has_smartmips	0
 #define cpu_has_tlb		1
 #define cpu_has_tx39_cache	0
-#define cpu_has_userlocal	0
 #define cpu_has_vce		0
 #define cpu_has_veic		0
 #define cpu_has_vint		0
@@ -57,5 +46,6 @@
 #define cpu_has_local_ebase	0
 
 #define cpu_has_wsbh		IS_ENABLED(CONFIG_CPU_LOONGSON3)
+#define cpu_hwrena_impl_bits	0xc0000000
 
 #endif /* __ASM_MACH_LOONGSON64_CPU_FEATURE_OVERRIDES_H */
diff --git a/arch/mips/include/asm/mach-loongson64/kernel-entry-init.h b/arch/mips/include/asm/mach-loongson64/kernel-entry-init.h
index 3f2f84f..da83482 100644
--- a/arch/mips/include/asm/mach-loongson64/kernel-entry-init.h
+++ b/arch/mips/include/asm/mach-loongson64/kernel-entry-init.h
@@ -23,7 +23,8 @@
 	or	t0, (0x1 << 7)
 	mtc0	t0, $16, 3
 	/* Set ELPA on LOONGSON3 pagegrain */
-	li	t0, (0x1 << 29)
+	mfc0	t0, $5, 1
+	or	t0, (0x1 << 29)
 	mtc0	t0, $5, 1
 	_ehb
 	.set	pop
@@ -42,7 +43,8 @@
 	or	t0, (0x1 << 7)
 	mtc0	t0, $16, 3
 	/* Set ELPA on LOONGSON3 pagegrain */
-	li	t0, (0x1 << 29)
+	mfc0	t0, $5, 1
+	or	t0, (0x1 << 29)
 	mtc0	t0, $5, 1
 	_ehb
 	.set	pop
diff --git a/arch/mips/include/asm/mipsregs.h b/arch/mips/include/asm/mipsregs.h
index 3ad19ad..9290fd4 100644
--- a/arch/mips/include/asm/mipsregs.h
+++ b/arch/mips/include/asm/mipsregs.h
@@ -633,6 +633,8 @@
 #define MIPS_CONF6_SYND		(_ULCAST_(1) << 13)
 /* proAptiv FTLB on/off bit */
 #define MIPS_CONF6_FTLBEN	(_ULCAST_(1) << 15)
+/* Loongson-3 FTLB on/off bit */
+#define MIPS_CONF6_FTLBDIS	(_ULCAST_(1) << 22)
 /* FTLB probability bits */
 #define MIPS_CONF6_FTLBP_SHIFT	(16)
 
diff --git a/arch/mips/include/asm/pgtable-bits.h b/arch/mips/include/asm/pgtable-bits.h
index 97b3138..32b77bd 100644
--- a/arch/mips/include/asm/pgtable-bits.h
+++ b/arch/mips/include/asm/pgtable-bits.h
@@ -113,7 +113,7 @@
 #define _PAGE_PRESENT_SHIFT	0
 #define _PAGE_PRESENT		(1 << _PAGE_PRESENT_SHIFT)
 /* R2 or later cores check for RI/XI support to determine _PAGE_READ */
-#if defined(CONFIG_CPU_MIPSR2) || defined(CONFIG_CPU_MIPSR6)
+#if defined(CONFIG_CPU_MIPSR2) || defined(CONFIG_CPU_MIPSR6) || defined(CONFIG_CPU_LOONGSON3)
 #define _PAGE_WRITE_SHIFT	(_PAGE_PRESENT_SHIFT + 1)
 #define _PAGE_WRITE		(1 << _PAGE_WRITE_SHIFT)
 #else
@@ -133,7 +133,7 @@
 #define _PAGE_HUGE		(1 << _PAGE_HUGE_SHIFT)
 #endif	/* CONFIG_64BIT && CONFIG_MIPS_HUGE_TLB_SUPPORT */
 
-#if defined(CONFIG_CPU_MIPSR2) || defined(CONFIG_CPU_MIPSR6)
+#if defined(CONFIG_CPU_MIPSR2) || defined(CONFIG_CPU_MIPSR6) || defined(CONFIG_CPU_LOONGSON3)
 /* XI - page cannot be executed */
 #ifdef _PAGE_HUGE_SHIFT
 #define _PAGE_NO_EXEC_SHIFT	(_PAGE_HUGE_SHIFT + 1)
@@ -147,7 +147,7 @@
 #define _PAGE_READ		(cpu_has_rixi ? 0 : (1 << _PAGE_READ_SHIFT))
 #define _PAGE_NO_READ_SHIFT	_PAGE_READ_SHIFT
 #define _PAGE_NO_READ		(cpu_has_rixi ? (1 << _PAGE_READ_SHIFT) : 0)
-#endif	/* defined(CONFIG_CPU_MIPSR2) || defined(CONFIG_CPU_MIPSR6) */
+#endif	/* defined(CONFIG_CPU_MIPSR2) || defined(CONFIG_CPU_MIPSR6) || defined(CONFIG_CPU_LOONGSON3) */
 
 #if defined(_PAGE_NO_READ_SHIFT)
 #define _PAGE_GLOBAL_SHIFT	(_PAGE_NO_READ_SHIFT + 1)
@@ -198,7 +198,7 @@
  */
 static inline uint64_t pte_to_entrylo(unsigned long pte_val)
 {
-#if defined(CONFIG_CPU_MIPSR2) || defined(CONFIG_CPU_MIPSR6)
+#if defined(CONFIG_CPU_MIPSR2) || defined(CONFIG_CPU_MIPSR6) || defined(CONFIG_CPU_LOONGSON3)
 	if (cpu_has_rixi) {
 		int sa;
 #ifdef CONFIG_32BIT
diff --git a/arch/mips/include/asm/pgtable.h b/arch/mips/include/asm/pgtable.h
index 9a4fe01..35cd713 100644
--- a/arch/mips/include/asm/pgtable.h
+++ b/arch/mips/include/asm/pgtable.h
@@ -353,7 +353,7 @@ static inline pte_t pte_mkdirty(pte_t pte)
 static inline pte_t pte_mkyoung(pte_t pte)
 {
 	pte_val(pte) |= _PAGE_ACCESSED;
-#if defined(CONFIG_CPU_MIPSR2) || defined(CONFIG_CPU_MIPSR6)
+#if defined(CONFIG_CPU_MIPSR2) || defined(CONFIG_CPU_MIPSR6) || defined(CONFIG_CPU_LOONGSON3)
 	if (!(pte_val(pte) & _PAGE_NO_READ))
 		pte_val(pte) |= _PAGE_SILENT_READ;
 	else
@@ -542,7 +542,7 @@ static inline pmd_t pmd_mkyoung(pmd_t pmd)
 {
 	pmd_val(pmd) |= _PAGE_ACCESSED;
 
-#if defined(CONFIG_CPU_MIPSR2) || defined(CONFIG_CPU_MIPSR6)
+#if defined(CONFIG_CPU_MIPSR2) || defined(CONFIG_CPU_MIPSR6) || defined(CONFIG_CPU_LOONGSON3)
 	if (!(pmd_val(pmd) & _PAGE_NO_READ))
 		pmd_val(pmd) |= _PAGE_SILENT_READ;
 	else
diff --git a/arch/mips/kernel/cpu-probe.c b/arch/mips/kernel/cpu-probe.c
index 9ad6157..2a2ae86 100644
--- a/arch/mips/kernel/cpu-probe.c
+++ b/arch/mips/kernel/cpu-probe.c
@@ -561,6 +561,16 @@ static int set_ftlb_enable(struct cpuinfo_mips *c, int enable)
 		write_c0_config7(config | (calculate_ftlb_probability(c)
 					   << MIPS_CONF7_FTLBP_SHIFT));
 		break;
+	case CPU_LOONGSON3:
+		/* Loongson-3 cores use Config6 to enable the FTLB */
+		config = read_c0_config6();
+		if (enable)
+			/* Enable FTLB */
+			write_c0_config6(config & ~MIPS_CONF6_FTLBDIS);
+		else
+			/* Disable FTLB */
+			write_c0_config6(config | MIPS_CONF6_FTLBDIS);
+		break;
 	default:
 		return 1;
 	}
@@ -1172,7 +1182,7 @@ static inline void cpu_probe_legacy(struct cpuinfo_mips *c, unsigned int cpu)
 			set_isa(c, MIPS_CPU_ISA_III);
 			c->fpu_msk31 |= FPU_CSR_CONDX;
 			break;
-		case PRID_REV_LOONGSON3A:
+		case PRID_REV_LOONGSON3A_R1:
 			c->cputype = CPU_LOONGSON3;
 			__cpu_name[cpu] = "ICT Loongson-3";
 			set_elf_platform(cpu, "loongson3a");
@@ -1495,6 +1505,29 @@ platform:
 	}
 }
 
+static inline void cpu_probe_loongson(struct cpuinfo_mips *c, unsigned int cpu)
+{
+	switch (c->processor_id & PRID_IMP_MASK) {
+	case PRID_IMP_LOONGSON_64:  /* Loongson-2/3 */
+		switch (c->processor_id & PRID_REV_MASK) {
+		case PRID_REV_LOONGSON3A_R2:
+			c->cputype = CPU_LOONGSON3;
+			__cpu_name[cpu] = "ICT Loongson-3";
+			set_elf_platform(cpu, "loongson3a");
+			set_isa(c, MIPS_CPU_ISA_M64R2);
+			break;
+		}
+
+		decode_configs(c);
+		c->options |= MIPS_CPU_TLBINV;
+		c->writecombine = _CACHE_UNCACHED_ACCELERATED;
+		break;
+	default:
+		panic("Unknown Loongson Processor ID!");
+		break;
+	}
+}
+
 static inline void cpu_probe_ingenic(struct cpuinfo_mips *c, unsigned int cpu)
 {
 	decode_configs(c);
@@ -1642,6 +1675,9 @@ void cpu_probe(void)
 	case PRID_COMP_CAVIUM:
 		cpu_probe_cavium(c, cpu);
 		break;
+	case PRID_COMP_LOONGSON:
+		cpu_probe_loongson(c, cpu);
+		break;
 	case PRID_COMP_INGENIC_D0:
 	case PRID_COMP_INGENIC_D1:
 	case PRID_COMP_INGENIC_E1:
diff --git a/arch/mips/kernel/idle.c b/arch/mips/kernel/idle.c
index 46794d6..60ab4c4 100644
--- a/arch/mips/kernel/idle.c
+++ b/arch/mips/kernel/idle.c
@@ -181,6 +181,11 @@ void __init check_wait(void)
 	case CPU_XLP:
 		cpu_wait = r4k_wait;
 		break;
+	case CPU_LOONGSON3:
+		if ((c->processor_id & PRID_REV_MASK) >= PRID_REV_LOONGSON3A_R2)
+			cpu_wait = r4k_wait;
+		break;
+
 	case CPU_BMIPS5000:
 		cpu_wait = r4k_wait_irqoff;
 		break;
diff --git a/arch/mips/kernel/traps.c b/arch/mips/kernel/traps.c
index ae790c5..a4cca8a 100644
--- a/arch/mips/kernel/traps.c
+++ b/arch/mips/kernel/traps.c
@@ -1770,7 +1770,8 @@ asmlinkage void do_ftlb(void)
 
 	/* For the moment, report the problem and hang. */
 	if ((cpu_has_mips_r2_r6) &&
-	    ((current_cpu_data.processor_id & 0xff0000) == PRID_COMP_MIPS)) {
+	    (((current_cpu_data.processor_id & 0xff0000) == PRID_COMP_MIPS) ||
+	    ((current_cpu_data.processor_id & 0xff0000) == PRID_COMP_LOONGSON))) {
 		pr_err("FTLB error exception, cp0_ecc=0x%08x:\n",
 		       read_c0_ecc());
 		pr_err("cp0_errorepc == %0*lx\n", field, read_c0_errorepc());
diff --git a/arch/mips/loongson64/common/env.c b/arch/mips/loongson64/common/env.c
index d6d07ad..57d590a 100644
--- a/arch/mips/loongson64/common/env.c
+++ b/arch/mips/loongson64/common/env.c
@@ -105,6 +105,10 @@ void __init prom_init_env(void)
 		loongson_chiptemp[1] = 0x900010001fe0019c;
 		loongson_chiptemp[2] = 0x900020001fe0019c;
 		loongson_chiptemp[3] = 0x900030001fe0019c;
+		loongson_freqctrl[0] = 0x900000001fe001d0;
+		loongson_freqctrl[1] = 0x900010001fe001d0;
+		loongson_freqctrl[2] = 0x900020001fe001d0;
+		loongson_freqctrl[3] = 0x900030001fe001d0;
 		loongson_sysconf.ht_control_base = 0x90000EFDFB000000;
 		loongson_sysconf.workarounds = WORKAROUND_CPUFREQ;
 	} else if (ecpu->cputype == Loongson_3B) {
@@ -187,7 +191,8 @@ void __init prom_init_env(void)
 		case PRID_REV_LOONGSON2F:
 			cpu_clock_freq = 797000000;
 			break;
-		case PRID_REV_LOONGSON3A:
+		case PRID_REV_LOONGSON3A_R1:
+		case PRID_REV_LOONGSON3A_R2:
 			cpu_clock_freq = 900000000;
 			break;
 		case PRID_REV_LOONGSON3B_R1:
diff --git a/arch/mips/loongson64/loongson-3/smp.c b/arch/mips/loongson64/loongson-3/smp.c
index b913cd2..e59759a 100644
--- a/arch/mips/loongson64/loongson-3/smp.c
+++ b/arch/mips/loongson64/loongson-3/smp.c
@@ -439,7 +439,7 @@ static void loongson3_cpu_die(unsigned int cpu)
  * flush all L1 entries at first. Then, another core (usually Core 0) can
  * safely disable the clock of the target core. loongson3_play_dead() is
  * called via CKSEG1 (uncached and unmmaped) */
-static void loongson3a_play_dead(int *state_addr)
+static void loongson3a_r1_play_dead(int *state_addr)
 {
 	register int val;
 	register long cpuid, core, node, count;
@@ -501,6 +501,89 @@ static void loongson3a_play_dead(int *state_addr)
 		: "a1");
 }
 
+static void loongson3a_r2_play_dead(int *state_addr)
+{
+	register int val;
+	register long cpuid, core, node, count;
+	register void *addr, *base, *initfunc;
+
+	__asm__ __volatile__(
+		"   .set push                     \n"
+		"   .set noreorder                \n"
+		"   li %[addr], 0x80000000        \n" /* KSEG0 */
+		"1: cache 0, 0(%[addr])           \n" /* flush L1 ICache */
+		"   cache 0, 1(%[addr])           \n"
+		"   cache 0, 2(%[addr])           \n"
+		"   cache 0, 3(%[addr])           \n"
+		"   cache 1, 0(%[addr])           \n" /* flush L1 DCache */
+		"   cache 1, 1(%[addr])           \n"
+		"   cache 1, 2(%[addr])           \n"
+		"   cache 1, 3(%[addr])           \n"
+		"   addiu %[sets], %[sets], -1    \n"
+		"   bnez  %[sets], 1b             \n"
+		"   addiu %[addr], %[addr], 0x40  \n"
+		"   li %[addr], 0x80000000        \n" /* KSEG0 */
+		"2: cache 2, 0(%[addr])           \n" /* flush L1 VCache */
+		"   cache 2, 1(%[addr])           \n"
+		"   cache 2, 2(%[addr])           \n"
+		"   cache 2, 3(%[addr])           \n"
+		"   cache 2, 4(%[addr])           \n"
+		"   cache 2, 5(%[addr])           \n"
+		"   cache 2, 6(%[addr])           \n"
+		"   cache 2, 7(%[addr])           \n"
+		"   cache 2, 8(%[addr])           \n"
+		"   cache 2, 9(%[addr])           \n"
+		"   cache 2, 10(%[addr])          \n"
+		"   cache 2, 11(%[addr])          \n"
+		"   cache 2, 12(%[addr])          \n"
+		"   cache 2, 13(%[addr])          \n"
+		"   cache 2, 14(%[addr])          \n"
+		"   cache 2, 15(%[addr])          \n"
+		"   addiu %[vsets], %[vsets], -1  \n"
+		"   bnez  %[vsets], 2b            \n"
+		"   addiu %[addr], %[addr], 0x40  \n"
+		"   li    %[val], 0x7             \n" /* *state_addr = CPU_DEAD; */
+		"   sw    %[val], (%[state_addr]) \n"
+		"   sync                          \n"
+		"   cache 21, (%[state_addr])     \n" /* flush entry of *state_addr */
+		"   .set pop                      \n"
+		: [addr] "=&r" (addr), [val] "=&r" (val)
+		: [state_addr] "r" (state_addr),
+		  [sets] "r" (cpu_data[smp_processor_id()].dcache.sets),
+		  [vsets] "r" (cpu_data[smp_processor_id()].vcache.sets));
+
+	__asm__ __volatile__(
+		"   .set push                         \n"
+		"   .set noreorder                    \n"
+		"   .set mips64                       \n"
+		"   mfc0  %[cpuid], $15, 1            \n"
+		"   andi  %[cpuid], 0x3ff             \n"
+		"   dli   %[base], 0x900000003ff01000 \n"
+		"   andi  %[core], %[cpuid], 0x3      \n"
+		"   sll   %[core], 8                  \n" /* get core id */
+		"   or    %[base], %[base], %[core]   \n"
+		"   andi  %[node], %[cpuid], 0xc      \n"
+		"   dsll  %[node], 42                 \n" /* get node id */
+		"   or    %[base], %[base], %[node]   \n"
+		"1: li    %[count], 0x100             \n" /* wait for init loop */
+		"2: bnez  %[count], 2b                \n" /* limit mailbox access */
+		"   addiu %[count], -1                \n"
+		"   ld    %[initfunc], 0x20(%[base])  \n" /* get PC via mailbox */
+		"   beqz  %[initfunc], 1b             \n"
+		"   nop                               \n"
+		"   ld    $sp, 0x28(%[base])          \n" /* get SP via mailbox */
+		"   ld    $gp, 0x30(%[base])          \n" /* get GP via mailbox */
+		"   ld    $a1, 0x38(%[base])          \n"
+		"   jr    %[initfunc]                 \n" /* jump to initial PC */
+		"   nop                               \n"
+		"   .set pop                          \n"
+		: [core] "=&r" (core), [node] "=&r" (node),
+		  [base] "=&r" (base), [cpuid] "=&r" (cpuid),
+		  [count] "=&r" (count), [initfunc] "=&r" (initfunc)
+		: /* No Input */
+		: "a1");
+}
+
 static void loongson3b_play_dead(int *state_addr)
 {
 	register int val;
@@ -572,13 +655,18 @@ void play_dead(void)
 	void (*play_dead_at_ckseg1)(int *);
 
 	idle_task_exit();
-	switch (loongson_sysconf.cputype) {
-	case Loongson_3A:
+	switch (read_c0_prid() & PRID_REV_MASK) {
+	case PRID_REV_LOONGSON3A_R1:
 	default:
 		play_dead_at_ckseg1 =
-			(void *)CKSEG1ADDR((unsigned long)loongson3a_play_dead);
+			(void *)CKSEG1ADDR((unsigned long)loongson3a_r1_play_dead);
+		break;
+	case PRID_REV_LOONGSON3A_R2:
+		play_dead_at_ckseg1 =
+			(void *)CKSEG1ADDR((unsigned long)loongson3a_r2_play_dead);
 		break;
-	case Loongson_3B:
+	case PRID_REV_LOONGSON3B_R1:
+	case PRID_REV_LOONGSON3B_R2:
 		play_dead_at_ckseg1 =
 			(void *)CKSEG1ADDR((unsigned long)loongson3b_play_dead);
 		break;
@@ -593,9 +681,9 @@ void loongson3_disable_clock(int cpu)
 	uint64_t core_id = cpu_data[cpu].core;
 	uint64_t package_id = cpu_data[cpu].package;
 
-	if (loongson_sysconf.cputype == Loongson_3A) {
+	if ((read_c0_prid() & PRID_REV_MASK) == PRID_REV_LOONGSON3A_R1) {
 		LOONGSON_CHIPCFG(package_id) &= ~(1 << (12 + core_id));
-	} else if (loongson_sysconf.cputype == Loongson_3B) {
+	} else {
 		if (!(loongson_sysconf.workarounds & WORKAROUND_CPUHOTPLUG))
 			LOONGSON_FREQCTRL(package_id) &= ~(1 << (core_id * 4 + 3));
 	}
@@ -606,9 +694,9 @@ void loongson3_enable_clock(int cpu)
 	uint64_t core_id = cpu_data[cpu].core;
 	uint64_t package_id = cpu_data[cpu].package;
 
-	if (loongson_sysconf.cputype == Loongson_3A) {
+	if ((read_c0_prid() & PRID_REV_MASK) == PRID_REV_LOONGSON3A_R1) {
 		LOONGSON_CHIPCFG(package_id) |= 1 << (12 + core_id);
-	} else if (loongson_sysconf.cputype == Loongson_3B) {
+	} else {
 		if (!(loongson_sysconf.workarounds & WORKAROUND_CPUHOTPLUG))
 			LOONGSON_FREQCTRL(package_id) |= 1 << (core_id * 4 + 3);
 	}
diff --git a/arch/mips/mm/c-r4k.c b/arch/mips/mm/c-r4k.c
index 04a38d8..65fb28c 100644
--- a/arch/mips/mm/c-r4k.c
+++ b/arch/mips/mm/c-r4k.c
@@ -77,6 +77,7 @@ static inline void r4k_on_each_cpu(void (*func) (void *info), void *info)
  */
 static unsigned long icache_size __read_mostly;
 static unsigned long dcache_size __read_mostly;
+static unsigned long vcache_size __read_mostly;
 static unsigned long scache_size __read_mostly;
 
 /*
@@ -1349,6 +1350,31 @@ static void probe_pcache(void)
 	       c->dcache.linesz);
 }
 
+static void probe_vcache(void)
+{
+	struct cpuinfo_mips *c = &current_cpu_data;
+	unsigned int config2, lsize;
+
+	if (current_cpu_type() != CPU_LOONGSON3)
+		return;
+
+	config2 = read_c0_config2();
+	if ((lsize = ((config2 >> 20) & 15)))
+		c->vcache.linesz = 2 << lsize;
+	else
+		c->vcache.linesz = lsize;
+
+	c->vcache.sets = 64 << ((config2 >> 24) & 15);
+	c->vcache.ways = 1 + ((config2 >> 16) & 15);
+
+	vcache_size = c->vcache.sets * c->vcache.ways * c->vcache.linesz;
+
+	c->vcache.waybit = 0;
+
+	pr_info("Unified victim cache %ldkB %s, linesize %d bytes.\n",
+		vcache_size >> 10, way_string[c->vcache.ways], c->vcache.linesz);
+}
+
 /*
  * If you even _breathe_ on this function, look at the gcc output and make sure
  * it does not pop things on and off the stack for the cache sizing loop that
@@ -1671,6 +1697,7 @@ void r4k_cache_init(void)
 	struct cpuinfo_mips *c = &current_cpu_data;
 
 	probe_pcache();
+	probe_vcache();
 	setup_scache();
 
 	r4k_blast_dcache_page_setup();
diff --git a/arch/mips/mm/tlbex.c b/arch/mips/mm/tlbex.c
index 1ab9557..1e95151 100644
--- a/arch/mips/mm/tlbex.c
+++ b/arch/mips/mm/tlbex.c
@@ -247,7 +247,7 @@ static void output_pgtable_bits_defines(void)
 #ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
 	pr_define("_PAGE_HUGE_SHIFT %d\n", _PAGE_HUGE_SHIFT);
 #endif
-#if defined(CONFIG_CPU_MIPSR2) || defined(CONFIG_CPU_MIPSR6)
+#if defined(CONFIG_CPU_MIPSR2) || defined(CONFIG_CPU_MIPSR6) || defined(CONFIG_CPU_LOONGSON3)
 	if (cpu_has_rixi) {
 #ifdef _PAGE_NO_EXEC_SHIFT
 		pr_define("_PAGE_NO_EXEC_SHIFT %d\n", _PAGE_NO_EXEC_SHIFT);
diff --git a/drivers/platform/mips/cpu_hwmon.c b/drivers/platform/mips/cpu_hwmon.c
index 0f6c63e..8c5072b 100644
--- a/drivers/platform/mips/cpu_hwmon.c
+++ b/drivers/platform/mips/cpu_hwmon.c
@@ -20,9 +20,9 @@ int loongson3_cpu_temp(int cpu)
 	u32 reg;
 
 	reg = LOONGSON_CHIPTEMP(cpu);
-	if (loongson_sysconf.cputype == Loongson_3A)
+	if ((read_c0_prid() & PRID_REV_MASK) == PRID_REV_LOONGSON3A_R1)
 		reg = (reg >> 8) & 0xff;
-	else if (loongson_sysconf.cputype == Loongson_3B)
+	else
 		reg = ((reg >> 8) & 0xff) - 100;
 
 	return (int)reg * 1000;
-- 
2.7.0

[PATCH V3 2/5] MIPS: Loongson: Invalidate special TLBs when needed

[Huacai Chen]

Loongson-2 has a 4 entry itlb which is a subset of jtlb, Loongson-3 has
a 4 entry itlb and a 4 entry dtlb which are subsets of jtlb. We should
write diag register to invalidate itlb/dtlb when flushing jtlb because
itlb/dtlb are not totally transparent to software.

For Loongson-3A R2 (and newer), we should invalidate ITLB, DTLB, VTLB
and FTLB before we enable/disable FTLB.

Signed-off-by: Huacai Chen <chenhc@lemote.com>
---
 arch/mips/kernel/cpu-probe.c |  2 ++
 arch/mips/mm/tlb-r4k.c       | 27 +++++++++++++++------------
 2 files changed, 17 insertions(+), 12 deletions(-)

diff --git a/arch/mips/kernel/cpu-probe.c b/arch/mips/kernel/cpu-probe.c
index 2a2ae86..ef605e2 100644
--- a/arch/mips/kernel/cpu-probe.c
+++ b/arch/mips/kernel/cpu-probe.c
@@ -562,6 +562,8 @@ static int set_ftlb_enable(struct cpuinfo_mips *c, int enable)
 					   << MIPS_CONF7_FTLBP_SHIFT));
 		break;
 	case CPU_LOONGSON3:
+		/* Flush ITLB, DTLB, VTLB and FTLB */
+		write_c0_diag(1<<2 | 1<<3 | 1<<12 | 1<<13);
 		/* Loongson-3 cores use Config6 to enable the FTLB */
 		config = read_c0_config6();
 		if (enable)
diff --git a/arch/mips/mm/tlb-r4k.c b/arch/mips/mm/tlb-r4k.c
index c17d762..7593529 100644
--- a/arch/mips/mm/tlb-r4k.c
+++ b/arch/mips/mm/tlb-r4k.c
@@ -28,25 +28,28 @@
 extern void build_tlb_refill_handler(void);
 
 /*
- * LOONGSON2/3 has a 4 entry itlb which is a subset of dtlb,
- * unfortunately, itlb is not totally transparent to software.
+ * LOONGSON-2 has a 4 entry itlb which is a subset of jtlb, LOONGSON-3 has
+ * a 4 entry itlb and a 4 entry dtlb which are subsets of jtlb. Unfortunately,
+ * itlb/dtlb are not totally transparent to software.
  */
-static inline void flush_itlb(void)
+static inline void flush_spec_tlb(void)
 {
 	switch (current_cpu_type()) {
 	case CPU_LOONGSON2:
+		write_c0_diag(0x4);
+		break;
 	case CPU_LOONGSON3:
-		write_c0_diag(4);
+		write_c0_diag(0xc);
 		break;
 	default:
 		break;
 	}
 }
 
-static inline void flush_itlb_vm(struct vm_area_struct *vma)
+static inline void flush_spec_tlb_vm(struct vm_area_struct *vma)
 {
 	if (vma->vm_flags & VM_EXEC)
-		flush_itlb();
+		flush_spec_tlb();
 }
 
 void local_flush_tlb_all(void)
@@ -93,7 +96,7 @@ void local_flush_tlb_all(void)
 	tlbw_use_hazard();
 	write_c0_entryhi(old_ctx);
 	htw_start();
-	flush_itlb();
+	flush_spec_tlb();
 	local_irq_restore(flags);
 }
 EXPORT_SYMBOL(local_flush_tlb_all);
@@ -159,7 +162,7 @@ void local_flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
 		} else {
 			drop_mmu_context(mm, cpu);
 		}
-		flush_itlb();
+		flush_spec_tlb();
 		local_irq_restore(flags);
 	}
 }
@@ -205,7 +208,7 @@ void local_flush_tlb_kernel_range(unsigned long start, unsigned long end)
 	} else {
 		local_flush_tlb_all();
 	}
-	flush_itlb();
+	flush_spec_tlb();
 	local_irq_restore(flags);
 }
 
@@ -240,7 +243,7 @@ void local_flush_tlb_page(struct vm_area_struct *vma, unsigned long page)
 	finish:
 		write_c0_entryhi(oldpid);
 		htw_start();
-		flush_itlb_vm(vma);
+		flush_spec_tlb_vm(vma);
 		local_irq_restore(flags);
 	}
 }
@@ -274,7 +277,7 @@ void local_flush_tlb_one(unsigned long page)
 	}
 	write_c0_entryhi(oldpid);
 	htw_start();
-	flush_itlb();
+	flush_spec_tlb();
 	local_irq_restore(flags);
 }
 
@@ -357,7 +360,7 @@ void __update_tlb(struct vm_area_struct * vma, unsigned long address, pte_t pte)
 	}
 	tlbw_use_hazard();
 	htw_start();
-	flush_itlb_vm(vma);
+	flush_spec_tlb_vm(vma);
 	local_irq_restore(flags);
 }
 
-- 
2.7.0

[PATCH V3 3/5] MIPS: Loongson-3: Fast TLB refill handler

[Huacai Chen]

Loongson-3A R2 has pwbase/pwfield/pwsize/pwctl registers in CP0 (this
is very similar to HTW) and lwdir/lwpte/lddir/ldpte instructions which
can be used for fast TLB refill.

Signed-off-by: Huacai Chen <chenhc@lemote.com>
---
 arch/mips/include/asm/cpu-features.h |   3 +
 arch/mips/include/asm/cpu.h          |   1 +
 arch/mips/include/asm/mipsregs.h     |   6 ++
 arch/mips/include/asm/uasm.h         |   3 +-
 arch/mips/include/uapi/asm/inst.h    |  10 +++
 arch/mips/kernel/cpu-probe.c         |   2 +-
 arch/mips/mm/tlbex.c                 | 124 ++++++++++++++++++++++++++++++++++-
 arch/mips/mm/uasm-mips.c             |   2 +
 arch/mips/mm/uasm.c                  |   3 +
 9 files changed, 149 insertions(+), 5 deletions(-)

diff --git a/arch/mips/include/asm/cpu-features.h b/arch/mips/include/asm/cpu-features.h
index 0b7f8a5..1ec3dea 100644
--- a/arch/mips/include/asm/cpu-features.h
+++ b/arch/mips/include/asm/cpu-features.h
@@ -35,6 +35,9 @@
 #ifndef cpu_has_htw
 #define cpu_has_htw		(cpu_data[0].options & MIPS_CPU_HTW)
 #endif
+#ifndef cpu_has_ldpte
+#define cpu_has_ldpte		(cpu_data[0].options & MIPS_CPU_LDPTE)
+#endif
 #ifndef cpu_has_rixiex
 #define cpu_has_rixiex		(cpu_data[0].options & MIPS_CPU_RIXIEX)
 #endif
diff --git a/arch/mips/include/asm/cpu.h b/arch/mips/include/asm/cpu.h
index 858d036..0d9bf6c 100644
--- a/arch/mips/include/asm/cpu.h
+++ b/arch/mips/include/asm/cpu.h
@@ -393,6 +393,7 @@ enum cpu_type_enum {
 #define MIPS_CPU_NAN_LEGACY	0x40000000000ull /* Legacy NaN implemented */
 #define MIPS_CPU_NAN_2008	0x80000000000ull /* 2008 NaN implemented */
 #define MIPS_CPU_CACHE_COHERENT	0x100000000000ull /* CPU maintains cache coherency by hardware */
+#define MIPS_CPU_LDPTE		0x200000000000ull /* CPU has ldpte/lddir instructions */
 
 /*
  * CPU ASE encodings
diff --git a/arch/mips/include/asm/mipsregs.h b/arch/mips/include/asm/mipsregs.h
index 9290fd4..8affca2 100644
--- a/arch/mips/include/asm/mipsregs.h
+++ b/arch/mips/include/asm/mipsregs.h
@@ -1444,6 +1444,12 @@ do {									\
 #define read_c0_pwctl()		__read_32bit_c0_register($6, 6)
 #define write_c0_pwctl(val)	__write_32bit_c0_register($6, 6, val)
 
+#define read_c0_pgd()		__read_64bit_c0_register($9, 7)
+#define write_c0_pgd(val)	__write_64bit_c0_register($9, 7, val)
+
+#define read_c0_kpgd()		__read_64bit_c0_register($31, 7)
+#define write_c0_kpgd(val)	__write_64bit_c0_register($31, 7, val)
+
 /* Cavium OCTEON (cnMIPS) */
 #define read_c0_cvmcount()	__read_ulong_c0_register($9, 6)
 #define write_c0_cvmcount(val)	__write_ulong_c0_register($9, 6, val)
diff --git a/arch/mips/include/asm/uasm.h b/arch/mips/include/asm/uasm.h
index fc1cdd2..b6ecfee 100644
--- a/arch/mips/include/asm/uasm.h
+++ b/arch/mips/include/asm/uasm.h
@@ -171,7 +171,8 @@ Ip_u2u1(_wsbh);
 Ip_u3u1u2(_xor);
 Ip_u2u1u3(_xori);
 Ip_u2u1(_yield);
-
+Ip_u1u2(_ldpte);
+Ip_u2u1u3(_lddir);
 
 /* Handle labels. */
 struct uasm_label {
diff --git a/arch/mips/include/uapi/asm/inst.h b/arch/mips/include/uapi/asm/inst.h
index ddea53e..3bb8cd9 100644
--- a/arch/mips/include/uapi/asm/inst.h
+++ b/arch/mips/include/uapi/asm/inst.h
@@ -204,6 +204,16 @@ enum mad_func {
 };
 
 /*
+ * func field for page table walker (Loongson-3).
+ */
+enum ptw_func {
+	lwdir_op = 0x00,
+	lwpte_op = 0x01,
+	lddir_op = 0x02,
+	ldpte_op = 0x03,
+};
+
+/*
  * func field for special3 lx opcodes (Cavium Octeon).
  */
 enum lx_func {
diff --git a/arch/mips/kernel/cpu-probe.c b/arch/mips/kernel/cpu-probe.c
index ef605e2..aa2e615 100644
--- a/arch/mips/kernel/cpu-probe.c
+++ b/arch/mips/kernel/cpu-probe.c
@@ -1521,7 +1521,7 @@ static inline void cpu_probe_loongson(struct cpuinfo_mips *c, unsigned int cpu)
 		}
 
 		decode_configs(c);
-		c->options |= MIPS_CPU_TLBINV;
+		c->options |= MIPS_CPU_TLBINV | MIPS_CPU_LDPTE;
 		c->writecombine = _CACHE_UNCACHED_ACCELERATED;
 		break;
 	default:
diff --git a/arch/mips/mm/tlbex.c b/arch/mips/mm/tlbex.c
index 1e95151..da68ffb 100644
--- a/arch/mips/mm/tlbex.c
+++ b/arch/mips/mm/tlbex.c
@@ -290,7 +290,12 @@ static inline void dump_handler(const char *symbol, const u32 *handler, int coun
 #define C0_ENTRYLO1	3, 0
 #define C0_CONTEXT	4, 0
 #define C0_PAGEMASK	5, 0
+#define C0_PWBASE	5, 5
+#define C0_PWFIELD	5, 6
+#define C0_PWSIZE	5, 7
+#define C0_PWCTL	6, 6
 #define C0_BADVADDR	8, 0
+#define C0_PGD		9, 7
 #define C0_ENTRYHI	10, 0
 #define C0_EPC		14, 0
 #define C0_XCONTEXT	20, 0
@@ -830,7 +835,10 @@ build_get_pmde64(u32 **p, struct uasm_label **l, struct uasm_reloc **r,
 
 	if (pgd_reg != -1) {
 		/* pgd is in pgd_reg */
-		UASM_i_MFC0(p, ptr, c0_kscratch(), pgd_reg);
+		if (cpu_has_ldpte)
+			UASM_i_MFC0(p, ptr, C0_PWBASE);
+		else
+			UASM_i_MFC0(p, ptr, c0_kscratch(), pgd_reg);
 	} else {
 #if defined(CONFIG_MIPS_PGD_C0_CONTEXT)
 		/*
@@ -1443,6 +1451,108 @@ static void build_r4000_tlb_refill_handler(void)
 	dump_handler("r4000_tlb_refill", (u32 *)ebase, 64);
 }
 
+static void setup_pw(void)
+{
+	unsigned long pgd_i, pgd_w;
+#ifndef __PAGETABLE_PMD_FOLDED
+	unsigned long pmd_i, pmd_w;
+#endif
+	unsigned long pt_i, pt_w;
+	unsigned long pte_i, pte_w;
+#ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
+	unsigned long psn;
+
+	psn = ilog2(_PAGE_HUGE);     /* bit used to indicate huge page */
+#endif
+	pgd_i = PGDIR_SHIFT;  /* 1st level PGD */
+#ifndef __PAGETABLE_PMD_FOLDED
+	pgd_w = PGDIR_SHIFT - PMD_SHIFT + PGD_ORDER;
+
+	pmd_i = PMD_SHIFT;    /* 2nd level PMD */
+	pmd_w = PMD_SHIFT - PAGE_SHIFT;
+#else
+	pgd_w = PGDIR_SHIFT - PAGE_SHIFT + PGD_ORDER;
+#endif
+
+	pt_i  = PAGE_SHIFT;    /* 3rd level PTE */
+	pt_w  = PAGE_SHIFT - 3;
+
+	pte_i = ilog2(_PAGE_GLOBAL);
+	pte_w = 0;
+
+#ifndef __PAGETABLE_PMD_FOLDED
+	write_c0_pwfield(pgd_i << 24 | pmd_i << 12 | pt_i << 6 | pte_i);
+	write_c0_pwsize(1 << 30 | pgd_w << 24 | pmd_w << 12 | pt_w << 6 | pte_w);
+#else
+	write_c0_pwfield(pgd_i << 24 | pt_i << 6 | pte_i);
+	write_c0_pwsize(1 << 30 | pgd_w << 24 | pt_w << 6 | pte_w);
+#endif
+
+#ifdef CONFIG_MIPS_HUGE_TLB_SUPPORT
+	write_c0_pwctl(1 << 6 | psn);
+#endif
+	write_c0_kpgd(swapper_pg_dir);
+	kscratch_used_mask |= (1 << 7); /* KScratch6 is used for KPGD */
+}
+
+static void build_loongson3_tlb_refill_handler(void)
+{
+	u32 *p = tlb_handler;
+	struct uasm_label *l = labels;
+	struct uasm_reloc *r = relocs;
+
+	memset(labels, 0, sizeof(labels));
+	memset(relocs, 0, sizeof(relocs));
+	memset(tlb_handler, 0, sizeof(tlb_handler));
+
+	if (check_for_high_segbits) {
+		uasm_i_dmfc0(&p, K0, C0_BADVADDR);
+		uasm_i_dsrl_safe(&p, K1, K0, PGDIR_SHIFT + PGD_ORDER + PAGE_SHIFT - 3);
+		uasm_il_beqz(&p, &r, K1, label_vmalloc);
+		uasm_i_nop(&p);
+
+		uasm_il_bgez(&p, &r, K0, label_large_segbits_fault);
+		uasm_i_nop(&p);
+		uasm_l_vmalloc(&l, p);
+	}
+
+	uasm_i_dmfc0(&p, K1, C0_PGD);
+
+	uasm_i_lddir(&p, K0, K1, 3);  /* global page dir */
+#ifndef __PAGETABLE_PMD_FOLDED
+	uasm_i_lddir(&p, K1, K0, 1);  /* middle page dir */
+#endif
+	uasm_i_ldpte(&p, K1, 0);      /* even */
+	uasm_i_ldpte(&p, K1, 1);      /* odd */
+	uasm_i_tlbwr(&p);
+
+	/* restore page mask */
+	if (PM_DEFAULT_MASK >> 16) {
+		uasm_i_lui(&p, K0, PM_DEFAULT_MASK >> 16);
+		uasm_i_ori(&p, K0, K0, PM_DEFAULT_MASK & 0xffff);
+		uasm_i_mtc0(&p, K0, C0_PAGEMASK);
+	} else if (PM_DEFAULT_MASK) {
+		uasm_i_ori(&p, K0, 0, PM_DEFAULT_MASK);
+		uasm_i_mtc0(&p, K0, C0_PAGEMASK);
+	} else {
+		uasm_i_mtc0(&p, 0, C0_PAGEMASK);
+	}
+
+	uasm_i_eret(&p);
+
+	if (check_for_high_segbits) {
+		uasm_l_large_segbits_fault(&l, p);
+		UASM_i_LA(&p, K1, (unsigned long)tlb_do_page_fault_0);
+		uasm_i_jr(&p, K1);
+		uasm_i_nop(&p);
+	}
+
+	uasm_resolve_relocs(relocs, labels);
+	memcpy((void *)(ebase + 0x80), tlb_handler, 0x80);
+	local_flush_icache_range(ebase + 0x80, ebase + 0x100);
+	dump_handler("loongson3_tlb_refill", (u32 *)(ebase + 0x80), 32);
+}
+
 extern u32 handle_tlbl[], handle_tlbl_end[];
 extern u32 handle_tlbs[], handle_tlbs_end[];
 extern u32 handle_tlbm[], handle_tlbm_end[];
@@ -1490,7 +1600,10 @@ static void build_setup_pgd(void)
 	} else {
 		/* PGD in c0_KScratch */
 		uasm_i_jr(&p, 31);
-		UASM_i_MTC0(&p, a0, c0_kscratch(), pgd_reg);
+		if (cpu_has_ldpte)
+			UASM_i_MTC0(&p, a0, C0_PWBASE);
+		else
+			UASM_i_MTC0(&p, a0, c0_kscratch(), pgd_reg);
 	}
 #else
 #ifdef CONFIG_SMP
@@ -2466,13 +2579,18 @@ void build_tlb_refill_handler(void)
 		break;
 
 	default:
+		if (cpu_has_ldpte)
+			setup_pw();
+
 		if (!run_once) {
 			scratch_reg = allocate_kscratch();
 			build_setup_pgd();
 			build_r4000_tlb_load_handler();
 			build_r4000_tlb_store_handler();
 			build_r4000_tlb_modify_handler();
-			if (!cpu_has_local_ebase)
+			if (cpu_has_ldpte)
+				build_loongson3_tlb_refill_handler();
+			else if (!cpu_has_local_ebase)
 				build_r4000_tlb_refill_handler();
 			flush_tlb_handlers();
 			run_once++;
diff --git a/arch/mips/mm/uasm-mips.c b/arch/mips/mm/uasm-mips.c
index b4a83789..9c2220a 100644
--- a/arch/mips/mm/uasm-mips.c
+++ b/arch/mips/mm/uasm-mips.c
@@ -153,6 +153,8 @@ static struct insn insn_table[] = {
 	{ insn_xori,  M(xori_op, 0, 0, 0, 0, 0),  RS | RT | UIMM },
 	{ insn_xor,  M(spec_op, 0, 0, 0, 0, xor_op),  RS | RT | RD },
 	{ insn_yield, M(spec3_op, 0, 0, 0, 0, yield_op), RS | RD },
+	{ insn_ldpte, M(lwc2_op, 0, 0, 0, ldpte_op, mult_op), RS | RD },
+	{ insn_lddir, M(lwc2_op, 0, 0, 0, lddir_op, mult_op), RS | RT | RD },
 	{ insn_invalid, 0, 0 }
 };
 
diff --git a/arch/mips/mm/uasm.c b/arch/mips/mm/uasm.c
index 319051c..ad718de 100644
--- a/arch/mips/mm/uasm.c
+++ b/arch/mips/mm/uasm.c
@@ -60,6 +60,7 @@ enum opcode {
 	insn_sltiu, insn_sltu, insn_sra, insn_srl, insn_srlv, insn_subu,
 	insn_sw, insn_sync, insn_syscall, insn_tlbp, insn_tlbr, insn_tlbwi,
 	insn_tlbwr, insn_wait, insn_wsbh, insn_xor, insn_xori, insn_yield,
+	insn_lddir, insn_ldpte,
 };
 
 struct insn {
@@ -335,6 +336,8 @@ I_u1u2s3(_bbit0);
 I_u1u2s3(_bbit1);
 I_u3u1u2(_lwx)
 I_u3u1u2(_ldx)
+I_u1u2(_ldpte)
+I_u2u1u3(_lddir)
 
 #ifdef CONFIG_CPU_CAVIUM_OCTEON
 #include <asm/octeon/octeon.h>
-- 
2.7.0

[PATCH V3 4/5] MIPS: Loongson-3: Use cpu_has_coherent_cache feature

[Huacai Chen]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain, Size: 989 bytes --]

Loongson-3 maintains cache coherency by hardware, this means:
 1) Loongson-3's icache is coherent with dcache.
 2) Loongson-3's dcaches don't alias (if PAGE_SIZE>=16K).
 3) Loongson-3 maintains cache coherency across cores (and for DMA).

Signed-off-by: Huacai Chen <chenhc@lemote.com>
---
 arch/mips/include/asm/mach-loongson64/cpu-feature-overrides.h | 1 +
 1 file changed, 1 insertion(+)

diff --git a/arch/mips/include/asm/mach-loongson64/cpu-feature-overrides.h b/arch/mips/include/asm/mach-loongson64/cpu-feature-overrides.h
index c3406db..647d952 100644
--- a/arch/mips/include/asm/mach-loongson64/cpu-feature-overrides.h
+++ b/arch/mips/include/asm/mach-loongson64/cpu-feature-overrides.h
@@ -46,6 +46,7 @@
 #define cpu_has_local_ebase	0
 
 #define cpu_has_wsbh		IS_ENABLED(CONFIG_CPU_LOONGSON3)
+#define cpu_has_coherent_cache	IS_ENABLED(CONFIG_CPU_LOONGSON3)
 #define cpu_hwrena_impl_bits	0xc0000000
 
 #endif /* __ASM_MACH_LOONGSON64_CPU_FEATURE_OVERRIDES_H */
-- 
2.7.0


3h™\x03

[PATCH V3 5/5] MIPS: Loongson-3: Introduce CONFIG_LOONGSON3_ENHANCEMENT

[Huacai Chen]

New Loongson 3 CPU (since Loongson-3A R2, as opposed to Loongson-3A R1,
Loongson-3B R1 and Loongson-3B R2) has many enhancements, such as FTLB,
L1-VCache, EI/DI/Wait/Prefetch instruction, DSP/DSPv2 ASE, User Local
register, Read-Inhibit/Execute-Inhibit, SFB (Store Fill Buffer), Fast
TLB refill support, etc.

This patch introduce a config option, CONFIG_LOONGSON3_ENHANCEMENT, to
enable those enhancements which cannot be probed at run time. If you
want a generic kernel to run on all Loongson 3 machines, please say 'N'
here. If you want a high-performance kernel to run on new Loongson 3
machines only, please say 'Y' here.

Some additional explanations:
1) SFB locates between core and L1 cache, it causes memory access out
   of order, so writel/outl (and other similar functions) need a I/O
   reorder barrier.
2) Loongson 3 has a bug that di instruction can not save the irqflag,
   so arch_local_irq_save() is modified. Since CPU_MIPSR2 is selected
   by CONFIG_LOONGSON3_ENHANCEMENT, generic kernel doesn't use ei/di
   at all.
3) CPU_HAS_PREFETCH is selected by CONFIG_LOONGSON3_ENHANCEMENT, so
   MIPS_CPU_PREFETCH (used by uasm) probing is also put in this patch.

Signed-off-by: Huacai Chen <chenhc@lemote.com>
---
 arch/mips/Kconfig                                      | 18 ++++++++++++++++++
 arch/mips/include/asm/hazards.h                        |  7 ++++---
 arch/mips/include/asm/io.h                             | 10 +++++-----
 arch/mips/include/asm/irqflags.h                       |  5 +++++
 .../include/asm/mach-loongson64/kernel-entry-init.h    | 12 ++++++++++++
 arch/mips/mm/c-r4k.c                                   |  2 ++
 arch/mips/mm/page.c                                    |  9 +++++++++
 7 files changed, 55 insertions(+), 8 deletions(-)

diff --git a/arch/mips/Kconfig b/arch/mips/Kconfig
index 38c0b11..68ec57a 100644
--- a/arch/mips/Kconfig
+++ b/arch/mips/Kconfig
@@ -1350,6 +1350,24 @@ config CPU_LOONGSON3
 		The Loongson 3 processor implements the MIPS64R2 instruction
 		set with many extensions.
 
+config LOONGSON3_ENHANCEMENT
+	bool "New Loongson 3 CPU Enhancements"
+	default n
+	select CPU_MIPSR2
+	select CPU_HAS_PREFETCH
+	depends on CPU_LOONGSON3
+	help
+	  New Loongson 3 CPU (since Loongson-3A R2, as opposed to Loongson-3A
+	  R1, Loongson-3B R1 and Loongson-3B R2) has many enhancements, such as
+	  FTLB, L1-VCache, EI/DI/Wait/Prefetch instruction, DSP/DSPv2 ASE, User
+	  Local register, Read-Inhibit/Execute-Inhibit, SFB (Store Fill Buffer),
+	  Fast TLB refill support, etc.
+
+	  This option enable those enhancements which cannot be probed at run
+	  time. If you want a generic kernel to run on all Loongson 3 machines,
+	  please say 'N' here. If you want a high-performance kernel to run on
+	  new Loongson 3 machines only, please say 'Y' here.
+
 config CPU_LOONGSON2E
 	bool "Loongson 2E"
 	depends on SYS_HAS_CPU_LOONGSON2E
diff --git a/arch/mips/include/asm/hazards.h b/arch/mips/include/asm/hazards.h
index 7b99efd..dbb1eb6 100644
--- a/arch/mips/include/asm/hazards.h
+++ b/arch/mips/include/asm/hazards.h
@@ -22,7 +22,8 @@
 /*
  * TLB hazards
  */
-#if defined(CONFIG_CPU_MIPSR2) || defined(CONFIG_CPU_MIPSR6) && !defined(CONFIG_CPU_CAVIUM_OCTEON)
+#if (defined(CONFIG_CPU_MIPSR2) || defined(CONFIG_CPU_MIPSR6)) && \
+	!defined(CONFIG_CPU_CAVIUM_OCTEON) && !defined(CONFIG_LOONGSON3_ENHANCEMENT)
 
 /*
  * MIPSR2 defines ehb for hazard avoidance
@@ -155,8 +156,8 @@ do {									\
 } while (0)
 
 #elif defined(CONFIG_MIPS_ALCHEMY) || defined(CONFIG_CPU_CAVIUM_OCTEON) || \
-	defined(CONFIG_CPU_LOONGSON2) || defined(CONFIG_CPU_R10000) || \
-	defined(CONFIG_CPU_R5500) || defined(CONFIG_CPU_XLR)
+	defined(CONFIG_CPU_LOONGSON2) || defined(CONFIG_LOONGSON3_ENHANCEMENT) || \
+	defined(CONFIG_CPU_R10000) || defined(CONFIG_CPU_R5500) || defined(CONFIG_CPU_XLR)
 
 /*
  * R10000 rocks - all hazards handled in hardware, so this becomes a nobrainer.
diff --git a/arch/mips/include/asm/io.h b/arch/mips/include/asm/io.h
index 2b4dc7a..ecabc00 100644
--- a/arch/mips/include/asm/io.h
+++ b/arch/mips/include/asm/io.h
@@ -304,10 +304,10 @@ static inline void iounmap(const volatile void __iomem *addr)
 #undef __IS_KSEG1
 }
 
-#ifdef CONFIG_CPU_CAVIUM_OCTEON
-#define war_octeon_io_reorder_wmb()		wmb()
+#if defined(CONFIG_CPU_CAVIUM_OCTEON) || defined(CONFIG_LOONGSON3_ENHANCEMENT)
+#define war_io_reorder_wmb()		wmb()
 #else
-#define war_octeon_io_reorder_wmb()		do { } while (0)
+#define war_io_reorder_wmb()		do { } while (0)
 #endif
 
 #define __BUILD_MEMORY_SINGLE(pfx, bwlq, type, irq)			\
@@ -318,7 +318,7 @@ static inline void pfx##write##bwlq(type val,				\
 	volatile type *__mem;						\
 	type __val;							\
 									\
-	war_octeon_io_reorder_wmb();					\
+	war_io_reorder_wmb();					\
 									\
 	__mem = (void *)__swizzle_addr_##bwlq((unsigned long)(mem));	\
 									\
@@ -387,7 +387,7 @@ static inline void pfx##out##bwlq##p(type val, unsigned long port)	\
 	volatile type *__addr;						\
 	type __val;							\
 									\
-	war_octeon_io_reorder_wmb();					\
+	war_io_reorder_wmb();					\
 									\
 	__addr = (void *)__swizzle_addr_##bwlq(mips_io_port_base + port); \
 									\
diff --git a/arch/mips/include/asm/irqflags.h b/arch/mips/include/asm/irqflags.h
index 65c351e..9d3610b 100644
--- a/arch/mips/include/asm/irqflags.h
+++ b/arch/mips/include/asm/irqflags.h
@@ -41,7 +41,12 @@ static inline unsigned long arch_local_irq_save(void)
 	"	.set	push						\n"
 	"	.set	reorder						\n"
 	"	.set	noat						\n"
+#if defined(CONFIG_CPU_LOONGSON3)
+	"	mfc0	%[flags], $12					\n"
+	"	di							\n"
+#else
 	"	di	%[flags]					\n"
+#endif
 	"	andi	%[flags], 1					\n"
 	"	" __stringify(__irq_disable_hazard) "			\n"
 	"	.set	pop						\n"
diff --git a/arch/mips/include/asm/mach-loongson64/kernel-entry-init.h b/arch/mips/include/asm/mach-loongson64/kernel-entry-init.h
index da83482..8393bc54 100644
--- a/arch/mips/include/asm/mach-loongson64/kernel-entry-init.h
+++ b/arch/mips/include/asm/mach-loongson64/kernel-entry-init.h
@@ -26,6 +26,12 @@
 	mfc0	t0, $5, 1
 	or	t0, (0x1 << 29)
 	mtc0	t0, $5, 1
+#ifdef CONFIG_LOONGSON3_ENHANCEMENT
+	/* Enable STFill Buffer */
+	mfc0	t0, $16, 6
+	or	t0, 0x100
+	mtc0	t0, $16, 6
+#endif
 	_ehb
 	.set	pop
 #endif
@@ -46,6 +52,12 @@
 	mfc0	t0, $5, 1
 	or	t0, (0x1 << 29)
 	mtc0	t0, $5, 1
+#ifdef CONFIG_LOONGSON3_ENHANCEMENT
+	/* Enable STFill Buffer */
+	mfc0	t0, $16, 6
+	or	t0, 0x100
+	mtc0	t0, $16, 6
+#endif
 	_ehb
 	.set	pop
 #endif
diff --git a/arch/mips/mm/c-r4k.c b/arch/mips/mm/c-r4k.c
index 65fb28c..dc05dc3 100644
--- a/arch/mips/mm/c-r4k.c
+++ b/arch/mips/mm/c-r4k.c
@@ -1170,6 +1170,8 @@ static void probe_pcache(void)
 					  c->dcache.ways *
 					  c->dcache.linesz;
 		c->dcache.waybit = 0;
+		if ((prid & PRID_REV_MASK) >= PRID_REV_LOONGSON3A_R2)
+			c->options |= MIPS_CPU_PREFETCH;
 		break;
 
 	case CPU_CAVIUM_OCTEON3:
diff --git a/arch/mips/mm/page.c b/arch/mips/mm/page.c
index 885d73f..c41953c 100644
--- a/arch/mips/mm/page.c
+++ b/arch/mips/mm/page.c
@@ -188,6 +188,15 @@ static void set_prefetch_parameters(void)
 			}
 			break;
 
+		case CPU_LOONGSON3:
+			/* Loongson-3 only support the Pref_Load/Pref_Store. */
+			pref_bias_clear_store = 128;
+			pref_bias_copy_load = 128;
+			pref_bias_copy_store = 128;
+			pref_src_mode = Pref_Load;
+			pref_dst_mode = Pref_Store;
+			break;
+
 		default:
 			pref_bias_clear_store = 128;
 			pref_bias_copy_load = 256;
-- 
2.7.0

Re: [PATCH V3 5/5] MIPS: Loongson-3: Introduce CONFIG_LOONGSON3_ENHANCEMENT

[Maciej W. Rozycki]

On Wed, 24 Feb 2016, Huacai Chen wrote:

> This patch introduce a config option, CONFIG_LOONGSON3_ENHANCEMENT, to
> enable those enhancements which cannot be probed at run time. If you
> want a generic kernel to run on all Loongson 3 machines, please say 'N'
> here. If you want a high-performance kernel to run on new Loongson 3
> machines only, please say 'Y' here.
[...]
> diff --git a/arch/mips/include/asm/irqflags.h b/arch/mips/include/asm/irqflags.h
> index 65c351e..9d3610b 100644
> --- a/arch/mips/include/asm/irqflags.h
> +++ b/arch/mips/include/asm/irqflags.h
> @@ -41,7 +41,12 @@ static inline unsigned long arch_local_irq_save(void)
>  	"	.set	push						\n"
>  	"	.set	reorder						\n"
>  	"	.set	noat						\n"
> +#if defined(CONFIG_CPU_LOONGSON3)
> +	"	mfc0	%[flags], $12					\n"
> +	"	di							\n"
> +#else
>  	"	di	%[flags]					\n"
> +#endif
>  	"	andi	%[flags], 1					\n"
>  	"	" __stringify(__irq_disable_hazard) "			\n"
>  	"	.set	pop						\n"

 This part does not appear related to CONFIG_LOONGSON3_ENHANCEMENT -- 
please either fold it into one of the other changes in the set, if there's 
one it really belongs to, or make it an entirely separate change 
otherwise.

  Maciej

Re: [PATCH V3 5/5] MIPS: Loongson-3: Introduce CONFIG_LOONGSON3_ENHANCEMENT

[Maciej W. Rozycki]

On Wed, 24 Feb 2016, Huacai Chen wrote:

> This patch introduce a config option, CONFIG_LOONGSON3_ENHANCEMENT, to
> enable those enhancements which cannot be probed at run time. If you
> want a generic kernel to run on all Loongson 3 machines, please say 'N'
> here. If you want a high-performance kernel to run on new Loongson 3
> machines only, please say 'Y' here.
[...]
> diff --git a/arch/mips/include/asm/irqflags.h b/arch/mips/include/asm/irqflags.h
> index 65c351e..9d3610b 100644
> --- a/arch/mips/include/asm/irqflags.h
> +++ b/arch/mips/include/asm/irqflags.h
> @@ -41,7 +41,12 @@ static inline unsigned long arch_local_irq_save(void)
>  	"	.set	push						\n"
>  	"	.set	reorder						\n"
>  	"	.set	noat						\n"
> +#if defined(CONFIG_CPU_LOONGSON3)
> +	"	mfc0	%[flags], $12					\n"
> +	"	di							\n"
> +#else
>  	"	di	%[flags]					\n"
> +#endif
>  	"	andi	%[flags], 1					\n"
>  	"	" __stringify(__irq_disable_hazard) "			\n"
>  	"	.set	pop						\n"

 This part does not appear related to CONFIG_LOONGSON3_ENHANCEMENT -- 
please either fold it into one of the other changes in the set, if there's 
one it really belongs to, or make it an entirely separate change 
otherwise.

  Maciej

Re: [PATCH V3 5/5] MIPS: Loongson-3: Introduce CONFIG_LOONGSON3_ENHANCEMENT

[Huacai Chen]

 Hi, Maciej,

Please read my explaination: New Loongson 3 has a bug that di
instruction can not save the irqflag, so arch_local_irq_save() is
modified. Since CPU_MIPSR2 is selected by
CONFIG_LOONGSON3_ENHANCEMENT, generic kernel doesn't use ei/di at all
(because old Loongson 3 doesn't support ei/di at all).

Huacai

On Thu, Feb 25, 2016 at 8:49 AM, Maciej W. Rozycki <macro@imgtec.com> wrote:
> On Wed, 24 Feb 2016, Huacai Chen wrote:
>
>> This patch introduce a config option, CONFIG_LOONGSON3_ENHANCEMENT, to
>> enable those enhancements which cannot be probed at run time. If you
>> want a generic kernel to run on all Loongson 3 machines, please say 'N'
>> here. If you want a high-performance kernel to run on new Loongson 3
>> machines only, please say 'Y' here.
> [...]
>> diff --git a/arch/mips/include/asm/irqflags.h b/arch/mips/include/asm/irqflags.h
>> index 65c351e..9d3610b 100644
>> --- a/arch/mips/include/asm/irqflags.h
>> +++ b/arch/mips/include/asm/irqflags.h
>> @@ -41,7 +41,12 @@ static inline unsigned long arch_local_irq_save(void)
>>       "       .set    push                                            \n"
>>       "       .set    reorder                                         \n"
>>       "       .set    noat                                            \n"
>> +#if defined(CONFIG_CPU_LOONGSON3)
>> +     "       mfc0    %[flags], $12                                   \n"
>> +     "       di                                                      \n"
>> +#else
>>       "       di      %[flags]                                        \n"
>> +#endif
>>       "       andi    %[flags], 1                                     \n"
>>       "       " __stringify(__irq_disable_hazard) "                   \n"
>>       "       .set    pop                                             \n"
>
>  This part does not appear related to CONFIG_LOONGSON3_ENHANCEMENT --
> please either fold it into one of the other changes in the set, if there's
> one it really belongs to, or make it an entirely separate change
> otherwise.
>
>   Maciej
>

Re: [PATCH V3 5/5] MIPS: Loongson-3: Introduce CONFIG_LOONGSON3_ENHANCEMENT

[Maciej W. Rozycki]

On Thu, 25 Feb 2016, Huacai Chen wrote:

> Please read my explaination: New Loongson 3 has a bug that di
> instruction can not save the irqflag, so arch_local_irq_save() is
> modified. Since CPU_MIPSR2 is selected by
> CONFIG_LOONGSON3_ENHANCEMENT, generic kernel doesn't use ei/di at all
> (because old Loongson 3 doesn't support ei/di at all).

 Thanks.  This opens an interesting question though.  You've written that 
the enhancements controlled by CONFIG_LOONGSON3_ENHANCEMENT cannot be 
determined at the run time, but it looks to me like they can, either by 
checking for MIPSr2 in the CP0.Config.AR or, if you did something wrong 
about that, by trying to execute EI or DI early on and seeing if it's 
triggered an RI exception or not.

 Have I missed anything?

  Maciej

Re: [PATCH V3 5/5] MIPS: Loongson-3: Introduce CONFIG_LOONGSON3_ENHANCEMENT

[Huacai Chen]

Hi, Maciej,

Yes, we can probe whether the platform support ei/di or not. But
whether use ei/di is not depend on runtime flag, but depend on
CONFIG_* in current kernel. So I have no idea how to use ei/di by
probing.

Huacai

On Thu, Feb 25, 2016 at 8:25 PM, Maciej W. Rozycki <macro@imgtec.com> wrote:
> On Thu, 25 Feb 2016, Huacai Chen wrote:
>
>> Please read my explaination: New Loongson 3 has a bug that di
>> instruction can not save the irqflag, so arch_local_irq_save() is
>> modified. Since CPU_MIPSR2 is selected by
>> CONFIG_LOONGSON3_ENHANCEMENT, generic kernel doesn't use ei/di at all
>> (because old Loongson 3 doesn't support ei/di at all).
>
>  Thanks.  This opens an interesting question though.  You've written that
> the enhancements controlled by CONFIG_LOONGSON3_ENHANCEMENT cannot be
> determined at the run time, but it looks to me like they can, either by
> checking for MIPSr2 in the CP0.Config.AR or, if you did something wrong
> about that, by trying to execute EI or DI early on and seeing if it's
> triggered an RI exception or not.
>
>  Have I missed anything?
>
>   Maciej
>

Re: [PATCH V3 5/5] MIPS: Loongson-3: Introduce CONFIG_LOONGSON3_ENHANCEMENT

[Maciej W. Rozycki]

On Thu, 25 Feb 2016, Huacai Chen wrote:

> Yes, we can probe whether the platform support ei/di or not. But
> whether use ei/di is not depend on runtime flag, but depend on
> CONFIG_* in current kernel. So I have no idea how to use ei/di by
> probing.

 OK, I think I have a full picture now.

 There's no need to probe, but I think the commit description and config 
text option are a bit misleading.  This looks like the next architecture 
revision to me, much like MIPS32r1 vs MIPS32r2 for example, and it is 
quite normal that you don't probe for the new features it provides (even 
though -- sticking to the example -- we could probe for MIPS32r2 
additions), but you just require whoever configures the kernel to choose 
the intended setup.  IOW it's a design decision rather than a technical 
limitation.

 So may I suggest that you rewrite the relevant part of the help text to 
read:

"This option enables those enhancements which are not probed at run time."

or suchlike?  And then similarly in the commit description.

 Thanks,

  Maciej

Re: [PATCH V3 5/5] MIPS: Loongson-3: Introduce CONFIG_LOONGSON3_ENHANCEMENT

[Huacai Chen]

Yes, you are right, I'll update the commit message.

Huacai

On Tue, Mar 1, 2016 at 9:06 AM, Maciej W. Rozycki <macro@imgtec.com> wrote:
> On Thu, 25 Feb 2016, Huacai Chen wrote:
>
>> Yes, we can probe whether the platform support ei/di or not. But
>> whether use ei/di is not depend on runtime flag, but depend on
>> CONFIG_* in current kernel. So I have no idea how to use ei/di by
>> probing.
>
>  OK, I think I have a full picture now.
>
>  There's no need to probe, but I think the commit description and config
> text option are a bit misleading.  This looks like the next architecture
> revision to me, much like MIPS32r1 vs MIPS32r2 for example, and it is
> quite normal that you don't probe for the new features it provides (even
> though -- sticking to the example -- we could probe for MIPS32r2
> additions), but you just require whoever configures the kernel to choose
> the intended setup.  IOW it's a design decision rather than a technical
> limitation.
>
>  So may I suggest that you rewrite the relevant part of the help text to
> read:
>
> "This option enables those enhancements which are not probed at run time."
>
> or suchlike?  And then similarly in the commit description.
>
>  Thanks,
>
>   Maciej
>