diff mbox series

[v14,8/9] xen/riscv: introduce ANDN_INSN

Message ID 07c628e8552f6e31a07e4261b273553cb4a3669b.1719917348.git.oleksii.kurochko@gmail.com (mailing list archive)
State New
Headers show
Series Enable build of full Xen for RISC-V | expand

Commit Message

Oleksii July 2, 2024, 11:01 a.m. UTC
RISC-V does a conditional toolchain for the Zbb extension
(xen/arch/riscv/rules.mk), but unconditionally uses the
ANDN instruction in emulate_xchg_1_2().

Fixes: 51dabd6312c ("xen/riscv: introduce cmpxchg.h")

Suggested-by: Andrew Cooper <andrew.cooper3@citrix.com>
Signed-off-by: Oleksii Kurochko <oleksii.kurochko@gmail.com>
Suggested-By: Jan Beulich <jbeulich@suse.com>
---
Andrew,

Could you please take a look if you are okay with suggested
changes.

Thanks in advance.
---
 Changes in V14:
  - update the commit on top of ANDN definition.
  - use .insn instead of andn to support gas which doesn't understand
    andn instruction.
  - add Suggested-By: Jan Beulich <jbeulich@suse.com>
---
 Changes in V13:
  - new patch
---
 xen/arch/riscv/include/asm/cmpxchg.h | 15 ++++++++++++++-
 1 file changed, 14 insertions(+), 1 deletion(-)

Comments

Jan Beulich July 2, 2024, 12:08 p.m. UTC | #1
On 02.07.2024 13:01, Oleksii Kurochko wrote:
> RISC-V does a conditional toolchain for the Zbb extension
> (xen/arch/riscv/rules.mk), but unconditionally uses the
> ANDN instruction in emulate_xchg_1_2().
> 
> Fixes: 51dabd6312c ("xen/riscv: introduce cmpxchg.h")
> 
> Suggested-by: Andrew Cooper <andrew.cooper3@citrix.com>
> Signed-off-by: Oleksii Kurochko <oleksii.kurochko@gmail.com>
> Suggested-By: Jan Beulich <jbeulich@suse.com>

Hmm, meant to say that for the earlier patch already: A Suggested-by:
after an S-o-b is somewhat odd. You can't have signed off a change
before it was suggested, as you can't possibly have written it then.
Otoh I'm not sure tagging me is appropriate here anyway.

> --- a/xen/arch/riscv/include/asm/cmpxchg.h
> +++ b/xen/arch/riscv/include/asm/cmpxchg.h
> @@ -18,6 +18,19 @@
>          : "r" (new) \
>          : "memory" );
>  
> +/*
> + * To not face an issue that gas doesn't understand ANDN instruction
> + * it is encoded using .insn directive.
> + */
> +#ifdef __riscv_zbb
> +#define ANDN_INSN(rd, rs1, rs2)                 \
> +    ".insn r 0x33, 0x7, 0x20, " rd ", " rs1 ", " rs2 "\n"

Why 0x33? Just like in the other patch you used MISC_MEM (as suggested,
for being more descriptive), you want to use OP here. Then
Reviewed-by: Jan Beulich <jbeulich@suse.com>
(and I'd certainly be okay making the change while committing)

Jan
Oleksii July 2, 2024, 3:51 p.m. UTC | #2
On Tue, 2024-07-02 at 14:08 +0200, Jan Beulich wrote:
> > --- a/xen/arch/riscv/include/asm/cmpxchg.h
> > +++ b/xen/arch/riscv/include/asm/cmpxchg.h
> > @@ -18,6 +18,19 @@
> >           : "r" (new) \
> >           : "memory" );
> >   
> > +/*
> > + * To not face an issue that gas doesn't understand ANDN
> > instruction
> > + * it is encoded using .insn directive.
> > + */
> > +#ifdef __riscv_zbb
> > +#define ANDN_INSN(rd, rs1, rs2)                 \
> > +    ".insn r 0x33, 0x7, 0x20, " rd ", " rs1 ", " rs2 "\n"
> 
> Why 0x33? Just like in the other patch you used MISC_MEM (as
> suggested,
> for being more descriptive), you want to use OP here. Then
> Reviewed-by: Jan Beulich <jbeulich@suse.com>
> (and I'd certainly be okay making the change while committing)

According to Volume 1, Unprivileged Specification version 20240411 the
andn instruction looks
like:
31          25 24 20 19 15 14   12 11 7 6           0
0 1 0 0 0 0 0   rs2   rs1   1 1 1   rd  0 1 1 0 0 1 1
                                               OP

This instruction is R-type. According to the gcc manual
( https://sourceware.org/binutils/docs/as/RISC_002dV_002dFormats.html
):
R type: .insn r opcode6, func3, func7, rd, rs1, rs2
+-------+-----+-----+-------+----+---------+
| func7 | rs2 | rs1 | func3 | rd | opcode6 |
+-------+-----+-----+-------+----+---------+
31      25    20    15      12   7        0

so opcode6 in case of andn is 0x33 (0110011)

~ Oleksii
Jan Beulich July 2, 2024, 3:53 p.m. UTC | #3
On 02.07.2024 17:51, Oleksii wrote:
> On Tue, 2024-07-02 at 14:08 +0200, Jan Beulich wrote:
>>> --- a/xen/arch/riscv/include/asm/cmpxchg.h
>>> +++ b/xen/arch/riscv/include/asm/cmpxchg.h
>>> @@ -18,6 +18,19 @@
>>>           : "r" (new) \
>>>           : "memory" );
>>>   
>>> +/*
>>> + * To not face an issue that gas doesn't understand ANDN
>>> instruction
>>> + * it is encoded using .insn directive.
>>> + */
>>> +#ifdef __riscv_zbb
>>> +#define ANDN_INSN(rd, rs1, rs2)                 \
>>> +    ".insn r 0x33, 0x7, 0x20, " rd ", " rs1 ", " rs2 "\n"
>>
>> Why 0x33? Just like in the other patch you used MISC_MEM (as
>> suggested,
>> for being more descriptive), you want to use OP here. Then
>> Reviewed-by: Jan Beulich <jbeulich@suse.com>
>> (and I'd certainly be okay making the change while committing)
> 
> According to Volume 1, Unprivileged Specification version 20240411 the
> andn instruction looks
> like:
> 31          25 24 20 19 15 14   12 11 7 6           0
> 0 1 0 0 0 0 0   rs2   rs1   1 1 1   rd  0 1 1 0 0 1 1
>                                                OP
> 
> This instruction is R-type. According to the gcc manual
> ( https://sourceware.org/binutils/docs/as/RISC_002dV_002dFormats.html
> ):
> R type: .insn r opcode6, func3, func7, rd, rs1, rs2
> +-------+-----+-----+-------+----+---------+
> | func7 | rs2 | rs1 | func3 | rd | opcode6 |
> +-------+-----+-----+-------+----+---------+
> 31      25    20    15      12   7        0
> 
> so opcode6 in case of andn is 0x33 (0110011)

Right. And "OP" is the identifier you can use with .insn to get 0x33
without needing to write 0x33. As said - the requested replacement is
solely to help readability a little.

Jan
diff mbox series

Patch

diff --git a/xen/arch/riscv/include/asm/cmpxchg.h b/xen/arch/riscv/include/asm/cmpxchg.h
index d5e678c036..34d5bd48b3 100644
--- a/xen/arch/riscv/include/asm/cmpxchg.h
+++ b/xen/arch/riscv/include/asm/cmpxchg.h
@@ -18,6 +18,19 @@ 
         : "r" (new) \
         : "memory" );
 
+/*
+ * To not face an issue that gas doesn't understand ANDN instruction
+ * it is encoded using .insn directive.
+ */
+#ifdef __riscv_zbb
+#define ANDN_INSN(rd, rs1, rs2)                 \
+    ".insn r 0x33, 0x7, 0x20, " rd ", " rs1 ", " rs2 "\n"
+#else
+#define ANDN_INSN(rd, rs1, rs2)                 \
+    "not " rd ", " rs2 "\n"                     \
+    "and " rd ", " rs1 ", " rd "\n"
+#endif
+
 /*
  * For LR and SC, the A extension requires that the address held in rs1 be
  * naturally aligned to the size of the operand (i.e., eight-byte aligned
@@ -48,7 +61,7 @@ 
     \
     asm volatile ( \
         "0: lr.w" lr_sfx " %[old], %[ptr_]\n" \
-        "   andn  %[scratch], %[old], %[mask]\n" \
+        ANDN_INSN("%[scratch]", "%[old]", "%[mask]") \
         "   or   %[scratch], %[scratch], %z[new_]\n" \
         "   sc.w" sc_sfx " %[scratch], %[scratch], %[ptr_]\n" \
         "   bnez %[scratch], 0b\n" \