Received-SPF: pass (google.com: domain of linux-kernel+bounces-30599-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45e3:2400::1 as permitted sender) client-ip=2604:1380:45e3:2400::1;
Date: Thu, 18 Jan 2024 14:20:42 -0800 (PST)
Subject:     Re: [PATCH] riscv: lib: Support csum on GCC <11
In-Reply-To: <20240118-museum-ample-11d649f95de8@spud>
CC: charlie@rivosinc.com, Paul Walmsley <paul.walmsley@sifive.com>,
  aou@eecs.berkeley.edu, linux-riscv@lists.infradead.org, linux-kernel@vger.kernel.org
From: Palmer Dabbelt <palmer@dabbelt.com>
To: Conor Dooley <conor@kernel.org>
Message-ID: <mhng-6127df88-9768-4f9d-8d78-a1aeb4213451@palmer-ri-x1c9>
Precedence: bulk
Mime-Version: 1.0 (MHng)
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Transfer-Encoding: 8bit

On Thu, 18 Jan 2024 14:05:44 PST (-0800), Conor Dooley wrote:
> On Thu, Jan 18, 2024 at 01:53:59PM -0800, Charlie Jenkins wrote:
>> The OutputOperands field for asm goto statements is only supported
>> starting from GCC 11. Split the asm goto to remove the use of this
>> feature.
>> 
>> Signed-off-by: Charlie Jenkins <charlie@rivosinc.com>
>> Fixes: a04c192eabfb ("riscv: Add checksum library")
>> ---
>> The OutputOperands field for asm goto statements is only supported
>> starting from GCC 11. Split the asm goto to remove the use of this
>> feature.
>
> Maybe this is a super naive question, but is it possible to just not
> use the custom csum code for gcc older than 11?

Charlie and I were talking, these old GCC versions also don't support 
ZBB.  So I think we can get away with something like

diff --git a/arch/riscv/lib/csum.c b/arch/riscv/lib/csum.c
index 06ce8e7250d9..17f883b612c9 100644
--- a/arch/riscv/lib/csum.c
+++ b/arch/riscv/lib/csum.c
@@ -158,10 +158,16 @@ do_csum_with_alignment(const unsigned char *buff, int len)
 
 	/*
 	 * Zbb support saves 6 instructions, so not worth checking without
-	 * alternatives if supported
+	 * alternatives if supported.
+	 *
+	 * Note that we pull the check for ZBB into the preprocessor proper
+	 * here, as otherwise GCC will attempt to compile the code inside the
+	 * "if (IS_ENABLED(ZBB)" block which fails because GCC10 doesn't
+	 * support ASM goto output operands.  GCC 10 also doesn't support ZBB,
+	 * so we're safe with that check here.
 	 */
-	if (IS_ENABLED(CONFIG_RISCV_ISA_ZBB) &&
-	    IS_ENABLED(CONFIG_RISCV_ALTERNATIVE)) {
+#if defined(CONFIG_RISCV_ISA_ZBB)
+	if (IS_ENABLED(CONFIG_RISCV_ALTERNATIVE)) {
 		unsigned long fold_temp;
 
 		/*
@@ -213,6 +219,7 @@ do_csum_with_alignment(const unsigned char *buff, int len)
 end:
 		return csum >> 16;
 	}
+#endif /*CONFIG_RISCV_ISA_ZBB*/
 no_zbb:
 #ifndef CONFIG_32BIT
 	csum += ror64(csum, 32);
@@ -244,10 +251,11 @@ do_csum_no_alignment(const unsigned char *buff, int len)
 
 	/*
 	 * Zbb support saves 6 instructions, so not worth checking without
-	 * alternatives if supported
+	 * alternatives if supported.  See above for the ZBB preprocessor
+	 * check.
 	 */
-	if (IS_ENABLED(CONFIG_RISCV_ISA_ZBB) &&
-	    IS_ENABLED(CONFIG_RISCV_ALTERNATIVE)) {
+#if defined(CONFIG_RISCV_ISA_ZBB)
+	if (IS_ENABLED(CONFIG_RISCV_ALTERNATIVE)) {
 		unsigned long fold_temp;
 
 		/*
@@ -287,6 +295,7 @@ do_csum_no_alignment(const unsigned char *buff, int len)
 #endif /* !CONFIG_32BIT */
 		return csum >> 16;
 	}
+#endif /*CONFIG_RISCV_ISA_ZBB*/
 no_zbb:
 #ifndef CONFIG_32BIT
 	csum += ror64(csum, 32);

which is building for me on GCC-10/defconfig.

>
>> ---
>>  arch/riscv/lib/csum.c | 42 ++++++++++++++++++++++++++++++------------
>>  1 file changed, 30 insertions(+), 12 deletions(-)
>> 
>> diff --git a/arch/riscv/lib/csum.c b/arch/riscv/lib/csum.c
>> index 06ce8e7250d9..23be289f52b6 100644
>> --- a/arch/riscv/lib/csum.c
>> +++ b/arch/riscv/lib/csum.c
>> @@ -177,22 +177,35 @@ do_csum_with_alignment(const unsigned char *buff, int len)
>>  				  : no_zbb);
>>  
>>  #ifdef CONFIG_32BIT
>> -		asm_volatile_goto(".option push			\n\
>> +		/*
>> +		 * OutputOperands in asm goto is not supported until GCC 11, so
>> +		 * this asm has to be split to be compatible.
>> +		 */
>> +		asm (".option push				\n\
>>  		.option arch,+zbb				\n\
>>  			rori	%[fold_temp], %[csum], 16	\n\
>>  			andi	%[offset], %[offset], 1		\n\
>>  			add	%[csum], %[fold_temp], %[csum]	\n\
>> -			beq	%[offset], zero, %l[end]	\n\
>> -			rev8	%[csum], %[csum]		\n\
>>  		.option pop"
>>  			: [csum] "+r" (csum), [fold_temp] "=&r" (fold_temp)
>> -			: [offset] "r" (offset)
>> -			:
>> -			: end);
>> +			: [offset] "r" (offset));
>> +
>> +		if (offset == 0)
>> +			goto end;
>> +
>> +		asm (".option push				\n\
>> +		.option arch, +zbb				\n\
>> +			rev8	%[csum], %[csum]		\n\
>> +		.option pop"
>> +			: [csum] "+r" (csum));
>>  
>>  		return (unsigned short)csum;
>>  #else /* !CONFIG_32BIT */
>> -		asm_volatile_goto(".option push			\n\
>> +		/*
>> +		 * OutputOperands in asm goto is not supported until GCC 11, so
>> +		 * this asm has to be split to be compatible.
>> +		 */
>> +		asm (".option push				\n\
>>  		.option arch,+zbb				\n\
>>  			rori	%[fold_temp], %[csum], 32	\n\
>>  			add	%[csum], %[fold_temp], %[csum]	\n\
>> @@ -200,13 +213,18 @@ do_csum_with_alignment(const unsigned char *buff, int len)
>>  			roriw	%[fold_temp], %[csum], 16	\n\
>>  			addw	%[csum], %[fold_temp], %[csum]	\n\
>>  			andi	%[offset], %[offset], 1		\n\
>> -			beq	%[offset], zero, %l[end]	\n\
>> -			rev8	%[csum], %[csum]		\n\
>>  		.option pop"
>>  			: [csum] "+r" (csum), [fold_temp] "=&r" (fold_temp)
>> -			: [offset] "r" (offset)
>> -			:
>> -			: end);
>> +			: [offset] "r" (offset));
>> +
>> +		if (offset == 0)
>> +			goto end;
>> +
>> +		asm (".option push				\n\
>> +		.option arch, +zbb				\n\
>> +			rev8	%[csum], %[csum]		\n\
>> +		.option pop"
>> +			: [csum] "+r" (csum));
>>  
>>  		return (csum << 16) >> 48;
>>  #endif /* !CONFIG_32BIT */
>> 
>> ---
>> base-commit: 080c4324fa5e81ff3780206a138223abfb57a68e
>> change-id: 20240118-csum_remove_output_operands_asm_goto-49922c141ce7
>> -- 
>> - Charlie
>> 
>>