Hi Thomas,
-----Original Message-----
From: Thomas Monjalon [mailto:thomas.monja...@6wind.com]
Sent: Tuesday, June 24, 2014 4:47 PM
To: Chen, Jing D
Cc: dev at dpdk.org
Subject: Re: [dpdk-dev] [PATCH] i40e: fix shared code compile warning
Hi,
2014-06-24 13:22, Chen Jing D:
> Fix a compile warning in shared code on 32-bits RHEL6.3/6.5.
>
> Signed-off-by: Chen Jing D(Mark) <jing.d.chen at intel.com>
In such case you should show the error message in the log.
lib/librte_pmd_i40e/i40e/i40e_lan_hmc.c:917:
error: integer constant is too large for ?long? type
> --- a/lib/librte_pmd_i40e/Makefile
> +++ b/lib/librte_pmd_i40e/Makefile
> @@ -52,6 +52,7 @@ CFLAGS_SHARED_DRIVERS +=
> -Wno-missing-field-initializers CFLAGS_SHARED_DRIVERS +=
> -Wno-pointer-to-int-cast CFLAGS_SHARED_DRIVERS +=
> -Wno-format-nonliteral CFLAGS_SHARED_DRIVERS += -Wno-format-security
> +CFLAGS_i40e_lan_hmc.o += -Wno-error
I know we shouldn't modify base drivers. But this one seems to be an important
error. In such case, we already modified base driver. Recently:
http://dpdk.org/ml/archives/dev/2014-June/003498.html
I think it's different. The logic is right after adding the patch. Below is my
finding.
In this case, it met the error when compile on 32-bits OS. The message is :
/jenkins/workspace/DPDK_AUTO_IDT_VM_RHEL65_32_BUILD/DPDK/lib/librte_pmd_i40e/i40e/i40e_lan_hmc.c:
In function ?i40e_write_qword?:
/jenkins/workspace/DPDK_AUTO_IDT_VM_RHEL65_32_BUILD/DPDK/lib/librte_pmd_i40e/i40e/i40e_lan_hmc.c:917:
error: integer constant is too large for ?long? type
/jenkins/workspace/DPDK_AUTO_IDT_VM_RHEL65_32_BUILD/DPDK/lib/librte_pmd_i40e/i40e/i40e_lan_hmc.c:
In function ?i40e_read_qword?:
/jenkins/workspace/DPDK_AUTO_IDT_VM_RHEL65_32_BUILD/DPDK/lib/librte_pmd_i40e/i40e/i40e_lan_hmc.c:1097:
error: integer constant is too large for ?long? type
I found the code that cause errors. 'mask' is 'uint64_t' type and is assigned
to value 0Xffff_ffff_ffff_ffff. Compiler assumes the constant is 'int' type by
default. If changed it to oxffff_ffff_ffff_ffffULL, the warning will be gone.
if (ce_info->width < 64)
mask = ((u64)1 << ce_info->width) - 1;
else
mask = 0xFFFFFFFFFFFFFFFF;
besides that, I dis-assembler the code with the patch and get below segment. It
seems right.
if (ce_info->width < 64)
1946: 8b 45 0c mov 0xc(%ebp),%eax
1949: 0f b7 40 04 movzwl 0x4(%eax),%eax
194d: 66 83 f8 3f cmp $0x3f,%ax
1951: 77 30 ja 1983 <i40e_write_qword+0x62>
mask = ((u64)1 << ce_info->width) - 1;
1953: 8b 45 0c mov 0xc(%ebp),%eax
1956: 0f b7 40 04 movzwl 0x4(%eax),%eax
195a: 0f b7 c8 movzwl %ax,%ecx
195d: b8 01 00 00 00 mov $0x1,%eax
1962: ba 00 00 00 00 mov $0x0,%edx
1967: 0f a5 c2 shld %cl,%eax,%edx
196a: d3 e0 shl %cl,%eax
196c: f6 c1 20 test $0x20,%cl
196f: 74 04 je 1975 <i40e_write_qword+0x54>
1971: 89 c2 mov %eax,%edx
1973: 31 c0 xor %eax,%eax
1975: 83 c0 ff add $0xffffffff,%eax
1978: 83 d2 ff adc $0xffffffff,%edx
197b: 89 45 e0 mov %eax,-0x20(%ebp)
197e: 89 55 e4 mov %edx,-0x1c(%ebp)
1981: eb 0e jmp 1991 <i40e_write_qword+0x70>
else
mask = 0xFFFFFFFFFFFFFFFF;
1983: c7 45 e0 ff ff ff ff movl $0xffffffff,-0x20(%ebp)
198a: c7 45 e4 ff ff ff ff movl $0xffffffff,-0x1c(%ebp)
--
Thomas
