Messages in this thread | | | Subject | Re: Avoiding unnecessary jump relocations in gas? | From | "H. Peter Anvin" <> | Date | Wed, 20 May 2015 14:58:49 -0700 |
| |
Okay, let's go with it. that also let us merge this constant with the length of the array in the C header file.
On May 20, 2015 2:47:10 PM PDT, Andy Lutomirski <luto@amacapital.net> wrote: >On Wed, May 20, 2015 at 1:59 PM, H. Peter Anvin <hpa@zytor.com> wrote: >> On 05/20/2015 01:53 PM, Andy Lutomirski wrote: >>> Egads. Now I understand what that code is. I don't like the >balign, >>> since this has nothing to do with alignment -- we're creating an >array >>> of functions. >> >> Actually it does... we align to the beginning of each slot. If >.balign >> could be something other than a power of 2 that would work, too. > >When I see "align", I think that we want to align to a multiple of X >but we don't particularly care which multiple of X. Here we want a >specific address and any other address would be an error. > >> >> I was mostly looking to minimize the amount of gas magic we rely on. >> >>> Can't we make it explicit? >>> >>> #define EARLY_IDT_HANDLER_STRIDE 9 >>> >>> ... >>> >>> .rept NUM_EXCEPTION_VECTORS >>> . = early_idt_handlers + i * EARLY_IDT_HANDLER_STRIDE >>> .if (EXCEPTION_ERRCODE_MASK >> i) & 1 >>> ASM_NOP2 >>> .else >>> pushl $0 # Dummy error code, to make stack frame uniform >>> .endif >>> pushl $i # 20(%esp) Vector number >>> jmp early_idt_handler >>> i = i + 1 >>> .endr >>> >>> gas will error out if we try to move . backwards, so this should be >safe. >> >> If that works too with all the versions of gas we care about, that >would >> be fine (and I do appreciate the explicitness.) However, .[b]align >is >> something that will have been well exercised in every version of gas, >so >> I do feel slightly safer with it. > >I have no idea how to easily test my approach on really old binutils >versions, but I'm reasonably confident that assigning to . is common. > >--Andy
-- Sent from my mobile phone. Please pardon brevity and lack of formatting.
| |