On Wed, 25 Mar 2020 at 23:10, Arvind Sankar <nivedita@xxxxxxxxxxxx> wrote: > > On Wed, Mar 25, 2020 at 05:41:43PM +0100, Ard Biesheuvel wrote: > > On Fri, 20 Mar 2020 at 03:00, Arvind Sankar <nivedita@xxxxxxxxxxxx> wrote: > > > > > > This series is against tip:efi/core. > > > > > > Patches 1-9 are small cleanups and refactoring of the code in > > > libstub/gop.c. > > > > > > The rest of the patches add the ability to use a command-line option to > > > switch the gop's display mode. > > > > > > The options supported are: > > > video=efifb:mode=n > > > Choose a specific mode number > > > video=efifb:<xres>x<yres>[-(rgb|bgr|<bpp>)] > > > Specify mode by resolution and optionally color depth > > > video=efifb:auto > > > Let the EFI stub choose the highest resolution mode available. > > > > > > The mode-setting additions increase code size of gop.o by about 3k on > > > x86-64 with EFI_MIXED enabled. > > > > > > Changes in v2 (HT lkp@xxxxxxxxx): > > > - Fix __efistub_global attribute to be after the variable. > > > (NB: bunch of other places should ideally be fixed, those I guess > > > don't matter as they are scalars?) > > > - Silence -Wmaybe-uninitialized warning in set_mode function. > > > > > > > These look good to me. The only question I have is whether it would be > > possible to use the existing next_arg() and parse_option_str() > > functions to replace some of the open code parsing that goes on in > > patches 11 - 14. > > > > I don't think so -- next_arg is for parsing space-separated param=value > pairs, so efi_parse_options can use it, but it doesn't work for the > comma-separated options we'll have within the value. > > parse_option_str would only work for the "auto" option, but it scans the > entire option string and just returns whether it was there or not, so it > wouldn't be too useful either, since we have to check for the other > possibilities anyway. > > It would be nice to have a more generic library for cmdline parsing, > there are a lot of places that have to open-code option parsing like > this. > > There's one thing I noticed while working at this, btw. The Makefile > specifies -ffreestanding, but at least x86 builds without having to > specify that. With -ffreestanding, the compiler doesn't optimize string > functions -- strlen(string literal) into a compile-time constant, for > eg. A couple hundred bytes or so can be saved by removing that option, > if it also works for ARM. Yes, -ffreestanding implies -fno-builtin, which means that the compiler cannot assume it knows (and can optimize away) the behavior of strlen(), memset(), etc.
