Re: [RFC PATCH 01/11] efi: Fix gcc error around __umoddi3 for 32 bit builds

From: Ard Biesheuvel <ardb@kernel.org>
To: Atish Patra <atishp@atishpatra.org>
Cc: Heinrich Schuchardt <xypron.glpk@gmx.de>,
	Atish Patra <atish.patra@wdc.com>,
	"linux-kernel@vger.kernel.org List"
	<linux-kernel@vger.kernel.org>,
	linux-efi <linux-efi@vger.kernel.org>,
	linux-riscv <linux-riscv@lists.infradead.org>,
	Masahiro Yamada <masahiroy@kernel.org>,
	Palmer Dabbelt <palmer@dabbelt.com>,
	"linux-arm-kernel@lists.infradead.org" 
	<linux-arm-kernel@lists.infradead.org>
Subject: Re: [RFC PATCH 01/11] efi: Fix gcc error around __umoddi3 for 32 bit builds
Date: Sat, 27 Jun 2020 11:22:46 +0200	[thread overview]
Message-ID: <CAMj1kXFd3wpob228WOYOu3SG0r2SEbT0cZF1JkF6uwjzrmf2EQ@mail.gmail.com> (raw)
In-Reply-To: <CAMj1kXFyQiYPYB-81POq7agRW1ROt=2j3nN9wcpHGmr4YjmFCQ@mail.gmail.com>

On Sat, 27 Jun 2020 at 00:03, Ard Biesheuvel <ardb@kernel.org> wrote:
>
> On Fri, 26 Jun 2020 at 23:56, Atish Patra <atishp@atishpatra.org> wrote:
> >
> > On Thu, Jun 25, 2020 at 7:43 PM Heinrich Schuchardt <xypron.glpk@gmx.de> wrote:
> > >
> > > On 6/26/20 1:45 AM, Atish Patra wrote:
> > > > 32bit gcc doesn't support modulo operation on 64 bit data. It results in
> > > > a __umoddi3 error while building EFI for 32 bit.
> > > >
> > > > Use bitwise operations instead of modulo operations to fix the issue.
> > > >
> > > > Signed-off-by: Atish Patra <atish.patra@wdc.com>
> > > > ---
> > > >  drivers/firmware/efi/libstub/alignedmem.c | 2 +-
> > > >  1 file changed, 1 insertion(+), 1 deletion(-)
> > > >
> > > > diff --git a/drivers/firmware/efi/libstub/alignedmem.c b/drivers/firmware/efi/libstub/alignedmem.c
> > > > index cc89c4d6196f..1de9878ddd3a 100644
> > > > --- a/drivers/firmware/efi/libstub/alignedmem.c
> > > > +++ b/drivers/firmware/efi/libstub/alignedmem.c
> > > > @@ -44,7 +44,7 @@ efi_status_t efi_allocate_pages_aligned(unsigned long size, unsigned long *addr,
> > > >       *addr = ALIGN((unsigned long)alloc_addr, align);
> > > >
> > > >       if (slack > 0) {
> > > > -             int l = (alloc_addr % align) / EFI_PAGE_SIZE;
> > > > +             int l = (alloc_addr & (align - 1)) / EFI_PAGE_SIZE;
> > >
> > > Here you rely on the compiler to silently convert the division by
> > > EFI_PAGE_SIZE into a right shift. Wouldn't it be cleaner to use
> > > EFI_PAGE_SHIFT to explicitly avoid a dependency on __udivdi3()?
> > >
> > > slack = (align >> EFI_PAGE_SHIFT) - 1;
> > > ...
> > > int l = (alloc_addr & (align - 1)) >> EFI_PAGE_SHIFT;
> > >
> >
> > Sure. I will update it in the next version. Thanks for the suggestion.
>
> Please don't. Dividing by EFI_PAGE_SIZE is perfectly fine, and is more readable.

Actually, I will take this patch as a fix right away - 32-bit ARM has
the same issue (although it does not actually incorporate this object
file)

In the meantime, please check how the stub gets pulled into your
kernel: the idea of a static library is that only the objects that are
used are included in the final build, but this turned out to be broken
for arm64 [0]. IOW, simply applying a similar change might fix your
build issue as well (assuming RISC-V does not actually call
efi_allocate_pages_aligned() anywhere)

[0] https://lore.kernel.org/linux-arm-kernel/20200604022031.164207-1-masahiroy@kernel.org/