| Message ID | 20201011142012.96493-1-nivedita@alum.mit.edu |
|---|---|
| State | Accepted |
| Commit | 688eb28211abdf82a3f51e8997f1c8137947227d |
| Headers | show |
| Series | efi/x86: Only copy the compressed kernel image in efi_relocate_kernel() | expand |
On Sun, Oct 11, 2020 at 10:20:12AM -0400, Arvind Sankar wrote: > The image_size argument to efi_relocate_kernel() is currently specified > as init_size, but this is unnecessarily large. The compressed kernel is > much smaller, in fact, its image only extends up to the start of _bss, > since at this point, the .bss section is still uninitialized. > > Depending on compression level, this can reduce the amount of data > copied by 4-5x. > > Signed-off-by: Arvind Sankar <nivedita@alum.mit.edu> Ping > --- > drivers/firmware/efi/libstub/x86-stub.c | 5 ++++- > 1 file changed, 4 insertions(+), 1 deletion(-) > > diff --git a/drivers/firmware/efi/libstub/x86-stub.c b/drivers/firmware/efi/libstub/x86-stub.c > index 3672539cb96e..f14c4ff5839f 100644 > --- a/drivers/firmware/efi/libstub/x86-stub.c > +++ b/drivers/firmware/efi/libstub/x86-stub.c > @@ -715,8 +715,11 @@ unsigned long efi_main(efi_handle_t handle, > (IS_ENABLED(CONFIG_X86_32) && buffer_end > KERNEL_IMAGE_SIZE) || > (IS_ENABLED(CONFIG_X86_64) && buffer_end > MAXMEM_X86_64_4LEVEL) || > (image_offset == 0)) { > + extern char _bss[]; > + > status = efi_relocate_kernel(&bzimage_addr, > - hdr->init_size, hdr->init_size, > + (unsigned long)_bss - bzimage_addr, > + hdr->init_size, > hdr->pref_address, > hdr->kernel_alignment, > LOAD_PHYSICAL_ADDR); > -- > 2.26.2 >
On Sun, 25 Oct 2020 at 17:19, Arvind Sankar <nivedita@alum.mit.edu> wrote: > > On Sun, Oct 11, 2020 at 10:20:12AM -0400, Arvind Sankar wrote: > > The image_size argument to efi_relocate_kernel() is currently specified > > as init_size, but this is unnecessarily large. The compressed kernel is > > much smaller, in fact, its image only extends up to the start of _bss, > > since at this point, the .bss section is still uninitialized. > > > > Depending on compression level, this can reduce the amount of data > > copied by 4-5x. > > > > Signed-off-by: Arvind Sankar <nivedita@alum.mit.edu> > > Ping > I'll pick this up as a fix once the merge window closes. > > --- > > drivers/firmware/efi/libstub/x86-stub.c | 5 ++++- > > 1 file changed, 4 insertions(+), 1 deletion(-) > > > > diff --git a/drivers/firmware/efi/libstub/x86-stub.c b/drivers/firmware/efi/libstub/x86-stub.c > > index 3672539cb96e..f14c4ff5839f 100644 > > --- a/drivers/firmware/efi/libstub/x86-stub.c > > +++ b/drivers/firmware/efi/libstub/x86-stub.c > > @@ -715,8 +715,11 @@ unsigned long efi_main(efi_handle_t handle, > > (IS_ENABLED(CONFIG_X86_32) && buffer_end > KERNEL_IMAGE_SIZE) || > > (IS_ENABLED(CONFIG_X86_64) && buffer_end > MAXMEM_X86_64_4LEVEL) || > > (image_offset == 0)) { > > + extern char _bss[]; > > + > > status = efi_relocate_kernel(&bzimage_addr, > > - hdr->init_size, hdr->init_size, > > + (unsigned long)_bss - bzimage_addr, > > + hdr->init_size, > > hdr->pref_address, > > hdr->kernel_alignment, > > LOAD_PHYSICAL_ADDR); > > -- > > 2.26.2 > >
On Sun, Oct 25, 2020 at 05:28:06PM +0100, Ard Biesheuvel wrote: > On Sun, 25 Oct 2020 at 17:19, Arvind Sankar <nivedita@alum.mit.edu> wrote: > > > > On Sun, Oct 11, 2020 at 10:20:12AM -0400, Arvind Sankar wrote: > > > The image_size argument to efi_relocate_kernel() is currently specified > > > as init_size, but this is unnecessarily large. The compressed kernel is > > > much smaller, in fact, its image only extends up to the start of _bss, > > > since at this point, the .bss section is still uninitialized. > > > > > > Depending on compression level, this can reduce the amount of data > > > copied by 4-5x. > > > > > > Signed-off-by: Arvind Sankar <nivedita@alum.mit.edu> > > > > Ping > > > > I'll pick this up as a fix once the merge window closes. Thanks! > > > > --- > > > drivers/firmware/efi/libstub/x86-stub.c | 5 ++++- > > > 1 file changed, 4 insertions(+), 1 deletion(-) > > > > > > diff --git a/drivers/firmware/efi/libstub/x86-stub.c b/drivers/firmware/efi/libstub/x86-stub.c > > > index 3672539cb96e..f14c4ff5839f 100644 > > > --- a/drivers/firmware/efi/libstub/x86-stub.c > > > +++ b/drivers/firmware/efi/libstub/x86-stub.c > > > @@ -715,8 +715,11 @@ unsigned long efi_main(efi_handle_t handle, > > > (IS_ENABLED(CONFIG_X86_32) && buffer_end > KERNEL_IMAGE_SIZE) || > > > (IS_ENABLED(CONFIG_X86_64) && buffer_end > MAXMEM_X86_64_4LEVEL) || > > > (image_offset == 0)) { > > > + extern char _bss[]; > > > + > > > status = efi_relocate_kernel(&bzimage_addr, > > > - hdr->init_size, hdr->init_size, > > > + (unsigned long)_bss - bzimage_addr, > > > + hdr->init_size, > > > hdr->pref_address, > > > hdr->kernel_alignment, > > > LOAD_PHYSICAL_ADDR); > > > -- > > > 2.26.2 > > >
diff --git a/drivers/firmware/efi/libstub/x86-stub.c b/drivers/firmware/efi/libstub/x86-stub.c index 3672539cb96e..f14c4ff5839f 100644 --- a/drivers/firmware/efi/libstub/x86-stub.c +++ b/drivers/firmware/efi/libstub/x86-stub.c @@ -715,8 +715,11 @@ unsigned long efi_main(efi_handle_t handle, (IS_ENABLED(CONFIG_X86_32) && buffer_end > KERNEL_IMAGE_SIZE) || (IS_ENABLED(CONFIG_X86_64) && buffer_end > MAXMEM_X86_64_4LEVEL) || (image_offset == 0)) { + extern char _bss[]; + status = efi_relocate_kernel(&bzimage_addr, - hdr->init_size, hdr->init_size, + (unsigned long)_bss - bzimage_addr, + hdr->init_size, hdr->pref_address, hdr->kernel_alignment, LOAD_PHYSICAL_ADDR);
The image_size argument to efi_relocate_kernel() is currently specified as init_size, but this is unnecessarily large. The compressed kernel is much smaller, in fact, its image only extends up to the start of _bss, since at this point, the .bss section is still uninitialized. Depending on compression level, this can reduce the amount of data copied by 4-5x. Signed-off-by: Arvind Sankar <nivedita@alum.mit.edu> --- drivers/firmware/efi/libstub/x86-stub.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-)