On Fri, Apr 10, 2015 at 5:56 AM, Ard Biesheuvel <ard.biesheuvel@xxxxxxxxxx> wrote: > This changes the allocation for the ASCII-converted command > line to use an ordinary memory pool rather than a separate > page based allocation. > > Pool allocations are generally preferred over page based > allocations due to the fact that they cause less fragmentation, > but in the particular case of arm64, where page allocations are > rounded up to 64 KB and where this allocation happens to be the > only explicit low allocation, it results in the lowest 64 KB of > memory to always be taken up by this particular allocation. > > So allocate from the EFI_LOADER_DATA pool instead. > > Signed-off-by: Ard Biesheuvel <ard.biesheuvel@xxxxxxxxxx> > --- > drivers/firmware/efi/libstub/efi-stub-helper.c | 3 ++- > 1 file changed, 2 insertions(+), 1 deletion(-) > > diff --git a/drivers/firmware/efi/libstub/efi-stub-helper.c b/drivers/firmware/efi/libstub/efi-stub-helper.c > index f07d4a67fa76..c95a567ca132 100644 > --- a/drivers/firmware/efi/libstub/efi-stub-helper.c > +++ b/drivers/firmware/efi/libstub/efi-stub-helper.c > @@ -684,7 +684,8 @@ char *efi_convert_cmdline(efi_system_table_t *sys_table_arg, > > options_bytes++; /* NUL termination */ > > - status = efi_low_alloc(sys_table_arg, options_bytes, 0, &cmdline_addr); > + status = efi_call_early(allocate_pool, EFI_LOADER_DATA, > + options_bytes, (void **)&cmdline_addr); > if (status != EFI_SUCCESS) > return NULL; > > -- > 1.8.3.2 > Hi Ard, We can't do this without also changing the frees on the error paths in arm-stub.c (line 289) and eboot.c (line 1148) to be a pool free as well. Also, for x86/x86_64 the address of the command line is put into a boot_params structure, which only has a __u32 field for the address of the command line, so we could get an address that won't fit if we use a pool allocation. Looking at that bit of if it, I don't think we handle the case of no 32 bit addressable memory existing at the time of the efi_low_alloc() for x86_64 systems, so if a >32 bit address is returned here, this won't be detected as a failure, and the cmdline address will be the 32 bit truncated address. I don't think that there is a way to control the address range returned by pool allocations, so I think we are stuck with the page based allocations if we have address restrictions. Roy -- To unsubscribe from this list: send the line "unsubscribe linux-efi" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
