On 10 March 2016 at 15:25, Ingo Molnar [off-list ref] wrote:

* Ard Biesheuvel [off-list ref] wrote:

quoted

diff --git a/drivers/firmware/efi/libstub/arm32-stub.c b/drivers/firmware/efi/libstub/arm32-stub.c
index 495ebd657e38..0fbe00f13186 100644
--- a/drivers/firmware/efi/libstub/arm32-stub.c
+++ b/drivers/firmware/efi/libstub/arm32-stub.c

@@ -9,6 +9,44 @@
 #include <linux/efi.h>
 #include <asm/efi.h>

+static const efi_guid_t screen_info_guid = LINUX_ARM32_SCREEN_INFO_TABLE_GUID;
+
+struct screen_info *alloc_screen_info(efi_system_table_t *sys_table_arg)
+{
+     struct screen_info *si;
+     efi_status_t status;
+
+     /*
+      * Unlike on arm64, where we can directly fill out the screen_info
+      * structure from the stub, we need to allocate a buffer to hold
+      * its contents while we hand over to the kernel proper from the
+      * decompressor.
+      */
+     status = efi_call_early(allocate_pool, EFI_RUNTIME_SERVICES_DATA,
+                             sizeof(*si), (void **)&si);
+
+     if (status != EFI_SUCCESS)
+             return NULL;
+
+     status = efi_call_early(install_configuration_table,
+                             (efi_guid_t *)&screen_info_guid, si);
+     if (status == EFI_SUCCESS)
+             return si;
+
+     efi_call_early(free_pool, si);
+     return NULL;
+}
+
+void free_screen_info(efi_system_table_t *sys_table_arg, struct screen_info *si)
+{
+     if (!si)
+             return;
+
+     efi_call_early(install_configuration_table,
+                    (efi_guid_t *)&screen_info_guid, NULL);
+     efi_call_early(free_pool, si);
+}
+
 efi_status_t handle_kernel_image(efi_system_table_t *sys_table,
                               unsigned long *image_addr,
                               unsigned long *image_size,

So screen_info_guid should probably not be a 'const': you have to cast it away
anyway, adding artificial linebreaks and uglifying the code. It's also a bad
practice to cast away const-ness, it hinders move-consts-to-readonly-sections
efforts.

The problem here is that the UEFI spec never uses const qualifiers in
its APIs for by-ref parameters that are obviously never modified by
the caller, such as these GUIDs. This is __init code regardless
(either due to the fact that it is linked into the stub rather than
the core kernel, or because we __init-ize it using objcopy) so I don't
care deeply, and I am happy to remove it.

Otherwise the series looks mostly good to me. Time for new v4.6 merges is short,
but if Matt acks it I can still take it into tip:efi/core.

Thank you. I am perfectly happy exercising some more caution here,
though, considering that our past experiences involving reshuffling
stub code for x86 were not a huge success (and broke Linus's laptop,
among others')

I was very careful not to introduce cross-object data symbol
references (which were the culprit before), and I tested both i386 and
x86_64 to the extent feasible to me, but we'd need to confirm that
those things are still fully correct before queuing this.

Thanks,
Ard.