Skip to contentSkip to conversation

EFI memmap fix

PATCH In review Borislav Petkov 4 participants 12 messages Dec 16, 2013 – Dec 18, 2013

Subsystems (3)
lore
Share & tools
lore

Borislav Petkov

author maintainer
#0 lore raw

cover EFI memmap fix

From: Borislav Petkov <[email protected]>

Hi guys,

this is the result of Toshi and me debugging a #GP on one of his big HP boxes sporting UEFI. Each commit message should be self-explanatory so please look there.

This has more or less an RFC nature thus I'm sending it now to collect feedback. It is going to wait in the EFI queue anyway after the kexec stuff gets sorted out first.

Comments and suggestions as always are very much appreciated.

Thanks.

Borislav Petkov (3):
x86, ptdump: Add the functionality to dump an arbitrary pagetable
efi: Dump the EFI page table
efi: Make efi virtual runtime map passing more robust

 arch/x86/include/asm/pgtable.h |  3 +-
 arch/x86/mm/dump_pagetables.c  | 77 ++++++++++++++++++++++++++++--------------
 arch/x86/platform/efi/efi.c    | 68 ++++++++++++++++++++++++++++++++-----
 3 files changed, 112 insertions(+), 36 deletions(-)
Signature · 1 line
1.8.4
author maintainer
#1 lore raw

PATCH 1/3 x86, ptdump: Add the functionality to dump an arbitrary pagetable

From: Borislav Petkov <[email protected]>

With reusing the ->trampoline_pgd page table for mapping EFI regions in order to use them after having switched to EFI virtual mode, it is very useful to be able to dump aforementioned page table in dmesg. This adds that functionality through the walk_pgd_level() interface which can be called from somewhere else.

The original functionality of dumping to debugfs remains untouched.

Cc Arjan van de Ven <[email protected]>
---
 arch/x86/include/asm/pgtable.h |  3 +-
 arch/x86/mm/dump_pagetables.c  | 77 ++++++++++++++++++++++++++++--------------
 2 files changed, 53 insertions(+), 27 deletions(-)
arch/x86/include/asm/pgtable.h +2 -1
index 3d1999458709..595305f59da0 100644
@@ -15,9 +15,10 @@
 	 : (prot))
 
 #ifndef __ASSEMBLY__
-
 #include <asm/x86_init.h>
 
+void walk_pgd_level(struct seq_file *m, pgd_t *pgd);
+
 /*
  * ZERO_PAGE is a global shared page that is always zero: used
  * for zero-mapped memory areas etc..
arch/x86/mm/dump_pagetables.c +51 -26

160 lines not on this page · raw message

Signature · 1 line
1.8.4
author maintainer
#2 lore raw

PATCH 2/3 efi: Dump the EFI page table

In reply to Borislav Petkov

From: Borislav Petkov <[email protected]>

This is very useful for debugging issues with the recently added pagetable switching code for EFI virtual mode.

---
 arch/x86/platform/efi/efi.c | 11 +++++++++++
 1 file changed, 11 insertions(+)
arch/x86/platform/efi/efi.c +11 -0
index 7d8caccbf2f3..51d6285701e9 100644
@@ -740,6 +740,15 @@ void efi_memory_uc(u64 addr, unsigned long size)
 	set_memory_uc(addr, npages);
 }
 
+static void efi_dump_pagetable(void)
+{
+#if defined(EFI_DEBUG) && defined(CONFIG_X86_PTDUMP)
+	pgd_t *pgd = (pgd_t *)__va(real_mode_header->trampoline_pgd);
+
+	walk_pgd_level(NULL, pgd);
+#endif
+}
+
 void __init old_map_region(efi_memory_desc_t *md)
 {
 	u64 start_pfn, end_pfn, end;
@@ -869,6 +878,8 @@ void __init efi_enter_virtual_mode(void)
 	efi_setup_page_tables();
 	efi_sync_low_kernel_mappings();
 
+	efi_dump_pagetable();
+
 	status = phys_efi_set_virtual_address_map(
 		memmap.desc_size * count,
 		memmap.desc_size,
Signature · 1 line
1.8.4
author maintainer
#3 lore raw

PATCH 3/3 efi: Make efi virtual runtime map passing more robust

In reply to Borislav Petkov

From: Borislav Petkov <[email protected]>

Currently, running SetVirtualAddressMap() and passing the physical address of the virtual map array was working only by a lucky coincidence because the memory was present in the EFI page table too. Until Toshi went and booted this on a big HP box - the krealloc() manner of resizing the memmap we're doing did allocate from such physical addresses which were not mapped anymore and boom:

http://lkml.kernel.org/r/[email protected]

One way to take care of that issue is to reimplement the krealloc thing but with pages. We start with contiguous pages of order 1, i.e. 2 pages, and when we deplete that memory (shouldn't happen all that often but you know firmware) we realloc the next power-of-two pages.

Having the pages, it is much more handy and easy to map them into the EFI page table with the already existing mapping code which we're using for building the virtual mappings.

And, it doesn't matter all that much how much pages we've used as we're freeing them right after they've fulfilled their purpose at the end of the function anyway.

---
 arch/x86/platform/efi/efi.c | 57 ++++++++++++++++++++++++++++++++++++++-------
 1 file changed, 48 insertions(+), 9 deletions(-)
arch/x86/platform/efi/efi.c +48 -9
index 51d6285701e9..39c52cc9b63a 100644
@@ -112,7 +112,6 @@ static int __init setup_storage_paranoia(char *arg)
 }
 early_param("efi_no_storage_paranoia", setup_storage_paranoia);
 
-
 static efi_status_t virt_efi_get_time(efi_time_t *tm, efi_time_cap_t *tc)
 {
 	unsigned long flags;
@@ -775,6 +774,27 @@ void __init old_map_region(efi_memory_desc_t *md)
 		       (unsigned long long)md->phys_addr);
 }
 
+static void *realloc_pages(void *old_memmap, int old_shift)
+{
+	void *ret;
+
+	ret = (void *)__get_free_pages(GFP_KERNEL, old_shift + 1);
+	if (!ret)
+		goto out;
+
+	/*
+	 * A first-time allocation doesn't have anything to copy.
+	 */
+	if (!old_memmap)
+		return ret;
+
+	memcpy(ret, old_memmap, PAGE_SIZE << old_shift);
+
+out:
+	__free_pages(old_memmap, old_shift);
+	return ret;
+}
+
 /*
  * This function will switch the EFI runtime services to virtual mode.
  * Essentially, we look through the EFI memmap and map every region that
@@ -794,12 +814,13 @@ void __init old_map_region(efi_memory_desc_t *md)
  */
 void __init efi_enter_virtual_mode(void)
 {
+	pgd_t *pgd = (pgd_t *)__va(real_mode_header->trampoline_pgd);
+	unsigned long size, new_memmap_left = 0;
 	efi_memory_desc_t *md, *prev_md = NULL;
+	int count = 0, new_memmap_shift = 0;
 	void *p, *new_memmap = NULL;
-	unsigned long size;
 	efi_status_t status;
 	u64 end, systab;
-	int count = 0;
 
 	efi.systab = NULL;
 
@@ -862,14 +883,19 @@ void __init efi_enter_virtual_mode(void)
 			efi.systab = (efi_system_table_t *) (unsigned long) systab;
 		}
 
-		new_memmap = krealloc(new_memmap,
-				      (count + 1) * memmap.desc_size,
-				      GFP_KERNEL);
-		if (!new_memmap)
-			goto err_out;
+		if (new_memmap_left < memmap.desc_size) {
+			new_memmap = realloc_pages(new_memmap, new_memmap_shift);
+			if (!new_memmap)
+				goto err_out;
+
+			new_memmap_shift++;
+			new_memmap_left += PAGE_SIZE << new_memmap_shift;
+		}
 
 		memcpy(new_memmap + (count * memmap.desc_size), md,
 		       memmap.desc_size);
+
+		new_memmap_left -= memmap.desc_size;
 		count++;
 	}
 
@@ -880,6 +906,19 @@ void __init efi_enter_virtual_mode(void)
 
 	efi_dump_pagetable();
 
+	/*
+	 * It can happen that the physical address of new_memmap lands in memory
+	 * which is not mapped in the EFI page table. Therefore we need to go
+	 * and ident-map those pages containing the map before calling
+	 * phys_efi_set_virtual_address_map().
+	 */
+	if (kernel_map_pages_in_pgd(pgd, __pa(new_memmap), __pa(new_memmap),
+				    1 << new_memmap_shift, _PAGE_NX)) {
+		pr_err("Error ident-mapping new memmap (0x%lx)!\n",
+		       __pa(new_memmap));
+		goto err_out;
+	}
+
 	status = phys_efi_set_virtual_address_map(
 		memmap.desc_size * count,
 		memmap.desc_size,
@@ -916,7 +955,7 @@ void __init efi_enter_virtual_mode(void)
 	if (efi_enabled(EFI_OLD_MEMMAP) && (__supported_pte_mask & _PAGE_NX))
 		runtime_code_page_mkexec();
 
-	kfree(new_memmap);
+	__free_pages(new_memmap, new_memmap_shift);
 
 	/* clean DUMMY object */
 	efi.set_variable(efi_dummy_name, &EFI_DUMMY_GUID,
Signature · 1 line
1.8.4
author maintainer
#4 lore raw

Re: [PATCH 3/3] efi: Make efi virtual runtime map passing more robust

Quoted 21 lines · Borislav Petkov
arch/x86/platform/efi/efi.c +13 -0
@@ -880,6 +906,19 @@ void __init efi_enter_virtual_mode(void)
 
 	efi_dump_pagetable();
 
+	/*
+	 * It can happen that the physical address of new_memmap lands in memory
+	 * which is not mapped in the EFI page table. Therefore we need to go
+	 * and ident-map those pages containing the map before calling
+	 * phys_efi_set_virtual_address_map().
+	 */
+	if (kernel_map_pages_in_pgd(pgd, __pa(new_memmap), __pa(new_memmap),
+				    1 << new_memmap_shift, _PAGE_NX)) {
+		pr_err("Error ident-mapping new memmap (0x%lx)!\n",
+		       __pa(new_memmap));
+		goto err_out;
+	}
+
 	status = phys_efi_set_virtual_address_map(
 		memmap.desc_size * count,
 		memmap.desc_size,

arch/x86/platform/efi/efi.c void __init efi_enter_virtual_mode(void) · line 958

Quoted 5 lines
arch/x86/platform/efi/efi.c +1 -1
@@ -916,7 +955,7 @@ void __init efi_enter_virtual_mode(void)
 	if (efi_enabled(EFI_OLD_MEMMAP) && (__supported_pte_mask & _PAGE_NX))
 		runtime_code_page_mkexec();
 
-	kfree(new_memmap);
+	__free_pages(new_memmap, new_memmap_shift);

Note to self: carve out the error path of populate_pgd() into a separate function and call it here to unmap the pages previously mapped above as we're freeing them here.

Signature · 5 lines
Regards/Gruss,
    Boris.

Sent from a fat crate under my desk. Formatting is fine.
--
#5 lore raw

Re: [PATCH 3/3] efi: Make efi virtual runtime map passing more robust

In reply to Borislav Petkov
Quoted 29 lines · Borislav Petkov

From: Borislav Petkov <[email protected]>

Currently, running SetVirtualAddressMap() and passing the physical address of the virtual map array was working only by a lucky coincidence because the memory was present in the EFI page table too. Until Toshi went and booted this on a big HP box - the krealloc() manner of resizing the memmap we're doing did allocate from such physical addresses which were not mapped anymore and boom:

http://lkml.kernel.org/r/1386806463.1791.295.camel-RbGIw1UOYPVo/[email protected]

One way to take care of that issue is to reimplement the krealloc thing but with pages. We start with contiguous pages of order 1, i.e. 2 pages, and when we deplete that memory (shouldn't happen all that often but you know firmware) we realloc the next power-of-two pages.

Having the pages, it is much more handy and easy to map them into the EFI page table with the already existing mapping code which we're using for building the virtual mappings.

And, it doesn't matter all that much how much pages we've used as we're freeing them right after they've fulfilled their purpose at the end of the function anyway.

---
 arch/x86/platform/efi/efi.c | 57 ++++++++++++++++++++++++++++++++++++++-------
 1 file changed, 48 insertions(+), 9 deletions(-)

[...]

Quoted 5 lines
arch/x86/platform/efi/efi.c +1 -0
@@ -794,12 +814,13 @@ void __init old_map_region(efi_memory_desc_t *md)
  */
 void __init efi_enter_virtual_mode(void)
 {
+	pgd_t *pgd = (pgd_t *)__va(real_mode_header->trampoline_pgd);

arch/x86/platform/efi/efi.c void __init old_map_region(efi_memory_desc_t *md) · line 825

Quoted 10 lines
arch/x86/platform/efi/efi.c +2 -2
+	unsigned long size, new_memmap_left = 0;
 	efi_memory_desc_t *md, *prev_md = NULL;
+	int count = 0, new_memmap_shift = 0;
 	void *p, *new_memmap = NULL;
-	unsigned long size;
 	efi_status_t status;
 	u64 end, systab;
-	int count = 0;
 
 	efi.systab = NULL;

You sunk my i386 battleship,

/home/build/git/efi/arch/x86/platform/efi/efi.c:824:24: error: ‘struct real_mode_header’ has no member named ‘trampoline_pgd’ make[4]: *** [arch/x86/platform/efi/efi.o] Error 1
make[3]: *** [arch/x86/platform/efi] Error 2

Signature · 1 line
Matt Fleming, Intel Open Source Technology Center
author maintainer
#6 lore raw

Re: [PATCH 3/3] efi: Make efi virtual runtime map passing more robust

Quoted 5 lines · Matt Fleming

You sunk my i386 battleship,

/home/build/git/efi/arch/x86/platform/efi/efi.c:824:24: error: ‘struct real_mode_header’ has no member named ‘trampoline_pgd’ make[4]: *** [arch/x86/platform/efi/efi.o] Error 1
make[3]: *** [arch/x86/platform/efi] Error 2

Right, this is a recurring issue with the trampoline_pgd. What do you
prefer: an #ifdef CONFIG_X86_64 or a separate function in efi_$(BITS).c?

Signature · 5 lines
Regards/Gruss,
    Boris.

Sent from a fat crate under my desk. Formatting is fine.
--
#7 lore raw

Re: [PATCH 3/3] efi: Make efi virtual runtime map passing more robust

Quoted 9 lines · Borislav Petkov
Quoted 5 lines · Matt Fleming

You sunk my i386 battleship,

/home/build/git/efi/arch/x86/platform/efi/efi.c:824:24: error: ‘struct real_mode_header’ has no member named ‘trampoline_pgd’ make[4]: *** [arch/x86/platform/efi/efi.o] Error 1
make[3]: *** [arch/x86/platform/efi] Error 2

Right, this is a recurring issue with the trampoline_pgd. What do you
prefer: an #ifdef CONFIG_X86_64 or a separate function in efi_$(BITS).c?

Couldn't we do the mapping in efi_setup_page_tables() if we pass __pa(new_memmap) and 1 << new_memmap_shift as arguments?

Signature · 1 line
Matt Fleming, Intel Open Source Technology Center
author maintainer
#8 lore raw

Re: [PATCH 3/3] efi: Make efi virtual runtime map passing more robust

Quoted 2 lines · Matt Fleming

Couldn't we do the mapping in efi_setup_page_tables() if we pass __pa(new_memmap) and 1 << new_memmap_shift as arguments?

Yep, not a bad idea. I'll give it a try.

Thanks.

Signature · 5 lines
Regards/Gruss,
    Boris.

Sent from a fat crate under my desk. Formatting is fine.
--
#9 lore raw

Re: [PATCH 3/3] efi: Make efi virtual runtime map passing more robust

In reply to Borislav Petkov
Quoted 9 lines · Borislav Petkov

From: Borislav Petkov <[email protected]>

Currently, running SetVirtualAddressMap() and passing the physical address of the virtual map array was working only by a lucky coincidence because the memory was present in the EFI page table too. Until Toshi went and booted this on a big HP box - the krealloc() manner of resizing the memmap we're doing did allocate from such physical addresses which were not mapped anymore and boom:

Quoted 6 lines

http://lkml.kernel.org/r/[email protected]

One way to take care of that issue is to reimplement the krealloc thing but with pages. We start with contiguous pages of order 1, i.e. 2 pages, and when we deplete that memory (shouldn't happen all that often but you know firmware) we realloc the next power-of-two pages.

How about firstly count the md numbers in the 1st loop, then get/roundup the total size, alloc the pages, map the mds one by one in another loop.

Thanks
Dave

author maintainer
#10 lore raw

Re: [PATCH 3/3] efi: Make efi virtual runtime map passing more robust

Quoted 3 lines · Dave Young

How about firstly count the md numbers in the 1st loop, then get/roundup the total size, alloc the pages, map the mds one by one in another loop.

No need as most systems would never need to reallocate. Even on Toshi's big system, the mds fit in less than one page - having a second one is an additional safeguard, i.e. 2 pages should be more than enough in the majority of the cases.

Signature · 5 lines
Regards/Gruss,
    Boris.

Sent from a fat crate under my desk. Formatting is fine.
--
Tested-by on the v1 series
#11 lore raw
In reply to Borislav Petkov
Quoted 21 lines · Borislav Petkov

From: Borislav Petkov <[email protected]>

Hi guys,

this is the result of Toshi and me debugging a #GP on one of his big HP boxes sporting UEFI. Each commit message should be self-explanatory so please look there.

This has more or less an RFC nature thus I'm sending it now to collect feedback. It is going to wait in the EFI queue anyway after the kexec stuff gets sorted out first.

Comments and suggestions as always are very much appreciated.

Thanks.

Borislav Petkov (3):
x86, ptdump: Add the functionality to dump an arbitrary pagetable
efi: Dump the EFI page table
efi: Make efi virtual runtime map passing more robust

Quoted 5 lines
 arch/x86/include/asm/pgtable.h |  3 +-
 arch/x86/mm/dump_pagetables.c  | 77 ++++++++++++++++++++++++++++--------------
 arch/x86/platform/efi/efi.c    | 68 ++++++++++++++++++++++++++++++++-----
 3 files changed, 112 insertions(+), 36 deletions(-)

Thanks for the quick turnaround! For the series:

Toshi