Skip to contentSkip to conversation

efi: get_memory_map: add sufficient slack for memory descriptors

PATCH New Ard Biesheuvel 4 participants 13 messages Feb 12, 2015 – Feb 13, 2015 EFI

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 efi: get_memory_map: add sufficient slack for memory descriptors

As it turns out, when allocating room for the UEFI memory map using UEFI's AllocatePool (), it may result in two new memory map entries being created, for instance, when using Tianocore's preallocated region feature. For example, the following region

0x00005ead5000-0x00005ebfffff [Conventional Memory| | | | | |WB|WT|WC|UC]

may be split like this

0x00005ead5000-0x00005eae2fff [Conventional Memory|   |  |  |  |   |WB|WT|WC|UC]
0x00005eae3000-0x00005eae4fff [Loader Data        |   |  |  |  |   |WB|WT|WC|UC]
0x00005eae5000-0x00005ebfffff [Conventional Memory|   |  |  |  |   |WB|WT|WC|UC]

if the preallocated Loader Data region was chosen to be right in the middle of the original free space.

After patch d1a8d66b9177 ("efi/libstub: Call get_memory_map() to obtain map and desc sizes"), this is not being dealt with correctly anymore, as the existing logic to allocate room for a single additional entry has become insufficient.

So instead, add room for two additional entries instead.

Fixes d1a8d66b9177 ("efi/libstub: Call get_memory_map() to obtain map and desc sizes")
---
 drivers/firmware/efi/libstub/efi-stub-helper.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
drivers/firmware/efi/libstub/efi-stub-helper.c +3 -3
index af5d63c7cc53..ca0b07ed3b14 100644
@@ -84,10 +84,10 @@ efi_status_t efi_get_memory_map(efi_system_table_t *sys_table_arg,
 		return EFI_LOAD_ERROR;
 
 	/*
-	 * Add an additional efi_memory_desc_t because we're doing an
-	 * allocation which may be in a new descriptor region.
+	 * Add room for two additional efi_memory_desc_t entries because we're
+	 * doing an allocation which may be in a new descriptor region.
 	 */
-	*map_size += *desc_size;
+	*map_size += *desc_size * 2;
 	status = efi_call_early(allocate_pool, EFI_LOADER_DATA,
 				*map_size, (void **)&m);
 	if (status != EFI_SUCCESS)
Signature · 1 line
1.8.3.2
#1 lore raw

PATCH 1/1 efi: get_memory_map: add sufficient slack for memory descriptors

Quoted 48 lines · Ard Biesheuvel

As it turns out, when allocating room for the UEFI memory map using UEFI's AllocatePool (), it may result in two new memory map entries being created, for instance, when using Tianocore's preallocated region feature. For example, the following region

0x00005ead5000-0x00005ebfffff [Conventional Memory| | | | | |WB|WT|WC|UC]

may be split like this

0x00005ead5000-0x00005eae2fff [Conventional Memory|   |  |  |  |   |WB|WT|WC|UC]
0x00005eae3000-0x00005eae4fff [Loader Data        |   |  |  |  |   |WB|WT|WC|UC]
0x00005eae5000-0x00005ebfffff [Conventional Memory|   |  |  |  |   |WB|WT|WC|UC]

if the preallocated Loader Data region was chosen to be right in the middle of the original free space.

After patch d1a8d66b9177 ("efi/libstub: Call get_memory_map() to obtain map and desc sizes"), this is not being dealt with correctly anymore, as the existing logic to allocate room for a single additional entry has become insufficient.

So instead, add room for two additional entries instead.

Fixes d1a8d66b9177 ("efi/libstub: Call get_memory_map() to obtain map and desc sizes")
---
 drivers/firmware/efi/libstub/efi-stub-helper.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
drivers/firmware/efi/libstub/efi-stub-helper.c +3 -4
index af5d63c7cc53..ca0b07ed3b14 100644
@@ -84,10 +84,10 @@ efi_status_t efi_get_memory_map(efi_system_table_t *sys_table_arg,
                return EFI_LOAD_ERROR;
 
        /*
-        * Add an additional efi_memory_desc_t because we're doing an
-        * allocation which may be in a new descriptor region.
+        * Add room for two additional efi_memory_desc_t entries because we're
+        * doing an allocation which may be in a new descriptor region.
         */
-       *map_size += *desc_size;
+       *map_size += *desc_size * 2;
        status = efi_call_early(allocate_pool, EFI_LOADER_DATA,
                                *map_size, (void **)&m);
        if (status != EFI_SUCCESS)
--
Quoted 2 lines

1.8.3.2

#2 lore raw

PATCH 1/1 efi: get_memory_map: add sufficient slack for memory descriptors

In reply to Ard Biesheuvel
Quoted 34 lines · Ard Biesheuvel

As it turns out, when allocating room for the UEFI memory map using UEFI's AllocatePool (), it may result in two new memory map entries being created, for instance, when using Tianocore's preallocated region feature. For example, the following region

0x00005ead5000-0x00005ebfffff [Conventional Memory| | | | | |WB|WT|WC|UC]

may be split like this

0x00005ead5000-0x00005eae2fff [Conventional Memory|   |  |  |  |   |WB|WT|WC|UC]
0x00005eae3000-0x00005eae4fff [Loader Data        |   |  |  |  |   |WB|WT|WC|UC]
0x00005eae5000-0x00005ebfffff [Conventional Memory|   |  |  |  |   |WB|WT|WC|UC]

if the preallocated Loader Data region was chosen to be right in the middle of the original free space.

After patch d1a8d66b9177 ("efi/libstub: Call get_memory_map() to obtain map and desc sizes"), this is not being dealt with correctly anymore, as the existing logic to allocate room for a single additional entry has become insufficient.

So instead, add room for two additional entries instead.

Fixes d1a8d66b9177 ("efi/libstub: Call get_memory_map() to obtain map and desc sizes")
---
 drivers/firmware/efi/libstub/efi-stub-helper.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
drivers/firmware/efi/libstub/efi-stub-helper.c +0 -0
index af5d63c7cc53..ca0b07ed3b14 100644

drivers/firmware/efi/libstub/efi-stub-helper.c efi_status_t efi_get_memory_map(efi_system_table_t *sys_table_arg, · line 88

Quoted 7 lines
drivers/firmware/efi/libstub/efi-stub-helper.c +2 -2
@@ -84,10 +84,10 @@ efi_status_t efi_get_memory_map(efi_system_table_t *sys_table_arg,
 		return EFI_LOAD_ERROR;
 
 	/*
-	 * Add an additional efi_memory_desc_t because we're doing an
-	 * allocation which may be in a new descriptor region.
+	 * Add room for two additional efi_memory_desc_t entries because we're
+	 * doing an allocation which may be in a new descriptor region.

It might be worth noting that a existing regions can be split/reorganised here, otherwise it's a little difficult to deduce from the comment why to regions are needed.

drivers/firmware/efi/libstub/efi-stub-helper.c efi_status_t efi_get_memory_map(efi_system_table_t *sys_table_arg, · line 90

Quoted 3 lines
drivers/firmware/efi/libstub/efi-stub-helper.c +1 -1
 	 */
-	*map_size += *desc_size;
+	*map_size += *desc_size * 2;

Can we forsee any cases where we might need more than two additional descs? Is it perhaps adding a little more slack now?

Otherwise this looks fine to me.

Thanks,
Mark.

Quoted 7 lines
drivers/firmware/efi/libstub/efi-stub-helper.c +0 -0
 	status = efi_call_early(allocate_pool, EFI_LOADER_DATA,
 				*map_size, (void **)&m);
 	if (status != EFI_SUCCESS)
Signature · 1 line
1.8.3.2
#3 lore raw

PATCH 1/1 efi: get_memory_map: add sufficient slack for memory descriptors

Quoted 37 lines · Mark Rutland
Quoted 36 lines · Ard Biesheuvel

As it turns out, when allocating room for the UEFI memory map using UEFI's AllocatePool (), it may result in two new memory map entries being created, for instance, when using Tianocore's preallocated region feature. For example, the following region

0x00005ead5000-0x00005ebfffff [Conventional Memory| | | | | |WB|WT|WC|UC]

may be split like this

0x00005ead5000-0x00005eae2fff [Conventional Memory|   |  |  |  |   |WB|WT|WC|UC]
0x00005eae3000-0x00005eae4fff [Loader Data        |   |  |  |  |   |WB|WT|WC|UC]
0x00005eae5000-0x00005ebfffff [Conventional Memory|   |  |  |  |   |WB|WT|WC|UC]

if the preallocated Loader Data region was chosen to be right in the middle of the original free space.

After patch d1a8d66b9177 ("efi/libstub: Call get_memory_map() to obtain map and desc sizes"), this is not being dealt with correctly anymore, as the existing logic to allocate room for a single additional entry has become insufficient.

So instead, add room for two additional entries instead.

Fixes d1a8d66b9177 ("efi/libstub: Call get_memory_map() to obtain map and desc sizes")
---
 drivers/firmware/efi/libstub/efi-stub-helper.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
drivers/firmware/efi/libstub/efi-stub-helper.c +0 -0
index af5d63c7cc53..ca0b07ed3b14 100644
@@ -84,10 +84,10 @@ efi_status_t efi_get_memory_map(efi_system_table_t *sys_table_arg,
              return EFI_LOAD_ERROR;
 
Quoted 10 lines

drivers/firmware/efi/libstub/efi-stub-helper.c efi_status_t efi_get_memory_map(efi_system_table_t *sys_table_arg, · line 88

Quoted 5 lines
drivers/firmware/efi/libstub/efi-stub-helper.c +2 -2
      /*
-      * Add an additional efi_memory_desc_t because we're doing an
-      * allocation which may be in a new descriptor region.
+      * Add room for two additional efi_memory_desc_t entries because we're
+      * doing an allocation which may be in a new descriptor region.

It might be worth noting that a existing regions can be split/reorganised here, otherwise it's a little difficult to deduce from the comment why to regions are needed.

OK

Quoted 7 lines

drivers/firmware/efi/libstub/efi-stub-helper.c efi_status_t efi_get_memory_map(efi_system_table_t *sys_table_arg, · line 90

Quoted 3 lines
drivers/firmware/efi/libstub/efi-stub-helper.c +1 -1
       */
-     *map_size += *desc_size;
+     *map_size += *desc_size * 2;

Can we forsee any cases where we might need more than two additional descs? Is it perhaps adding a little more slack now?

I don't see how doing a single allocation could result in a single free region to be split into more than 1 occupied region + 2 free regions.
So no, I don't think it is ...

Signature · 1 line
Ard.
Quoted 12 lines

Otherwise this looks fine to me.

Thanks,
Mark.

Quoted 7 lines
drivers/firmware/efi/libstub/efi-stub-helper.c +0 -1
      status = efi_call_early(allocate_pool, EFI_LOADER_DATA,
                              *map_size, (void **)&m);
      if (status != EFI_SUCCESS)
--

1.8.3.2

#4 lore raw

PATCH 1/1 efi: get_memory_map: add sufficient slack for memory descriptors

Quoted 55 lines · Ard Biesheuvel
Quoted 37 lines · Mark Rutland
Quoted 36 lines · Ard Biesheuvel

As it turns out, when allocating room for the UEFI memory map using UEFI's AllocatePool (), it may result in two new memory map entries being created, for instance, when using Tianocore's preallocated region feature. For example, the following region

0x00005ead5000-0x00005ebfffff [Conventional Memory| | | | | |WB|WT|WC|UC]

may be split like this

0x00005ead5000-0x00005eae2fff [Conventional Memory|   |  |  |  |   |WB|WT|WC|UC]
0x00005eae3000-0x00005eae4fff [Loader Data        |   |  |  |  |   |WB|WT|WC|UC]
0x00005eae5000-0x00005ebfffff [Conventional Memory|   |  |  |  |   |WB|WT|WC|UC]

if the preallocated Loader Data region was chosen to be right in the middle of the original free space.

After patch d1a8d66b9177 ("efi/libstub: Call get_memory_map() to obtain map and desc sizes"), this is not being dealt with correctly anymore, as the existing logic to allocate room for a single additional entry has become insufficient.

So instead, add room for two additional entries instead.

Fixes d1a8d66b9177 ("efi/libstub: Call get_memory_map() to obtain map and desc sizes")
---
 drivers/firmware/efi/libstub/efi-stub-helper.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
drivers/firmware/efi/libstub/efi-stub-helper.c +0 -0
index af5d63c7cc53..ca0b07ed3b14 100644
@@ -84,10 +84,10 @@ efi_status_t efi_get_memory_map(efi_system_table_t *sys_table_arg,
              return EFI_LOAD_ERROR;
 
Quoted 10 lines

drivers/firmware/efi/libstub/efi-stub-helper.c efi_status_t efi_get_memory_map(efi_system_table_t *sys_table_arg, · line 88

Quoted 5 lines
drivers/firmware/efi/libstub/efi-stub-helper.c +2 -2
      /*
-      * Add an additional efi_memory_desc_t because we're doing an
-      * allocation which may be in a new descriptor region.
+      * Add room for two additional efi_memory_desc_t entries because we're
+      * doing an allocation which may be in a new descriptor region.

It might be worth noting that a existing regions can be split/reorganised here, otherwise it's a little difficult to deduce from the comment why to regions are needed.

OK

Quoted 4 lines
Quoted 3 lines
drivers/firmware/efi/libstub/efi-stub-helper.c +1 -1
       */
-     *map_size += *desc_size;
+     *map_size += *desc_size * 2;
Quoted 8 lines
Quoted 3 lines

Can we forsee any cases where we might need more than two additional descs? Is it perhaps adding a little more slack now?

I don't see how doing a single allocation could result in a single free region to be split into more than 1 occupied region + 2 free regions.
So no, I don't think it is ...

Fair enough. If we're surprised later we can always fix things up.

With the comment fix up:

Thanks,
Mark.

#5 lore raw

PATCH 1/1 efi: get_memory_map: add sufficient slack for memory descriptors

In reply to Ard Biesheuvel
Quoted 5 lines · Ard Biesheuvel

I don't see how doing a single allocation could result in a single free region to be split into more than 1 occupied region + 2 free regions.
So no, I don't think it is ...

I don't think that's a guarantee we can make, nor is it something we should rely upon.

Please explain the user-visible failure that this patch fixes. Does your machine refuse to boot? Why is the 'goto again' loop insufficient in handling this scenario?

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

PATCH 1/1 efi: get_memory_map: add sufficient slack for memory descriptors

Quoted 7 lines · Matt Fleming
Quoted 5 lines · Ard Biesheuvel

I don't see how doing a single allocation could result in a single free region to be split into more than 1 occupied region + 2 free regions.
So no, I don't think it is ...

Quoted 5 lines

I don't think that's a guarantee we can make, nor is it something we should rely upon.

Please explain the user-visible failure that this patch fixes. Does your machine refuse to boot?

I am running UEFI under QEMU and Xen primarily at the moment, and experimenting with various build options in Tianocore, One of the options is preallocating and freeing blocks of various memory types, in a way that should result in the final number of distinct regions to be much lower. It could result however in a free memory region to be carved up in three instead of two, and that is a failure I have seen occur.

Quoted 3 lines

Why is the 'goto again' loop insufficient in
handling this scenario?

Yes, that should solve it as well, so if you prefer I reinstate that, I can respin the patch. There is a theoretical possibility that it would take more than just one more iteration, but that is highly unlikely and it should still always complete.

#7 lore raw

PATCH 1/1 efi: get_memory_map: add sufficient slack for memory descriptors

Quoted 14 lines · Ard Biesheuvel
Quoted 7 lines · Matt Fleming
Quoted 5 lines · Ard Biesheuvel

I don't see how doing a single allocation could result in a single free region to be split into more than 1 occupied region + 2 free regions.
So no, I don't think it is ...

Quoted 5 lines

I don't think that's a guarantee we can make, nor is it something we should rely upon.

Please explain the user-visible failure that this patch fixes. Does your machine refuse to boot?

Quoted 7 lines

I am running UEFI under QEMU and Xen primarily at the moment, and experimenting with various build options in Tianocore, One of the options is preallocating and freeing blocks of various memory types, in a way that should result in the final number of distinct regions to be much lower. It could result however in a free memory region to be carved up in three instead of two, and that is a failure I have seen occur.

The simple answer is that the machine will fail to boot, beause the efi_get_memory_map helper will give up after one go, and propagate the error. The arm-stub will give up when the error is encountered.

Quoted 8 lines
Quoted 3 lines

Why is the 'goto again' loop insufficient in
handling this scenario?

Yes, that should solve it as well, so if you prefer I reinstate that, I can respin the patch. There is a theoretical possibility that it would take more than just one more iteration, but that is highly unlikely and it should still always complete.

Please reinstate the loop. It will make this far less fragile.

Thanks,
Mark.

#8 lore raw

PATCH 1/1 efi: get_memory_map: add sufficient slack for memory descriptors

Quoted 23 lines · Mark Rutland
Quoted 14 lines · Ard Biesheuvel
Quoted 7 lines · Matt Fleming
Quoted 5 lines · Ard Biesheuvel

I don't see how doing a single allocation could result in a single free region to be split into more than 1 occupied region + 2 free regions.
So no, I don't think it is ...

Quoted 5 lines

I don't think that's a guarantee we can make, nor is it something we should rely upon.

Please explain the user-visible failure that this patch fixes. Does your machine refuse to boot?

Quoted 7 lines

I am running UEFI under QEMU and Xen primarily at the moment, and experimenting with various build options in Tianocore, One of the options is preallocating and freeing blocks of various memory types, in a way that should result in the final number of distinct regions to be much lower. It could result however in a free memory region to be carved up in three instead of two, and that is a failure I have seen occur.

Quoted 4 lines

The simple answer is that the machine will fail to boot, beause the efi_get_memory_map helper will give up after one go, and propagate the error. The arm-stub will give up when the error is encountered.

Indeed.

Quoted 11 lines
Quoted 8 lines
Quoted 3 lines

Why is the 'goto again' loop insufficient in
handling this scenario?

Yes, that should solve it as well, so if you prefer I reinstate that, I can respin the patch. There is a theoretical possibility that it would take more than just one more iteration, but that is highly unlikely and it should still always complete.

Please reinstate the loop. It will make this far less fragile.

Actually, looking again at the original patch, it appears that my analysis was incorrect regarding the possibility that the loop would never terminate. The only thing that could happen if desc_size > sizeof(efi_memory_desc_t) is that you need two iterations instead of one to get a pool allocation that is of sufficient size. So perhaps it is better to just revert the patch.

My apologies for the hassle.

Signature · 1 line
Ard.
#9 lore raw

PATCH 1/1 efi: get_memory_map: add sufficient slack for memory descriptors

Quoted 9 lines · Ard Biesheuvel

Actually, looking again at the original patch, it appears that my analysis was incorrect regarding the possibility that the loop would never terminate. The only thing that could happen if desc_size > sizeof(efi_memory_desc_t) is that you need two iterations instead of one to get a pool allocation that is of sufficient size. So perhaps it is better to just revert the patch.

My apologies for the hassle.

This is what I've got queued up,

---

Quoted 1 line

From 3f281b98ffc99e604a3988aa93304a3a591eeeb5 Mon Sep 17 00:00:00 2001

From: Matt Fleming <[email protected]>
Date: Fri, 13 Feb 2015 15:46:56 +0000
Subject: [PATCH] Revert "efi/libstub: Call get_memory_map() to obtain map and
desc sizes"

This reverts commit d1a8d66b9177105e898e73716f97eb61842c457a.

Ard reported a boot failure when running UEFI under Qemu and Xen and experimenting with various Tianocore build options,

"As it turns out, when allocating room for the UEFI memory map using
UEFI's AllocatePool (), it may result in two new memory map entries being created, for instance, when using Tianocore's preallocated region feature. For example, the following region

0x00005ead5000-0x00005ebfffff [Conventional Memory| | | | | |WB|WT|WC|UC]

may be split like this

  0x00005ead5000-0x00005eae2fff [Conventional Memory|   |  |  |  |  |WB|WT|WC|UC]
  0x00005eae3000-0x00005eae4fff [Loader Data        |   |  |  |  |  |WB|WT|WC|UC]
  0x00005eae5000-0x00005ebfffff [Conventional Memory|   |  |  |  |  |WB|WT|WC|UC]

if the preallocated Loader Data region was chosen to be right in the middle of the original free space.

After patch d1a8d66b9177 ("efi/libstub: Call get_memory_map() to obtain map and desc sizes"), this is not being dealt with correctly anymore, as the existing logic to allocate room for a single additional entry has become insufficient."

Mark requested to reinstate the old loop we had before commit d1a8d66b9177, which grows the memory map buffer until it's big enough to hold the EFI memory map.

Cc Ard Biesheuvel <[email protected]>
Cc Mark Rutland <[email protected]>
---
 drivers/firmware/efi/libstub/efi-stub-helper.c | 16 ++++++----------
 1 file changed, 6 insertions(+), 10 deletions(-)
drivers/firmware/efi/libstub/efi-stub-helper.c +6 -10
index d073e3946383..9bd9fbb5bea8 100644
@@ -66,29 +66,25 @@ efi_status_t efi_get_memory_map(efi_system_table_t *sys_table_arg,
 	unsigned long key;
 	u32 desc_version;
 
-	*map_size = 0;
-	*desc_size = 0;
-	key = 0;
-	status = efi_call_early(get_memory_map, map_size, NULL,
-				&key, desc_size, &desc_version);
-	if (status != EFI_BUFFER_TOO_SMALL)
-		return EFI_LOAD_ERROR;
-
+	*map_size = sizeof(*m) * 32;
+again:
 	/*
 	 * Add an additional efi_memory_desc_t because we're doing an
 	 * allocation which may be in a new descriptor region.
 	 */
-	*map_size += *desc_size;
+	*map_size += sizeof(*m);
 	status = efi_call_early(allocate_pool, EFI_LOADER_DATA,
 				*map_size, (void **)&m);
 	if (status != EFI_SUCCESS)
 		goto fail;
 
+	*desc_size = 0;
+	key = 0;
 	status = efi_call_early(get_memory_map, map_size, m,
 				&key, desc_size, &desc_version);
 	if (status == EFI_BUFFER_TOO_SMALL) {
 		efi_call_early(free_pool, m);
-		return EFI_LOAD_ERROR;
+		goto again;
 	}
 
 	if (status != EFI_SUCCESS)
Signature · 4 lines
1.9.3

-- 
Matt Fleming, Intel Open Source Technology Center
#10 lore raw

PATCH 1/1 efi: get_memory_map: add sufficient slack for memory descriptors

Quoted 12 lines · Matt Fleming
Quoted 10 lines · Ard Biesheuvel

Actually, looking again at the original patch, it appears that my analysis was incorrect regarding the possibility that the loop would never terminate. The only thing that could happen if desc_size > sizeof(efi_memory_desc_t) is that you need two iterations instead of one to get a pool allocation that is of sufficient size. So perhaps it is better to just revert the patch.

Quoted 4 lines
Quoted 1 line

My apologies for the hassle.

This is what I've got queued up,

Thanks Matt

Quoted 90 lines

---

From 3f281b98ffc99e604a3988aa93304a3a591eeeb5 Mon Sep 17 00:00:00 2001
From: Matt Fleming <[email protected]>
Date: Fri, 13 Feb 2015 15:46:56 +0000
Subject: [PATCH] Revert "efi/libstub: Call get_memory_map() to obtain map and desc sizes"

This reverts commit d1a8d66b9177105e898e73716f97eb61842c457a.

Ard reported a boot failure when running UEFI under Qemu and Xen and experimenting with various Tianocore build options,

"As it turns out, when allocating room for the UEFI memory map using
UEFI's AllocatePool (), it may result in two new memory map entries being created, for instance, when using Tianocore's preallocated region feature. For example, the following region

0x00005ead5000-0x00005ebfffff [Conventional Memory| | | | | |WB|WT|WC|UC]

may be split like this

 0x00005ead5000-0x00005eae2fff [Conventional Memory|   |  |  |  |  |WB|WT|WC|UC]
 0x00005eae3000-0x00005eae4fff [Loader Data        |   |  |  |  |  |WB|WT|WC|UC]
 0x00005eae5000-0x00005ebfffff [Conventional Memory|   |  |  |  |  |WB|WT|WC|UC]

if the preallocated Loader Data region was chosen to be right in the middle of the original free space.

After patch d1a8d66b9177 ("efi/libstub: Call get_memory_map() to obtain map and desc sizes"), this is not being dealt with correctly anymore, as the existing logic to allocate room for a single additional entry has become insufficient."

Mark requested to reinstate the old loop we had before commit d1a8d66b9177, which grows the memory map buffer until it's big enough to hold the EFI memory map.

Cc Ard Biesheuvel <[email protected]>
Cc Mark Rutland <[email protected]>

---
drivers/firmware/efi/libstub/efi-stub-helper.c | 16 ++++++----------

1 file changed, 6 insertions(+), 10 deletions(-)
drivers/firmware/efi/libstub/efi-stub-helper.c +6 -10
index d073e3946383..9bd9fbb5bea8 100644
@@ -66,29 +66,25 @@ efi_status_t efi_get_memory_map(efi_system_table_t *sys_table_arg,
   unsigned long key;
   u32 desc_version;
 
-    *map_size = 0;
-    *desc_size = 0;
-    key = 0;
-    status = efi_call_early(get_memory_map, map_size, NULL,
-                &key, desc_size, &desc_version);
-    if (status != EFI_BUFFER_TOO_SMALL)
-        return EFI_LOAD_ERROR;
-
+    *map_size = sizeof(*m) * 32;
+again:
   /*
    * Add an additional efi_memory_desc_t because we're doing an
    * allocation which may be in a new descriptor region.
    */
-    *map_size += *desc_size;
+    *map_size += sizeof(*m);
   status = efi_call_early(allocate_pool, EFI_LOADER_DATA,
               *map_size, (void **)&m);
   if (status != EFI_SUCCESS)
       goto fail;
 
+    *desc_size = 0;
+    key = 0;
   status = efi_call_early(get_memory_map, map_size, m,
               &key, desc_size, &desc_version);
   if (status == EFI_BUFFER_TOO_SMALL) {
       efi_call_early(free_pool, m);
-        return EFI_LOAD_ERROR;
+        goto again;
   }
 
   if (status != EFI_SUCCESS)
Signature · 4 lines
1.9.3

-- 
Matt Fleming, Intel Open Source Technology Center
#11 lore raw

PATCH 1/1 efi: get_memory_map: add sufficient slack for memory descriptors

In reply to Matt Fleming
Quoted 9 lines · Matt Fleming
Quoted 8 lines · Ard Biesheuvel

Actually, looking again at the original patch, it appears that my analysis was incorrect regarding the possibility that the loop would never terminate. The only thing that could happen if desc_size > sizeof(efi_memory_desc_t) is that you need two iterations instead of one to get a pool allocation that is of sufficient size. So perhaps it is better to just revert the patch.

Quoted 3 lines
Quoted 1 line

My apologies for the hassle.

This is what I've got queued up,

Looks sane to me:

Thanks,
Mark.

Quoted 92 lines

---

From 3f281b98ffc99e604a3988aa93304a3a591eeeb5 Mon Sep 17 00:00:00 2001
From: Matt Fleming <[email protected]>
Date: Fri, 13 Feb 2015 15:46:56 +0000
Subject: [PATCH] Revert "efi/libstub: Call get_memory_map() to obtain map and
desc sizes"

This reverts commit d1a8d66b9177105e898e73716f97eb61842c457a.

Ard reported a boot failure when running UEFI under Qemu and Xen and experimenting with various Tianocore build options,

"As it turns out, when allocating room for the UEFI memory map using
UEFI's AllocatePool (), it may result in two new memory map entries being created, for instance, when using Tianocore's preallocated region feature. For example, the following region

0x00005ead5000-0x00005ebfffff [Conventional Memory| | | | | |WB|WT|WC|UC]

may be split like this

  0x00005ead5000-0x00005eae2fff [Conventional Memory|   |  |  |  |  |WB|WT|WC|UC]
  0x00005eae3000-0x00005eae4fff [Loader Data        |   |  |  |  |  |WB|WT|WC|UC]
  0x00005eae5000-0x00005ebfffff [Conventional Memory|   |  |  |  |  |WB|WT|WC|UC]

if the preallocated Loader Data region was chosen to be right in the middle of the original free space.

After patch d1a8d66b9177 ("efi/libstub: Call get_memory_map() to obtain map and desc sizes"), this is not being dealt with correctly anymore, as the existing logic to allocate room for a single additional entry has become insufficient."

Mark requested to reinstate the old loop we had before commit d1a8d66b9177, which grows the memory map buffer until it's big enough to hold the EFI memory map.

Cc Ard Biesheuvel <[email protected]>
Cc Mark Rutland <[email protected]>
---
 drivers/firmware/efi/libstub/efi-stub-helper.c | 16 ++++++----------
 1 file changed, 6 insertions(+), 10 deletions(-)
drivers/firmware/efi/libstub/efi-stub-helper.c +6 -10
index d073e3946383..9bd9fbb5bea8 100644
@@ -66,29 +66,25 @@ efi_status_t efi_get_memory_map(efi_system_table_t *sys_table_arg,
 	unsigned long key;
 	u32 desc_version;
 
-	*map_size = 0;
-	*desc_size = 0;
-	key = 0;
-	status = efi_call_early(get_memory_map, map_size, NULL,
-				&key, desc_size, &desc_version);
-	if (status != EFI_BUFFER_TOO_SMALL)
-		return EFI_LOAD_ERROR;
-
+	*map_size = sizeof(*m) * 32;
+again:
 	/*
 	 * Add an additional efi_memory_desc_t because we're doing an
 	 * allocation which may be in a new descriptor region.
 	 */
-	*map_size += *desc_size;
+	*map_size += sizeof(*m);
 	status = efi_call_early(allocate_pool, EFI_LOADER_DATA,
 				*map_size, (void **)&m);
 	if (status != EFI_SUCCESS)
 		goto fail;
 
+	*desc_size = 0;
+	key = 0;
 	status = efi_call_early(get_memory_map, map_size, m,
 				&key, desc_size, &desc_version);
 	if (status == EFI_BUFFER_TOO_SMALL) {
 		efi_call_early(free_pool, m);
-		return EFI_LOAD_ERROR;
+		goto again;
 	}
 
 	if (status != EFI_SUCCESS)
Signature · 4 lines
1.9.3

-- 
Matt Fleming, Intel Open Source Technology Center
#12 lore raw

PATCH 1/1 efi: get_memory_map: add sufficient slack for memory descriptors

In reply to Ard Biesheuvel
Quoted 43 lines · Ard Biesheuvel
Quoted 23 lines · Mark Rutland
Quoted 14 lines · Ard Biesheuvel
Quoted 7 lines · Matt Fleming
Quoted 5 lines · Ard Biesheuvel

I don't see how doing a single allocation could result in a single free region to be split into more than 1 occupied region + 2 free regions.
So no, I don't think it is ...

Quoted 5 lines

I don't think that's a guarantee we can make, nor is it something we should rely upon.

Please explain the user-visible failure that this patch fixes. Does your machine refuse to boot?

Quoted 7 lines

I am running UEFI under QEMU and Xen primarily at the moment, and experimenting with various build options in Tianocore, One of the options is preallocating and freeing blocks of various memory types, in a way that should result in the final number of distinct regions to be much lower. It could result however in a free memory region to be carved up in three instead of two, and that is a failure I have seen occur.

Quoted 4 lines

The simple answer is that the machine will fail to boot, beause the efi_get_memory_map helper will give up after one go, and propagate the error. The arm-stub will give up when the error is encountered.

Indeed.

Quoted 11 lines
Quoted 8 lines
Quoted 3 lines

Why is the 'goto again' loop insufficient in
handling this scenario?

Yes, that should solve it as well, so if you prefer I reinstate that, I can respin the patch. There is a theoretical possibility that it would take more than just one more iteration, but that is highly unlikely and it should still always complete.

Please reinstate the loop. It will make this far less fragile.

Quoted 6 lines

Actually, looking again at the original patch, it appears that my analysis was incorrect regarding the possibility that the loop would never terminate. The only thing that could happen if desc_size > sizeof(efi_memory_desc_t) is that you need two iterations instead of one to get a pool allocation that is of sufficient size. So perhaps it is better to just revert the patch.

In general we have no idea how many iterations will be required, depending on the layout and fragmentation of memory.

The important part is that with the original loop we will either manage to get a big enough allocation for the memory map or we will fail to allocate memory and we'll terminate.

Thanks,
Mark.