Skip to contentSkip to conversation

hw/block/pflash_cfi01: Limit maximum flash size to 256 MiB

PATCH In review Philippe Mathieu-Daudé 3 participants 7 messages May 25, 2020 – Jun 4, 2020

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 hw/block/pflash_cfi01: Limit maximum flash size to 256 MiB

As of this commit, the biggest CFI01 NOR flash documented is the Micron PC28F00BP33EF. Its size is 2 GiB (256 MiB).

Actually this "2Gb device employs a virtual chip enable feature, which combines two 1Gb die with a common chip enable".

Since we do not want to model unrealistic hardware, cap the current model to this maximum. At least we have a datasheet to refer.

If a bigger flash is provided, the user get this warning:

qemu-system-aarch64: Initialization of device cfi.pflash01 failed: Maximum supported CFI flash size is 16 MiB.

Note, the sbsa-ref ARM machine introduced in commit 64580903c2b already uses a pair of 256 MiB flash devices.

---
 hw/block/pflash_cfi01.c | 10 ++++++++++
 1 file changed, 10 insertions(+)
hw/block/pflash_cfi01.c +10 -0
index 11922c0f96..40f145dde7 100644
@@ -37,6 +37,8 @@
  */
 
 #include "qemu/osdep.h"
+#include "qemu/units.h"
+#include "qemu/cutils.h"
 #include "hw/block/block.h"
 #include "hw/block/flash.h"
 #include "hw/qdev-properties.h"
@@ -68,6 +70,8 @@ do {                                                        \
 #define PFLASH_BE          0
 #define PFLASH_SECURE      1
 
+#define PFLASH_SIZE_MAX     (256 * MiB) /* Micron PC28F00BP33EF */
+
 struct PFlashCFI01 {
     /*< private >*/
     SysBusDevice parent_obj;
@@ -717,6 +721,12 @@ static void pflash_cfi01_realize(DeviceState *dev, Error **errp)
     }
 
     total_len = pfl->sector_len * pfl->nb_blocs;
+    if (total_len > PFLASH_SIZE_MAX) {
+        char *maxsz = size_to_str(PFLASH_SIZE_MAX);
+        error_setg(errp, "Maximum supported CFI flash size is %s.", maxsz);
+        g_free(maxsz);
+        return;
+    }
 
     /* These are only used to expose the parameters of each device
      * in the cfi_table[].
Signature · 1 line
2.21.3
Reviewed-by on v1 1/1
#1 lore raw
Quoted 2 lines · Philippe Mathieu-Daudé

As of this commit, the biggest CFI01 NOR flash documented is the Micron PC28F00BP33EF. Its size is 2 GiB (256 MiB).

I don't understand what "2 GiB (256 MiB)" means; please clarify.

Quoted 46 lines

Actually this "2Gb device employs a virtual chip enable feature, which combines two 1Gb die with a common chip enable".

Since we do not want to model unrealistic hardware, cap the current model to this maximum. At least we have a datasheet to refer.

If a bigger flash is provided, the user get this warning:

qemu-system-aarch64: Initialization of device cfi.pflash01 failed: Maximum supported CFI flash size is 16 MiB.

Note, the sbsa-ref ARM machine introduced in commit 64580903c2b already uses a pair of 256 MiB flash devices.

---
 hw/block/pflash_cfi01.c | 10 ++++++++++
 1 file changed, 10 insertions(+)
hw/block/pflash_cfi01.c +4 -0
index 11922c0f96..40f145dde7 100644
@@ -37,6 +37,8 @@
  */
 
 #include "qemu/osdep.h"
+#include "qemu/units.h"
+#include "qemu/cutils.h"
 #include "hw/block/block.h"
 #include "hw/block/flash.h"
 #include "hw/qdev-properties.h"
@@ -68,6 +70,8 @@ do {                                                        \
 #define PFLASH_BE          0
 #define PFLASH_SECURE      1
 
+#define PFLASH_SIZE_MAX     (256 * MiB) /* Micron PC28F00BP33EF */
+
 struct PFlashCFI01 {
     /*< private >*/
     SysBusDevice parent_obj;
@@ -717,6 +721,12 @@ static void pflash_cfi01_realize(DeviceState *dev, Error **errp)
     }
 

hw/block/pflash_cfi01.c static void pflash_cfi01_realize(DeviceState *dev, Error **errp) · line 732

Quoted 10 lines
hw/block/pflash_cfi01.c +6 -0
     total_len = pfl->sector_len * pfl->nb_blocs;
+    if (total_len > PFLASH_SIZE_MAX) {
+        char *maxsz = size_to_str(PFLASH_SIZE_MAX);
+        error_setg(errp, "Maximum supported CFI flash size is %s.", maxsz);
+        g_free(maxsz);
+        return;
+    }
 
     /* These are only used to expose the parameters of each device
      * in the cfi_table[].

I'm unsure how strong the argument is, "we do not want to model unrealistic hardware", but I do think the patch does what it says on the tin (modulo the one part in the commit message that I don't understand).

With the commit message clarified

(I expect & hope people will provide better reviews than mine.)

Thanks
Laszlo

#2 lore raw
In reply to Philippe Mathieu-Daudé
Quoted 11 lines · Philippe Mathieu-Daudé

As of this commit, the biggest CFI01 NOR flash documented is the Micron PC28F00BP33EF. Its size is 2 GiB (256 MiB).

Actually this "2Gb device employs a virtual chip enable feature, which combines two 1Gb die with a common chip enable".

Since we do not want to model unrealistic hardware, cap the current model to this maximum. At least we have a datasheet to refer.

Quoted 6 lines

If a bigger flash is provided, the user get this warning:

qemu-system-aarch64: Initialization of device cfi.pflash01 failed: Maximum supported CFI flash size is 16 MiB.

Note, the sbsa-ref ARM machine introduced in commit 64580903c2b already uses a pair of 256 MiB flash devices.

What problem is this check solving? Is there some implementation imposed limitation or a limit within the flash header format that means larger sizes don't work? If so, what's the restriction?

thanks
-- PMM

#3 lore raw

Hi Peter,

Quoted 4 lines · Peter Maydell
Quoted 3 lines · Philippe Mathieu-Daudé

As of this commit, the biggest CFI01 NOR flash documented is the Micron PC28F00BP33EF. Its size is 2 GiB (256 MiB).

Oops, read as "2 Gbit (256 Mbyte)".

Quoted 11 lines
Quoted 11 lines

Actually this "2Gb device employs a virtual chip enable feature, which combines two 1Gb die with a common chip enable".

Since we do not want to model unrealistic hardware, cap the current model to this maximum. At least we have a datasheet to refer.

If a bigger flash is provided, the user get this warning:

qemu-system-aarch64: Initialization of device cfi.pflash01 failed: Maximum supported CFI flash size is 16 MiB.

Also here read "256 MiB" (I capped to 16MiB to test the patch).

Quoted 7 lines
Quoted 3 lines

Note, the sbsa-ref ARM machine introduced in commit 64580903c2b already uses a pair of 256 MiB flash devices.

What problem is this check solving? Is there some implementation imposed limitation or a limit within the flash header format that means larger sizes don't work? If so, what's the restriction?

I am not confident maintaining virtual device with no specifications. If someone come with a datasheet for a pflash > 256MB then we can add another commit to relax this restriction.
If someone is forced to use a >256MB and such hardware does not exist, I'd rather have a document in docs/specs/pflash_cfi01.rst describing the CFI fields.

IOW this is not a hardware limitation, but a maintenance protection.

I am worried with the recent EDK2 activity with the SBSA-ref, and I don't want to give false hope to developers that they can create any kind of pflash with the current device model.

If I reword this better in the commit description, is my argument acceptable?

Thanks,

Phil.

Quoted 4 lines

thanks
-- PMM

#4 lore raw
Quoted 6 lines · Philippe Mathieu-Daudé
Quoted 3 lines · Peter Maydell

What problem is this check solving? Is there some implementation imposed limitation or a limit within the flash header format that means larger sizes don't work? If so, what's the restriction?

I am not confident maintaining virtual device with no specifications.

This isn't a virtual device, is it? The CFI spec exists and defines what these flash devices behave like.

Quoted 6 lines

If someone come with a datasheet for a pflash > 256MB then we can add another commit to relax this restriction.
If someone is forced to use a >256MB and such hardware does not exist, I'd rather have a document in docs/specs/pflash_cfi01.rst describing the CFI fields.

Quoted 8 lines

IOW this is not a hardware limitation, but a maintenance protection.

I am worried with the recent EDK2 activity with the SBSA-ref, and I don't want to give false hope to developers that they can create any kind of pflash with the current device model.

If I reword this better in the commit description, is my argument acceptable?

Not really; I think we should know what we're limiting against. Currently you're checking total_len, but this is just sector_len * nb_blocs, so if there's a problem with silly large values then it's probably actually a problem with one of those being over-sized which would still show up even if the total_len was less than 256MB.
(I suspect the underlying limit here is what the cfi_table entries 0x2D..0x30 impose on blocks_per_device and sector_len_per_device.)

thanks
-- PMM

#5 lore raw
Quoted 26 lines · Peter Maydell
Quoted 6 lines · Philippe Mathieu-Daudé
Quoted 3 lines · Peter Maydell

What problem is this check solving? Is there some implementation imposed limitation or a limit within the flash header format that means larger sizes don't work? If so, what's the restriction?

I am not confident maintaining virtual device with no specifications.

This isn't a virtual device, is it? The CFI spec exists and defines what these flash devices behave like.

Quoted 6 lines

If someone come with a datasheet for a pflash > 256MB then we can add another commit to relax this restriction.
If someone is forced to use a >256MB and such hardware does not exist, I'd rather have a document in docs/specs/pflash_cfi01.rst describing the CFI fields.

Quoted 8 lines

IOW this is not a hardware limitation, but a maintenance protection.

I am worried with the recent EDK2 activity with the SBSA-ref, and I don't want to give false hope to developers that they can create any kind of pflash with the current device model.

If I reword this better in the commit description, is my argument acceptable?

Quoted 7 lines

Not really; I think we should know what we're limiting against. Currently you're checking total_len, but this is just sector_len * nb_blocs, so if there's a problem with silly large values then it's probably actually a problem with one of those being over-sized which would still show up even if the total_len was less than 256MB.
(I suspect the underlying limit here is what the cfi_table entries 0x2D..0x30 impose on blocks_per_device and sector_len_per_device.)

What I'm working on is a whitelist of the few models our machines really use, but it is taking time. Meanwhile I wanted to at least limit the total size.

I'll try to finish my whitelist effort before someone try to use a

Quoted 6 lines

256MB flash.

thanks
-- PMM

#6 lore raw
Quoted 10 lines · Philippe Mathieu-Daudé
Quoted 7 lines · Peter Maydell

Not really; I think we should know what we're limiting against. Currently you're checking total_len, but this is just sector_len * nb_blocs, so if there's a problem with silly large values then it's probably actually a problem with one of those being over-sized which would still show up even if the total_len was less than 256MB.
(I suspect the underlying limit here is what the cfi_table entries 0x2D..0x30 impose on blocks_per_device and sector_len_per_device.)

Quoted 3 lines

What I'm working on is a whitelist of the few models our machines really use, but it is taking time. Meanwhile I wanted to at least limit the total size.

I don't see what we would be whitelisting, though. The only way to create a flash device is from hand-written C code in the board model. If a new board model does something weird we can catch that in code review. Sanity checks on whether the properties supplied by the board code make sense might be useful; randomly saying "you can't have a flash device unless it's one we've seen before" makes less sense to me, because it just means we'll end up adding to the whitelist every time.

thanks
-- PMM