Skip to contentSkip to conversation

ARM: mvebu: don't dereference null bd pointer

PATCH New Chris Packham 2 participants 5 messages Oct 22, 2019 – Nov 14, 2019

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ARM: mvebu: don't dereference null bd pointer

As mentioned in doc/README.arm-relocation gd->bd is not available in dram_init() so we shouldn't attempt to access it.

---

 arch/arm/mach-mvebu/dram.c | 10 ----------
 1 file changed, 10 deletions(-)
arch/arm/mach-mvebu/dram.c +0 -10
index fa8c799a462e..ba8ebc62887f 100644
@@ -281,16 +281,6 @@ int dram_init(void)
 			size = MVEBU_SDRAM_SIZE_MAX;
 	}
 
-	for (; i < CONFIG_NR_DRAM_BANKS; i++) {
-		/* If above loop terminated prematurely, we need to set
-		 * remaining banks' start address & size as 0. Otherwise other
-		 * u-boot functions and Linux kernel gets wrong values which
-		 * could result in crash */
-		gd->bd->bi_dram[i].start = 0;
-		gd->bd->bi_dram[i].size = 0;
-	}
-
-
 	if (ecc_enabled())
 		dram_ecc_scrubbing();
 
Signature · 1 line
2.23.0
#1 lore raw

PATCH 1/1 ARM: mvebu: don't dereference null bd pointer

Hi Chris,

Quoted 17 lines · Chris Packham

As mentioned in doc/README.arm-relocation gd->bd is not available in dram_init() so we shouldn't attempt to access it.

---

  arch/arm/mach-mvebu/dram.c | 10 ----------
  1 file changed, 10 deletions(-)
arch/arm/mach-mvebu/dram.c +0 -0
index fa8c799a462e..ba8ebc62887f 100644
@@ -281,16 +281,6 @@ int dram_init(void)
  			size = MVEBU_SDRAM_SIZE_MAX;
  	}
 

arch/arm/mach-mvebu/dram.c int dram_init(void) · line 293

Quoted 10 lines
arch/arm/mach-mvebu/dram.c +0 -10
-	for (; i < CONFIG_NR_DRAM_BANKS; i++) {
-		/* If above loop terminated prematurely, we need to set
-		 * remaining banks' start address & size as 0. Otherwise other
-		 * u-boot functions and Linux kernel gets wrong values which
-		 * could result in crash */
-		gd->bd->bi_dram[i].start = 0;
-		gd->bd->bi_dram[i].size = 0;
-	}
-
-

How did you spot this issue? Did you see some crash / oops on one of your boards lately? I'm asking since I've never seen any issues with this code.

Thanks,
Stefan

#2 lore raw

PATCH 1/1 ARM: mvebu: don't dereference null bd pointer

Quoted 21 lines · Stefan Roese

Hi Chris,

Quoted 17 lines · Chris Packham

As mentioned in doc/README.arm-relocation gd->bd is not available in dram_init() so we shouldn't attempt to access it.

---

  arch/arm/mach-mvebu/dram.c | 10 ----------
  1 file changed, 10 deletions(-)
arch/arm/mach-mvebu/dram.c +0 -0
index fa8c799a462e..ba8ebc62887f 100644
@@ -281,16 +281,6 @@ int dram_init(void)
                      size = MVEBU_SDRAM_SIZE_MAX;
      }
 
Quoted 15 lines

arch/arm/mach-mvebu/dram.c int dram_init(void) · line 293

Quoted 10 lines
arch/arm/mach-mvebu/dram.c +0 -10
-     for (; i < CONFIG_NR_DRAM_BANKS; i++) {
-             /* If above loop terminated prematurely, we need to set
-              * remaining banks' start address & size as 0. Otherwise other
-              * u-boot functions and Linux kernel gets wrong values which
-              * could result in crash */
-             gd->bd->bi_dram[i].start = 0;
-             gd->bd->bi_dram[i].size = 0;
-     }
-
-

How did you spot this issue? Did you see some crash / oops on one of your boards lately? I'm asking since I've never seen any issues with this code.

Actually spotted this on another board.

Part of the $dayjob customisations include a diagnostic menu used for production testing. For mostly historical reasons we try and run the RAM test from the pre-relocation parts of u-boot. This lets us test all of RAM but as a consequence the bank start/size get overwritten with the test pattern so we hit a bug where the test would work the first time but then run off into the weeds the second time.

Aside from the wrongness I'm not aware of any problem this causes.

#3 lore raw

PATCH 1/1 ARM: mvebu: don't dereference null bd pointer

Quoted 40 lines · Chris Packham
Quoted 21 lines · Stefan Roese

Hi Chris,

Quoted 17 lines · Chris Packham

As mentioned in doc/README.arm-relocation gd->bd is not available in dram_init() so we shouldn't attempt to access it.

---

   arch/arm/mach-mvebu/dram.c | 10 ----------
   1 file changed, 10 deletions(-)
arch/arm/mach-mvebu/dram.c +0 -0
index fa8c799a462e..ba8ebc62887f 100644
@@ -281,16 +281,6 @@ int dram_init(void)
                       size = MVEBU_SDRAM_SIZE_MAX;
       }
 
Quoted 15 lines

arch/arm/mach-mvebu/dram.c int dram_init(void) · line 293

Quoted 10 lines
arch/arm/mach-mvebu/dram.c +0 -10
-     for (; i < CONFIG_NR_DRAM_BANKS; i++) {
-             /* If above loop terminated prematurely, we need to set
-              * remaining banks' start address & size as 0. Otherwise other
-              * u-boot functions and Linux kernel gets wrong values which
-              * could result in crash */
-             gd->bd->bi_dram[i].start = 0;
-             gd->bd->bi_dram[i].size = 0;
-     }
-
-

How did you spot this issue? Did you see some crash / oops on one of your boards lately? I'm asking since I've never seen any issues with this code.

Actually spotted this on another board.

Quoted 8 lines

Part of the $dayjob customisations include a diagnostic menu used for production testing. For mostly historical reasons we try and run the RAM test from the pre-relocation parts of u-boot. This lets us test all of RAM but as a consequence the bank start/size get overwritten with the test pattern so we hit a bug where the test would work the first time but then run off into the weeds the second time.

Aside from the wrongness I'm not aware of any problem this causes.

Thanks for the explanation.

Thanks,
Stefan