cfi_flash: Fix devicetree address determination
PATCH 1/1 cfi_flash: Fix devicetree address determination
The cfi-flash driver uses an open-coded version of the generic algorithm to decode and translate multiple frames of a "reg" property.
This starts off the wrong foot by using the address-cells and size-cells properties of *this* very node, and not of the parent. This somewhat happened to work back when we were using a wrong default size of 2, but broke about a year ago with commit 0ba41ce1b781 ("libfdt: return correct value if #size-cells property is not present").
Instead of fixing the reinvented wheel, just use the generic function that does all of this properly.
This fixes U-Boot on QEMU (outside of EL1), which was crashing due to decoding a wrong start address:
DRAM: 1 GiB
Flash: "Synchronous Abort" handler, esr 0x96000044
elr: 00000000000211dc lr : 00000000000211b0 (reloc)
elr: 000000007ff5e1dc lr : 000000007ff5e1b0
x0 : 00000000000000f0 x1 : 000000007ff5e1d8
x2 : 000000007edfbc48 x3 : 0000000000000000
x4 : 0000000000000000 x5 : 00000000000000f0
x6 : 000000007edfbc2c x7 : 0000000000000000
x8 : 000000007ffd8d70 x9 : 000000000000000c
x10: 0400000000000003 x11: 0000000000000055
^^^^^^^^^^^^^^^^
--- drivers/mtd/cfi_flash.c | 25 +++++++------------------ 1 file changed, 7 insertions(+), 18 deletions(-)
drivers/mtd/cfi_flash.c +7 -18
@@ -2468,29 +2468,18 @@ unsigned long flash_init(void)#ifdef CONFIG_CFI_FLASH /* for driver model */ static int cfi_flash_probe (struct udevice *dev) { -const fdt32_t *cell; -int addrc, sizec; -int len, idx; +fdt_addr_t addr; +fdt_size_t size; +int idx; - addrc =dev_read_addr_cells (dev); - sizec =dev_read_size_cells (dev); - -/* decode regs; there may be multiple reg tuples. */ - cell =dev_read_prop (dev,"reg" , &len); -if (!cell) -return -ENOENT ; - idx =0 ; - len /=sizeof (fdt32_t); -while (idx < len) { -phys_addr_t addr; - - addr =dev_translate_address (dev, cell + idx); +for (idx =0 ; idx <CFI_MAX_FLASH_BANKS ; idx++) { + addr =devfdt_get_addr_size_index (dev, idx, &size); +if (addr ==FDT_ADDR_T_NONE ) +break ; flash_info[cfi_flash_num_flash_banks].dev = dev; flash_info[cfi_flash_num_flash_banks].base = addr; cfi_flash_num_flash_banks++; - - idx += addrc + sizec; } gd->bd->bi_flashstart = flash_info[0 ].base;
Signature · 1 line
2.17.5
PATCH 1/1 cfi_flash: Fix devicetree address determination
Hi Andre,
(added Simon)
Quoted 63 lines · Andre Przywara
The cfi-flash driver uses an open-coded version of the generic algorithm to decode and translate multiple frames of a "reg" property.
This starts off the wrong foot by using the address-cells and size-cells properties of *this* very node, and not of the parent. This somewhat happened to work back when we were using a wrong default size of 2, but broke about a year ago with commit
0ba41ce1b781("libfdt: return correct value if #size-cells property is not present").Instead of fixing the reinvented wheel, just use the generic function that does all of this properly.
This fixes U-Boot on QEMU (outside of EL1), which was crashing due to decoding a wrong start address:
DRAM: 1 GiB
Flash: "Synchronous Abort" handler, esr0x96000044
elr:00000000000211dclr :00000000000211b0(reloc)
elr:000000007ff5e1dclr :000000007ff5e1b0
x0 :00000000000000f0x1 :000000007ff5e1d8
x2 :000000007edfbc48x3 : 0000000000000000
x4 : 0000000000000000 x5 :00000000000000f0
x6 :000000007edfbc2cx7 : 0000000000000000
x8 :000000007ffd8d70x9 :000000000000000c
x10: 0400000000000003 x11: 0000000000000055^^^^^^^^^^^^^^^^Signed-off-by Andre Przywara <[email protected]>--- drivers/mtd/cfi_flash.c | 25 +++++++------------------ 1 file changed, 7 insertions(+), 18 deletions(-)drivers/mtd/cfi_flash.c +6
-16@@ -2468,29 +2468,18 @@ unsigned long flash_init(void)#ifdef CONFIG_CFI_FLASH /* for driver model */ static int cfi_flash_probe (struct udevice *dev) { -const fdt32_t *cell; -int addrc, sizec; -int len, idx; +fdt_addr_t addr; +fdt_size_t size; +int idx; - addrc =dev_read_addr_cells (dev); - sizec =dev_read_size_cells (dev); - -/* decode regs; there may be multiple reg tuples. */ - cell =dev_read_prop (dev,"reg" , &len); -if (!cell) -return -ENOENT ; - idx =0 ; - len /=sizeof (fdt32_t); -while (idx < len) { -phys_addr_t addr; - - addr =dev_translate_address (dev, cell + idx); +for (idx =0 ; idx <CFI_MAX_FLASH_BANKS ; idx++) { + addr =devfdt_get_addr_size_index (dev, idx, &size); +if (addr ==FDT_ADDR_T_NONE )
drivers/mtd/cfi_flash.c unsigned long flash_init(void) · line 2485
Quoted 10 lines
drivers/mtd/cfi_flash.c +1
-2+break ; flash_info[cfi_flash_num_flash_banks].dev = dev; flash_info[cfi_flash_num_flash_banks].base = addr; cfi_flash_num_flash_banks++; - - idx += addrc + sizec; } gd->bd->bi_flashstart = flash_info[0 ].base;
This fails on my Octeon MIPS64 platform "octeon_ebb7304". I did some debugging and found that here "of_offset" is a 64 bit value (type long) which gets truncated in dev_of_offset() to 32 bit (type int).
This problem only arises when of_live_active() is set. Here, "of_offset" holds a pointer AFACT and truncating it to 32 bits breaks things.
I'm wondering why this did not hit me earlier on this 64bit platform. Simon, do you have a quick idea how to solve this?
Thanks,
Stefan
PATCH 1/1 cfi_flash: Fix devicetree address determination
Hi Stefan,
Quoted 84 lines · Stefan Roese
Hi Andre,
(added Simon)
Quoted 63 lines · Andre Przywara
The cfi-flash driver uses an open-coded version of the generic algorithm to decode and translate multiple frames of a "reg" property.
This starts off the wrong foot by using the address-cells and size-cells properties of *this* very node, and not of the parent. This somewhat happened to work back when we were using a wrong default size of 2, but broke about a year ago with commit
0ba41ce1b781("libfdt: return correct value if #size-cells property is not present").Instead of fixing the reinvented wheel, just use the generic function that does all of this properly.
This fixes U-Boot on QEMU (outside of EL1), which was crashing due to decoding a wrong start address:
DRAM: 1 GiB
Flash: "Synchronous Abort" handler, esr0x96000044
elr:00000000000211dclr :00000000000211b0(reloc)
elr:000000007ff5e1dclr :000000007ff5e1b0
x0 :00000000000000f0x1 :000000007ff5e1d8
x2 :000000007edfbc48x3 : 0000000000000000
x4 : 0000000000000000 x5 :00000000000000f0
x6 :000000007edfbc2cx7 : 0000000000000000
x8 :000000007ffd8d70x9 :000000000000000c
x10: 0400000000000003 x11: 0000000000000055^^^^^^^^^^^^^^^^Signed-off-by Andre Przywara <[email protected]>--- drivers/mtd/cfi_flash.c | 25 +++++++------------------ 1 file changed, 7 insertions(+), 18 deletions(-)drivers/mtd/cfi_flash.c +6
-16@@ -2468,29 +2468,18 @@ unsigned long flash_init(void)#ifdef CONFIG_CFI_FLASH /* for driver model */ static int cfi_flash_probe (struct udevice *dev) { -const fdt32_t *cell; -int addrc, sizec; -int len, idx; +fdt_addr_t addr; +fdt_size_t size; +int idx; - addrc =dev_read_addr_cells (dev); - sizec =dev_read_size_cells (dev); - -/* decode regs; there may be multiple reg tuples. */ - cell =dev_read_prop (dev,"reg" , &len); -if (!cell) -return -ENOENT ; - idx =0 ; - len /=sizeof (fdt32_t); -while (idx < len) { -phys_addr_t addr; - - addr =dev_translate_address (dev, cell + idx); +for (idx =0 ; idx <CFI_MAX_FLASH_BANKS ; idx++) { + addr =devfdt_get_addr_size_index (dev, idx, &size); +if (addr ==FDT_ADDR_T_NONE )drivers/mtd/cfi_flash.c unsigned long flash_init(void) · line 2485
Quoted 10 lines
drivers/mtd/cfi_flash.c +1
-2+break ; flash_info[cfi_flash_num_flash_banks].dev = dev; flash_info[cfi_flash_num_flash_banks].base = addr; cfi_flash_num_flash_banks++; - - idx += addrc + sizec; } gd->bd->bi_flashstart = flash_info[0 ].base;This fails on my Octeon MIPS64 platform "
octeon_ebb7304". I did some debugging and found that here "of_offset" is a 64 bit value (type long) which gets truncated indev_of_offset()to 32 bit (type int).
Quoted 5 lines
This problem only arises when
of_live_active()is set. Here, "of_offset" holds a pointer AFACT and truncating it to 32 bits breaks things.I'm wondering why this did not hit me earlier on this 64bit platform. Simon, do you have a quick idea how to solve this?
Well I don't think ofnode should use long for of_offset, since int should be enough.
ofnode_to_offset() converts an ofnode to a DT offset but only if it is not using livetree. With livetree there are no offsets so this is not going to work. If you define OF_CHECKS you will see that.
Note that an ofnode can either hold a pointer or an offset. There are detailed comments on ofnode_union to explain how it is supposed to work.
This patch looks correct to me, but perhaps there is something else going on?
Regards,
Simon
PATCH 1/1 cfi_flash: Fix devicetree address determination
Hi Simon,
Quoted 93 lines · Simon Glass
Hi Stefan,
Quoted 84 lines · Stefan Roese
Hi Andre,
(added Simon)
Quoted 63 lines · Andre Przywara
The cfi-flash driver uses an open-coded version of the generic algorithm to decode and translate multiple frames of a "reg" property.
This starts off the wrong foot by using the address-cells and size-cells properties of *this* very node, and not of the parent. This somewhat happened to work back when we were using a wrong default size of 2, but broke about a year ago with commit
0ba41ce1b781("libfdt: return correct value if #size-cells property is not present").Instead of fixing the reinvented wheel, just use the generic function that does all of this properly.
This fixes U-Boot on QEMU (outside of EL1), which was crashing due to decoding a wrong start address:
DRAM: 1 GiB
Flash: "Synchronous Abort" handler, esr0x96000044
elr:00000000000211dclr :00000000000211b0(reloc)
elr:000000007ff5e1dclr :000000007ff5e1b0
x0 :00000000000000f0x1 :000000007ff5e1d8
x2 :000000007edfbc48x3 : 0000000000000000
x4 : 0000000000000000 x5 :00000000000000f0
x6 :000000007edfbc2cx7 : 0000000000000000
x8 :000000007ffd8d70x9 :000000000000000c
x10: 0400000000000003 x11: 0000000000000055^^^^^^^^^^^^^^^^Signed-off-by Andre Przywara <[email protected]>--- drivers/mtd/cfi_flash.c | 25 +++++++------------------ 1 file changed, 7 insertions(+), 18 deletions(-)drivers/mtd/cfi_flash.c +6
-16@@ -2468,29 +2468,18 @@ unsigned long flash_init(void)#ifdef CONFIG_CFI_FLASH /* for driver model */ static int cfi_flash_probe (struct udevice *dev) { -const fdt32_t *cell; -int addrc, sizec; -int len, idx; +fdt_addr_t addr; +fdt_size_t size; +int idx; - addrc =dev_read_addr_cells (dev); - sizec =dev_read_size_cells (dev); - -/* decode regs; there may be multiple reg tuples. */ - cell =dev_read_prop (dev,"reg" , &len); -if (!cell) -return -ENOENT ; - idx =0 ; - len /=sizeof (fdt32_t); -while (idx < len) { -phys_addr_t addr; - - addr =dev_translate_address (dev, cell + idx); +for (idx =0 ; idx <CFI_MAX_FLASH_BANKS ; idx++) { + addr =devfdt_get_addr_size_index (dev, idx, &size); +if (addr ==FDT_ADDR_T_NONE )drivers/mtd/cfi_flash.c unsigned long flash_init(void) · line 2485
Quoted 10 lines
drivers/mtd/cfi_flash.c +1
-2+break ; flash_info[cfi_flash_num_flash_banks].dev = dev; flash_info[cfi_flash_num_flash_banks].base = addr; cfi_flash_num_flash_banks++; - - idx += addrc + sizec; } gd->bd->bi_flashstart = flash_info[0 ].base;This fails on my Octeon MIPS64 platform "
octeon_ebb7304". I did some debugging and found that here "of_offset" is a 64 bit value (type long) which gets truncated indev_of_offset()to 32 bit (type int).Quoted 5 lines
This problem only arises when
of_live_active()is set. Here, "of_offset" holds a pointer AFACT and truncating it to 32 bits breaks things.I'm wondering why this did not hit me earlier on this 64bit platform. Simon, do you have a quick idea how to solve this?
Quoted 6 lines
Well I don't think ofnode should use long for
of_offset, since int should be enough.
ofnode_to_offset()converts an ofnode to a DT offset but only if it is not using livetree. With livetree there are no offsets so this is not going to work. If you defineOF_CHECKSyou will see that.
This does not work right now. I'll send a patch fixing compiling with OF_CHECK enabled shortly.
Quoted 3 lines
Note that an ofnode can either hold a pointer or an offset. There are detailed comments on
ofnode_unionto explain how it is supposed to work.
Right. Thanks for all the detailed infos in the header. The main issue seems to be, that this CFI patch uses a function from fdtaddr.c (devfdt_get_addr_size_index), which unconditionally uses dev_of_offset() without checking if livetree is enabled or not. This breaks on my 64 bit platform (see below).
Quoted 1 line
This patch looks correct to me, but perhaps there is something else going on?
Making this change below, works for me:
- addr = devfdt_get_addr_size_index(dev, idx, &size);
+ addr = dev_read_addr_index(dev, idx);
Maybe we should make sure, that all functions from fdtaddr.c are not used with livetree active? To prevent similar issues using devfdt_foo() functions with livetree active.
Thanks,
Stefan
PATCH 1/1 cfi_flash: Fix devicetree address determination
Quoted 117 lines · Stefan Roese
Hi Simon,
Quoted 94 lines · Simon Glass
Hi Stefan,
Quoted 85 lines · Stefan Roese
Hi Andre,
(added Simon)
Quoted 65 lines · Andre Przywara
The cfi-flash driver uses an open-coded version of the generic algorithm to decode and translate multiple frames of a "reg" property.
This starts off the wrong foot by using the address-cells and size-cells
properties of *this* very node, and not of the parent. This somewhat happened to work back when we were using a wrong default size of 2, but broke about a year ago with commit0ba41ce1b781("libfdt: return correct value if #size-cells property is not present").Instead of fixing the reinvented wheel, just use the generic function that does all of this properly.
This fixes U-Boot on QEMU (outside of EL1), which was crashing due to decoding a wrong start address:
DRAM:? 1 GiB
Flash: "Synchronous Abort" handler, esr0x96000044
elr:00000000000211dclr :00000000000211b0(reloc)
elr:000000007ff5e1dclr :000000007ff5e1b0
x0 :00000000000000f0x1 :000000007ff5e1d8
x2 :000000007edfbc48x3 : 0000000000000000
x4 : 0000000000000000 x5 :00000000000000f0
x6 :000000007edfbc2cx7 : 0000000000000000
x8 :000000007ffd8d70x9 :000000000000000c
x10: 0400000000000003 x11: 0000000000000055
?????? ^^^^^^^^^^^^^^^^Signed-off-by Andre Przywara <[email protected]>---
??drivers/mtd/cfi_flash.c| 25 +++++++------------------ ?? 1 file changed, 7 insertions(+), 18 deletions(-)drivers/mtd/cfi_flash.c +0
-0@@ -2468,29 +2468,18 @@ unsigned long flash_init(void)?? #ifdef
CONFIG_CFI_FLASH/* for driver model */ ?? static intcfi_flash_probe(struct udevice *dev)?? {
-???? constfdt32_t*cell;
-???? int addrc, sizec;
-???? int len, idx;
+????fdt_addr_taddr;
+????fdt_size_tsize;
+???? int idx;-???? addrc =
dev_read_addr_cells(dev);
-???? sizec =dev_read_size_cells(dev);drivers/mtd/cfi_flash.c +4
-11- -???? /* decode regs; there may be multiple reg tuples. */ -???? cell = dev_read_prop(dev, "reg", &len); -???? if (!cell) -???????????? return -ENOENT; -???? idx = 0; -???? len /= sizeof(fdt32_t); -???? while (idx < len) { -???????????? phys_addr_t addr; - -???????????? addr = dev_translate_address(dev, cell + idx); +???? for (idx = 0; idx < CFI_MAX_FLASH_BANKS; idx++) { +???????????? addr = devfdt_get_addr_size_index(dev, idx, &size); +???????????? if (addr == FDT_ADDR_T_NONE) +???????????????????? break;Quoted 9 lines
?????????????? flash_info[cfi_flash_num_flash_banks].dev = dev; ?????????????? flash_info[cfi_flash_num_flash_banks].base = addr; ?????????????? cfi_flash_num_flash_banks++;
-
-???????????? idx += addrc + sizec;
?????? }
??????gd->bd->bi_flashstart= flash_info[0].base;This fails on my Octeon MIPS64 platform "
octeon_ebb7304". I did some debugging and found that here "of_offset" is a 64 bit value (type long) which gets truncated indev_of_offset()to 32 bit (type int).Quoted 5 lines
This problem only arises when
of_live_active()is set. Here, "of_offset" holds a pointer AFACT and truncating it to 32 bits breaks things.I'm wondering why this did not hit me earlier on this 64bit platform. Simon, do you have a quick idea how to solve this?
Quoted 6 lines
Well I don't think ofnode should use long for
of_offset, since int should be enough.
ofnode_to_offset()converts an ofnode to a DT offset but only if it is not using livetree. With livetree there are no offsets so this is not going to work. If you defineOF_CHECKSyou will see that.This does not work right now. I'll send a patch fixing compiling with
OF_CHECKenabled shortly.Quoted 3 lines
Note that an ofnode can either hold a pointer or an offset. There are detailed comments on
ofnode_unionto explain how it is supposed to work.Right. Thanks for all the detailed infos in the header. The main issue seems to be, that this CFI patch uses a function from
fdtaddr.c(devfdt_get_addr_size_index), which unconditionally usesdev_of_offset()without checking if livetree is enabled or not. This breaks on my 64 bit platform (see below).
Quoted 7 lines
Quoted 2 lines
This patch looks correct to me, but perhaps there is something else going on?
Making this change below, works for me:
-??????? addr =
devfdt_get_addr_size_index(dev, idx, &size); +??????? addr =dev_read_addr_index(dev, idx);
Ouch, sorry for that!
One thing I noticed: Technically this fix is no longer needed, since Heinrich's patch ae6b33dcc342 ("dm: fix ofnode_read_addr/size_cells()") recently fixed that particular issue (I missed that one when doing the bisect earlier).
However I still consider this patch here useful, since it removes code duplication (and the original bug gives a good rationale for that!). So I will repost this one here, but leave it up to you whether to merge it or not.
Also: this function was the only user of dev_read_{addr,size}_cells(). Shall we consequently remove them? They have this somewhat surprising feature of querying the parent now, which prevents them from being used when someone want to determine the current #a-c and #s-c applicable for subnodes, for instance.
Quoted 3 lines
Maybe we should make sure, that all functions from
fdtaddr.care not used with livetree active? To prevent similar issues usingdevfdt_foo()functions with livetree active.
That sounds useful, but can this be done easily?
Cheers,
Andre
PATCH 1/1 cfi_flash: Fix devicetree address determination
Hi Andr?,
Quoted 139 lines · Andr? Przywara
Quoted 122 lines · Stefan Roese
Hi Simon,
Quoted 94 lines · Simon Glass
Hi Stefan,
Quoted 85 lines · Stefan Roese
Hi Andre,
(added Simon)
Quoted 64 lines · Andre Przywara
The cfi-flash driver uses an open-coded version of the generic algorithm to decode and translate multiple frames of a "reg" property.
This starts off the wrong foot by using the address-cells and size-cells
properties of *this* very node, and not of the parent. This somewhat happened to work back when we were using a wrong default size of 2, but broke about a year ago with commit0ba41ce1b781("libfdt: return correct value if #size-cells property is not present").Instead of fixing the reinvented wheel, just use the generic function that does all of this properly.
This fixes U-Boot on QEMU (outside of EL1), which was crashing due to decoding a wrong start address:
DRAM: 1 GiB
Flash: "Synchronous Abort" handler, esr0x96000044
elr:00000000000211dclr :00000000000211b0(reloc)
elr:000000007ff5e1dclr :000000007ff5e1b0
x0 :00000000000000f0x1 :000000007ff5e1d8
x2 :000000007edfbc48x3 : 0000000000000000
x4 : 0000000000000000 x5 :00000000000000f0
x6 :000000007edfbc2cx7 : 0000000000000000
x8 :000000007ffd8d70x9 :000000000000000c
x10: 0400000000000003 x11: 0000000000000055^^^^^^^^^^^^^^^^Signed-off-by Andre Przywara <[email protected]>--- drivers/mtd/cfi_flash.c | 25 +++++++------------------ 1 file changed, 7 insertions(+), 18 deletions(-)drivers/mtd/cfi_flash.c +6
-16@@ -2468,29 +2468,18 @@ unsigned long flash_init(void)#ifdef CONFIG_CFI_FLASH /* for driver model */ static int cfi_flash_probe (struct udevice *dev) { -const fdt32_t *cell; -int addrc, sizec; -int len, idx; +fdt_addr_t addr; +fdt_size_t size; +int idx; - addrc =dev_read_addr_cells (dev); - sizec =dev_read_size_cells (dev); - -/* decode regs; there may be multiple reg tuples. */ - cell =dev_read_prop (dev,"reg" , &len); -if (!cell) -return -ENOENT ; - idx =0 ; - len /=sizeof (fdt32_t); -while (idx < len) { -phys_addr_t addr; - - addr =dev_translate_address (dev, cell + idx); +for (idx =0 ; idx <CFI_MAX_FLASH_BANKS ; idx++) { + addr =devfdt_get_addr_size_index (dev, idx, &size); +if (addr ==FDT_ADDR_T_NONE )drivers/mtd/cfi_flash.c unsigned long flash_init(void) · line 2485
Quoted 10 lines
drivers/mtd/cfi_flash.c +1
-2+break ; flash_info[cfi_flash_num_flash_banks].dev = dev; flash_info[cfi_flash_num_flash_banks].base = addr; cfi_flash_num_flash_banks++; - - idx += addrc + sizec; } gd->bd->bi_flashstart = flash_info[0 ].base;This fails on my Octeon MIPS64 platform "
octeon_ebb7304". I did some debugging and found that here "of_offset" is a 64 bit value (type long) which gets truncated indev_of_offset()to 32 bit (type int).Quoted 5 lines
This problem only arises when
of_live_active()is set. Here, "of_offset" holds a pointer AFACT and truncating it to 32 bits breaks things.I'm wondering why this did not hit me earlier on this 64bit platform. Simon, do you have a quick idea how to solve this?
Quoted 6 lines
Well I don't think ofnode should use long for
of_offset, since int should be enough.
ofnode_to_offset()converts an ofnode to a DT offset but only if it is not using livetree. With livetree there are no offsets so this is not going to work. If you defineOF_CHECKSyou will see that.This does not work right now. I'll send a patch fixing compiling with
OF_CHECKenabled shortly.Quoted 3 lines
Note that an ofnode can either hold a pointer or an offset. There are detailed comments on
ofnode_unionto explain how it is supposed to work.Right. Thanks for all the detailed infos in the header. The main issue seems to be, that this CFI patch uses a function from
fdtaddr.c(devfdt_get_addr_size_index), which unconditionally usesdev_of_offset()without checking if livetree is enabled or not. This breaks on my 64 bit platform (see below).Quoted 2 lines
This patch looks correct to me, but perhaps there is something else going on?
Making this change below, works for me:
Review
Quoted 2 lines
(quoted diff) +1
-1- addr = devfdt_get_addr_size_index(dev, idx, &size); + addr = dev_read_addr_index(dev, idx);Ouch, sorry for that!
One thing I noticed: Technically this fix is no longer needed, since Heinrich's patch
ae6b33dcc342("dm: fix ofnode_read_addr/size_cells()") recently fixed that particular issue (I missed that one when doing the bisect earlier).However I still consider this patch here useful, since it removes code duplication (and the original bug gives a good rationale for that!). So I will repost this one here, but leave it up to you whether to merge it or not.
Quoted 5 lines
Also: this function was the only user of dev_read_{addr,size}_cells(). Shall we consequently remove them? They have this somewhat surprising feature of querying the parent now, which prevents them from being used when someone want to determine the current #a-c and #s-c applicable for subnodes, for instance.
Yes I agree that is surprising. If we don't think they will be useful in future I think removing them is OK.
Quoted 5 lines
Quoted 4 lines
Maybe we should make sure, that all functions from
fdtaddr.care not used with livetree active? To prevent similar issues usingdevfdt_foo()functions with livetree active.That sounds useful, but can this be done easily?
We could add a check at the start of each function, perhaps.
Regards,
Simon