Skip to contentSkip to conversation

riscv: Fix 32-bit RISC-V boot failure

PATCH In review Bin Meng 3 participants 12 messages Jun 27, 2021 – Jul 22, 2021

Subsystems (2)
lore
Share & tools
lore

Bin Meng

author
#0 lore raw

PATCH 1/1 riscv: Fix 32-bit RISC-V boot failure

Commit dd2d082b5760 ("riscv: Cleanup setup_bootmem()") adjusted the calling sequence in setup_bootmem(), which invalidates the fix commit de043da0b9e7 ("RISC-V: Fix usage of memblock_enforce_memory_limit") did for 32-bit RISC-V unfortunately.

So now 32-bit RISC-V does not boot again when testing booting kernel on QEMU 'virt' with '-m 2G', which was exactly what the original commit de043da0b9e7 ("RISC-V: Fix usage of memblock_enforce_memory_limit") tried to fix.

Fixes dd2d082b5760 ("riscv: Cleanup setup_bootmem()")
Cc [email protected] # v5.12+

---

 arch/riscv/mm/init.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
arch/riscv/mm/init.c +3 -1
index 4c4c92ce0bb8..9b23b95c50cf 100644
@@ -123,7 +123,7 @@ void __init setup_bootmem(void)
 {
 	phys_addr_t vmlinux_end = __pa_symbol(&_end);
 	phys_addr_t vmlinux_start = __pa_symbol(&_start);
-	phys_addr_t dram_end = memblock_end_of_DRAM();
+	phys_addr_t dram_end;
 	phys_addr_t max_mapped_addr = __pa(~(ulong)0);
 
 #ifdef CONFIG_XIP_KERNEL
@@ -146,6 +146,8 @@ void __init setup_bootmem(void)
 #endif
 	memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start);
 
+	dram_end = memblock_end_of_DRAM();
+
 	/*
 	 * memblock allocator is not aware of the fact that last 4K bytes of
 	 * the addressable memory can not be mapped because of IS_ERR_VALUE
Signature · 1 line
2.25.1
#1 lore raw

Hi, sorry for the mistake,the bug is fixed by

https://lore.kernel.org/linux-riscv/[email protected]/

Quoted 40 lines · Bin Meng

Commit dd2d082b5760 ("riscv: Cleanup setup_bootmem()") adjusted the calling sequence in setup_bootmem(), which invalidates the fix commit de043da0b9e7 ("RISC-V: Fix usage of memblock_enforce_memory_limit") did for 32-bit RISC-V unfortunately.

So now 32-bit RISC-V does not boot again when testing booting kernel on QEMU 'virt' with '-m 2G', which was exactly what the original commit de043da0b9e7 ("RISC-V: Fix usage of memblock_enforce_memory_limit") tried to fix.

Fixes dd2d082b5760 ("riscv: Cleanup setup_bootmem()")
Cc [email protected] # v5.12+

---

  arch/riscv/mm/init.c | 4 +++-
  1 file changed, 3 insertions(+), 1 deletion(-)
arch/riscv/mm/init.c +3 -1
index 4c4c92ce0bb8..9b23b95c50cf 100644
@@ -123,7 +123,7 @@ void __init setup_bootmem(void)
  {
  	phys_addr_t vmlinux_end = __pa_symbol(&_end);
  	phys_addr_t vmlinux_start = __pa_symbol(&_start);
-	phys_addr_t dram_end = memblock_end_of_DRAM();
+	phys_addr_t dram_end;
  	phys_addr_t max_mapped_addr = __pa(~(ulong)0);
 
  #ifdef CONFIG_XIP_KERNEL
@@ -146,6 +146,8 @@ void __init setup_bootmem(void)
  #endif
  	memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start);
 
+	dram_end = memblock_end_of_DRAM();
+
  	/*
  	 * memblock allocator is not aware of the fact that last 4K bytes of
  	 * the addressable memory can not be mapped because of IS_ERR_VALUE
#3 lore raw
Quoted 10 lines · Bin Meng
Quoted 3 lines · Kefeng Wang

Hi, sorry for the mistake,the bug is fixed by

https://lore.kernel.org/linux-riscv/[email protected]/

What are we on the patch you mentioned?

I don't see it applied in the linux/master.

Also there should be a "Fixes" tag and [email protected] cc'ed because 32-bit is broken since v5.12.

https://kernel.googlesource.com/pub/scm/linux/kernel/git/riscv/linux/+/c9811e379b211c67ba29fb09d6f644dd44cfcff2

it's on Palmer' riscv-next.

Hi Palmer, should I resend or could you help me to add the fixes tag?

Quoted 8 lines

Regards,
Bin

_______________________________________________
linux-riscv mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-riscv

author
#4 lore raw
Quoted 9 lines · Kefeng Wang
Quoted 6 lines · Bin Meng
Quoted 3 lines · Kefeng Wang

Hi, sorry for the mistake,the bug is fixed by

https://lore.kernel.org/linux-riscv/[email protected]/

What are we on the patch you mentioned?

Quoted 8 lines
Quoted 4 lines

I don't see it applied in the linux/master.

Also there should be a "Fixes" tag and [email protected] cc'ed because 32-bit is broken since v5.12.

https://kernel.googlesource.com/pub/scm/linux/kernel/git/riscv/linux/+/c9811e379b211c67ba29fb09d6f644dd44cfcff2

it's on Palmer' riscv-next.

Not sure riscv-next is for which release? This is a regression and should be on 5.13.

Quoted 2 lines

Hi Palmer, should I resend or could you help me to add the fixes tag?

Regards,
Bin

author
#5 lore raw
Quoted 16 lines · Bin Meng
Quoted 14 lines · Kefeng Wang
Quoted 10 lines · Bin Meng
Quoted 3 lines · Kefeng Wang

Hi, sorry for the mistake,the bug is fixed by

https://lore.kernel.org/linux-riscv/[email protected]/

What are we on the patch you mentioned?

I don't see it applied in the linux/master.

Also there should be a "Fixes" tag and [email protected] cc'ed because 32-bit is broken since v5.12.

Quoted 9 lines
Quoted 3 lines

https://kernel.googlesource.com/pub/scm/linux/kernel/git/riscv/linux/+/c9811e379b211c67ba29fb09d6f644dd44cfcff2

it's on Palmer' riscv-next.

Not sure riscv-next is for which release? This is a regression and should be on 5.13.

Quoted 2 lines

Hi Palmer, should I resend or could you help me to add the fixes tag?

Your patch mixed 2 things (fix plus one feature) together, so it is not proper to back port your patch.

Here is my 2 cents:

1. Drop your patch from riscv-next
2. Apply my patch as it is a simple fix to previous commit. This allows stable kernel to cherry-pick the fix to v5.12 and v5.13.
3. Rebase your patch against mine, and resend v2

Let me know if this makes sense.

Regards,
Bin

#6 lore raw
Quoted 16 lines · Bin Meng
Quoted 15 lines · Bin Meng
Quoted 14 lines · Kefeng Wang
Quoted 10 lines · Bin Meng
Quoted 3 lines · Kefeng Wang

Hi, sorry for the mistake,the bug is fixed by

https://lore.kernel.org/linux-riscv/[email protected]/

What are we on the patch you mentioned?

I don't see it applied in the linux/master.

Also there should be a "Fixes" tag and [email protected] cc'ed because 32-bit is broken since v5.12.

https://kernel.googlesource.com/pub/scm/linux/kernel/git/riscv/linux/+/c9811e379b211c67ba29fb09d6f644dd44cfcff2

Quoted 7 lines
Quoted 5 lines
Quoted 1 line

it's on Palmer' riscv-next.

Not sure riscv-next is for which release? This is a regression and should be on 5.13.

Quoted 1 line

Hi Palmer, should I resend or could you help me to add the fixes tag?

Your patch mixed 2 things (fix plus one feature) together, so it is not proper to back port your patch.

"mem=" will change the range of memblock, so the fix part must be included.

Quoted 9 lines

Here is my 2 cents:

1. Drop your patch from riscv-next
2. Apply my patch as it is a simple fix to previous commit. This allows stable kernel to cherry-pick the fix to v5.12 and v5.13.
3. Rebase your patch against mine, and resend v2

Let me know if this makes sense.

It is not a big problem for me, but I have no right abourt riscv-next,

let's wait Palmer's advise.

Quoted 8 lines

Regards,
Bin

_______________________________________________
linux-riscv mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-riscv

author
#7 lore raw
Quoted 20 lines · Kefeng Wang
Quoted 17 lines · Bin Meng
Quoted 16 lines · Bin Meng
Quoted 15 lines · Kefeng Wang
Quoted 10 lines · Bin Meng
Quoted 3 lines · Kefeng Wang

Hi, sorry for the mistake,the bug is fixed by

https://lore.kernel.org/linux-riscv/[email protected]/

What are we on the patch you mentioned?

I don't see it applied in the linux/master.

Also there should be a "Fixes" tag and [email protected] cc'ed because 32-bit is broken since v5.12.

https://kernel.googlesource.com/pub/scm/linux/kernel/git/riscv/linux/+/c9811e379b211c67ba29fb09d6f644dd44cfcff2

it's on Palmer' riscv-next.

Quoted 9 lines
Quoted 6 lines
Quoted 4 lines

Not sure riscv-next is for which release? This is a regression and should be on 5.13.

Quoted 1 line

Hi Palmer, should I resend or could you help me to add the fixes tag?

Your patch mixed 2 things (fix plus one feature) together, so it is not proper to back port your patch.

"mem=" will change the range of memblock, so the fix part must be included.

Yes, so you can rebase the "mem=" changes on top of my patch.

The practice is that we should not mix 2 things in one patch. I can imagine that you wanted to add "mem=" to RISC-V and suddenly found the existing logic was broken, so you sent one patch to do both.

Quoted 9 lines
Quoted 9 lines

Here is my 2 cents:

1. Drop your patch from riscv-next
2. Apply my patch as it is a simple fix to previous commit. This allows stable kernel to cherry-pick the fix to v5.12 and v5.13.
3. Rebase your patch against mine, and resend v2

Quoted 6 lines
Quoted 1 line

Let me know if this makes sense.

It is not a big problem for me, but I have no right abourt riscv-next,

let's wait Palmer's advise.

Sure. Palmer, let me know your thoughts.

Regards,
Bin

author
#8 lore raw

Hi Palmer,

Quoted 47 lines · Bin Meng
Quoted 20 lines · Kefeng Wang
Quoted 17 lines · Bin Meng
Quoted 16 lines · Bin Meng
Quoted 15 lines · Kefeng Wang
Quoted 10 lines · Bin Meng
Quoted 3 lines · Kefeng Wang

Hi, sorry for the mistake,the bug is fixed by

https://lore.kernel.org/linux-riscv/[email protected]/

What are we on the patch you mentioned?

I don't see it applied in the linux/master.

Also there should be a "Fixes" tag and [email protected] cc'ed because 32-bit is broken since v5.12.

https://kernel.googlesource.com/pub/scm/linux/kernel/git/riscv/linux/+/c9811e379b211c67ba29fb09d6f644dd44cfcff2

it's on Palmer' riscv-next.

Quoted 9 lines
Quoted 6 lines
Quoted 4 lines

Not sure riscv-next is for which release? This is a regression and should be on 5.13.

Quoted 1 line

Hi Palmer, should I resend or could you help me to add the fixes tag?

Your patch mixed 2 things (fix plus one feature) together, so it is not proper to back port your patch.

"mem=" will change the range of memblock, so the fix part must be included.

Yes, so you can rebase the "mem=" changes on top of my patch.

The practice is that we should not mix 2 things in one patch. I can imagine that you wanted to add "mem=" to RISC-V and suddenly found the existing logic was broken, so you sent one patch to do both.

Quoted 9 lines
Quoted 9 lines

Here is my 2 cents:

1. Drop your patch from riscv-next
2. Apply my patch as it is a simple fix to previous commit. This allows stable kernel to cherry-pick the fix to v5.12 and v5.13.
3. Rebase your patch against mine, and resend v2

Quoted 8 lines
Quoted 6 lines
Quoted 1 line

Let me know if this makes sense.

It is not a big problem for me, but I have no right abourt riscv-next,

let's wait Palmer's advise.

Sure. Palmer, let me know your thoughts.

Ping?

Regards,
Bin

author
#9 lore raw
Quoted 53 lines · Bin Meng

Hi Palmer,

Quoted 49 lines · Bin Meng
Quoted 20 lines · Kefeng Wang
Quoted 17 lines · Bin Meng
Quoted 16 lines · Bin Meng
Quoted 15 lines · Kefeng Wang
Quoted 10 lines · Bin Meng
Quoted 3 lines · Kefeng Wang

Hi, sorry for the mistake,the bug is fixed by

https://lore.kernel.org/linux-riscv/[email protected]/

What are we on the patch you mentioned?

I don't see it applied in the linux/master.

Also there should be a "Fixes" tag and [email protected] cc'ed because 32-bit is broken since v5.12.

https://kernel.googlesource.com/pub/scm/linux/kernel/git/riscv/linux/+/c9811e379b211c67ba29fb09d6f644dd44cfcff2

it's on Palmer' riscv-next.

Quoted 9 lines
Quoted 6 lines
Quoted 4 lines

Not sure riscv-next is for which release? This is a regression and should be on 5.13.

Quoted 1 line

Hi Palmer, should I resend or could you help me to add the fixes tag?

Your patch mixed 2 things (fix plus one feature) together, so it is not proper to back port your patch.

"mem=" will change the range of memblock, so the fix part must be included.

Yes, so you can rebase the "mem=" changes on top of my patch.

The practice is that we should not mix 2 things in one patch. I can imagine that you wanted to add "mem=" to RISC-V and suddenly found the existing logic was broken, so you sent one patch to do both.

Quoted 11 lines
Quoted 10 lines

Here is my 2 cents:

1. Drop your patch from riscv-next
2. Apply my patch as it is a simple fix to previous commit. This allows stable kernel to cherry-pick the fix to v5.12 and v5.13.
3. Rebase your patch against mine, and resend v2

Let me know if this makes sense.

Quoted 8 lines
Quoted 6 lines
Quoted 4 lines

It is not a big problem for me, but I have no right abourt riscv-next,

let's wait Palmer's advise.

Sure. Palmer, let me know your thoughts.

Ping?

Ping?

#10 lore raw
Quoted 56 lines · On Thu, 15 Jul 2021 19:14:20 PDT (-0700), [email protected] wrote:
Quoted 55 lines · Bin Meng

Hi Palmer,

Quoted 51 lines · Bin Meng
Quoted 20 lines · Kefeng Wang
Quoted 17 lines · Bin Meng
Quoted 16 lines · Bin Meng
Quoted 15 lines · Kefeng Wang
Quoted 10 lines · Bin Meng
Quoted 3 lines · Kefeng Wang

Hi, sorry for the mistake,the bug is fixed by

https://lore.kernel.org/linux-riscv/[email protected]/

What are we on the patch you mentioned?

I don't see it applied in the linux/master.

Also there should be a "Fixes" tag and [email protected] cc'ed because 32-bit is broken since v5.12.

https://kernel.googlesource.com/pub/scm/linux/kernel/git/riscv/linux/+/c9811e379b211c67ba29fb09d6f644dd44cfcff2

it's on Palmer' riscv-next.

Quoted 9 lines
Quoted 6 lines
Quoted 4 lines

Not sure riscv-next is for which release? This is a regression and should be on 5.13.

Quoted 1 line

Hi Palmer, should I resend or could you help me to add the fixes tag?

Your patch mixed 2 things (fix plus one feature) together, so it is not proper to back port your patch.

"mem=" will change the range of memblock, so the fix part must be included.

Yes, so you can rebase the "mem=" changes on top of my patch.

The practice is that we should not mix 2 things in one patch. I can imagine that you wanted to add "mem=" to RISC-V and suddenly found the existing logic was broken, so you sent one patch to do both.

Quoted 13 lines
Quoted 10 lines

Here is my 2 cents:

1. Drop your patch from riscv-next
2. Apply my patch as it is a simple fix to previous commit. This allows stable kernel to cherry-pick the fix to v5.12 and v5.13.
3. Rebase your patch against mine, and resend v2

Let me know if this makes sense.

It is not a big problem for me, but I have no right abourt riscv-next,

Quoted 8 lines
Quoted 6 lines
Quoted 4 lines
Quoted 2 lines

let's wait Palmer's advise.

Sure. Palmer, let me know your thoughts.

Ping?

Ping?

Sorry, I missed this one. It looks like the patch that adds mem= and fixes the bug has already been merged, so I'm not really quite sure what the right thing to do is here: we don't really want the mem= code on stable, but we do want the fix. I went ahead and did

commit 444818b599189fd8b6c814da542ff8cfc9fe67d4 (HEAD -> fixes, palmer/fixes)

gpg: Signature made Wed 21 Jul 2021 10:21:05 PM PDT
gpg:                using RSA key 2B3C3747446843B24A943A7A2E1319F35FBB1889
gpg:                issuer "[email protected]"
gpg: Good signature from "Palmer Dabbelt <[email protected]>" [ultimate]
gpg:                 aka "Palmer Dabbelt <[email protected]>" [ultimate]
Merge: e73f0f0ee754 d0e4dae74470
Author: Palmer Dabbelt <[email protected]>
Date:   Wed Jul 21 22:18:58 2021 -0700

Merge remote-tracking branch 'riscv/riscv-fix-32bit' into fixes

This contains a single fix for 32-bit boot. It happens this was already fixed by c9811e379b21 ("riscv: Add mem kernel parameter support"), but the bug existed before that feature addition so I've applied the patch earlier and then merged it in (which results in a conflict, which is fixed via not changing the resulting tree).

* riscv/riscv-fix-32bit:
riscv: Fix 32-bit RISC-V boot failure

as that"s the best I could come up with -- then the fix will land on master, which should cause it to get pulled onto stable.

Greg: is there a better way to make something like this get to stable?

author
#11 lore raw
Quoted 90 lines · Palmer Dabbelt
Quoted 56 lines · On Thu, 15 Jul 2021 19:14:20 PDT (-0700), [email protected] wrote:
Quoted 55 lines · Bin Meng

Hi Palmer,

Quoted 51 lines · Bin Meng
Quoted 20 lines · Kefeng Wang
Quoted 17 lines · Bin Meng
Quoted 16 lines · Bin Meng
Quoted 15 lines · Kefeng Wang
Quoted 10 lines · Bin Meng
Quoted 3 lines · Kefeng Wang

Hi, sorry for the mistake,the bug is fixed by

https://lore.kernel.org/linux-riscv/[email protected]/

What are we on the patch you mentioned?

I don't see it applied in the linux/master.

Also there should be a "Fixes" tag and [email protected] cc'ed because 32-bit is broken since v5.12.

https://kernel.googlesource.com/pub/scm/linux/kernel/git/riscv/linux/+/c9811e379b211c67ba29fb09d6f644dd44cfcff2

it's on Palmer' riscv-next.

Quoted 9 lines
Quoted 6 lines
Quoted 4 lines

Not sure riscv-next is for which release? This is a regression and should be on 5.13.

Quoted 1 line

Hi Palmer, should I resend or could you help me to add the fixes tag?

Your patch mixed 2 things (fix plus one feature) together, so it is not proper to back port your patch.

"mem=" will change the range of memblock, so the fix part must be included.

Yes, so you can rebase the "mem=" changes on top of my patch.

The practice is that we should not mix 2 things in one patch. I can imagine that you wanted to add "mem=" to RISC-V and suddenly found the existing logic was broken, so you sent one patch to do both.

Quoted 13 lines
Quoted 10 lines

Here is my 2 cents:

1. Drop your patch from riscv-next
2. Apply my patch as it is a simple fix to previous commit. This allows stable kernel to cherry-pick the fix to v5.12 and v5.13.
3. Rebase your patch against mine, and resend v2

Let me know if this makes sense.

It is not a big problem for me, but I have no right abourt riscv-next,

Quoted 8 lines
Quoted 6 lines
Quoted 4 lines
Quoted 2 lines

let's wait Palmer's advise.

Sure. Palmer, let me know your thoughts.

Ping?

Ping?

Sorry, I missed this one. It looks like the patch that adds mem= and fixes the bug has already been merged, so I'm not really quite sure what the right thing to do is here: we don't really want the mem= code on stable, but we do want the fix. I went ahead and did

commit 444818b599189fd8b6c814da542ff8cfc9fe67d4 (HEAD -> fixes, palmer/fixes)

gpg: Signature made Wed 21 Jul 2021 10:21:05 PM PDT
gpg:                using RSA key 2B3C3747446843B24A943A7A2E1319F35FBB1889
gpg:                issuer "[email protected]"
gpg: Good signature from "Palmer Dabbelt <[email protected]>" [ultimate]
gpg:                 aka "Palmer Dabbelt <[email protected]>" [ultimate]
Merge: e73f0f0ee754 d0e4dae74470
Author: Palmer Dabbelt <[email protected]>
Date:   Wed Jul 21 22:18:58 2021 -0700

Merge remote-tracking branch 'riscv/riscv-fix-32bit' into fixes

This contains a single fix for 32-bit boot. It happens this was already fixed by c9811e379b21 ("riscv: Add mem kernel parameter support"), but the bug existed before that feature addition so I've applied the patch earlier and then merged it in (which results in a conflict, which is fixed via not changing the resulting tree).

Quoted 7 lines

* riscv/riscv-fix-32bit:
riscv: Fix 32-bit RISC-V boot failure

as that"s the best I could come up with -- then the fix will land on master, which should cause it to get pulled onto stable.

Greg: is there a better way to make something like this get to stable?

We can:

1. git revert c9811e379b21 ("riscv: Add mem kernel parameter support")
2. git am <this patch>, cc stable
3. Kefeng resent a new patch that adds the mem kernel parameter support, on top of mine

Regards,
Bin