Skip to contentSkip to conversation

powerpc/powermac: Use early_* IO variants in via_calibrate_decr

PATCH In review Benjamin Gray 3 participants 4 messages Jul 6, 2023 – Aug 2, 2023

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 powerpc/powermac: Use early_* IO variants in via_calibrate_decr

On a powermac platform, under the call path

start_kernel
time_init
ppc_md.calibrate_decr (pmac_calibrate_decr)
via_calibrate_decr

we run ioremap and iounmap. The unmap can enable interrupts unexpectedly (cond_resched in vunmap_pmd_range), which is warned about later in the boot sequence in start_kernel.

Use the early_* variants of these IO functions to prevent this.

The issue is pre-existing, but is surfaced by commit 721255b9826b ("genirq: Use a maple tree for interrupt descriptor management"). It's not clear to me why this causes it to surface.

---
 arch/powerpc/platforms/powermac/time.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
arch/powerpc/platforms/powermac/time.c +3 -3
index 4c5790aff1b5..8633891b7aa5 100644
@@ -26,8 +26,8 @@
 #include <linux/rtc.h>
 #include <linux/of_address.h>
 
+#include <asm/early_ioremap.h>
 #include <asm/sections.h>
-#include <asm/io.h>
 #include <asm/machdep.h>
 #include <asm/time.h>
 #include <asm/nvram.h>
@@ -182,7 +182,7 @@ static int __init via_calibrate_decr(void)
 		return 0;
 	}
 	of_node_put(vias);
-	via = ioremap(rsrc.start, resource_size(&rsrc));
+	via = early_ioremap(rsrc.start, resource_size(&rsrc));
 	if (via == NULL) {
 		printk(KERN_ERR "Failed to map VIA for timer calibration !\n");
 		return 0;
@@ -207,7 +207,7 @@ static int __init via_calibrate_decr(void)
 
 	ppc_tb_freq = (dstart - dend) * 100 / 6;
 
-	iounmap(via);
+	early_iounmap((void *)via, resource_size(&rsrc));
 
 	return 1;
 }
Signature · 1 line
2.41.0
Reviewed-by on v1 1/1
#1 lore raw

Le 06/07/2023 à 03:08, Benjamin Gray a écrit :

Quoted 11 lines

On a powermac platform, under the call path

start_kernel
time_init
ppc_md.calibrate_decr (pmac_calibrate_decr)
via_calibrate_decr

we run ioremap and iounmap. The unmap can enable interrupts unexpectedly (cond_resched in vunmap_pmd_range), which is warned about later in the boot sequence in start_kernel.

Quoted 7 lines

Use the early_* variants of these IO functions to prevent this.

The issue is pre-existing, but is surfaced by commit 721255b9826b ("genirq: Use a maple tree for interrupt descriptor management"). It's not clear to me why this causes it to surface.

Quoted 36 lines
---
  arch/powerpc/platforms/powermac/time.c | 6 +++---
  1 file changed, 3 insertions(+), 3 deletions(-)
arch/powerpc/platforms/powermac/time.c +3 -3
index 4c5790aff1b5..8633891b7aa5 100644
@@ -26,8 +26,8 @@
  #include <linux/rtc.h>
  #include <linux/of_address.h>
 
+#include <asm/early_ioremap.h>
  #include <asm/sections.h>
-#include <asm/io.h>
  #include <asm/machdep.h>
  #include <asm/time.h>
  #include <asm/nvram.h>
@@ -182,7 +182,7 @@ static int __init via_calibrate_decr(void)
  		return 0;
  	}
  	of_node_put(vias);
-	via = ioremap(rsrc.start, resource_size(&rsrc));
+	via = early_ioremap(rsrc.start, resource_size(&rsrc));
  	if (via == NULL) {
  		printk(KERN_ERR "Failed to map VIA for timer calibration !\n");
  		return 0;
@@ -207,7 +207,7 @@ static int __init via_calibrate_decr(void)
 
  	ppc_tb_freq = (dstart - dend) * 100 / 6;
 
-	iounmap(via);
+	early_iounmap((void *)via, resource_size(&rsrc));
 
  	return 1;
  }
#2 lore raw
In reply to Benjamin Gray
Quoted 3 lines · Benjamin Gray

The issue is pre-existing, but is surfaced by commit 721255b9826b ("genirq: Use a maple tree for interrupt descriptor management"). It's not clear to me why this causes it to surface.

From the thread chain in [1], it looks like the maple tree implementation just has different allocation behaviour, which follows a pre-existing code path to kmem_cache_alloc_bulk(), which unconditionally enables interrupts.

(thanks Jordan Niethe for finding this thread)

[1]: https://lore.kernel.org/all/87o7qdzfay.ffs@tglx/

#3 lore raw
Quoted 9 lines · Benjamin Gray
Quoted 3 lines · Benjamin Gray

The issue is pre-existing, but is surfaced by commit 721255b9826b ("genirq: Use a maple tree for interrupt descriptor management"). It's not clear to me why this causes it to surface.

From the thread chain in [1], it looks like the maple tree implementation just has different allocation behaviour, which follows a pre-existing code path to kmem_cache_alloc_bulk(), which unconditionally enables interrupts.

That was a bug that was fixed before the series was merged.

See:
f5451547b831 ("mm, slab/slub: Ensure kmem_cache_alloc_bulk() is available early")

It looks like the trigger here is that the maple tree code uses call_rcu() in ma_free_rcu(), and call_rcu() can cause TIF_NEED_RESCHED to be set, which causes cond_resched() to actually reschedule, enabling interrupts.

cheers

Quoted 3 lines

(thanks Jordan Niethe for finding this thread)

[1]: https://lore.kernel.org/all/87o7qdzfay.ffs@tglx/