Skip to contentSkip to conversation

Adjustment after introducing ASSERT_ALLOC_CONTEXT

PATCH In review Henry Wang 2 participants 8 messages May 7, 2022 – Jun 8, 2022

lore
Share & tools
lore

Henry Wang

author
#0 lore raw

v3 cover Adjustment after introducing ASSERT_ALLOC_CONTEXT

With the enhanced ASSERT_ALLOC_CONTEXT, calling request_irq before local_irq_enable on secondary cores on Arm will lead to

(XEN) Xen call trace:
(XEN) [<000000000021d86c>] alloc_xenheap_pages+0x74/0x194 (PC) (XEN) [<000000000021d864>] alloc_xenheap_pages+0x6c/0x194 (LR) (XEN) [<0000000000229e90>] xmalloc_tlsf.c#xmalloc_pool_get+0x1c/0x28 (XEN) [<000000000022a270>] xmem_pool_alloc+0x21c/0x448 (XEN) [<000000000022a8dc>] _xmalloc+0x8c/0x290
(XEN) [<000000000026b57c>] request_irq+0x40/0xb8
(XEN) [<0000000000272780>] init_timer_interrupt+0x74/0xcc (XEN) [<000000000027212c>] start_secondary+0x1b4/0x238 (XEN) [<0000000084000200>] 0000000084000200
(XEN)
(XEN)
(XEN) ****************************************
(XEN) Panic on CPU 4:
(XEN) Assertion '!in_irq() && (local_irq_is_enabled() || num_online_cpus() <= 1)' failed at common/page_alloc.c:2212 (XEN) ****************************************

on systems without a big enough pool for xmalloc() to cater the requested size. To solve this issue, this series introduces two patches. The first one defers the calling of request_irq on secondary CPUs after local_irq_enable on Arm. The second one moves the definition of ASSERT_ALLOC_CONTEXT to header and uses the ASSERT_ALLOC_CONTEXT to replace the original assertion in xmalloc().

Also take the chance to enhance the assertion in xmalloc_tlsf.c.

Henry Wang (2):
xen/arm: Defer request_irq on secondary CPUs after local_irq_enable xen/common: Use enhanced ASSERT_ALLOC_CONTEXT in xmalloc()

 xen/arch/arm/smpboot.c    | 12 +++++++++---
 xen/common/page_alloc.c   |  7 -------
 xen/common/xmalloc_tlsf.c | 10 +++++++---
 xen/include/xen/irq.h     |  7 +++++++
 4 files changed, 23 insertions(+), 13 deletions(-)
Signature · 1 line
2.25.1
author
#1 lore raw

PATCH v3 1/2 xen/arm: Defer request_irq on secondary CPUs after local_irq_enable

With the enhanced ASSERT_ALLOC_CONTEXT, calling request_irq before local_irq_enable on secondary cores will lead to

(XEN) Xen call trace:
(XEN) [<000000000021d86c>] alloc_xenheap_pages+0x74/0x194 (PC) (XEN) [<000000000021d864>] alloc_xenheap_pages+0x6c/0x194 (LR) (XEN) [<0000000000229e90>] xmalloc_tlsf.c#xmalloc_pool_get+0x1c/0x28 (XEN) [<000000000022a270>] xmem_pool_alloc+0x21c/0x448 (XEN) [<000000000022a8dc>] _xmalloc+0x8c/0x290
(XEN) [<000000000026b57c>] request_irq+0x40/0xb8
(XEN) [<0000000000272780>] init_timer_interrupt+0x74/0xcc (XEN) [<000000000027212c>] start_secondary+0x1b4/0x238 (XEN) [<0000000084000200>] 0000000084000200
(XEN)
(XEN)
(XEN) ****************************************
(XEN) Panic on CPU 4:
(XEN) Assertion '!in_irq() && (local_irq_is_enabled() || num_online_cpus() <= 1)' failed at common/page_alloc.c:2212 (XEN) ****************************************

on systems without a big enough pool for xmalloc() to cater the requested size.

Moving the call of request_irq() past local_irq_enable() on secondary cores will make sure the assertion condition in alloc_xenheap_pages(), i.e. !in_irq && local_irq_enabled() is satisfied. It is also safe because the timer and GIC maintenance interrupt will not be used until the CPU is fully online.

---
v2 -> v3:
- No changes.
v1 -> v2:
- Explain why the moving of code is safe in the commit message and add comments.

---
 xen/arch/arm/smpboot.c | 12 +++++++++---
 1 file changed, 9 insertions(+), 3 deletions(-)
xen/arch/arm/smpboot.c +9 -3
index 7bfd0a73a7..9bb32a301a 100644
@@ -361,9 +361,6 @@ void start_secondary(void)
 
     init_secondary_IRQ();
 
-    init_maintenance_interrupt();
-    init_timer_interrupt();
-
     set_current(idle_vcpu[cpuid]);
 
     setup_cpu_sibling_map(cpuid);
@@ -380,6 +377,15 @@ void start_secondary(void)
     cpumask_set_cpu(cpuid, &cpu_online_map);
 
     local_irq_enable();
+
+    /*
+     * Calling request_irq() after local_irq_enable() on secondary cores
+     * will make sure the assertion condition in alloc_xenheap_pages(),
+     * i.e. !in_irq && local_irq_enabled() is satisfied.
+     */
+    init_maintenance_interrupt();
+    init_timer_interrupt();
+
     local_abort_enable();
 
     check_local_cpu_errata();
Signature · 1 line
2.25.1
Reviewed-by on v3 1/2
#2 lore raw

Re: [PATCH v3 1/2] xen/arm: Defer request_irq on secondary CPUs after local_irq_enable

Hi Henry,

Quoted 30 lines · Henry Wang

With the enhanced ASSERT_ALLOC_CONTEXT, calling request_irq before local_irq_enable on secondary cores will lead to

(XEN) Xen call trace:
(XEN) [<000000000021d86c>] alloc_xenheap_pages+0x74/0x194 (PC) (XEN) [<000000000021d864>] alloc_xenheap_pages+0x6c/0x194 (LR) (XEN) [<0000000000229e90>] xmalloc_tlsf.c#xmalloc_pool_get+0x1c/0x28 (XEN) [<000000000022a270>] xmem_pool_alloc+0x21c/0x448 (XEN) [<000000000022a8dc>] _xmalloc+0x8c/0x290
(XEN) [<000000000026b57c>] request_irq+0x40/0xb8
(XEN) [<0000000000272780>] init_timer_interrupt+0x74/0xcc (XEN) [<000000000027212c>] start_secondary+0x1b4/0x238 (XEN) [<0000000084000200>] 0000000084000200
(XEN)
(XEN)
(XEN) ****************************************
(XEN) Panic on CPU 4:
(XEN) Assertion '!in_irq() && (local_irq_is_enabled() || num_online_cpus() <= 1)' failed at common/page_alloc.c:2212 (XEN) ****************************************

on systems without a big enough pool for xmalloc() to cater the requested size.

Moving the call of request_irq() past local_irq_enable() on secondary cores will make sure the assertion condition in alloc_xenheap_pages(), i.e. !in_irq && local_irq_enabled() is satisfied. It is also safe because the timer and GIC maintenance interrupt will not be used until the CPU is fully online.

Quoted 3 lines

Cheers,

Signature · 1 line
Julien Grall
#3 lore raw

Re: [PATCH v3 1/2] xen/arm: Defer request_irq on secondary CPUs after local_irq_enable

Quoted 31 lines · Julien Grall
Quoted 30 lines · Henry Wang

With the enhanced ASSERT_ALLOC_CONTEXT, calling request_irq before local_irq_enable on secondary cores will lead to

(XEN) Xen call trace:
(XEN) [<000000000021d86c>] alloc_xenheap_pages+0x74/0x194 (PC) (XEN) [<000000000021d864>] alloc_xenheap_pages+0x6c/0x194 (LR) (XEN) [<0000000000229e90>] xmalloc_tlsf.c#xmalloc_pool_get+0x1c/0x28 (XEN) [<000000000022a270>] xmem_pool_alloc+0x21c/0x448 (XEN) [<000000000022a8dc>] _xmalloc+0x8c/0x290
(XEN) [<000000000026b57c>] request_irq+0x40/0xb8
(XEN) [<0000000000272780>] init_timer_interrupt+0x74/0xcc (XEN) [<000000000027212c>] start_secondary+0x1b4/0x238 (XEN) [<0000000084000200>] 0000000084000200
(XEN)
(XEN)
(XEN) ****************************************
(XEN) Panic on CPU 4:
(XEN) Assertion '!in_irq() && (local_irq_is_enabled() || num_online_cpus() <= 1)' failed at common/page_alloc.c:2212 (XEN) ****************************************

on systems without a big enough pool for xmalloc() to cater the requested size.

Moving the call of request_irq() past local_irq_enable() on secondary cores will make sure the assertion condition in alloc_xenheap_pages(), i.e. !in_irq && local_irq_enabled() is satisfied. It is also safe because the timer and GIC maintenance interrupt will not be used until the CPU is fully online.

Quoted 5 lines
Quoted 3 lines

I have committed this patch. The second patch will go in once "page_alloc: assert IRQs are enabled in heap alloc/free" has been re-committed.

Cheers,

Signature · 1 line
Julien Grall
author
#4 lore raw

PATCH v3 2/2 xen/common: Use enhanced ASSERT_ALLOC_CONTEXT in xmalloc()

In reply to Henry Wang

xmalloc() will use a pool for allocation smaller than a page. The pool is extended only when there are no more space. At which point, alloc_xenheap_pages() is called to add more memory.

xmalloc() must be protected by ASSERT_ALLOC_CONTEXT. It should not rely on pool expanding to trigger the ASSERT_ALLOC_CONTEXT in alloc_xenheap_pages(). Hence, this commit moves the definition of ASSERT_ALLOC_CONTEXT to header and uses the ASSERT_ALLOC_CONTEXT to replace the original assertion in xmalloc().

For consistency, the same assertion should be used in xfree(), and the position of the assertion should be at the beginning of the xfree().

Also take the opportunity to enhance the non-static functions xmem_pool_{alloc,free}() with the same assertion so that future callers of these two functions can be benefited.

---
v2 -> v3:
- Add ASSERT_ALLOC_CONTEXT in xmem_pool_{alloc,free}()
- Also change the assertion in xfree to ASSERT_ALLOC_CONTEXT and move the position of assertion to the beginning of the function. v1 -> v2:
- No changes

---
 xen/common/page_alloc.c   |  7 -------
 xen/common/xmalloc_tlsf.c | 10 +++++++---
 xen/include/xen/irq.h     |  7 +++++++
 3 files changed, 14 insertions(+), 10 deletions(-)
xen/common/page_alloc.c +0 -7
index e866e0d864..ea59cd1a4a 100644
@@ -162,13 +162,6 @@
 static char __initdata opt_badpage[100] = "";
 string_param("badpage", opt_badpage);
 
-/*
- * Heap allocations may need TLB flushes which may require IRQs to be
- * enabled (except when only 1 PCPU is online).
- */
-#define ASSERT_ALLOC_CONTEXT() \
-    ASSERT(!in_irq() && (local_irq_is_enabled() || num_online_cpus() <= 1))
-
 /*
  * no-bootscrub -> Free pages are not zeroed during boot.
  */
xen/common/xmalloc_tlsf.c +7 -3
index d2ad909502..75bdf18c4e 100644
@@ -378,6 +378,8 @@ void *xmem_pool_alloc(unsigned long size, struct xmem_pool *pool)
     int fl, sl;
     unsigned long tmp_size;
 
+    ASSERT_ALLOC_CONTEXT();
+
     if ( size < MIN_BLOCK_SIZE )
         size = MIN_BLOCK_SIZE;
     else
@@ -456,6 +458,8 @@ void xmem_pool_free(void *ptr, struct xmem_pool *pool)
     struct bhdr *b, *tmp_b;
     int fl = 0, sl = 0;
 
+    ASSERT_ALLOC_CONTEXT();
+
     if ( unlikely(ptr == NULL) )
         return;
 
@@ -594,7 +598,7 @@ void *_xmalloc(unsigned long size, unsigned long align)
 {
     void *p = NULL;
 
-    ASSERT(!in_irq());
+    ASSERT_ALLOC_CONTEXT();
 
     if ( !size )
         return ZERO_BLOCK_PTR;
@@ -697,11 +701,11 @@ void *_xrealloc(void *ptr, unsigned long size, unsigned long align)
 
 void xfree(void *p)
 {
+    ASSERT_ALLOC_CONTEXT();
+
     if ( p == NULL || p == ZERO_BLOCK_PTR )
         return;
 
-    ASSERT(!in_irq());
-
     if ( !((unsigned long)p & (PAGE_SIZE - 1)) )
     {
         unsigned long size = PFN_ORDER(virt_to_page(p));
xen/include/xen/irq.h +7 -0
index d8beadd16b..300625e56d 100644
@@ -10,6 +10,13 @@
 #include <asm/hardirq.h>
 #include <public/event_channel.h>
 
+/*
+ * Heap allocations may need TLB flushes which may require IRQs to be
+ * enabled (except when only 1 PCPU is online).
+ */
+#define ASSERT_ALLOC_CONTEXT() \
+    ASSERT(!in_irq() && (local_irq_is_enabled() || num_online_cpus() <= 1))
+
 struct irqaction {
     void (*handler)(int, void *, struct cpu_user_regs *);
     const char *name;
Signature · 1 line
2.25.1
Tested-by Acked-by on v3 2/2
#5 lore raw

Re: [PATCH v3 2/2] xen/common: Use enhanced ASSERT_ALLOC_CONTEXT in xmalloc()

Hi,

Quoted 5 lines · Henry Wang
xen/common/page_alloc.c +0 -0
index e866e0d864..ea59cd1a4a 100644

xen/common/page_alloc.c line 171

Quoted 10 lines
xen/common/page_alloc.c +0 -7
@@ -162,13 +162,6 @@
  static char __initdata opt_badpage[100] = "";
  string_param("badpage", opt_badpage);
 
-/*
- * Heap allocations may need TLB flushes which may require IRQs to be
- * enabled (except when only 1 PCPU is online).
- */
-#define ASSERT_ALLOC_CONTEXT() \
-    ASSERT(!in_irq() && (local_irq_is_enabled() || num_online_cpus() <= 1))
-

FYI, the patch introducing ASSERT_ALLOC_CONTEXT() has been reverted. I intend to re-introduce it once your previous patch and the one fixing the ITS (not yet formally sent) have been committed.

I have also checked that none of the ASSERTs() would be triggered on my x86 setup. So:

On a side note (no action expected for you), I noticed that the ASSERT()s would only trigger from CPU2 and onwards at least for Arm. This is because num_online_cpus() would still be 1 when bringing-up CPU1.

I went through the original discussion and I am not sure why we switched from < SYS_STATE_smp_boot to num_online_cpus() (aside that Arm doesn't set it). Aynway, this is not a major issue here as this is an ASSERT().

Cheers,

Signature · 1 line
Julien Grall
author
#6 lore raw

RE: [PATCH v3 2/2] xen/common: Use enhanced ASSERT_ALLOC_CONTEXT in xmalloc()

Hi Julien,

Quoted 23 lines

-----Original Message-----
From: Julien Grall <[email protected]>
Subject: Re: [PATCH v3 2/2] xen/common: Use enhanced ASSERT_ALLOC_CONTEXT in xmalloc()

Hi,

Quoted 5 lines · Henry Wang
xen/common/page_alloc.c +0 -0
index e866e0d864..ea59cd1a4a 100644

xen/common/page_alloc.c line 170

Quoted 9 lines
xen/common/page_alloc.c +0 -6
@@ -162,13 +162,6 @@
  static char __initdata opt_badpage[100] = "";
  string_param("badpage", opt_badpage);
 
-/*
- * Heap allocations may need TLB flushes which may require IRQs to be
- * enabled (except when only 1 PCPU is online).
- */
-#define ASSERT_ALLOC_CONTEXT() \
-    ASSERT(!in_irq() && (local_irq_is_enabled() || num_online_cpus() <=

1))

Quoted 4 lines

xen/common/page_alloc.c line 171

Quoted 1 line
xen/common/page_alloc.c +0 -1
-

FYI, the patch introducing ASSERT_ALLOC_CONTEXT() has been reverted. I intend to re-introduce it once your previous patch and the one fixing the ITS (not yet formally sent) have been committed.

Thanks for the information! IIUC the patch:
"xen/arm: gic-v3-lpi: Allocate the pending table while preparing the CPU" is merged. So I guess both "page_alloc: assert IRQs are enabled in heap alloc/free" and this patch can be re-introduced if everyone is happy with the patch?

Quoted 6 lines

I have also checked that none of the ASSERTs() would be triggered on my x86 setup. So:

Thank you very much for both testing and acking this patch!

Kind regards,
Henry

Quoted 14 lines

On a side note (no action expected for you), I noticed that the ASSERT()s would only trigger from CPU2 and onwards at least for Arm. This is because num_online_cpus() would still be 1 when bringing-up CPU1.

I went through the original discussion and I am not sure why we switched from < SYS_STATE_smp_boot to num_online_cpus() (aside that Arm doesn't
set it). Aynway, this is not a major issue here as this is an ASSERT().

Cheers,

--
Julien Grall

#7 lore raw

Re: [PATCH v3 2/2] xen/common: Use enhanced ASSERT_ALLOC_CONTEXT in xmalloc()

Hi Henry,

Quoted 24 lines · Henry Wang
Quoted 24 lines

-----Original Message-----
From: Julien Grall <[email protected]>
Subject: Re: [PATCH v3 2/2] xen/common: Use enhanced ASSERT_ALLOC_CONTEXT in xmalloc()

Hi,

Quoted 5 lines · Henry Wang
xen/common/page_alloc.c +0 -0
index e866e0d864..ea59cd1a4a 100644

xen/common/page_alloc.c line 170

Quoted 9 lines
xen/common/page_alloc.c +0 -6
@@ -162,13 +162,6 @@
   static char __initdata opt_badpage[100] = "";
   string_param("badpage", opt_badpage);
 
-/*
- * Heap allocations may need TLB flushes which may require IRQs to be
- * enabled (except when only 1 PCPU is online).
- */
-#define ASSERT_ALLOC_CONTEXT() \
-    ASSERT(!in_irq() && (local_irq_is_enabled() || num_online_cpus() <=

1))

Quoted 1 line
xen/common/page_alloc.c +0 -1
-
Quoted 8 lines
Quoted 3 lines

FYI, the patch introducing ASSERT_ALLOC_CONTEXT() has been reverted. I intend to re-introduce it once your previous patch and the one fixing the ITS (not yet formally sent) have been committed.

Thanks for the information! IIUC the patch:
"xen/arm: gic-v3-lpi: Allocate the pending table while preparing the CPU" is merged. So I guess both "page_alloc: assert IRQs are enabled in heap alloc/free" and this patch can be re-introduced if everyone is happy with the patch?

I have re-committed David's patch and committed yours.

Cheers,

Signature · 1 line
Julien Grall