[PATCH] always print note when trimming EFI memory map

[PATCH] always print note when trimming EFI memory map

[Bjorn Helgaas]

I think it would be a good idea to always print a note when trimming
the EFI memory map, even if it's not "available" memory.

I just debugged a problem where we trimmed out some memory that
contained ACPI tables.  When we went to look at a table later,
we used the wrong attribute because we couldn't find its address
in the EFI memory map.  This caused an MCA (uncached access to
memory).  Straightforward to debug, but the note would have
made it quicker.

This patch is against 2.5.

Bjorn

=== arch/ia64/kernel/efi.c 1.23 vs edited ==--- 1.23/arch/ia64/kernel/efi.c	Wed Aug 27 10:47:27 2003
+++ edited/arch/ia64/kernel/efi.c	Wed Sep 24 20:14:01 2003
@@ -249,11 +249,10 @@
 	if (num_skipped_pages > md->num_pages)
 		num_skipped_pages = md->num_pages;
 
-	if (is_available_memory(md))
-		printk(KERN_NOTICE "efi.%s: ignoring %luKB of memory at 0x%lx due to granule hole "
-		       "at 0x%lx\n", __FUNCTION__,
-		       (num_skipped_pages << EFI_PAGE_SHIFT) >> 10,
-		       md->phys_addr, start_addr - IA64_GRANULE_SIZE);
+	printk(KERN_NOTICE "efi.%s: ignoring %luKB of memory at 0x%lx due to granule hole "
+	       "at 0x%lx\n", __FUNCTION__,
+	       (num_skipped_pages << EFI_PAGE_SHIFT) >> 10,
+	       md->phys_addr, start_addr - IA64_GRANULE_SIZE);
 	/*
 	 * NOTE: Don't set md->phys_addr to START_ADDR because that could cause the memory
 	 * descriptor list to become unsorted.  In such a case, md->num_pages will be
@@ -277,11 +276,10 @@
 	if (num_dropped_pages > md->num_pages)
 		num_dropped_pages = md->num_pages;
 
-	if (is_available_memory(md))
-		printk(KERN_NOTICE "efi.%s: ignoring %luKB of memory at 0x%lx due to granule hole "
-		       "at 0x%lx\n", __FUNCTION__,
-		       (num_dropped_pages << EFI_PAGE_SHIFT) >> 10,
-		       md->phys_addr, end_addr);
+	printk(KERN_NOTICE "efi.%s: ignoring %luKB of memory at 0x%lx due to granule hole "
+	       "at 0x%lx\n", __FUNCTION__,
+	       (num_dropped_pages << EFI_PAGE_SHIFT) >> 10,
+	       md->phys_addr, end_addr);
 	md->num_pages -= num_dropped_pages;
 }
 

Re: [PATCH] always print note when trimming EFI memory map

[David Mosberger]

>>>>> On Wed, 24 Sep 2003 17:53:14 -0600, Bjorn Helgaas <bjorn.helgaas@hp.com> said:

  Bjorn> I think it would be a good idea to always print a note when
  Bjorn> trimming the EFI memory map, even if it's not "available"
  Bjorn> memory.

  Bjorn> I just debugged a problem where we trimmed out some memory
  Bjorn> that contained ACPI tables.  When we went to look at a table
  Bjorn> later, we used the wrong attribute because we couldn't find
  Bjorn> its address in the EFI memory map.  This caused an MCA
  Bjorn> (uncached access to memory).  Straightforward to debug, but
  Bjorn> the note would have made it quicker.

The intention was to give the user an idea how much real & available
memory is lost due to the trimming.  When reporting unavailable
memory, you'll lose that info.  However, maybe we should be more
verbose if EFI_DEBUG is enabled?

	--david

Re: [PATCH] always print note when trimming EFI memory map

[Bjorn Helgaas]

On Friday 26 September 2003 2:32 pm, David Mosberger wrote:
> The intention was to give the user an idea how much real & available
> memory is lost due to the trimming.  When reporting unavailable
> memory, you'll lose that info.  However, maybe we should be more
> verbose if EFI_DEBUG is enabled?

How about something like the following?  It still always prints a
note when trimming, but identifies "reserved" memory to distinguish
it from the normal case.

This is the second or third time I've had to debug the problem,
and the failure mode doesn't obviously point to EFI_DEBUG.  It
just MCAs because you try to access main memory with a UC access.
In fact, I think on most boxes you don't even get an MCA; it
just appears to work, even though it introduces attribute
aliasing.

=== arch/ia64/kernel/efi.c 1.23 vs edited ==--- 1.23/arch/ia64/kernel/efi.c	Wed Aug 27 10:47:27 2003
+++ edited/arch/ia64/kernel/efi.c	Fri Sep 26 17:33:36 2003
@@ -249,11 +249,11 @@
 	if (num_skipped_pages > md->num_pages)
 		num_skipped_pages = md->num_pages;
 
-	if (is_available_memory(md))
-		printk(KERN_NOTICE "efi.%s: ignoring %luKB of memory at 0x%lx due to granule hole "
-		       "at 0x%lx\n", __FUNCTION__,
-		       (num_skipped_pages << EFI_PAGE_SHIFT) >> 10,
-		       md->phys_addr, start_addr - IA64_GRANULE_SIZE);
+	printk(KERN_NOTICE "efi.%s: ignoring %luKB of %smemory at 0x%lx due to granule hole "
+	       "at 0x%lx\n", __FUNCTION__,
+	       (num_skipped_pages << EFI_PAGE_SHIFT) >> 10,
+	       is_available_memory(md) ? "" : "reserved ",
+	       md->phys_addr, start_addr - IA64_GRANULE_SIZE);
 	/*
 	 * NOTE: Don't set md->phys_addr to START_ADDR because that could cause the memory
 	 * descriptor list to become unsorted.  In such a case, md->num_pages will be
@@ -277,11 +277,11 @@
 	if (num_dropped_pages > md->num_pages)
 		num_dropped_pages = md->num_pages;
 
-	if (is_available_memory(md))
-		printk(KERN_NOTICE "efi.%s: ignoring %luKB of memory at 0x%lx due to granule hole "
-		       "at 0x%lx\n", __FUNCTION__,
-		       (num_dropped_pages << EFI_PAGE_SHIFT) >> 10,
-		       md->phys_addr, end_addr);
+	printk(KERN_NOTICE "efi.%s: ignoring %luKB of %smemory at 0x%lx due to granule hole "
+	       "at 0x%lx\n", __FUNCTION__,
+	       (num_dropped_pages << EFI_PAGE_SHIFT) >> 10,
+	       is_available_memory(md) ? "" : "reserved ",
+	       md->phys_addr, end_addr);
 	md->num_pages -= num_dropped_pages;
 }