Skip to contentSkip to conversation

PATCH 2.4.23-pre6 add kmap_types.h for CONFIG_CRYPTO

PATCH Grant Grundler 6 participants 12 messages Oct 16, 2003 – Oct 24, 2003

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

CONFIG_CRYPTO requires asm/kmaps_types.h.
patch below clones kmap_types.h from asm-i386.

This doesn't feel like a great solution.
Willy suggested the right fix might be to not require all arches to provide asm/kmap_types.h. I'll let the folks who know WTH kmap_types.h is really for debate and submit this with the hope of getting the ia64-linux to build with CONFIG_CRYPTO.

grant

include/asm-ia64/kmap_types.h +16 -0
diff -Nur --exclude=RCS --exclude=CVS --exclude=SCCS --exclude=BitKeeper --exclude=ChangeSet linux-ia64-2.4-orig/include/asm-ia64/kmap_types.h linux-ia64-2.4/include/asm-ia64/kmap_types.h
@@ -0,0 +1,16 @@
+#ifndef _ASM_KMAP_TYPES_H
+#define _ASM_KMAP_TYPES_H
+
+enum km_type {
+	KM_BOUNCE_READ,
+	KM_SKB_SUNRPC_DATA,
+	KM_SKB_DATA_SOFTIRQ,
+	KM_USER0,
+	KM_USER1,
+	KM_BH_IRQ,
+	KM_SOFTIRQ0,
+	KM_SOFTIRQ1,
+	KM_TYPE_NR
+};
+
+#endif
#1 lore raw
Quoted 5 lines · Grant Grundler

This doesn't feel like a great solution.
Willy suggested the right fix might be to not require all arches to provide asm/kmap_types.h. I'll let the folks who know WTH kmap_types.h is really for debate and submit this with the hope of getting the ia64-linux to build with CONFIG_CRYPTO.

Cloning kmap_types from i386 is a pretty bad idea since that architecture actually supports highmem. Better to clone from an arch which has no support for highmem such as ARM.

Signature · 3 lines
"It's not Hollywood.  War is real, war is primarily not about defeat or
victory, it is about death.  I've seen thousands and thousands of dead bodies.
Do you think I want to have an academic debate on this subject?" -- Robert Fisk
#2 lore raw
In reply to Grant Grundler
Quoted 3 lines · Matthew Wilcox

Cloning kmap_types from i386 is a pretty bad idea since that architecture actually supports highmem. Better to clone from an arch which has no support for highmem such as ARM.

uhm, I would, but...

grundler@gsyprf3:/usr/src/linux-ia64-2.4-orig$ ls -l include/asm-arm/kma*
ls: include/asm-arm/kma*: No such file or directory

Anyway, given ia64 doesn't otherwise need kmap_types.h, I suspect it doesn't matter. Except that it's misleading. Maybe a stripped down version of the file would be better.

grant

#3 lore raw
In reply to Grant Grundler
Quoted 2 lines · Grant Grundler

CONFIG_CRYPTO requires asm/kmaps_types.h.
patch below clones kmap_types.h from asm-i386.

That is an error in crypto/internal.h. Nothing should include asm/kmap_types.h directly, they should include linux/highmem.h and let that decide if the arch needs kmap or not.

#4 lore raw
In reply to Grant Grundler
Quoted 3 lines · Keith Owens

That is an error in crypto/internal.h. Nothing should include asm/kmap_types.h directly, they should include linux/highmem.h and let that decide if the arch needs kmap or not.

The would be the logical conclusion, but the crypto maintainers decided this is a feature and not a bug..

#5 lore raw
In reply to Grant Grundler
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

"Matthew" = Matthew Wilcox <[email protected]> writes:

Matthew> On Thu, Oct 16, 2003 at 12:29:58PM -0700, Grant Grundler

Quoted 5 lines · Matthew> wrote:
Quoted 5 lines

This doesn't feel like a great solution. Willy suggested the right fix might be to not require all arches to provide asm/kmap_types.h. I'll let the folks who know WTH kmap_types.h is really for debate and submit this with the hope of getting the ia64-linux to build with CONFIG_CRYPTO.

Matthew> Cloning kmap_types from i386 is a pretty bad idea since that Matthew> architecture actually supports highmem. Better to clone from Matthew> an arch which has no support for highmem such as ARM.

I'd argue we should fix the crypto code instead, it doesn't seem reasonable for them to require kmap types to be defined on archs that don't have highmem.

Cheers,
Jes

#6 lore raw
In reply to Grant Grundler
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

"Christoph" = Christoph Hellwig <[email protected]> writes:

Christoph> On Fri, Oct 17, 2003 at 09:20:33AM +1000, Keith Owens

Quoted 3 lines · Christoph> wrote:
Quoted 3 lines

That is an error in crypto/internal.h. Nothing should include asm/kmap_types.h directly, they should include linux/highmem.h and let that decide if the arch needs kmap or not.

Christoph> The would be the logical conclusion, but the crypto Christoph> maintainers decided this is a feature and not a bug..

Well then we send Linus a big-fix, clearly this is a bug, if the crypto maintainer doesn't want to fix it, we'll have to fix it for him/her.

Jes

#7 lore raw
In reply to Grant Grundler

Here's my proposal for a fix. I don't think I'll send it to Linus/Andrew yet, because it's really a cleanup, not an actual bug fix, but here it is for any comments.

Bjorn

Currently every architecture must supply kmap_types.h, even though it only makes sense for architectures that use highmem.

This patch removes the need for generic code to know about kmap_types.h by adding the appropriate #include to include/linux/highmem.h and defining a minimal set of KM_ enums that are used by generic code.

One might argue that we shouldn't need to define even the minimal set of KM_ enums, but generic code needs to be able to do, for example:

	kmap_atomic(page, KM_USER);

In many cases the KM_USER needn't be defined at all because the non-highmem kmap_atomic() never evaluates it, but leaving it undefined also restricts generic code from doing other things like defining an array of KM_ types. crypto/cipher.c defines such an array, and it seems non-intuitive to prevent that usage.

This allows the removal of the following dummy kmap_types.h files:

	include/asm-alpha/kmap_types.h
	include/asm-arm/kmap_types.h
	include/asm-arm26/kmap_types.h
	include/asm-cris/kmap_types.h
	include/asm-h8300/kmap_types.h
	include/asm-ia64/kmap_types.h
	include/asm-m68k/kmap_types.h
	include/asm-m68knommu/kmap_types.h
	include/asm-parisc/kmap_types.h
	include/asm-ppc64/kmap_types.h
	include/asm-s390/kmap_types.h
	include/asm-sh/kmap_types.h
	include/asm-sparc64/kmap_types.h
	include/asm-v850/kmap_types.h
	include/asm-x86_64/kmap_types.h

This is against 2.6 and has been boot tested on ia64 and x86.

Bjorn

# This is a BitKeeper generated patch for the following project:
# Project Name: Linux kernel tree
# This patch format is intended for GNU patch command version 2.5 or higher. # This patch includes the following deltas:
# ChangeSet 1.1368 -> 1.1369
# include/asm-m68knommu/kmap_types.h 1.1 -> (deleted)
# crypto/internal.h 1.19 -> 1.20

#	include/asm-m68k/kmap_types.h	1.3     ->         (deleted)      
#	include/asm-v850/kmap_types.h	1.1     ->         (deleted)      
#	include/asm-cris/kmap_types.h	1.1     ->         (deleted)      

# include/asm-arm26/kmap_types.h 1.1 -> (deleted)
# fs/aio.c 1.37 -> 1.38
# include/asm-sh/kmap_types.h 1.1 -> (deleted)
# include/asm-ia64/kmap_types.h 1.5 -> (deleted)
# include/asm-ppc64/kmap_types.h 1.4 -> (deleted)
# include/asm-parisc/kmap_types.h 1.3 -> (deleted)
# include/asm-s390/kmap_types.h 1.4 -> (deleted)
# include/asm-h8300/kmap_types.h 1.1 -> (deleted)
# include/asm-x86_64/kmap_types.h 1.7 -> (deleted)
# include/linux/highmem.h 1.27 -> 1.28
# include/asm-sparc64/kmap_types.h 1.4 -> (deleted)
# include/asm-arm/kmap_types.h 1.1 -> (deleted)
# include/asm-alpha/kmap_types.h 1.5 -> (deleted)
#
# The following is the BitKeeper ChangeSet Log
# --------------------------------------------
# 03/10/23 [email protected] 1.1369
# Remove dummy kmap_types.h files. Only architectures that use highmem

# really need those files.
# --------------------------------------------
#
diff -Nru a/crypto/internal.h b/crypto/internal.h
--- a/crypto/internal.h	Thu Oct 23 10:54:38 2003
+++ b/crypto/internal.h	Thu Oct 23 10:54:38 2003
@@ -17,7 +17,6 @@
 #include <linux/init.h>
 #include <linux/kmod.h>
 #include <asm/hardirq.h>
-#include <asm/kmap_types.h>

extern enum km_type crypto_km_types[];

fs/aio.c +0 -1
diff -Nru a/fs/aio.c b/fs/aio.c
@@ -28,7 +28,6 @@
 #include <linux/highmem.h>
 #include <linux/workqueue.h>
 
-#include <asm/kmap_types.h>
 #include <asm/uaccess.h>
 #include <asm/mmu_context.h>
 
include/asm-alpha/kmap_types.h deleted +0 -33
diff -Nru a/include/asm-alpha/kmap_types.h b/include/asm-alpha/kmap_types.h
@@ -1,33 +0,0 @@
-#ifndef _ASM_KMAP_TYPES_H
-#define _ASM_KMAP_TYPES_H
-
-/* Dummy header just to define km_type. */
-
-#include <linux/config.h>
-
-#ifdef CONFIG_DEBUG_HIGHMEM
-# define D(n) __KM_FENCE_##n ,
-#else
-# define D(n)
-#endif
-
-enum km_type {
-D(0)	KM_BOUNCE_READ,
-D(1)	KM_SKB_SUNRPC_DATA,
-D(2)	KM_SKB_DATA_SOFTIRQ,
-D(3)	KM_USER0,
-D(4)	KM_USER1,
-D(5)	KM_BIO_SRC_IRQ,
-D(6)	KM_BIO_DST_IRQ,
-D(7)	KM_PTE0,
-D(8)	KM_PTE1,
-D(9)	KM_IRQ0,
-D(10)	KM_IRQ1,
-D(11)	KM_SOFTIRQ0,
-D(12)	KM_SOFTIRQ1,
-D(13)	KM_TYPE_NR
-};
-
-#undef D
-
-#endif
include/asm-arm/kmap_types.h deleted +0 -12
diff -Nru a/include/asm-arm/kmap_types.h b/include/asm-arm/kmap_types.h
@@ -1,12 +0,0 @@
-#ifndef __ARM_KMAP_TYPES_H
-#define __ARM_KMAP_TYPES_H
-
-/*
- * This is the "bare minimum".  AIO seems to require this.
- */
-enum km_type {
-	KM_IRQ0,
-	KM_USER1
-};
-
-#endif
include/asm-arm26/kmap_types.h deleted +0 -12
diff -Nru a/include/asm-arm26/kmap_types.h b/include/asm-arm26/kmap_types.h
@@ -1,12 +0,0 @@
-#ifndef __ARM_KMAP_TYPES_H
-#define __ARM_KMAP_TYPES_H
-
-/*
- * This is the "bare minimum".  AIO seems to require this.
- */
-enum km_type {
-        KM_IRQ0,
-        KM_USER1
-};
-
-#endif
include/asm-cris/kmap_types.h deleted +0 -25
diff -Nru a/include/asm-cris/kmap_types.h b/include/asm-cris/kmap_types.h
@@ -1,25 +0,0 @@
-#ifndef _ASM_KMAP_TYPES_H
-#define _ASM_KMAP_TYPES_H
-
-/* Dummy header just to define km_type.  None of this
- * is actually used on cris. 
- */
-
-enum km_type {
-	KM_BOUNCE_READ,
-	KM_SKB_SUNRPC_DATA,
-	KM_SKB_DATA_SOFTIRQ,
-	KM_USER0,
-	KM_USER1,
-	KM_BIO_SRC_IRQ,
-	KM_BIO_DST_IRQ,
-	KM_PTE0,
-	KM_PTE1,
-	KM_IRQ0,
-	KM_IRQ1,
-	KM_CRYPTO_USER,
-	KM_CRYPTO_SOFTIRQ,
-	KM_TYPE_NR
-};
-
-#endif
include/asm-h8300/kmap_types.h deleted +0 -19
diff -Nru a/include/asm-h8300/kmap_types.h b/include/asm-h8300/kmap_types.h
@@ -1,19 +0,0 @@
-#ifndef _ASM_KMAP_TYPES_H
-#define _ASM_KMAP_TYPES_H
-
-enum km_type {
-	KM_BOUNCE_READ,
-	KM_SKB_SUNRPC_DATA,
-	KM_SKB_DATA_SOFTIRQ,
-	KM_USER0,
-	KM_USER1,
-	KM_BIO_SRC_IRQ,
-	KM_BIO_DST_IRQ,
-	KM_PTE0,
-	KM_PTE1,
-	KM_IRQ0,
-	KM_IRQ1,
-	KM_TYPE_NR
-};
-
-#endif
include/asm-ia64/kmap_types.h deleted +0 -31
diff -Nru a/include/asm-ia64/kmap_types.h b/include/asm-ia64/kmap_types.h
@@ -1,31 +0,0 @@
-#ifndef _ASM_IA64_KMAP_TYPES_H
-#define _ASM_IA64_KMAP_TYPES_H
-
-#include <linux/config.h>
-
-#ifdef CONFIG_DEBUG_HIGHMEM
-# define D(n) __KM_FENCE_##n ,
-#else
-# define D(n)
-#endif
-
-enum km_type {
-D(0)	KM_BOUNCE_READ,
-D(1)	KM_SKB_SUNRPC_DATA,
-D(2)	KM_SKB_DATA_SOFTIRQ,
-D(3)	KM_USER0,
-D(4)	KM_USER1,
-D(5)	KM_BIO_SRC_IRQ,
-D(6)	KM_BIO_DST_IRQ,
-D(7)	KM_PTE0,
-D(8)	KM_PTE1,
-D(9)	KM_IRQ0,
-D(10)	KM_IRQ1,
-D(11)	KM_SOFTIRQ0,
-D(12)	KM_SOFTIRQ1,
-D(13)	KM_TYPE_NR
-};
-
-#undef D
-
-#endif /* _ASM_IA64_KMAP_TYPES_H */
include/asm-m68k/kmap_types.h deleted +0 -21
diff -Nru a/include/asm-m68k/kmap_types.h b/include/asm-m68k/kmap_types.h
@@ -1,21 +0,0 @@
-#ifndef __ASM_M68K_KMAP_TYPES_H
-#define __ASM_M68K_KMAP_TYPES_H
-
-enum km_type {
-	KM_BOUNCE_READ,
-	KM_SKB_SUNRPC_DATA,
-	KM_SKB_DATA_SOFTIRQ,
-	KM_USER0,
-	KM_USER1,
-	KM_BIO_SRC_IRQ,
-	KM_BIO_DST_IRQ,
-	KM_PTE0,
-	KM_PTE1,
-	KM_IRQ0,
-	KM_IRQ1,
-	KM_SOFTIRQ0,
-	KM_SOFTIRQ1,
-	KM_TYPE_NR
-};
-
-#endif	/* __ASM_M68K_KMAP_TYPES_H */
include/asm-m68knommu/kmap_types.h deleted +0 -19
diff -Nru a/include/asm-m68knommu/kmap_types.h b/include/asm-m68knommu/kmap_types.h
@@ -1,19 +0,0 @@
-#ifndef _ASM_KMAP_TYPES_H
-#define _ASM_KMAP_TYPES_H
-
-enum km_type {
-	KM_BOUNCE_READ,
-	KM_SKB_SUNRPC_DATA,
-	KM_SKB_DATA_SOFTIRQ,
-	KM_USER0,
-	KM_USER1,
-	KM_BIO_SRC_IRQ,
-	KM_BIO_DST_IRQ,
-	KM_PTE0,
-	KM_PTE1,
-	KM_IRQ0,
-	KM_IRQ1,
-	KM_TYPE_NR
-};
-
-#endif
include/asm-parisc/kmap_types.h deleted +0 -31
diff -Nru a/include/asm-parisc/kmap_types.h b/include/asm-parisc/kmap_types.h
@@ -1,31 +0,0 @@
-#ifndef _ASM_KMAP_TYPES_H
-#define _ASM_KMAP_TYPES_H
-
-#include <linux/config.h>
-
-#ifdef CONFIG_DEBUG_HIGHMEM
-# define D(n) __KM_FENCE_##n ,
-#else
-# define D(n)
-#endif
-
-enum km_type {
-D(0)	KM_BOUNCE_READ,
-D(1)	KM_SKB_SUNRPC_DATA,
-D(2)	KM_SKB_DATA_SOFTIRQ,
-D(3)	KM_USER0,
-D(4)	KM_USER1,
-D(5)	KM_BIO_SRC_IRQ,
-D(6)	KM_BIO_DST_IRQ,
-D(7)	KM_PTE0,
-D(8)	KM_PTE1,
-D(9)	KM_IRQ0,
-D(10)	KM_IRQ1,
-D(11)	KM_SOFTIRQ0,
-D(12)	KM_SOFTIRQ1,
-D(13)	KM_TYPE_NR
-};
-
-#undef D
-
-#endif
include/asm-ppc64/kmap_types.h deleted +0 -23
diff -Nru a/include/asm-ppc64/kmap_types.h b/include/asm-ppc64/kmap_types.h
@@ -1,23 +0,0 @@
-#ifdef __KERNEL__
-#ifndef _ASM_KMAP_TYPES_H
-#define _ASM_KMAP_TYPES_H
-
-enum km_type {
-	KM_BOUNCE_READ,
-	KM_SKB_SUNRPC_DATA,
-	KM_SKB_DATA_SOFTIRQ,
-	KM_USER0,
-	KM_USER1,
-	KM_BIO_SRC_IRQ,
-	KM_BIO_DST_IRQ,
-	KM_PTE0,
-	KM_PTE1,
-	KM_IRQ0,
-	KM_IRQ1,
-	KM_SOFTIRQ0,
-	KM_SOFTIRQ1,	
-	KM_TYPE_NR
-};
-
-#endif
-#endif /* __KERNEL__ */
include/asm-s390/kmap_types.h deleted +0 -23
diff -Nru a/include/asm-s390/kmap_types.h b/include/asm-s390/kmap_types.h
@@ -1,23 +0,0 @@
-#ifdef __KERNEL__
-#ifndef _ASM_KMAP_TYPES_H
-#define _ASM_KMAP_TYPES_H
-
-enum km_type {
-	KM_BOUNCE_READ,
-	KM_SKB_SUNRPC_DATA,
-	KM_SKB_DATA_SOFTIRQ,
-	KM_USER0,
-	KM_USER1,
-	KM_BIO_SRC_IRQ,
-	KM_BIO_DST_IRQ,
-	KM_PTE0,
-	KM_PTE1,
-	KM_IRQ0,
-	KM_IRQ1,
-	KM_SOFTIRQ0,
-	KM_SOFTIRQ1,	
-	KM_TYPE_NR
-};
-
-#endif
-#endif /* __KERNEL__ */
include/asm-sh/kmap_types.h deleted +0 -33
diff -Nru a/include/asm-sh/kmap_types.h b/include/asm-sh/kmap_types.h
@@ -1,33 +0,0 @@
-#ifndef __SH_KMAP_TYPES_H
-#define __SH_KMAP_TYPES_H
-
-/* Dummy header just to define km_type. */
-
-#include <linux/config.h>
-
-#if CONFIG_DEBUG_HIGHMEM
-# define D(n) __KM_FENCE_##n ,
-#else
-# define D(n)
-#endif
-
-enum km_type {
-D(0)	KM_BOUNCE_READ,
-D(1)	KM_SKB_SUNRPC_DATA,
-D(2)	KM_SKB_DATA_SOFTIRQ,
-D(3)	KM_USER0,
-D(4)	KM_USER1,
-D(5)	KM_BIO_SRC_IRQ,
-D(6)	KM_BIO_DST_IRQ,
-D(7)	KM_PTE0,
-D(8)	KM_PTE1,
-D(9)	KM_IRQ0,
-D(10)	KM_IRQ1,
-D(11)	KM_SOFTIRQ0,
-D(12)	KM_SOFTIRQ1,
-D(13)	KM_TYPE_NR
-};
-
-#undef D
-
-#endif
include/asm-sparc64/kmap_types.h deleted +0 -25
diff -Nru a/include/asm-sparc64/kmap_types.h b/include/asm-sparc64/kmap_types.h
@@ -1,25 +0,0 @@
-#ifndef _ASM_KMAP_TYPES_H
-#define _ASM_KMAP_TYPES_H
-
-/* Dummy header just to define km_type.  None of this
- * is actually used on sparc64.  -DaveM
- */
-
-enum km_type {
-	KM_BOUNCE_READ,
-	KM_SKB_SUNRPC_DATA,
-	KM_SKB_DATA_SOFTIRQ,
-	KM_USER0,
-	KM_USER1,
-	KM_BIO_SRC_IRQ,
-	KM_BIO_DST_IRQ,
-	KM_PTE0,
-	KM_PTE1,
-	KM_IRQ0,
-	KM_IRQ1,
-	KM_SOFTIRQ0,
-	KM_SOFTIRQ1,
-	KM_TYPE_NR
-};
-
-#endif
include/asm-v850/kmap_types.h deleted +0 -19
diff -Nru a/include/asm-v850/kmap_types.h b/include/asm-v850/kmap_types.h
@@ -1,19 +0,0 @@
-#ifndef __V850_KMAP_TYPES_H__
-#define __V850_KMAP_TYPES_H__
-
-enum km_type {
-	KM_BOUNCE_READ,
-	KM_SKB_SUNRPC_DATA,
-	KM_SKB_DATA_SOFTIRQ,
-	KM_USER0,
-	KM_USER1,
-	KM_BIO_SRC_IRQ,
-	KM_BIO_DST_IRQ,
-	KM_PTE0,
-	KM_PTE1,
-	KM_IRQ0,
-	KM_IRQ1,
-	KM_TYPE_NR
-};
-
-#endif /* __V850_KMAP_TYPES_H__ */
include/asm-x86_64/kmap_types.h deleted +0 -19
diff -Nru a/include/asm-x86_64/kmap_types.h b/include/asm-x86_64/kmap_types.h
@@ -1,19 +0,0 @@
-#ifndef _ASM_KMAP_TYPES_H
-#define _ASM_KMAP_TYPES_H
-
-enum km_type {
-	KM_BOUNCE_READ,
-	KM_SKB_SUNRPC_DATA,
-	KM_SKB_DATA_SOFTIRQ,
-	KM_USER0,
-	KM_USER1,
-	KM_BIO_SRC_IRQ,
-	KM_BIO_DST_IRQ,
-	KM_IRQ0,
-	KM_IRQ1,
-	KM_SOFTIRQ0,
-	KM_SOFTIRQ1,
-	KM_TYPE_NR
-};
-
-#endif
include/linux/highmem.h +11 -0
diff -Nru a/include/linux/highmem.h b/include/linux/highmem.h
@@ -12,6 +12,7 @@
 extern struct page *highmem_start_page;
 
 #include <asm/highmem.h>
+#include <asm/kmap_types.h>
 
 /* declarations for linux/mm/highmem.c */
 unsigned int nr_free_highpages(void);
@@ -25,6 +26,16 @@
 	might_sleep();
 	return page_address(page);
 }
+
+/* generic code may use these, even when CONFIG_HIGHMEM is not set */
+enum km_type {
+	KM_USER0,
+	KM_USER1,
+	KM_IRQ0,
+	KM_IRQ1,
+	KM_SOFTIRQ0,
+	KM_SOFTIRQ1,
+};
 
 #define kunmap(page) do { (void) (page); } while (0)
 
#8 lore raw
In reply to Grant Grundler
Quoted 15 lines · Bjorn Helgaas

Here's my proposal for a fix. I don't think I'll send it to Linus/Andrew yet, because it's really a cleanup, not an actual bug fix, but here it is for any comments.

Currently every architecture must supply kmap_types.h, even though it only makes sense for architectures that use highmem.

This patch removes the need for generic code to know about kmap_types.h by adding the appropriate #include to include/linux/highmem.h and defining a minimal set of KM_ enums that are used by generic code.

One might argue that we shouldn't need to define even the minimal set of KM_ enums, but generic code needs to be able to do, for example:

Quoted 7 lines
	kmap_atomic(page, KM_USER);

In many cases the KM_USER needn't be defined at all because the non-highmem kmap_atomic() never evaluates it, but leaving it undefined also restricts generic code from doing other things like defining an array of KM_ types. crypto/cipher.c defines such an array, and it seems non-intuitive to prevent that usage.

I disagree with this fix. All uses of KM_ variables are restricted to headers that make their use conditional on CONFIG_HIGHMEM. crypto is the only expection and the correct fix is to change crypto to test CONFIG_HIGHMEM, not to change every architecture. Untested.

Index: 23-pre7.1/crypto/internal.h

crypto/internal.h Fri, 24 Oct 2003 07:00:24 +1000 kaos (linux-2.4/Y/g/31_internal.h 1.1 644) renamed from crypto/internal.h Sun, 06 Jul 2003 14:16:24 +1000 kaos (linux-2.4/Y/g/31_internal.h 1.1 644) +4 -1
@@ -16,14 +16,17 @@
 #include <linux/init.h>
 #include <asm/hardirq.h>
 #include <asm/softirq.h>
-#include <asm/kmap_types.h>
 
+#ifdef CONFIG_HIGHMEM
 extern enum km_type crypto_km_types[];
 
 static inline enum km_type crypto_kmap_type(int out)
 {
 	return crypto_km_types[(in_softirq() ? 2 : 0) + out];
 }
+#else
+#define crypto_kmap_type(out) crypto_kmap_type_not_used_on_this_arch
+#endif
 
 static inline void *crypto_kmap(struct page *page, int out)
 {

Index: 23-pre7.1/crypto/cipher.c

crypto/cipher.c Fri, 24 Oct 2003 06:58:11 +1000 kaos (linux-2.4/Y/g/38_cipher.c 1.1 644) renamed from crypto/cipher.c Sun, 06 Jul 2003 14:16:24 +1000 kaos (linux-2.4/Y/g/38_cipher.c 1.1 644) +2 -0
@@ -35,12 +35,14 @@ struct scatter_walk {
 	unsigned int		offset;
 };
 
+#ifdef CONFIG_HIGHMEM
 enum km_type crypto_km_types[] = {
 	KM_USER0,
 	KM_USER1,
 	KM_SOFTIRQ0,
 	KM_SOFTIRQ1,
 };
+#endif
 
 static inline void xor_64(u8 *a, const u8 *b)
 {
#9 lore raw
In reply to Grant Grundler
Quoted 2 lines · Keith Owens

I disagree with this fix. All uses of KM_ variables are restricted to headers that make their use conditional on CONFIG_HIGHMEM.

It's not true that all uses of KM_ variables are restricted to headers. As one counter-example, file_read_actor() in mm/filemap.c does this:

	kaddr = kmap_atomic(page, KM_USER0);

I guess your argument is that KM_ variables should *only* be used in places where they'll never be evaluated, i.e., as arguments to kmap_atomic() and kunmap_atomic().

I think that if you can pass KM_USER0 to a function, you ought to be able to copy KM_USER0 somewhere and pass the copy to the function. Otherwise we'll just have to keep explaining this funny wart on the kmap interface.

Quoted 3 lines

crypto is
the only expection and the correct fix is to change crypto to test CONFIG_HIGHMEM, not to change every architecture.

I only changed so many architectures because the whole point was to avoid requiring a dummy kmap_types.h file if you don't use highmem. So I just removed the now-unnecessary kmap_types.h files, and I assume you would do the same. (Although you didn't mention changes to include/linux/highmem.h and fs/aio.c to remove generic knowledge of <asm/kmap_types.h>, so I could be misunderstanding your proposal.)

Bjorn

#10 lore raw
In reply to Grant Grundler
Quoted 4 lines · Bjorn Helgaas
Quoted 2 lines · Keith Owens

I disagree with this fix. All uses of KM_ variables are restricted to headers that make their use conditional on CONFIG_HIGHMEM.

Quoted 8 lines

It's not true that all uses of KM_ variables are restricted to headers. As one counter-example, file_read_actor() in mm/filemap.c does this:

	kaddr = kmap_atomic(page, KM_USER0);

I guess your argument is that KM_ variables should *only* be used in places where they'll never be evaluated, i.e., as arguments to kmap_atomic() and kunmap_atomic().

Right.

Quoted 4 lines

I think that if you can pass KM_USER0 to a function, you ought to be able to copy KM_USER0 somewhere and pass the copy to the function. Otherwise we'll just have to keep explaining this funny wart on the kmap interface.

Disagree. kmap is currently nicely encapsulated. Letting code play with km types just makes that code more fragile. crypto wants to use different km types based on context (softirq or not). If crypto needs that functionallity then I expect other code to need it as well, which means the facility should be part of kmap, not implemented by each code area.

Failing that, crypto should only use km types on builds that have CONFIG_HIGHMEM defined.

Quoted 6 lines

I only changed so many architectures because the whole point was to avoid requiring a dummy kmap_types.h file if you don't use highmem. So I just removed the now-unnecessary kmap_types.h files, and I assume you would do the same. (Although you didn't mention changes to include/linux/highmem.h and fs/aio.c to remove generic knowledge of <asm/kmap_types.h>, so I could be misunderstanding your proposal.)

Fix crypto to test CONFIG_HIGHMEM and no other changes are required.

#11 lore raw
In reply to Grant Grundler
Quoted 6 lines · Keith Owens

Letting code play
with km types just makes that code more fragile. crypto wants to use different km types based on context (softirq or not). If crypto needs that functionallity then I expect other code to need it as well, which means the facility should be part of kmap, not implemented by each code area.

Can you give an example of what such a new kmap interface would look like? The crypto code looks reasonable to me, and I don't know enough about kmap to guess what abstraction you might have in mind.

I agree it's not optimal to define the KM_ enums in highmem.h for the non-highmem case, but I think it's better than having to use #ifdef CONFIG_HIGHMEM in the crypto code.

You seem to be concerned about the fact that my patch touches so many architectures, but you said earlier:

Quoted 3 lines

That is an error in crypto/internal.h. Nothing should include asm/kmap_types.h directly, they should include linux/highmem.h and let that decide if the arch needs kmap or not.

So I assume that if you did a complete patch it would
1) Remove "#include <asm/kmap_types.h" from crypto/internal.h
2) Remove "#include <asm/kmap_types.h" from fs/aio.c
3) Add "#include <asm/kmap_types.h" to include/linux/highmem.h (under #ifdef CONFIG_HIGHMEM).
4) Remove kmap_types.h from non-highmem architectures, since they're no longer referenced.

Which means that your approach would touch just as many arches.

Bjorn