0/18 2.4.19 architecture updates for s390 following

PATCH New Arnd Bergmann 5 participants 20 messages Aug 5, 2002 – Aug 10, 2002

Subsystems (11)

Memory management VFS SCSI Block layer +7 more subsystems

lore

Share & tools

lore

Arnd Bergmann

author

2002-08-05 16:12 UTC #0 lore raw

Last friday, we had a new code drop on developerworks with s390 patches against 2.4.17. The patches following this mail contain that code, updated to 2.4.19. They were prepared by Martin Schwidefsky, but he's on holidays now, so I'm sending them instead.

The first eleven patches contains large updates for s390 architecture dependent files that are maintained here, so I'm not sending them to lkml. The rest are smaller bits, some of which are more controversial.

	Arnd <><

Arnd Bergmann

author

2002-08-05 16:09 UTC #1 lore raw

PATCH 1/1 12/18 console_unblank bug fix

Problem: console_unblank can cause an endless trap loop if the console
semaphore is already taken and console_unblank is called in interrupt context.
Solution: To avoid the trap loop just try to take the console semaphore
with down_trylock and exit if it fails. It is not perfect since c->unblank() is not called in this case but way better than an endless trap loop. s390/s390x are the only two architectures that still use console_unblank. This patch has been accepted to 2.5.

kernel/printk.c +8 -1

diff -urN linux-2.4.19-rc3/kernel/printk.c linux-2.4.19-s390/kernel/printk.c
@@ -556,7 +556,14 @@
 {
 	struct console *c;
 
-	acquire_console_sem();
+	/*
+	 * Try to get the console semaphore. If someone else owns it
+	 * we have to return without unblanking because console_unblank
+	 * may be called in interrupt context.
+	 */
+	if (down_trylock(&console_sem) != 0)
+		return;
+	console_may_schedule = 0;
 	for (c = console_drivers; c != NULL; c = c->next)
 		if ((c->flags & CON_ENABLED) && c->unblank)
 			c->unblank();

Arnd Bergmann

author

2002-08-05 16:09 UTC #2 lore raw

PATCH 1/1 13/18 Export elevator_init

In reply to Arnd Bergmann

Problem: The dasd driver wants to switch to the noop elevator function but
it can do that if the driver is compiled as a module because elevator_init is not exported.
Solution: Export elevator_init. Has been accepted to 2.5.

drivers/block/Makefile +1 -1

diff -urN linux-2.4.19-rc3/drivers/block/Makefile linux-2.4.19-s390/drivers/block/Makefile
@@ -10,7 +10,7 @@
 
 O_TARGET := block.o
 
-export-objs	:= ll_rw_blk.o blkpg.o loop.o DAC960.o genhd.o
+export-objs	:= ll_rw_blk.o blkpg.o elevator.o loop.o DAC960.o genhd.o
 
 obj-y	:= ll_rw_blk.o blkpg.o genhd.o elevator.o

drivers/block/elevator.c +2 -0

diff -urN linux-2.4.19-rc3/drivers/block/elevator.c linux-2.4.19-s390/drivers/block/elevator.c
@@ -220,3 +220,5 @@
 	*elevator = type;
 	elevator->queue_ID = queue_ID++;
 }
+
+EXPORT_SYMBOL(elevator_init);

Arnd Bergmann

author

2002-08-05 16:10 UTC #3 lore raw

PATCH 1/1 16/18 add more possible root devices

In reply to Arnd Bergmann

This patch adds some more s/390 specific devices to the list of root devices.

init/do_mounts.c +52 -0

diff -urN linux-2.4.19-rc3/init/do_mounts.c linux-2.4.19-s390/init/do_mounts.c
@@ -162,6 +162,58 @@
 	{ "dasdf", (DASD_MAJOR << MINORBITS) + (5 << 2) },
 	{ "dasdg", (DASD_MAJOR << MINORBITS) + (6 << 2) },
 	{ "dasdh", (DASD_MAJOR << MINORBITS) + (7 << 2) },
+        { "dasdi", (DASD_MAJOR << MINORBITS) + (8 << 2) },
+        { "dasdj", (DASD_MAJOR << MINORBITS) + (9 << 2) },
+        { "dasdk", (DASD_MAJOR << MINORBITS) + (10 << 2) },
+        { "dasdl", (DASD_MAJOR << MINORBITS) + (11 << 2) },
+        { "dasdm", (DASD_MAJOR << MINORBITS) + (12 << 2) },
+        { "dasdn", (DASD_MAJOR << MINORBITS) + (13 << 2) },
+        { "dasdo", (DASD_MAJOR << MINORBITS) + (14 << 2) },
+        { "dasdp", (DASD_MAJOR << MINORBITS) + (15 << 2) },
+        { "dasdq", (DASD_MAJOR << MINORBITS) + (16 << 2) },
+        { "dasdr", (DASD_MAJOR << MINORBITS) + (17 << 2) },
+        { "dasds", (DASD_MAJOR << MINORBITS) + (18 << 2) },
+        { "dasdt", (DASD_MAJOR << MINORBITS) + (19 << 2) },
+        { "dasdu", (DASD_MAJOR << MINORBITS) + (20 << 2) },
+        { "dasdv", (DASD_MAJOR << MINORBITS) + (21 << 2) },
+        { "dasdw", (DASD_MAJOR << MINORBITS) + (22 << 2) },
+        { "dasdx", (DASD_MAJOR << MINORBITS) + (23 << 2) },
+        { "dasdy", (DASD_MAJOR << MINORBITS) + (24 << 2) },
+        { "dasdz", (DASD_MAJOR << MINORBITS) + (25 << 2) },
+#endif
+#ifdef CONFIG_BLK_DEV_XPRAM
+	{ "xpram0", (XPRAM_MAJOR << MINORBITS) },
+	{ "xpram1", (XPRAM_MAJOR << MINORBITS) + 1 },
+	{ "xpram2", (XPRAM_MAJOR << MINORBITS) + 2 },
+	{ "xpram3", (XPRAM_MAJOR << MINORBITS) + 3 },
+	{ "xpram4", (XPRAM_MAJOR << MINORBITS) + 4 },
+	{ "xpram5", (XPRAM_MAJOR << MINORBITS) + 5 },
+	{ "xpram6", (XPRAM_MAJOR << MINORBITS) + 6 },
+	{ "xpram7", (XPRAM_MAJOR << MINORBITS) + 7 },
+	{ "xpram8", (XPRAM_MAJOR << MINORBITS) + 8 },
+	{ "xpram9", (XPRAM_MAJOR << MINORBITS) + 9 },
+	{ "xpram10", (XPRAM_MAJOR << MINORBITS) + 10 },
+	{ "xpram11", (XPRAM_MAJOR << MINORBITS) + 11 },
+	{ "xpram12", (XPRAM_MAJOR << MINORBITS) + 12 },
+	{ "xpram13", (XPRAM_MAJOR << MINORBITS) + 13 },
+	{ "xpram14", (XPRAM_MAJOR << MINORBITS) + 14 },
+	{ "xpram15", (XPRAM_MAJOR << MINORBITS) + 15 },
+	{ "xpram16", (XPRAM_MAJOR << MINORBITS) + 16 },
+	{ "xpram17", (XPRAM_MAJOR << MINORBITS) + 17 },
+	{ "xpram18", (XPRAM_MAJOR << MINORBITS) + 18 },
+	{ "xpram19", (XPRAM_MAJOR << MINORBITS) + 19 },
+	{ "xpram20", (XPRAM_MAJOR << MINORBITS) + 20 },
+	{ "xpram21", (XPRAM_MAJOR << MINORBITS) + 21 },
+	{ "xpram22", (XPRAM_MAJOR << MINORBITS) + 22 },
+	{ "xpram23", (XPRAM_MAJOR << MINORBITS) + 23 },
+	{ "xpram24", (XPRAM_MAJOR << MINORBITS) + 24 },
+	{ "xpram25", (XPRAM_MAJOR << MINORBITS) + 25 },
+	{ "xpram26", (XPRAM_MAJOR << MINORBITS) + 26 },
+	{ "xpram27", (XPRAM_MAJOR << MINORBITS) + 27 },
+	{ "xpram28", (XPRAM_MAJOR << MINORBITS) + 28 },
+	{ "xpram29", (XPRAM_MAJOR << MINORBITS) + 29 },
+	{ "xpram30", (XPRAM_MAJOR << MINORBITS) + 30 },
+	{ "xpram31", (XPRAM_MAJOR << MINORBITS) + 31 },
 #endif
 #if defined(CONFIG_BLK_CPQ_DA) || defined(CONFIG_BLK_CPQ_DA_MODULE)
 	{ "ida/c0d0p",0x4800 },

Christoph Hellwig

2002-08-05 17:11 UTC #4 lore raw

Re: [PATCH] 16/18 add more possible root devices

Quoted 1 line · Arnd Bergmann

This patch adds some more s/390 specific devices to the list of root devices.

Indentation is b0rked.

Arnd Bergmann

2002-08-06 09:21 UTC #5 lore raw

Re: [PATCH] 16/18 add more possible root devices

Quoted 5 lines · Christoph Hellwig

Quoted 2 lines · Arnd Bergmann
This patch adds some more s/390 specific devices to the list of root devices.
Indentation is b0rked.

Thanks, this fixes indentation and removes the extra xpram entries noticed by Bernd.

init/do_mounts.c +21 -0

diff -ur linux-2.4.19-s390/init/do_mounts.c linux-2.4.19-s390-bootdev/init/do_mounts.c
@@ -162,6 +162,27 @@
 	{ "dasdf", (DASD_MAJOR << MINORBITS) + (5 << 2) },
 	{ "dasdg", (DASD_MAJOR << MINORBITS) + (6 << 2) },
 	{ "dasdh", (DASD_MAJOR << MINORBITS) + (7 << 2) },
+	{ "dasdi", (DASD_MAJOR << MINORBITS) + (8 << 2) },
+	{ "dasdj", (DASD_MAJOR << MINORBITS) + (9 << 2) },
+	{ "dasdk", (DASD_MAJOR << MINORBITS) + (10 << 2) },
+	{ "dasdl", (DASD_MAJOR << MINORBITS) + (11 << 2) },
+	{ "dasdm", (DASD_MAJOR << MINORBITS) + (12 << 2) },
+	{ "dasdn", (DASD_MAJOR << MINORBITS) + (13 << 2) },
+	{ "dasdo", (DASD_MAJOR << MINORBITS) + (14 << 2) },
+	{ "dasdp", (DASD_MAJOR << MINORBITS) + (15 << 2) },
+	{ "dasdq", (DASD_MAJOR << MINORBITS) + (16 << 2) },
+	{ "dasdr", (DASD_MAJOR << MINORBITS) + (17 << 2) },
+	{ "dasds", (DASD_MAJOR << MINORBITS) + (18 << 2) },
+	{ "dasdt", (DASD_MAJOR << MINORBITS) + (19 << 2) },
+	{ "dasdu", (DASD_MAJOR << MINORBITS) + (20 << 2) },
+	{ "dasdv", (DASD_MAJOR << MINORBITS) + (21 << 2) },
+	{ "dasdw", (DASD_MAJOR << MINORBITS) + (22 << 2) },
+	{ "dasdx", (DASD_MAJOR << MINORBITS) + (23 << 2) },
+	{ "dasdy", (DASD_MAJOR << MINORBITS) + (24 << 2) },
+	{ "dasdz", (DASD_MAJOR << MINORBITS) + (25 << 2) },
+#endif
+#ifdef CONFIG_BLK_DEV_XPRAM
+	{ "xpram", (XPRAM_MAJOR << MINORBITS) },
 #endif
 #if defined(CONFIG_BLK_CPQ_DA) || defined(CONFIG_BLK_CPQ_DA_MODULE)
 	{ "ida/c0d0p",0x4800 },

Arnd Bergmann

author

2002-08-05 16:11 UTC #6 lore raw

PATCH 1/1 15/18 better pte invalidation

In reply to Arnd Bergmann

This is a patch that changes the memory management to allow s390/s390x to make best use of its memory subsystem. There are three problems:

1) the dirty and the referenced bit is kept in the storage key (there is one for every physical page) and not in the page table entries. Because of this difference the following piece of code is suboptimal:

        if (pte_present(pte)) {
                struct page *page = pte_page(pte);
                if (VALID_PAGE(page) && !PageReserved(page) &&
                    ptep_test_and_clear_dirty(ptep)) {
                        flush_tlb_page(vma, address);
                        set_page_dirty(page);
                }
        }

The dirty bit is tested and cleared with ptep_test_and_clear_dirty and if it is set the tlb for the page is flushed. But the tlb flush is completly superflous on s/390 because ptep_test_and_clear_dirty has already done everything that is needed (the magic behind that is the sske instruction - see include/asm-s390/pgtable.h). To allow s/390 to optimize the ptep_test_and_clear_and_flush_dirty function is introduced that does what its name implies.

2) s/390 has an instruction called ipte (invalidate page table entry) that is used to get rid of virtual pages. It does two things, first it set the invalid bit in the pte and second it flushes the tlbs for this page on all cpus. Very handy but it requires that the pte it should flush is still valid. The introduction of establish_pte was a step into the right direction but it is defined in mm/memory.c. We need to be able to replace this function with a special s/390 variant. I tried to move establish_pte to include/asm/pgtable.h but this failed because of include order conflicts. Therefore I moved it to include/asm/pgalloc.h (and renamed it to ptep_establish). To make ptep_establish available on architectures that do not want to replace it (all others), I added asm-generic/pgalloc.h and an include in asm-<arch>/pgalloc.h for all architectures != s390. The same problem as with establish_pte exists with invalidating ptes. The sequence used in e.g. mm/vmscan.c is

        flush_cache_page(vma, address);
        pte = ptep_get_and_clear(page_table);
        flush_tlb_page(vma, addess);

Again this makes it impossible to use the ipte in flush_tlb_page. The sequence of these tree lines is replaced by ptep_invalidate. The "old" implementation is moved to asm-generic/pgalloc.h and s/390 has its own implementation with ipte.

3) The 4 pages middle directory problem. This is related to the order2 patch for 64 bit s/390. The page size for 64 bit is 4096K but a full page middle directories requires 16K = 4 pages. This is bad because get_free_pages won't guarantee us that we can get 4 consecutive pages. The idea behind the order2 patch is to use a hardware feature of 64 bit s/390 that allows to use incomplete page middle directories. To pages is enough in 99.9% of all cases. To make use of this feature we have to be able to "replace" the pmd __pmd_alloc allocated by a large unit that contains a full page middle directory:

        #if defined(CONFIG_ARCH_S390X)
                new = pgd_populate(mm, pgd, new);
                if (!new)
                        return NULL;
        #else
                pgd_populate(mm, pgd, new);
        #endif

See arch/s390x/mm/init.c and include/asm-s390x/pgalloc.h for the details.

And finally the last change is a pure pure compiler optimization: move pte_alloc and make in extern inline. The compiler will generate a copy that can be called from other modules AND inline it in mm/memory.c. This saves a function call on the minor fault path and function calling is expensive on s/390.

include/asm-alpha/pgalloc.h +2 -0

diff -urN linux-2.4.19-rc3/include/asm-alpha/pgalloc.h linux-2.4.19-s390/include/asm-alpha/pgalloc.h
@@ -347,4 +347,6 @@
 
 extern int do_check_pgt_cache(int, int);
 
+#include <asm-generic/pgalloc.h>
+
 #endif /* _ALPHA_PGALLOC_H */

include/asm-arm/pgalloc.h +2 -0

diff -urN linux-2.4.19-rc3/include/asm-arm/pgalloc.h linux-2.4.19-s390/include/asm-arm/pgalloc.h
@@ -138,4 +138,6 @@
 
 extern int do_check_pgt_cache(int, int);
 
+#include <asm-generic/pgalloc.h>
+
 #endif

include/asm-generic/pgalloc.h +37 -0

diff -urN linux-2.4.19-rc3/include/asm-generic/pgalloc.h linux-2.4.19-s390/include/asm-generic/pgalloc.h
@@ -0,0 +1,37 @@
+#ifndef _ASM_GENERIC_PGALLOC_H
+#define _ASM_GENERIC_PGALLOC_H
+
+static inline int ptep_test_and_clear_and_flush_young(struct vm_area_struct *vma, unsigned long address, pte_t *ptep)
+{
+	if (!ptep_test_and_clear_young(ptep))
+		return 0;
+	flush_tlb_page(vma, address);
+	return 1;
+}
+
+static inline int ptep_test_and_clear_and_flush_dirty(struct vm_area_struct *vma, unsigned long address, pte_t *ptep)
+{
+	if (!ptep_test_and_clear_dirty(ptep))
+		return 0;
+	flush_tlb_page(vma, address);
+	return 1;
+}
+
+static inline void ptep_establish(struct vm_area_struct *vma, unsigned long address, pte_t *ptep, pte_t entry)
+{
+	set_pte(ptep, entry);
+	flush_tlb_page(vma, address);
+	update_mmu_cache(vma, address, entry);
+}
+
+static inline pte_t ptep_invalidate(struct vm_area_struct *vma, unsigned long address, pte_t *ptep)
+{
+	pte_t pte;
+
+	flush_cache_page(vma, address);
+        pte = ptep_get_and_clear(ptep);
+	flush_tlb_page(vma, address);
+	return pte;
+}
+
+#endif /* _ASM_GENERIC_PGALLOC_H */

include/asm-i386/pgalloc.h +2 -0

diff -urN linux-2.4.19-rc3/include/asm-i386/pgalloc.h linux-2.4.19-s390/include/asm-i386/pgalloc.h
@@ -235,4 +235,6 @@
 	flush_tlb_mm(mm);
 }
 
+#include <asm-generic/pgalloc.h>
+
 #endif /* _I386_PGALLOC_H */

include/asm-ia64/pgalloc.h +2 -0

diff -urN linux-2.4.19-rc3/include/asm-ia64/pgalloc.h linux-2.4.19-s390/include/asm-ia64/pgalloc.h
@@ -268,4 +268,6 @@
 	set_bit(PG_arch_1, &page->flags);	/* mark page as clean */
 }
 
+#include <asm-generic/pgalloc.h>
+
 #endif /* _ASM_IA64_PGALLOC_H */

include/asm-m68k/pgalloc.h +2 -0

diff -urN linux-2.4.19-rc3/include/asm-m68k/pgalloc.h linux-2.4.19-s390/include/asm-m68k/pgalloc.h
@@ -164,4 +164,6 @@
 #include <asm/motorola_pgalloc.h>
 #endif
 
+#include <asm-generic/pgalloc.h>
+
 #endif /* M68K_PGALLOC_H */

include/asm-mips/pgalloc.h +2 -0

diff -urN linux-2.4.19-rc3/include/asm-mips/pgalloc.h linux-2.4.19-s390/include/asm-mips/pgalloc.h
@@ -194,4 +194,6 @@
 
 extern int do_check_pgt_cache(int, int);
 
+#include <asm-generic/pgalloc.h>
+
 #endif /* _ASM_PGALLOC_H */

include/asm-mips64/pgalloc.h +2 -0

diff -urN linux-2.4.19-rc3/include/asm-mips64/pgalloc.h linux-2.4.19-s390/include/asm-mips64/pgalloc.h
@@ -198,4 +198,6 @@
 
 extern int do_check_pgt_cache(int, int);
 
+#include <asm-generic/pgalloc.h>
+
 #endif /* _ASM_PGALLOC_H */

include/asm-parisc/pgalloc.h +2 -0

diff -urN linux-2.4.19-rc3/include/asm-parisc/pgalloc.h linux-2.4.19-s390/include/asm-parisc/pgalloc.h
@@ -404,4 +404,6 @@
 
 extern int do_check_pgt_cache(int, int);
 
+#include <asm-generic/pgalloc.h>
+
 #endif

include/asm-ppc/pgalloc.h +2 -0

diff -urN linux-2.4.19-rc3/include/asm-ppc/pgalloc.h linux-2.4.19-s390/include/asm-ppc/pgalloc.h
@@ -146,5 +146,7 @@
 
 extern int do_check_pgt_cache(int, int);
 
+#include <asm-generic/pgalloc.h>
+
 #endif /* _PPC_PGALLOC_H */
 #endif /* __KERNEL__ */

include/asm-sh/pgalloc.h +3 -0

diff -urN linux-2.4.19-rc3/include/asm-sh/pgalloc.h linux-2.4.19-s390/include/asm-sh/pgalloc.h
@@ -153,4 +153,7 @@
 	pte_t old_pte = *ptep;
 	set_pte(ptep, pte_mkdirty(old_pte));
 }
+
+#include <asm-generic/pgalloc.h>
+
 #endif /* __ASM_SH_PGALLOC_H */

include/asm-sparc/pgalloc.h +2 -0

diff -urN linux-2.4.19-rc3/include/asm-sparc/pgalloc.h linux-2.4.19-s390/include/asm-sparc/pgalloc.h
@@ -141,4 +141,6 @@
 
 #define pte_free(pte)		free_pte_fast(pte)
 
+#include <asm-generic/pgalloc.h>
+
 #endif /* _SPARC_PGALLOC_H */

include/asm-sparc64/pgalloc.h +2 -0

diff -urN linux-2.4.19-rc3/include/asm-sparc64/pgalloc.h linux-2.4.19-s390/include/asm-sparc64/pgalloc.h
@@ -304,4 +304,6 @@
 
 extern int do_check_pgt_cache(int, int);
 
+#include <asm-generic/pgalloc.h>
+
 #endif /* _SPARC64_PGALLOC_H */

mm/filemap.c +3 -3

diff -urN linux-2.4.19-rc3/mm/filemap.c linux-2.4.19-s390/mm/filemap.c
@@ -1296,7 +1296,7 @@
  * bit. Otherwise, just mark it for future
  * action..
  */
-void mark_page_accessed(struct page *page)
+inline void mark_page_accessed(struct page *page)
 {
 	if (!PageActive(page) && PageReferenced(page)) {
 		activate_page(page);

@@ -2062,8 +2062,8 @@
 
 	if (pte_present(pte)) {
 		struct page *page = pte_page(pte);
-		if (VALID_PAGE(page) && !PageReserved(page) && ptep_test_and_clear_dirty(ptep)) {
-			flush_tlb_page(vma, address);
+		if (VALID_PAGE(page) && !PageReserved(page) &&
+                    ptep_test_and_clear_and_flush_dirty(vma, address, ptep)) {
 			set_page_dirty(page);
 		}
 	}

mm/memory.c +83 ~~-88~~

207 lines not on this page · raw message

mm/vmscan.c +1 -3

diff -urN linux-2.4.19-rc3/mm/vmscan.c linux-2.4.19-s390/mm/vmscan.c
@@ -68,9 +68,7 @@
 	 * is needed on CPUs which update the accessed and dirty
 	 * bits in hardware.
 	 */
-	flush_cache_page(vma, address);
-	pte = ptep_get_and_clear(page_table);
-	flush_tlb_page(vma, address);
+	pte = ptep_invalidate(vma, address, page_table);
 
 	if (pte_dirty(pte))
 		set_page_dirty(page);

Christoph Hellwig

2002-08-05 16:57 UTC #7 lore raw

Re: [PATCH] 15/18 better pte invalidation

Quoted 2 lines · Arnd Bergmann

This is a patch that changes the memory management to allow s390/s390x to make best use of its memory subsystem. There are three problems:

Implementation comment: make pgd_populate return something on all architectures, not just on s390.

Otherwise please try to get all this in 2.5 first, it's a major VM change.

Arnd Bergmann

author

2002-08-06 09:03 UTC #8 lore raw

Re: [PATCH] 15/18 better pte invalidation

Quoted 2 lines · Christoph Hellwig

Implementation comment: make pgd_populate return something on all architectures, not just on s390.

Makes sense to me. Martin?

Quoted 1 line

Otherwise please try to get all this in 2.5 first, it's a major VM change.

Right. I guess this one falls more in the category "If anyone wants to build an s390 2.4.19 kernel, use this patch, because it's what we have tested at IBM".

Since the 18 patches reflect what our recommended 2.4.17 patch contains (roughly sorted by relevance), not all of it is necessarily good for 2.4.20. You can see it as "this is all we have -- take what you like". Of course, anything that gets merged makes life easier for us.

	Arnd <><

Christoph Hellwig

2002-08-06 09:40 UTC #9 lore raw

Re: [PATCH] 15/18 better pte invalidation

Quoted 4 lines · Arnd Bergmann

Quoted 1 line
Otherwise please try to get all this in 2.5 first, it's a major VM change.
Right. I guess this one falls more in the category "If anyone wants to build an s390 2.4.19 kernel, use this patch, because it's what we have tested at IBM".

Umm. 2.4.19 _does_ build on S/390. By getting in a truckload of patches that update everything to IBM's latest & greatest that just require a few unacceptable / not yet acceptable intrusive core changes you break that.

You probably don't care a lot, but people trying to test S/390 compatiblity on Hercules (or a Multiprise in the basement..) really want to compile kernels out-of-the-box.

Paul Mackerras

2002-08-10 12:27 UTC #10 lore raw

Re: [PATCH] 15/18 better pte invalidation

In reply to Arnd Bergmann

Quoted 7 lines · Arnd Bergmann

2) s/390 has an instruction called ipte (invalidate page table entry) that is used to get rid of virtual pages. It does two things, first it set the invalid bit in the pte and second it flushes the tlbs for this page on all cpus. Very handy but it requires that the pte it should flush is still valid. The introduction of establish_pte was a step into the right direction but it is defined in mm/memory.c. We need to be able to replace this function with a special s/390 variant.

It's now possible in 2.5 to go from a PTE pointer to the mm and virtual address that it maps - see ptep_to_mm and ptep_to_address in include/asm-generic/rmap.h. I'm planning to use this to do MMU hashtable management on PPC in set_pte et al. and thus avoid the second walk of the Linux page tables that we currently do in flush_tlb_* on PPC. You could use those functions to let you use the ipte instruction in set_pte et al. and then make the flush_tlb_* functions be no-ops.

The ptep_to_mm and ptep_to_address functions came in with rmap, but all they rely on is having the page->mapping and page->index fields filled in for pagetable pages. It should only be a minor addition to the 2.4 MM system to set those fields, and it would be useful on ppc, ppc64, s390 and s390x.

Paul.

Arnd Bergmann

author

2002-08-05 16:12 UTC #11 lore raw

PATCH 1/1 14/18 support for partition labels in devfs

In reply to Arnd Bergmann

This patch adds a new feature to the partition code. It will automatically generate links in the device filesystem from the label name of a dasd to the device node. This is only needed if the device filesystem is used.

fs/partitions/check.c +102 -3

diff -urN linux-2.4.19-rc3/fs/partitions/check.c linux-2.4.19-s390/fs/partitions/check.c
@@ -85,7 +85,7 @@
 #ifdef CONFIG_ARCH_S390
 int (*genhd_dasd_name)(char*,int,int,struct gendisk*) = NULL;
 int (*genhd_dasd_ioctl)(struct inode *inp, struct file *filp,
-			    unsigned int no, unsigned long data);
+			unsigned int no, unsigned long data);
 EXPORT_SYMBOL(genhd_dasd_name);
 EXPORT_SYMBOL(genhd_dasd_ioctl);
 #endif

@@ -277,6 +277,7 @@
 	int devnum = minor >> dev->minor_shift;
 	devfs_handle_t dir;
 	unsigned int devfs_flags = DEVFS_FL_DEFAULT;
+	umode_t devfs_perm  = S_IFBLK | S_IRUSR | S_IWUSR;
 	char devname[16];
 
 	if (dev->part[minor + part].de) return;

@@ -284,11 +285,14 @@
 	if (!dir) return;
 	if ( dev->flags && (dev->flags[devnum] & GENHD_FL_REMOVABLE) )
 		devfs_flags |= DEVFS_FL_REMOVABLE;
+	if (is_read_only(MKDEV(dev->major, minor+part))) {
+	        devfs_perm &= ~(S_IWUSR);
+	}
 	sprintf (devname, "part%d", part);
 	dev->part[minor + part].de =
 	    devfs_register (dir, devname, devfs_flags,
 			    dev->major, minor + part,
-			    S_IFBLK | S_IRUSR | S_IWUSR,
+			    devfs_perm,
 			    dev->fops, NULL);
 }

@@ -300,12 +304,16 @@
 	int devnum = minor >> dev->minor_shift;
 	devfs_handle_t dir, slave;
 	unsigned int devfs_flags = DEVFS_FL_DEFAULT;
+	umode_t devfs_perm  = S_IFBLK | S_IRUSR | S_IWUSR;
 	char dirname[64], symlink[16];
 	static devfs_handle_t devfs_handle;
 
 	if (dev->part[minor].de) return;
 	if ( dev->flags && (dev->flags[devnum] & GENHD_FL_REMOVABLE) )
 		devfs_flags |= DEVFS_FL_REMOVABLE;
+	if (is_read_only(MKDEV(dev->major, minor))) {
+	        devfs_perm &= ~(S_IWUSR);
+	}
 	if (dev->de_arr) {
 		dir = dev->de_arr[devnum];
 		if (!dir)  /*  Aware driver wants to block disc management  */

@@ -327,7 +335,7 @@
 			  dirname + pos, &slave, NULL);
 	dev->part[minor].de =
 	    devfs_register (dir, "disc", devfs_flags, dev->major, minor,
-			    S_IFBLK | S_IRUSR | S_IWUSR, dev->fops, NULL);
+			    devfs_perm, dev->fops, NULL);
 	devfs_auto_unregister (dev->part[minor].de, slave);
 	if (!dev->de_arr)
 		devfs_auto_unregister (slave, dir);

@@ -355,6 +363,10 @@
 		dev->part[minor].de = NULL;
 		devfs_dealloc_unique_number (&disc_numspace,
 					     dev->part[minor].number);
+		if(dev->label_arr && dev->label_arr[minor >> dev->minor_shift]) {
+			devfs_unregister(dev->label_arr[minor >> dev->minor_shift]);
+			dev->label_arr[minor >> dev->minor_shift] = NULL;
+		}
 	}
 #endif  /*  CONFIG_DEVFS_FS  */
 }

@@ -409,6 +421,93 @@
 		for (i = first_minor; i < end_minor; i++)
 			dev->sizes[i] = dev->part[i].nr_sects >> (BLOCK_SIZE_BITS - 9);
 	}
+}
+
+/*
+ * This function creates a link from /dev/labels/<labelname> to the devfs
+ * device directory. The device driver must allocate memory to the label_arr
+ * for this to work.
+ * This enables devices/partition tables that support labels to be accessed
+ * by that name instead of the device name (which can change if devices are
+ * moved around).
+ *
+ * Current restictions:
+ *   - Only the first device that uses a certain label creates the link
+ *     (which can also be good in case there is a backup device)
+ *   - When removing devices that created labels previously suppressed
+ *     devices won't show up.
+ */
+void register_disk_label(struct gendisk *hd, int minor, char *label) {
+#ifdef CONFIG_DEVFS_FS
+	int                   disknum         = minor >> hd->minor_shift;
+	static devfs_handle_t devfs_label_dir = NULL;
+	int                   i;
+
+	/*
+	 * Check the given label. Trailing whitespaces are removed. Otherwise
+	 * only alphanumeric characters are allowed. (fstab)
+	 * Added [$#%@] since these are allowed by fdasd and seem
+	 * to work in fstab.
+	 */
+	for(i=0; label[i] != '\0'; i++);
+	for(i--; i >= 0; i--) {
+		if(label[i] == ' ' && label[i+1] == '\0') {
+			label[i] = '\0';
+			continue;
+		}
+		if(
+			label[i] == '$' || label[i] == '#' ||
+			label[i] == '@' || label[i] == '%'
+		)
+			continue;
+		if(label[i] >= 'a' && label[i] <= 'z')
+			continue;
+		if(label[i] >= 'A' && label[i] <= 'Z')
+			continue;
+		if(label[i] >= '0' && label[i] <= '9')
+			continue;
+
+		printk(KERN_WARNING "\nregister_disk_label: invalid character(s)"
+			" in label <%s>\n", label);
+		printk(KERN_WARNING "register_label: refusing to create devfs entry.\n");
+		return;
+	}
+
+	if(!hd->label_arr)
+		return;
+
+	if(!devfs_label_dir)
+		if(!(devfs_label_dir = devfs_mk_dir(NULL, "labels", NULL)))
+			return;
+
+	if(hd->label_arr[disknum]) {
+		if(strcmp(devfs_get_name(hd->label_arr[disknum], NULL), label) == 0)
+			return;
+
+		devfs_unregister(hd->label_arr[disknum]);
+		hd->label_arr[disknum] = NULL;
+	}
+	if(!devfs_find_handle(devfs_label_dir, label, 0, 0, 0, 0)) {
+		int  pos      = 0;
+		char path[64];
+
+		if(hd->de_arr) {
+			if(!hd->de_arr[disknum])
+				return;
+
+			pos = devfs_generate_path(hd->de_arr[disknum], path+3, sizeof(path)-3);
+			if(pos < 0)
+				return;
+
+			strncpy(path+pos, "../", 3);
+		} else {
+			sprintf(path, "../%s/disc/%d", hd->major_name, disknum);
+		}
+		devfs_mk_symlink(
+			devfs_label_dir, label, DEVFS_FL_DEFAULT, path+pos,
+			&hd->label_arr[disknum], NULL);
+	}
+#endif
 }
 
 unsigned char *read_dev_sector(struct block_device *bdev, unsigned long n, Sector *p)

fs/partitions/check.h +2 -0

diff -urN linux-2.4.19-rc3/fs/partitions/check.h linux-2.4.19-s390/fs/partitions/check.h
@@ -13,4 +13,6 @@
 	page_cache_release(p.v);
 }
 
+void register_disk_label(struct gendisk *hd, int minor, char *label);
+
 extern int warn_no_part;

fs/partitions/ibm.c +5 -1

diff -urN linux-2.4.19-rc3/fs/partitions/ibm.c linux-2.4.19-s390/fs/partitions/ibm.c
@@ -123,6 +123,7 @@
 	data = read_dev_sector(bdev, info->label_block*(blocksize/512), &sect);
 	if (data == NULL)
 		goto out_readerr;
+
 	strncpy (type, data, 4);
 	if ((!info->FBA_layout) && (!strcmp(info->type, "ECKD")))
 		strncpy(name, data + 8, 6);

@@ -134,6 +135,9 @@
 	EBCASC(type, 4);
 	EBCASC(name, 6);
 
+	if(name[0] != '\0')
+		register_disk_label(hd, MINOR(to_kdev_t(bdev->bd_dev)), name);
+
 	/*
 	 * Three different types: CMS1, VOL1 and LNX1/unlabeled
 	 */

@@ -214,7 +218,7 @@
 		// add_gd_partition(hd, first_part_minor - 1, 0, size);
 		add_gd_partition(hd, first_part_minor,
 				 offset*(blocksize >> 9),
-				  size-offset*(blocksize >> 9));
+				 size-offset*(blocksize >> 9));
 	}
 
 	printk("\n");

include/linux/genhd.h +1 -0

diff -urN linux-2.4.19-rc3/include/linux/genhd.h linux-2.4.19-s390/include/linux/genhd.h
@@ -100,6 +100,7 @@
 
 	devfs_handle_t *de_arr;         /* one per physical disc */
 	char *flags;                    /* one per physical disc */
+	devfs_handle_t *label_arr;      /* one per physical disc */
 };
 
 /* drivers/block/genhd.c */

Christoph Hellwig

2002-08-05 17:11 UTC #12 lore raw

Re: [PATCH] 14/18 support for partition labels in devfs

Quoted 3 lines · Arnd Bergmann

This patch adds a new feature to the partition code. It will automatically generate links in the device filesystem from the label name of a dasd to the device node. This is only needed if the device filesystem is used.

Sorry, a s390-only /dev/labels/ is not acceptable. There are more than enough storage devices that can support labels. And even a generic implementation should go into 2.5 first.

Arnd Bergmann

author

2002-08-05 16:12 UTC #13 lore raw

PATCH 1/1 17/18 support arbitrary devfs names for tty devices

In reply to Arnd Bergmann

This patch introduces two functions tty_io. They are needed by the 3270 driver to generate the correct entries in the device file system. We do not want the 3270 device to get device names according to their minor number but we want to see the real device number. If the device file system is not used this can be replaced by a call to tty_register_devfs.

drivers/char/tty_io.c +55 -0

diff -urN linux-2.4.19-rc3/drivers/char/tty_io.c linux-2.4.19-s390/drivers/char/tty_io.c
@@ -2058,8 +2058,63 @@
 #endif /* CONFIG_DEVFS_FS */
 }
 
+/*
+ * Register a tty device described by <driver>, with minor number <minor>,
+ * device name <name> and in the /dev directory given by <dir>.
+ */
+void tty_register_devfs_name (struct tty_driver *driver, unsigned int flags,
+			      unsigned minor, devfs_handle_t dir,
+			      const char *name)
+{
+#ifdef CONFIG_DEVFS_FS
+	umode_t mode = S_IFCHR | S_IRUSR | S_IWUSR;
+	kdev_t device = MKDEV (driver->major, minor);
+
+	switch (device) {
+		case TTY_DEV:
+		case PTMX_DEV:
+			mode |= S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH;
+			break;
+		default:
+			if (driver->major == PTY_MASTER_MAJOR)
+				flags |= DEVFS_FL_AUTO_OWNER;
+			break;
+	}
+	if ( (minor <  driver->minor_start) || 
+	     (minor >= driver->minor_start + driver->num) ) {
+		printk(KERN_ERR "Attempt to register invalid minor number "
+		       "with devfs (%d:%d).\n", (int)driver->major,(int)minor);
+		return;
+	}
+#  ifdef CONFIG_UNIX98_PTYS
+	if ( (driver->major >= UNIX98_PTY_SLAVE_MAJOR) &&
+	     (driver->major < UNIX98_PTY_SLAVE_MAJOR + UNIX98_NR_MAJORS) )
+		flags |= DEVFS_FL_CURRENT_OWNER;
+#  endif
+	devfs_register (dir, name, flags | DEVFS_FL_DEFAULT,
+			driver->major, minor, mode, &tty_fops, NULL);
+#endif /* CONFIG_DEVFS_FS */
+}
+
+void tty_unregister_devfs_name (struct tty_driver *driver, unsigned minor,
+				devfs_handle_t dir, const char *name)
+{
+#ifdef CONFIG_DEVFS_FS
+	void * handle;
+
+	handle = devfs_find_handle (dir, name, driver->major, minor,
+				    DEVFS_SPECIAL_CHR, 0);
+	devfs_unregister (handle);
+#endif /* CONFIG_DEVFS_FS */
+}
+
+extern void tty_unregister_devfs_name (struct tty_driver *driver,
+				       unsigned minor, devfs_handle_t dir,
+				       const char *name);
 EXPORT_SYMBOL(tty_register_devfs);
 EXPORT_SYMBOL(tty_unregister_devfs);
+EXPORT_SYMBOL(tty_register_devfs_name);
+EXPORT_SYMBOL(tty_unregister_devfs_name);
 
 /*
  * Called by a tty driver to register itself.

include/linux/tty.h +7 -0

diff -urN linux-2.4.19-rc3/include/linux/tty.h linux-2.4.19-s390/include/linux/tty.h
@@ -381,6 +381,13 @@
 extern void tty_register_devfs (struct tty_driver *driver, unsigned int flags,
 				unsigned minor);
 extern void tty_unregister_devfs (struct tty_driver *driver, unsigned minor);
+struct devfs_entry;
+extern void tty_register_devfs_name (struct tty_driver *driver,
+				     unsigned int flags, unsigned minor,
+				     struct devfs_entry *dir, const char *name);
+extern void tty_unregister_devfs_name (struct tty_driver *driver,
+				       unsigned minor, struct devfs_entry *dir,
+				       const char *name);
 extern int tty_read_raw_data(struct tty_struct *tty, unsigned char *bufp,
 			     int buflen);
 extern void tty_write_message(struct tty_struct *tty, char *msg);

Arnd Bergmann

author

2002-08-05 16:12 UTC #14 lore raw

PATCH 1/1 18/18 scsi core changes

In reply to Arnd Bergmann

This patch does not look fit for inclusion to me, but appearantly it's needed if anyone actually wants to use the new zfcp driver I sent with patch 10/18...

drivers/scsi/Config.in +8 -0

diff -urN linux-2.4.19-rc3/drivers/scsi/Config.in linux-2.4.19-s390/drivers/scsi/Config.in
@@ -45,6 +45,8 @@
 if [ "$CONFIG_PCI" = "y" ]; then
    dep_tristate '3ware Hardware ATA-RAID support' CONFIG_BLK_DEV_3W_XXXX_RAID $CONFIG_SCSI
 fi
+
+if [ "$CONFIG_ARCH_S390" != "y" ]; then
 dep_tristate '7000FASST SCSI support' CONFIG_SCSI_7000FASST $CONFIG_SCSI
 dep_tristate 'ACARD SCSI support' CONFIG_SCSI_ACARD $CONFIG_SCSI
 dep_tristate 'Adaptec AHA152X/2825 support' CONFIG_SCSI_AHA152X $CONFIG_SCSI

@@ -251,6 +253,12 @@
 #      bool 'Cyberstorm Mk III SCSI support (EXPERIMENTAL)' CONFIG_CYBERSTORMIII_SCSI
 #      bool 'GVP Turbo 040/060 SCSI support (EXPERIMENTAL)' CONFIG_GVP_TURBO_SCSI
    fi
+fi
+
+fi
+
+if [ "$CONFIG_ARCH_S390" = "y" ]; then
+dep_tristate 'IBM z900 FCP host bus adapter driver' CONFIG_ZFCP $CONFIG_QDIO
 fi
 
 endmenu

drivers/scsi/hosts.c +64 -2

102 lines not on this page · raw message

drivers/scsi/hosts.h +4 -1

diff -urN linux-2.4.19-rc3/drivers/scsi/hosts.h linux-2.4.19-s390/drivers/scsi/hosts.h
@@ -464,7 +464,10 @@
 
 extern Scsi_Host_Template * scsi_hosts;
 
-extern void build_proc_dir_entries(Scsi_Host_Template  *);
+#ifdef CONFIG_PROC_FS
+extern void build_proc_dir(Scsi_Host_Template *);
+extern void build_proc_dir_entry(struct Scsi_Host *);
+#endif
 
 /*
  *  scsi_init initializes the scsi hosts.

drivers/scsi/scsi.c +37 ~~-88~~

177 lines not on this page · raw message

drivers/scsi/scsi.h +11 -0

diff -urN linux-2.4.19-rc3/drivers/scsi/scsi.h linux-2.4.19-s390/drivers/scsi/scsi.h
@@ -390,6 +390,17 @@
 #include <asm/pgtable.h>
 #define CONTIGUOUS_BUFFERS(X,Y) \
 	(virt_to_phys((X)->b_data+(X)->b_size-1)+1==virt_to_phys((Y)->b_data))
+#elif defined(CONFIG_ARCH_S390) || defined(CONFIG_ARCH_S390X)
+#define _CONTIGUOUS_BUFFERS(xd, xs, yd, ys) \
+     (((xd + xs) == yd) \
+      && ((xd & PAGE_MASK) == ((yd + ys - 1) & PAGE_MASK)))
+
+#define CONTIGUOUS_BUFFERS(X,Y) \
+	_CONTIGUOUS_BUFFERS( \
+		(unsigned long)(X)->b_data, \
+		(unsigned long)(X)->b_size, \
+		(unsigned long)(Y)->b_data, \
+		(unsigned long)(Y)->b_size)
 #else
 #define CONTIGUOUS_BUFFERS(X,Y) ((X->b_data+X->b_size) == Y->b_data)
 #endif

drivers/scsi/scsi_lib.c +9 -9

diff -urN linux-2.4.19-rc3/drivers/scsi/scsi_lib.c linux-2.4.19-s390/drivers/scsi/scsi_lib.c
@@ -262,6 +262,15 @@
 		 * the bad sector.
 		 */
 		SCpnt->request.special = (void *) SCpnt;
+                /*
+                 * We need to recount the number of
+                 * scatter-gather segments here - the
+                 * normal case code assumes this to be
+                 * correct, as it would be a performance
+                 * loss to always recount.  Handling
+                 * errors is always unusual, of course.
+                 */
+                recount_segments(SCpnt);
 		list_add(&SCpnt->request.queue, &q->queue_head);
 	}

@@ -969,15 +978,6 @@
 			 */
 			if( req->special != NULL ) {
 				SCpnt = (Scsi_Cmnd *) req->special;
-				/*
-				 * We need to recount the number of
-				 * scatter-gather segments here - the
-				 * normal case code assumes this to be
-				 * correct, as it would be a performance
-				 * lose to always recount.  Handling
-				 * errors is always unusual, of course.
-				 */
-				recount_segments(SCpnt);
 			} else {
 				SCpnt = scsi_allocate_device(SDpnt, FALSE, FALSE);
 			}

drivers/scsi/scsi_proc.c +25 ~~-26~~

60 lines not on this page · raw message

Christoph Hellwig

2002-08-05 17:09 UTC #15 lore raw

Re: [PATCH] 18/18 scsi core changes

Quoted 1 line · Arnd Bergmann

This patch does not look fit for inclusion to me,

It absolutely does not look good for inclusion.

Quoted 3 lines

but appearantly it's
needed if anyone actually wants to use the new zfcp driver I sent with patch 10/18...

the zfcp driver itself is so ugly that I wonder you even show it publically..

Quoted 4 lines

+if [ "$CONFIG_ARCH_S390" != "y" ]; then
dep_tristate '7000FASST SCSI support' CONFIG_SCSI_7000FASST $CONFIG_SCSI dep_tristate 'ACARD SCSI support' CONFIG_SCSI_ACARD $CONFIG_SCSI dep_tristate 'Adaptec AHA152X/2825 support' CONFIG_SCSI_AHA152X $CONFIG_SCSI

all these entries want indentation by three spaces. See Documentation/CodingStyle.

drivers/scsi/Config.in line 261

Quoted 2 lines

drivers/scsi/Config.in +2 -0

+if [ "$CONFIG_ARCH_S390" = "y" ]; then
+dep_tristate 'IBM z900 FCP host bus adapter driver' CONFIG_ZFCP $CONFIG_QDIO

dito.

Quoted 11 lines

 scsi_unregister(struct Scsi_Host * sh){
     struct Scsi_Host * shpnt;
     Scsi_Host_Name *shn;

drivers/scsi/hosts.c +8 -0

+    char name[10];
+
+    /* kill error handling thread */
+    if (sh->hostt->use_new_eh_code
+        && sh->ehandler != NULL) {
+            DECLARE_MUTEX_LOCKED(sem);
+            
+            sh->eh_notify = &sem;

drivers/scsi/hosts.c line 108

Quoted 10 lines

drivers/scsi/hosts.c +10 -0

+            send_sig(SIGHUP, sh->ehandler, 1);
+            down(&sem);
+            sh->eh_notify = NULL;
+    }
+   
+    /* remove proc entry */
+#ifdef CONFIG_PROC_FS
+    sprintf(name, "%d", sh->host_no);
+    remove_proc_entry(name, sh->hostt->proc_dir);
+#endif

this change actually looks useful. but I don't think it's suitable for stable series.

Quoted 8 lines

diff -urN linux-2.4.19-rc3/drivers/scsi/scsi.h linux-2.4.19-s390/drivers/scsi/scsi.h

drivers/scsi/scsi.h +1 -0

@@ -390,6 +390,17 @@
 #include <asm/pgtable.h>
 #define CONTIGUOUS_BUFFERS(X,Y) \
 	(virt_to_phys((X)->b_data+(X)->b_size-1)+1==virt_to_phys((Y)->b_data))
+#elif defined(CONFIG_ARCH_S390) || defined(CONFIG_ARCH_S390X)

drivers/scsi/scsi.h line 403

Quoted 10 lines

drivers/scsi/scsi.h +10 -0

+#define _CONTIGUOUS_BUFFERS(xd, xs, yd, ys) \
+     (((xd + xs) == yd) \
+      && ((xd & PAGE_MASK) == ((yd + ys - 1) & PAGE_MASK)))
+
+#define CONTIGUOUS_BUFFERS(X,Y) \
+	_CONTIGUOUS_BUFFERS( \
+		(unsigned long)(X)->b_data, \
+		(unsigned long)(X)->b_size, \
+		(unsigned long)(Y)->b_data, \
+		(unsigned long)(Y)->b_size)

This area is going to change heavily with block-highmem in 2.4.20. Please don't touch it just now.

drivers/scsi/scsi.h

Quoted 9 lines

drivers/scsi/scsi.h +9 -0

+                /*
+                 * We need to recount the number of
+                 * scatter-gather segments here - the
+                 * normal case code assumes this to be
+                 * correct, as it would be a performance
+                 * loss to always recount.  Handling
+                 * errors is always unusual, of course.
+                 */
+                recount_segments(SCpnt);

might be worth to get the indentation right, heh?

What is the exact reason to move it around?

Arnd Bergmann

2002-08-06 09:03 UTC #16 lore raw

Re: [PATCH] 18/18 scsi core changes

Quoted 1 line · Christoph Hellwig

It absolutely does not look good for inclusion.

OK, thanks for looking at it. Maybe the zfcp developers can find a less intrusive way of getting the driver to work if they want it to work out of the box.

Quoted 2 lines

the zfcp driver itself is so ugly that I wonder you even show it publically..

The other new drivers I sent (lcs, z90crypt and qdio) and some of the s390 stuff that's already in 2.4 are not really any better. These three have all been closed source before and you can see them as a warning of what happens to your code if you skip public peer review ;-)
Still, it's the stuff IBM recommends for use and it's not going away (at least not in 2.4), so I guess it might just as well be included.

OTOH, if Marcello wants to the next patch size small, it might be a good idea to leave out the uglier drivers.

	Arnd <><

Christoph Hellwig

2002-08-06 09:14 UTC #17 lore raw

Re: [PATCH] 18/18 scsi core changes

Quoted 3 lines · Arnd Bergmann

Still, it's the stuff IBM recommends for use and it's not going away (at least not in 2.4), so I guess it might just as well be included.

Following that argumentation we could also include the broken qlogic driver and the nvidia glue..

Ingo Adlung

2002-08-06 23:03 UTC #18 lore raw

Re: [PATCH] 18/18 scsi core changes

Quoted 9 lines · Christoph Hellwig

Quoted 3 lines · Arnd Bergmann
Still, it's the stuff IBM recommends for use and it's not going away (at least not in 2.4), so I guess it might just as well be included.
Following that argumentation we could also include the broken qlogic driver and the nvidia glue..

Well, if the hardware implementation would always be straight forward to what you know from PCs, it wouldn't be required to apply additional patches to the common code, but it would be easy to stay within the architecture specifics. Unfortunately it is not always that easy, by large related to virtualization issues ... not an excuse, but an explanation. Those not intending to use Linux on IBM's mainframes, and what's worth in this context its FCP support can safely ignore the SCSI related patch. Anybody else will need it - but you're welcome to show alternatives, I'm sure the folks having written the code do very much appreciate any guidance you can provide 'em getting to a nicer/cleaner implementation finding everybody's appreciation :-)

Ingo

Christoph Hellwig

2002-08-06 23:08 UTC #19 lore raw

Re: [PATCH] 18/18 scsi core changes

Quoted 8 lines · Ingo Adlung

Quoted 3 lines
Following that argumentation we could also include the broken qlogic driver and the nvidia glue..
Well, if the hardware implementation would always be straight forward to what you know from PCs, it wouldn't be required to apply additional patches to the common code, but it would be easy to stay within the architecture specifics.

That above statement was about the horrible code quality of most if not all s390 drivers. Yes there are similarly bad vendor drivers around for intel hardware - buf fortunately they are not in the tree.