Skip to contentSkip to conversation

Problem booting Adaptec Ultra 320 driver with discontigmem support.

DISCUSSION Xavier Bru 3 participants 5 messages Dec 4, 2003 – Dec 5, 2003

lore
Share & tools
lore

Xavier Bru

author
#0 lore raw

Hello, all

Running 2.6.0-test11 with Adaptec Ultra 320 driver that provides full 64-bit support
we hit a BUG in blk_queue_bounce_limit() due to physical address greater than allowed by the dma mask.
Having a look into the code, it appears that the computation to allow full 64-bit support takes in account the size of the installed memory instead of the max physical address of the platform.
For example on a platform with 32GB of memory distributed in 1 TB physical address space, only 39-bits physical addresses are supported instead of the 40 bits needed.

IOSAPIC: vector 54 -> CPU 0x04e0, enabled
scsi0 : Adaptec AIC79XX PCI-X SCSI HBA DRIVER, Rev 1.3.9

        <Adaptec 39320D Ultra320 SCSI adapter>
        aic7902: Ultra320 Wide Channel A, SCSI Id=7, PCI-X 101-133Mhz, 512 SCBs

kernel BUG at drivers/block/ll_rw_blk.c:271!
swapper[1]: bugcheck! 0 [1]

Herafter a patch that fixes the problem:

--- linux-2.6.0-test11/drivers/scsi/aic7xxx/aic79xx_osm.c 2003-11-24
02:32:03.000000000 +0100
+++ linux-2.6.0-test11.new/drivers/scsi/aic7xxx/aic79xx_osm.c

2003-12-03 14:39:51.000000000 +0100

@@ -62,6 +62,7 @@

#include <linux/mm.h> /* For fetching system memory size */

+extern unsigned long blk_max_pfn;
/*
* Lock protecting manipulation of the ahd softc list. */
@@ -2158,10 +2159,9 @@
uint64_t
ahd_linux_get_memsize(void)
{
- struct sysinfo si;
-

-    si_meminfo(&si);
-    return ((uint64_t)si.totalram << PAGE_SHIFT);
+    /* Need to take in account the max physical address in case

+ * of discontiguous memory. */
+ return ((uint64_t)blk_max_low_pfn << PAGE_SHIFT);
}

/*

Signature · 8 lines
 Sincères salutations.
_____________________________________________________________________
 
Xavier BRU                 BULL ISD/R&D/INTEL office:     FREC B1-422
tel : +33 (0)4 76 29 77 45                    http://www-frec.bull.fr
fax : +33 (0)4 76 29 77 70                 mailto:[email protected]
addr: BULL, 1 rue de Provence, BP 208, 38432 Echirolles Cedex, FRANCE
_____________________________________________________________________
author
#1 lore raw

Sorry, this was the old one, this one should be better (use blk_max_pfn instead of blk_max_low_pfn):

--- linux-2.6.0-test11/drivers/scsi/aic7xxx/aic79xx_osm.c 2003-11-24
02:32:03.000000000 +0100
+++ linux-2.6.0-test11.new/drivers/scsi/aic7xxx/aic79xx_osm.c

2003-12-03 14:39:51.000000000 +0100

@@ -62,6 +62,7 @@

#include <linux/mm.h> /* For fetching system memory size */

+extern unsigned long blk_max_pfn;
/*
* Lock protecting manipulation of the ahd softc list. */
@@ -2158,10 +2159,9 @@
uint64_t
ahd_linux_get_memsize(void)
{
- struct sysinfo si;
-

-    si_meminfo(&si);
-    return ((uint64_t)si.totalram << PAGE_SHIFT);
+    /* Need to take in account the max physical address in case

+ * of discontiguous memory. */
+ return ((uint64_t)blk_max_pfn << PAGE_SHIFT);
}

/*

Quoted 51 lines · Xavier Bru

Hello, all

Running 2.6.0-test11 with Adaptec Ultra 320 driver that provides full 64-bit support
we hit a BUG in blk_queue_bounce_limit() due to physical address greater than allowed by the dma mask.
Having a look into the code, it appears that the computation to allow full 64-bit support takes in account the size of the installed memory instead of the max physical address of the platform.
For example on a platform with 32GB of memory distributed in 1 TB physical address space, only 39-bits physical addresses are supported instead of the 40 bits needed.

IOSAPIC: vector 54 -> CPU 0x04e0, enabled
scsi0 : Adaptec AIC79XX PCI-X SCSI HBA DRIVER, Rev 1.3.9

       <Adaptec 39320D Ultra320 SCSI adapter>
       aic7902: Ultra320 Wide Channel A, SCSI Id=7, PCI-X 101-133Mhz, 

512 SCBs

kernel BUG at drivers/block/ll_rw_blk.c:271!
swapper[1]: bugcheck! 0 [1]

Herafter a patch that fixes the problem:

--- linux-2.6.0-test11/drivers/scsi/aic7xxx/aic79xx_osm.c

2003-11-24 02:32:03.000000000 +0100
+++ linux-2.6.0-test11.new/drivers/scsi/aic7xxx/aic79xx_osm.c    
2003-12-03 14:39:51.000000000 +0100
(quoted diff) +1 -0
@@ -62,6 +62,7 @@
 
 #include <linux/mm.h>        /* For fetching system memory size */
 
+extern unsigned long blk_max_pfn;
 /*
 * Lock protecting manipulation of the ahd softc list.
 */
@@ -2158,10 +2159,9 @@

uint64_t
ahd_linux_get_memsize(void)
{

(quoted diff) +3 -4
-    struct sysinfo si;
-
-    si_meminfo(&si);
-    return ((uint64_t)si.totalram << PAGE_SHIFT);
+    /* Need to take in account the max physical address in case
+     * of discontiguous memory. */
+    return ((uint64_t)blk_max_low_pfn << PAGE_SHIFT);

}

/*

Signature · 8 lines
 Sincères salutations.
_____________________________________________________________________
 
Xavier BRU                 BULL ISD/R&D/INTEL office:     FREC B1-422
tel : +33 (0)4 76 29 77 45                    http://www-frec.bull.fr
fax : +33 (0)4 76 29 77 70                 mailto:[email protected]
addr: BULL, 1 rue de Provence, BP 208, 38432 Echirolles Cedex, FRANCE
_____________________________________________________________________
#2 lore raw
In reply to Xavier Bru
Quoted 12 lines · Xavier Bru

Hello, all

Running 2.6.0-test11 with Adaptec Ultra 320 driver that provides full 64-bit support
we hit a BUG in blk_queue_bounce_limit() due to physical address greater than allowed by the dma mask.
Having a look into the code, it appears that the computation to allow full 64-bit support takes in account the size of the installed memory instead of the max physical address of the platform.
For example on a platform with 32GB of memory distributed in 1 TB physical address space, only 39-bits physical addresses are supported instead of the 40 bits needed.

Still the wrong thing to do. It should just check sizeof(dma_addr_t), that the only sane check.

#3 lore raw
In reply to Xavier Bru
Quoted 6 lines · Xavier Bru

Hello, all

Running 2.6.0-test11 with Adaptec Ultra 320 driver that provides full 64-bit support
we hit a BUG in blk_queue_bounce_limit() due to physical address greater than allowed by the dma mask.

Can the driver provide "full 64-bit support" if it's setting the dma mask to something less than 64-bits?
Did you mean "full 40-bits DMA support"?

grant

author
#4 lore raw
In reply to Xavier Bru

The driver dynamically tests the size of the memory and sets the appropriate dma_mask to: 32, 39 or 64 bits (probably dues to historical reasons).
Problem is that with 32 GB of memory installed in a 1 TB physical address space the 39 bit mask is used instead
of the 64-bit mask.

Signature · 8 lines
 Sincères salutations.
_____________________________________________________________________
 
Xavier BRU                 BULL ISD/R&D/INTEL office:     FREC B1-422
tel : +33 (0)4 76 29 77 45                    http://www-frec.bull.fr
fax : +33 (0)4 76 29 77 70                 mailto:[email protected]
addr: BULL, 1 rue de Provence, BP 208, 38432 Echirolles Cedex, FRANCE
_____________________________________________________________________