pci issue - wrong detection of pci ressources

[Sergei Shtylyov]

Hello.

Christian Ehrhardt wrote:

>> Cheers,
>> Ben.

> For comparison I defined DEBUG in the good kernel (arch=ppc) and that is 
> what the initialization prints (pci ...:0a:1 is the secondary head of 
> the same graphic card an it's not an issue if thats not allocated):
> good case:
> PCI: Probing PCI hardware
> PCI: bridge rsrc 0..ffff (100), parent c0354624
> PCI: bridge rsrc 80000000..8fffffff (200), parent c0354608
> PCI:0000:00:0a.0: Resource 0: 0000000088000000-000000008fffffff (f=1208)
> PCI:0000:00:0a.0: Resource 1: 000000000000ff00-000000000000ffff (f=101)
> PCI:0000:00:0a.0: Resource 2: 0000000087ff0000-0000000087ffffff (f=200)
> PCI:0000:00:0a.1: Resource 0: 0000000078000000-000000007fffffff (f=1208)
> PCI: Cannot allocate resource region 0 of device 0000:00:0a.1
> PCI:0000:00:0a.1: Resource 1: 0000000077ff0000-0000000077ffffff (f=200)
> PCI: Cannot allocate resource region 1 of device 0000:00:0a.1
> PCI: Failed to allocate mem resource #0:8000000 at 90000000 for 0000:00:0a.1

> ------------------------------------------------------------------------

> PCI host bridge /plb/pci at 1ec000000 (primary) ranges:
>  MEM 0x0000000180000000..0x000000018fffffff -> 0x0000000080000000
>   IO 0x00000001e8000000..0x00000001e80fffff -> 0x0000000000000000
> 4xx PCI DMA offset set to 0x00000000
> PCI: Probing PCI hardware
> PCI: Hiding 4xx host bridge resources 0000:00:00.0
> Try to map irq for 0000:00:00.0...
>  -> got one, spec 2 cells (0x00000003 0x00000008...) on /interrupt-controller2
>  -> mapped to linux irq 16
> Try to map irq for 0000:00:0a.0...
>  -> got one, spec 2 cells (0x00000003 0x00000008...) on /interrupt-controller2
>  -> mapped to linux irq 16
> Try to map irq for 0000:00:0a.1...
> PCI: PHB (bus 0) bridge rsrc 0: 0000000000000000-00000000000fffff [0x100], parent c0365060 (PCI IO)
> __request_resource - request cf8045b0 name '/plb/pci at 1ec000000' start 0x0 end 0x0
> __request_resource - no conflict parent c0365060 sibling 00000000
> PCI: PHB (bus 0) bridge rsrc 1: 0000000180000000-000000018fffffff [0x200], parent c0365038 (PCI mem)
> __request_resource - request cf8045d8 name '/plb/pci at 1ec000000' start 0x1 end 0x80000000

    Bogus start/end value because you're not printing them right...

> __request_resource - no conflict parent c0365038 sibling 00000000
> PCI: Assigning unassigned resouces...
> pci_assign_resource - allocate with IORESOURCE_PREFETCH
> pci_bus_alloc_resource - enter
> pci_assign_resource - second pci_bus_alloc_resource call
> pci_bus_alloc_resource - enter
> pci_bus_alloc_resource - call allocate ressource size 0 startcalc 134217728, align -1
> find_resource - size 0, min 0x8000000, max 0x1
> find_resource - found start 0x1 end 0x80000000
> __request_resource - request cf810578 name '0000:00:0a.0' start 0x1 end 0x80000000

    The start/end values are bogus -- not printed right.

> __request_resource - no conflict parent cf8045d8 sibling 00000000
> pci_assign_resource - allocate with IORESOURCE_PREFETCH
> pci_bus_alloc_resource - enter
> pci_assign_resource - second pci_bus_alloc_resource call
> pci_bus_alloc_resource - enter
> pci_bus_alloc_resource - call allocate ressource size 0 startcalc 134217728, align -1
> find_resource - size 0, min 0x8000000, max 0x1
> find_resource - continue with start 0x1 on 88000000
> find_resource - found start 0x1 end 0x88000000
> __request_resource - request cf810178 name '0000:00:0a.1' start 0x1 end 0x88000000

    Same here.

> __request_resource - no conflict parent cf8045d8 sibling 00000000
> pci_assign_resource - allocate with IORESOURCE_PREFETCH

    It's not clear why the register space is prefetchable.

> pci_bus_alloc_resource - enter
> pci_assign_resource - second pci_bus_alloc_resource call
> pci_bus_alloc_resource - enter
> pci_bus_alloc_resource - call allocate ressource size 0 startcalc 131072, align -1
> find_resource - size 0, min 0x20000, max 0x1
> find_resource - continue with start 0x1 on 88000000
> find_resource - continue with start 0x1 on 90000000
> find_resource - no this - exit

    ... and here. It's not clear why it failed with 188000000... maybe we ran 
out of prefetchable space?

> PCI: pci_assign_resource - Failed to allocate mem resource #6:20000 at 190000000 for 0000:00:0a.0
> pci_assign_resource - allocate with IORESOURCE_PREFETCH
> pci_bus_alloc_resource - enter
> pci_bus_alloc_resource - call allocate ressource size 0 startcalc 65536, align -1
> find_resource - size 0, min 0x10000, max 0x1
> find_resource - continue with start 0x1 on 88000000
> find_resource - continue with start 0x1 on 90000000
> find_resource - no this - exit
> PCI: pci_assign_resource - Failed to allocate mem resource #2:10000 at 190000000 for 0000:00:0a.0
> pci_assign_resource - allocate with IORESOURCE_PREFETCH
> pci_bus_alloc_resource - enter
> pci_bus_alloc_resource - call allocate ressource size 0 startcalc 65536, align -1
> find_resource - size 0, min 0x10000, max 0x1
> find_resource - continue with start 0x1 on 88000000
> find_resource - continue with start 0x1 on 90000000
> find_resource - no this - exit
> PCI: pci_assign_resource - Failed to allocate mem resource #1:10000 at 190000000 for 0000:00:0a.1
> pci_assign_resource - allocate with IORESOURCE_PREFETCH
> pci_bus_alloc_resource - enter
> pci_bus_alloc_resource - call allocate ressource size 0 startcalc 256, align -1
> find_resource - size 0, min 0x100, max 0x0
> find_resource - found start 0x0 end 0x1000
> __request_resource - request cf8105a0 name '0000:00:0a.0' start 0x0 end 0x1000
> __request_resource - no conflict parent cf8045b0 sibling 00000000
> SCSI subsystem initialized
> [...]
> radeonfb_pci_register BEGIN
> radeonfb_pci_register - resource0 0x1 - 0x80000000
> radeonfb_pci_register - resource1 0x0 - 0x1000
> radeonfb_pci_register - resource2 0x0 - 0x0

    The value here are bugus because you're not printing them right.

> radeonfb_pci_register - got pci ressources fb_base_phys 0x1 mmio_base_phys 0x80000000
> __request_resource - request cf8bfd20 name 'radeonfb framebuffer' start 0x1 end 0x80000000
> __request_resource - request cf8bfd20 name 'radeonfb framebuffer' start 0x1 end 0x80000000
> __request_resource - request cf8bfd20 name 'radeonfb framebuffer' start 0x1 end 0x80000000
> __request_resource - no conflict parent cf810578 sibling 00000000
> radeonfb_pci_register - call ioremap for base 0 and size 0
> __ioremap(): phys addr 0x0 is RAM lr c029df88 - mem_init_done 1 highmem at 10000000
> radeonfb (0000:00:0a.0): cannot map MMIO
> radeonfb: probe of 0000:00:0a.0 failed with error -5


> ------------------------------------------------------------------------

> Subject: [PATCH] : ! debug only ! - printk's in pci ressource handling

> From: Christian Ehrhardt <ehrhardt at linux.vnet.ibm.com>

> This defines debug for pci and put's some printk's to keep track of the pci
> ressource handling. This patch is just for reference to understand the
> resulting boot message log.
> The patch is ugly - but for debugging only ;-)

    Moreover, it's inconsistent because you're not using the right format for 
printing throughout it...

> Signed-off-by: Christian Ehrhardt <ehrhardt at linux.vnet.ibm.com>
[...]
> diff --git a/arch/powerpc/kernel/pci-common.c b/arch/powerpc/kernel/pci-common.c
> --- a/arch/powerpc/kernel/pci-common.c
> +++ b/arch/powerpc/kernel/pci-common.c
> @@ -37,6 +37,8 @@
>  #include <asm/machdep.h>
>  #include <asm/ppc-pci.h>
>  #include <asm/firmware.h>
> +
> +#define DEBUG
>  
>  #ifdef DEBUG
>  #include <asm/udbg.h>
> diff --git a/arch/powerpc/mm/pgtable_32.c b/arch/powerpc/mm/pgtable_32.c
> --- a/arch/powerpc/mm/pgtable_32.c
> +++ b/arch/powerpc/mm/pgtable_32.c
> @@ -183,8 +183,8 @@ __ioremap(phys_addr_t addr, unsigned lon
>  	 * mem_init() sets high_memory so only do the check after that.
>  	 */
>  	if (mem_init_done && (p < virt_to_phys(high_memory))) {
> -		printk("__ioremap(): phys addr 0x%llx is RAM lr %p\n",
> -		       (unsigned long long)p, __builtin_return_address(0));
> +		printk("__ioremap(): phys addr 0x%llx is RAM lr %p - mem_init_done %d highmem@%p\n",
> +		       (unsigned long long)p, __builtin_return_address(0), mem_init_done, virt_to_phys(high_memory));
>  		return NULL;
>  	}
>  
> diff --git a/drivers/pci/bus.c b/drivers/pci/bus.c
> --- a/drivers/pci/bus.c
> +++ b/drivers/pci/bus.c
> @@ -42,6 +42,8 @@ pci_bus_alloc_resource(struct pci_bus *b
>  {
>  	int i, ret = -ENOMEM;
>  
> +printk(KERN_ERR"%s - enter\n", __func__);
> +
>  	type_mask |= IORESOURCE_IO | IORESOURCE_MEM;
>  
>  	for (i = 0; i < PCI_BUS_NUM_RESOURCES; i++) {
> @@ -59,6 +61,7 @@ pci_bus_alloc_resource(struct pci_bus *b
>  		    !(res->flags & IORESOURCE_PREFETCH))
>  			continue;
>  
> +printk(KERN_ERR"%s - call allocate ressource size %d startcalc %d, align %d\n", __func__, size, (r->start ? : min,-1), align);
>  		/* Ok, try it out.. */
>  		ret = allocate_resource(r, res, size,
>  					r->start ? : min,
> diff --git a/drivers/pci/setup-res.c b/drivers/pci/setup-res.c
> --- a/drivers/pci/setup-res.c
> +++ b/drivers/pci/setup-res.c
> @@ -142,10 +142,13 @@ int pci_assign_resource(struct pci_dev *
>  	   required alignment in the "start" field. */
>  	align = (resno < PCI_BRIDGE_RESOURCES) ? size : res->start;
>  
> +printk(KERN_ERR"%s - allocate with IORESOURCE_PREFETCH\n", __func__);
> +
>  	/* First, try exact prefetching match.. */
>  	ret = pci_bus_alloc_resource(bus, res, size, align, min,
>  				     IORESOURCE_PREFETCH,
>  				     pcibios_align_resource, dev);
> +
>  
>  	if (ret < 0 && (res->flags & IORESOURCE_PREFETCH)) {
>  		/*
> @@ -154,13 +157,14 @@ int pci_assign_resource(struct pci_dev *
>  		 * But a prefetching area can handle a non-prefetching
>  		 * window (it will just not perform as well).
>  		 */
> +printk(KERN_ERR"%s - second pci_bus_alloc_resource call\n", __func__);
>  		ret = pci_bus_alloc_resource(bus, res, size, align, min, 0,
>  					     pcibios_align_resource, dev);
>  	}
>  
>  	if (ret) {
> -		printk(KERN_ERR "PCI: Failed to allocate %s resource "
> -			"#%d:%llx@%llx for %s\n",
> +		printk(KERN_ERR "PCI: %s - Failed to allocate %s resource "
> +			"#%d:%llx@%llx for %s\n", __func__,
>  			res->flags & IORESOURCE_IO ? "I/O" : "mem",
>  			resno, (unsigned long long)size,
>  			(unsigned long long)res->start, pci_name(dev));
> @@ -197,8 +201,8 @@ int pci_assign_resource_fixed(struct pci
>  	}
>  
>  	if (ret) {
> -		printk(KERN_ERR "PCI: Failed to allocate %s resource "
> -				"#%d:%llx@%llx for %s\n",
> +		printk(KERN_ERR "PCI: %s - Failed to allocate %s resource "
> +				"#%d:%llx@%llx for %s\n", __func__,
>  			res->flags & IORESOURCE_IO ? "I/O" : "mem",
>  			resno, (unsigned long long)(res->end - res->start + 1),
>  			(unsigned long long)res->start, pci_name(dev));
> diff --git a/drivers/video/aty/radeon_base.c b/drivers/video/aty/radeon_base.c
> --- a/drivers/video/aty/radeon_base.c
> +++ b/drivers/video/aty/radeon_base.c
> @@ -2160,6 +2160,10 @@ static int __devinit radeonfb_pci_regist
>  	int ret;
>  
>  	RTRACE("radeonfb_pci_register BEGIN\n");
> +
> +printk(KERN_ERR"%s - resource0 0x%0lx - 0x%0lx\n", __func__, pdev->resource[0].start, pdev->resource[0].end);
> +printk(KERN_ERR"%s - resource1 0x%0lx - 0x%0lx\n", __func__, pdev->resource[1].start, pdev->resource[1].end);
> +printk(KERN_ERR"%s - resource2 0x%0lx - 0x%0lx\n", __func__, pdev->resource[2].start, pdev->resource[2].end);

    You should have used %llx here -- the resources are 64-bit on 44x.

>  	/* Enable device in PCI config */
>  	ret = pci_enable_device(pdev);
> @@ -2198,6 +2202,8 @@ static int __devinit radeonfb_pci_regist
>  	rinfo->fb_base_phys = pci_resource_start (pdev, 0);
>  	rinfo->mmio_base_phys = pci_resource_start (pdev, 2);
>  
> +printk(KERN_ERR"%s - got pci ressources fb_base_phys 0x%0lx mmio_base_phys 0x%0lx\n", __func__,rinfo->fb_base_phys,rinfo->mmio_base_phys);
> +

    You've changed fb_base_phys and mmio_base_phys to resource_size_t which is 
64-bit, so use %llx to print them.

>  	/* request the mem regions */
>  	ret = pci_request_region(pdev, 0, "radeonfb framebuffer");
>  	if (ret < 0) {
> @@ -2214,6 +2220,7 @@ static int __devinit radeonfb_pci_regist
>  	}
>  
>  	/* map the regions */
> +printk(KERN_ERR "%s - call ioremap for base %ld and size %d\n",__func__, rinfo->mmio_base_phys, RADEON_REGSIZE);

    You've changed mmio_base_phys to resource_size_t which is 64-bit, so use 
%llx to print it.

>  	rinfo->mmio_base = ioremap(rinfo->mmio_base_phys, RADEON_REGSIZE);
>  	if (!rinfo->mmio_base) {
>  		printk(KERN_ERR "radeonfb (%s): cannot map MMIO\n",
> diff --git a/drivers/video/aty/radeonfb.h b/drivers/video/aty/radeonfb.h
> --- a/drivers/video/aty/radeonfb.h
> +++ b/drivers/video/aty/radeonfb.h
> @@ -287,8 +287,8 @@ struct radeonfb_info {
>  
>  	char			name[DEVICE_NAME_SIZE];
>  
> -	unsigned long		mmio_base_phys;
> -	unsigned long		fb_base_phys;
> +	resource_size_t		mmio_base_phys;
> +	resource_size_t		fb_base_phys;
>  
>  	void __iomem		*mmio_base;
>  	void __iomem		*fb_base;
> diff --git a/kernel/resource.c b/kernel/resource.c
> --- a/kernel/resource.c
> +++ b/kernel/resource.c
> @@ -152,6 +152,9 @@ static struct resource * __request_resou
>  	resource_size_t end = new->end;
>  	struct resource *tmp, **p;
>  
> +printk(KERN_ERR"%s - request %p name '%s' start 0x%0lx end 0x%0lx\n",
> +			__func__, new, new->name, start, end);
> +

    You should have used %llx here too...

>  	if (end < start)
>  		return root;
>  	if (start < root->start)
> @@ -165,6 +168,7 @@ static struct resource * __request_resou
>  			new->sibling = tmp;
>  			*p = new;
>  			new->parent = root;
> +printk(KERN_ERR"%s - no conflict parent %p sibling %p\n",__func__, new->parent, new->sibling);
>  			return NULL;
>  		}
>  		p = &tmp->sibling;
> @@ -305,6 +309,8 @@ static int find_resource(struct resource
>  {
>  	struct resource *this = root->child;
>  
> +printk(KERN_ERR"%s - size %lx, min 0x%0lx, max 0x%0lx\n", __func__, size, min, max);
> +

    Should have used %0llx here as well...

>  	new->start = root->start;
>  	/*
>  	 * Skip past an allocated resource that starts at 0, since the assignment
> @@ -328,10 +334,16 @@ static int find_resource(struct resource
>  			alignf(alignf_data, new, size, align);
>  		if (new->start < new->end && new->end - new->start >= size - 1) {
>  			new->end = new->start + size - 1;
> +printk(KERN_ERR"%s - found start 0x%0lx end 0x%0lx\n", __func__, new->start, new->end);

    And here too...

>  			return 0;
>  		}
> -		if (!this)
> +		if (!this) {
> +printk(KERN_ERR"%s - no this - exit\n", __func__);
>  			break;
> +}
> +else {
> +printk(KERN_ERR"%s - continue with start 0x%0lx on %p\n", __func__, (this->end + 1), this->sibling);

    And here. Yet it's not clear why you call resource's 'end' 'start'...

> +}
>  		new->start = this->end + 1;
>  		this = this->sibling;
>  	}
> @@ -385,6 +397,9 @@ int insert_resource(struct resource *par
>  {
>  	int result;
>  	struct resource *first, *next;
> +
> +printk(KERN_ERR"%s - insert %p (start 0x%0lx end 0x%0lx) under parent %p\n",
> +			__func__, new, new->start, new->end, parent);

    Use %llx hereas well...

>  
>  	write_lock(&resource_lock);
>  

WBR, Sergei