RE: [PATCH] Export msi_remove_pci_irq_vectors

RE: [PATCH] Export msi_remove_pci_irq_vectors

[Nguyen, Tom L]

On Monday, June 21, 2004 Roland Dreier wrote: 

>As a followup to my previous post about the request_mem_region in
>msi.c, I noticed that the region is only released in
>msi_remove_pci_irq_vectors().  Based on the fact that this function is
>declared in linux/pci.h (and stubbed out if CONFIG_PCI_USE_VECTOR is
>not defined), I'm guessing that the intent is for a device driver to
>unconditionally call this when exiting.

The intent of msi_remove_pci_irq_vectors() is to support hot-removed 
operation. This function is not for a device driver to call and
should not be exported. I acknowledged the problem of the MSI-X region 
being only released in msi_remove_pci_irq_vectors(). I'm in a progress 
of updating the existing MSI-X code.

Thanks,
Long

Re: [PATCH] Export msi_remove_pci_irq_vectors

[Roland Dreier]

    Tom> The intent of msi_remove_pci_irq_vectors() is to support
    Tom> hot-removed operation. This function is not for a device
    Tom> driver to call and should not be exported. I acknowledged the
    Tom> problem of the MSI-X region being only released in
    Tom> msi_remove_pci_irq_vectors(). I'm in a progress of updating
    Tom> the existing MSI-X code.

Do you have any plans for when this should be fixed?  Right now, with
the standard kernel, if I unload and then reload my driver module,
setting up MSI-X fails the second time through because the core has
not cleaned up the memory region from the first time.

 - Roland

Question on using MSI in PCI driver

[Roland Dreier]

I'm looking at implementing MSI/MSI-X support in a PCI device driver
I'm working on.  However, I've run into an issue with the MSI API that
I would like some clarification on.

When I call pci_enable_msi, since my device is MSI-X capable, the
kernel calls msix_capability_init, which works out the memory region
where vectors should be written and then calls request_region.  (In
fact it calls

             request_mem_region(phys_addr,
		dev_msi_cap * PCI_MSIX_ENTRY_SIZE,
		"MSI-X iomap Failure"))

which leads to a bizarre entry in /proc/iomem with the name "MSI-X
iomap Failure")

The problem is that if I follow the standard route in my driver and
call pci_request_regions() during init (since I want to claim my whole
device), the request_mem_region in msix_capability_init will fail.
Now, for my device, the MSI-X table happens to fall in the middle of a
BAR, and I need to access stuff on both sides of it in that BAR.  To
make things even worse for me, my device has two more BARs I want to claim.

So it seems I am forced to turn my nice clean pci_request_regions()
call into two calls to request_mem_region() (to get the beginning and
end of the BAR with the MSI-X table in it) and two more calls to
pci_request_region() (to get the other two BARs).

This isn't the end of the world but it feels suboptimal to me.  Anyone
have an idea for a better way to do this?  (I'm happy to write a patch
to the kernel if someone suggests how to change the MSI API)

Thanks,
  Roland

[PATCH] Export msi_remove_pci_irq_vectors

[Roland Dreier]

As a followup to my previous post about the request_mem_region in
msi.c, I noticed that the region is only released in
msi_remove_pci_irq_vectors().  Based on the fact that this function is
declared in linux/pci.h (and stubbed out if CONFIG_PCI_USE_VECTOR is
not defined), I'm guessing that the intent is for a device driver to
unconditionally call this when exiting.

However, a module can't call msi_remove_pci_irq_vectors unless the
symbol is exported... so if this is the way to do things, please apply
this patch.

On the other hand, MSI-HOWTO.txt seems to imply that the 0th MSI
vector should be cleaned up just by calling free_irq... so should
pci_disable_msi be calling msi_remove_pci_irq_vectors?

 - Roland

Index: linux-2.6.7/drivers/pci/msi.c
===================================================================
--- linux-2.6.7.orig/drivers/pci/msi.c	2004-06-15 22:20:03.000000000 -0700
+++ linux-2.6.7/drivers/pci/msi.c	2004-06-21 20:51:33.000000000 -0700
@@ -1011,3 +1011,4 @@
 EXPORT_SYMBOL(pci_enable_msi);
 EXPORT_SYMBOL(msi_alloc_vectors);
 EXPORT_SYMBOL(msi_free_vectors);
+EXPORT_SYMBOL(msi_remove_pci_irq_vectors);

Re: [PATCH] Export msi_remove_pci_irq_vectors

[Christoph Hellwig]

On Mon, Jun 21, 2004 at 09:03:22PM -0700, Roland Dreier wrote:
> On the other hand, MSI-HOWTO.txt seems to imply that the 0th MSI
> vector should be cleaned up just by calling free_irq... so should
> pci_disable_msi be calling msi_remove_pci_irq_vectors?

I think so.