Re: [PATCH 1/2] amd/iommu: Preserve domain ids inside the kdump kernel

[Sairaj Kodilkar]

On 11/19/2025 7:13 PM, Jason Gunthorpe wrote:
> On Wed, Nov 19, 2025 at 04:20:28PM +0530, Vasant Hegde wrote:
> > > diff --git a/drivers/iommu/amd/init.c b/drivers/iommu/amd/init.c
> > > index f2991c11867c..9375fba1071c 100644
> > > --- a/drivers/iommu/amd/init.c
> > > +++ b/drivers/iommu/amd/init.c
> > > @@ -1136,9 +1136,13 @@ static void set_dte_bit(struct dev_table_entry *dte, u8 bit)
> > >   static bool __reuse_device_table(struct amd_iommu *iommu)
> > >   {
> > >   	struct amd_iommu_pci_seg *pci_seg = iommu->pci_seg;
> > > -	u32 lo, hi, old_devtb_size;
> > > +	struct dev_table_entry *old_dev_tbl_entry;
> > > +	u32 lo, hi, old_devtb_size, devid;
> > >   	phys_addr_t old_devtb_phys;
> > > +	u16 dom_id;
> > > +	bool dte_v;
> > >   	u64 entry;
> > > +	int ret;
> > >   
> > >   	/* Each IOMMU use separate device table with the same size */
> > >   	lo = readl(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET);
> > > @@ -1173,6 +1177,25 @@ static bool __reuse_device_table(struct amd_iommu *iommu)
> > >   		return false;
> > >   	}
> > >   
> > > +	for (devid = 0; devid <= pci_seg->last_bdf; devid++) {
> > > +		old_dev_tbl_entry = &pci_seg->old_dev_tbl_cpy[devid];
> > > +		dte_v = old_dev_tbl_entry->data[0] & DTE_FLAG_V;
> > > +		dom_id = old_dev_tbl_entry->data[1] & DEV_DOMID_MASK;
> > > +
> > > +		if (!dte_v || !dom_id)
> > > +			continue;
> > > +		/*
> > > +		 * ID reseveration can fail with -ENOSPC when there
> > > +		 * are multiple devices present in the same domain,
> > > +		 * hence check only for -ENOMEM.
> > > +		 */
> > > +		ret = ida_alloc_range(&pdom_ids, dom_id, dom_id, GFP_ATOMIC);
> Is it really an atomic context? Why?

No, its not atomic context. My intension was to ensure that the memory 
allocation
succeeds. But I am not sure if it's the most appropriate choice for this 
scenario

Could you please suggest which GFP flag would be best suited for this use ?


> > > +		if (ret == -ENOMEM) {
> > > +			pr_err("Failed to reserve domain ID 0x%x\n", dom_id);
> > > +			return false;
> Please don't print on ENOMEM, there is already a print.
>
> I think you should also keep iterating as other dom_ids may still be
> fit in already allocated bitmaps. Though the system is probably toast
> if this happens anyhow.

Sure


Thanks
Sairaj
> Jason