Skip to contentSkip to conversation

Patch: Increase maximum bus memory window size (for Xeon Phi)

PATCH Bryce Lelbach 3 participants 4 messages May 24, 2013 – Oct 4, 2013 PCI

lore
Share & tools
lore
#0 lore raw

The attached patch affects pbus_size_mem() in drivers/pci/setup-bus.c. The patch increases the maximum memory window size for a PCI bus from 2 Gb to 2^63 Gb.

This change is necessary to support Intel Xeon Phi co-processors. These co-processors are PCIe devices used for high-performance computing applications. The device requires an 8Gb memory window to function (the co-processor has 8Gb of onboard memory).

This patch is a modified version of a patch from Intel's MPSS framework (specifically, from the "KNC_gold_update_1-2.1.4982-15-rhel-6.3" package), which will apply to a 3.7.8 kernel (I am about to try it on a 3.8 kernel). To the best of my knowledge, newer RHEL kernels are shipped with this patch.

Signature · 11 lines
Bryce Adelstein-Lelbach aka wash
STE||AR Group, Center for Computation and Technology, LSU
--
225-317-3866 - iPhone
225-578-6182 - Work (no voicemail)
--
stellar.cct.lsu.edu
boost-spirit.com
llvm.linuxfoundation.org
cppnow.org
--
drivers/pci/setup-bus.c +3 -2
diff -u -r -N linux-source-3.7/drivers/pci/setup-bus.c linux-source-3.7-xeon-phi/drivers/pci/setup-bus.c
@@ -852,7 +852,7 @@
 {
 	struct pci_dev *dev;
 	resource_size_t min_align, align, size, size0, size1;
-	resource_size_t aligns[12];	/* Alignments from 1Mb to 2Gb */
+	resource_size_t aligns[44];	/* Alignments from 1Mb to 2^63 */
 	int order, max_order;
 	struct resource *b_res = find_free_bus_resource(bus, type);
 	unsigned int mem64_mask = 0;
@@ -891,7 +891,8 @@
 			/* For bridges size != alignment */
 			align = pci_resource_alignment(dev, r);
 			order = __ffs(align) - 20;
-			if (order > 11) {
+			if ((sizeof(size_t) == 4 && order > 11) ||
+					(sizeof(size_t) == 8 && order > 43)) {
 				dev_warn(&dev->dev, "disabling BAR %d: %pR "
 					 "(bad alignment %#llx)\n", i, r,
 					 (unsigned long long) align);
#1 lore raw
Quoted 12 lines · Bryce Lelbach

The attached patch affects pbus_size_mem() in drivers/pci/setup-bus.c. The patch increases the maximum memory window size for a PCI bus from 2 Gb to 2^63 Gb.

This change is necessary to support Intel Xeon Phi co-processors. These co-processors are PCIe devices used for high-performance computing applications. The device requires an 8Gb memory window to function (the co-processor has 8Gb of onboard memory).

This patch is a modified version of a patch from Intel's MPSS framework (specifically, from the "KNC_gold_update_1-2.1.4982-15-rhel-6.3" package), which will apply to a 3.7.8 kernel (I am about to try it on a 3.8 kernel). To the best of my knowledge, newer RHEL kernels are shipped with this patch.

This patch is not in (the latest) RHEL6(.4) or the internal staging tree for the next RHEL6(.5) release.

#2 lore raw

Sorry, my mistake (re: the RHEL thing).

I'll see if I can dig up the original author's name.

Signature · 11 lines
Bryce Adelstein-Lelbach aka wash
STE||AR Group, Center for Computation and Technology, LSU
--
225-317-3866 - iPhone
225-578-6182 - Work (no voicemail)
--
stellar.cct.lsu.edu
boost-spirit.com
llvm.linuxfoundation.org
cppnow.org
--
maintainer
#3 lore raw
In reply to Bryce Lelbach
Quoted 11 lines · Bryce Lelbach

The attached patch affects pbus_size_mem() in drivers/pci/setup-bus.c. The patch increases the maximum memory window size for a PCI bus from 2 Gb to 2^63 Gb.

This change is necessary to support Intel Xeon Phi co-processors. These co-processors are PCIe devices used for high-performance computing applications. The device requires an 8Gb memory window to function (the co-processor has 8Gb of onboard memory).

This patch is a modified version of a patch from Intel's MPSS framework (specifically, from the "KNC_gold_update_1-2.1.4982-15-rhel-6.3" package), which will apply to a 3.7.8 kernel (I am about to try it on a 3.8 kernel). To the best of my knowledge, newer RHEL kernels are shipped with this patch.

This needs a Signed-off-by before it can be applied (see Documentation/SubmittingPatches for details about what this means).

Also, there's some confusion about this being in RHEL. I take it that it is not actually in RHEL, but perhaps in an Intel-supplied update that is commonly applied on top of RHEL?

IIRC, a very similar patch was posted last year. It'd be nice to dig that up and credit that author as well.

It's more convenient for me to receive patches as plain-text email (not attachments), also as described in SubmittingPatches. I *can* deal with attachments, but it's a hassle and I tend to keep putting them off ;)

Bjorn