From: Doug Ledford <dledford@redhat.com>
Subject: [Patch RHEL5] Fix 64k page table problems on ppc specific ehca	driver
Date: Mon, 13 Nov 2006 15:19:49 -0500
Bugzilla: 199765
Message-Id: <1163449189.2023.20.camel@fc6.xsintricity.com>
Changelog: Fix 64k page table problems on ppc specific ehca driver


The ehca driver is specific to ppc hardware (it only exists on the GX
bus that is part of IBM's ppc architecure).  That driver, which was
included because it is an Infiniband specific hardware driver on the ppc
platform is workable, but not specifically in the 64k pagetable case.
The attached patch is needed to work in the 64k pagetable case (I can't
test this, I don't have the hardware, but IBM is testing our stuff and
says this is necessary for it to work).  In addition to this patch, the
generic config file for ppc64 needs to enable the
CONFIG_INFINIBAND_EHCA_SCALING option (I had left this turned off
because it's labeled experimental, but if it's necessary to work with
64k pagetables, then we don't really have a choice but to enable it).
I've verified that the patch applies cleanly, but I can't run test this,
and haven't compile tested it (brew compile power is a precious resource
in my opinion and testing on x86/x86_64 doesn't help).

So, in order to properly close bz199765, we need the config change and
this patch.  As mentioned, I can't test it, I just have to assume IBM
did their job correctly and rely on their test results (we just don't
have the hardware, period, and although that may change in the future,
for now IBM is our eyes and ears on the issue).

-- 
Doug Ledford <dledford@redhat.com>
              GPG KeyID: CFBFF194
              http://people.redhat.com/dledford

Infiniband specific RPMs available at
              http://people.redhat.com/dledford/Infiniband

--- linux-2.6.18.noarch/drivers/infiniband/hw/ehca/ehca_iverbs.h.64k	2006-11-13 15:12:06.000000000 -0500
+++ linux-2.6.18.noarch/drivers/infiniband/hw/ehca/ehca_iverbs.h	2006-11-13 15:16:41.000000000 -0500
@@ -178,4 +178,12 @@ int ehca_mmap_register(u64 physical,void
 
 int ehca_munmap(unsigned long addr, size_t len);
 
+#ifdef CONFIG_PPC_64K_PAGES
+void *ehca_alloc_fw_ctrlblock(void);
+void ehca_free_fw_ctrlblock(void *ptr);
+#else
+#define ehca_alloc_fw_ctrlblock() ((void*)get_zeroed_page(GFP_KERNEL))
+#define ehca_free_fw_ctrlblock(ptr) free_page((unsigned long)(ptr))
+#endif
+
 #endif
--- linux-2.6.18.noarch/drivers/infiniband/hw/ehca/ehca_mrmw.c.64k	2006-11-13 15:12:06.000000000 -0500
+++ linux-2.6.18.noarch/drivers/infiniband/hw/ehca/ehca_mrmw.c	2006-11-13 15:16:41.000000000 -0500
@@ -1013,7 +1013,7 @@ int ehca_reg_mr_rpages(struct ehca_shca 
 	u32 i;
 	u64 *kpage;
 
-	kpage = kzalloc(H_CB_ALIGNMENT, GFP_KERNEL);
+	kpage = ehca_alloc_fw_ctrlblock();
 	if (!kpage) {
 		ehca_err(&shca->ib_device, "kpage alloc failed");
 		ret = -ENOMEM;
@@ -1092,7 +1092,7 @@ int ehca_reg_mr_rpages(struct ehca_shca 
 
 
 ehca_reg_mr_rpages_exit1:
-	kfree(kpage);
+	ehca_free_fw_ctrlblock(kpage);
 ehca_reg_mr_rpages_exit0:
 	if (ret)
 		ehca_err(&shca->ib_device, "ret=%x shca=%p e_mr=%p pginfo=%p "
@@ -1124,7 +1124,7 @@ inline int ehca_rereg_mr_rereg1(struct e
 	ehca_mrmw_map_acl(acl, &hipz_acl);
 	ehca_mrmw_set_pgsize_hipz_acl(&hipz_acl);
 
-	kpage = kzalloc(H_CB_ALIGNMENT, GFP_KERNEL);
+	kpage = ehca_alloc_fw_ctrlblock();
 	if (!kpage) {
 		ehca_err(&shca->ib_device, "kpage alloc failed");
 		ret = -ENOMEM;
@@ -1181,7 +1181,7 @@ inline int ehca_rereg_mr_rereg1(struct e
 	}
 
 ehca_rereg_mr_rereg1_exit1:
-	kfree(kpage);
+	ehca_free_fw_ctrlblock(kpage);
 ehca_rereg_mr_rereg1_exit0:
 	if ( ret && (ret != -EAGAIN) )
 		ehca_err(&shca->ib_device, "ret=%x lkey=%x rkey=%x "
--- linux-2.6.18.noarch/drivers/infiniband/hw/ehca/ehca_main.c.64k	2006-11-13 15:12:06.000000000 -0500
+++ linux-2.6.18.noarch/drivers/infiniband/hw/ehca/ehca_main.c	2006-11-13 15:16:41.000000000 -0500
@@ -40,6 +40,9 @@
  * POSSIBILITY OF SUCH DAMAGE.
  */
 
+#ifdef CONFIG_PPC_64K_PAGES
+#include <linux/slab.h>
+#endif
 #include "ehca_classes.h"
 #include "ehca_iverbs.h"
 #include "ehca_mrmw.h"
@@ -99,6 +102,25 @@ static spinlock_t shca_list_lock;
 
 static struct timer_list poll_eqs_timer;
 
+#ifdef CONFIG_PPC_64K_PAGES
+static struct kmem_cache *ctblk_cache = NULL;
+
+void *ehca_alloc_fw_ctrlblock(void)
+{
+	void *ret = kmem_cache_zalloc(ctblk_cache, SLAB_KERNEL);
+	if (!ret)
+		ehca_gen_err("Out of memory for ctblk");
+	return ret;
+}
+
+void ehca_free_fw_ctrlblock(void *ptr)
+{
+	if (ptr)
+		kmem_cache_free(ctblk_cache, ptr);
+
+}
+#endif
+
 static int ehca_create_slab_caches(void)
 {
 	int ret;
@@ -133,6 +155,17 @@ static int ehca_create_slab_caches(void)
 		goto create_slab_caches5;
 	}
 
+#ifdef CONFIG_PPC_64K_PAGES
+	ctblk_cache = kmem_cache_create("ehca_cache_ctblk",
+					EHCA_PAGESIZE, H_CB_ALIGNMENT,
+					SLAB_HWCACHE_ALIGN,
+					NULL, NULL);
+	if (!ctblk_cache) {
+		ehca_gen_err("Cannot create ctblk SLAB cache.");
+		ehca_cleanup_mrmw_cache();
+		goto create_slab_caches5;
+	}
+#endif
 	return 0;
 
 create_slab_caches5:
@@ -157,6 +190,10 @@ static void ehca_destroy_slab_caches(voi
 	ehca_cleanup_qp_cache();
 	ehca_cleanup_cq_cache();
 	ehca_cleanup_pd_cache();
+#ifdef CONFIG_PPC_64K_PAGES
+	if (ctblk_cache)
+		kmem_cache_destroy(ctblk_cache);
+#endif
 }
 
 #define EHCA_HCAAVER  EHCA_BMASK_IBM(32,39)
@@ -168,7 +205,7 @@ int ehca_sense_attributes(struct ehca_sh
 	u64 h_ret;
 	struct hipz_query_hca *rblock;
 
-	rblock = kzalloc(H_CB_ALIGNMENT, GFP_KERNEL);
+	rblock = ehca_alloc_fw_ctrlblock();
 	if (!rblock) {
 		ehca_gen_err("Cannot allocate rblock memory.");
 		return -ENOMEM;
@@ -211,7 +248,7 @@ int ehca_sense_attributes(struct ehca_sh
 	shca->sport[1].rate = IB_RATE_30_GBPS;
 
 num_ports1:
-	kfree(rblock);
+	ehca_free_fw_ctrlblock(rblock);
 	return ret;
 }
 
@@ -220,7 +257,7 @@ static int init_node_guid(struct ehca_sh
 	int ret = 0;
 	struct hipz_query_hca *rblock;
 
-	rblock = kzalloc(H_CB_ALIGNMENT, GFP_KERNEL);
+	rblock = ehca_alloc_fw_ctrlblock();
 	if (!rblock) {
 		ehca_err(&shca->ib_device, "Can't allocate rblock memory.");
 		return -ENOMEM;
@@ -235,7 +272,7 @@ static int init_node_guid(struct ehca_sh
 	memcpy(&shca->ib_device.node_guid, &rblock->node_guid, sizeof(u64));
 
 init_node_guid1:
-	kfree(rblock);
+	ehca_free_fw_ctrlblock(rblock);
 	return ret;
 }
 
@@ -431,7 +468,7 @@ static ssize_t  ehca_show_##name(struct 
 									   \
 	shca = dev->driver_data;					   \
 									   \
-	rblock = kzalloc(H_CB_ALIGNMENT, GFP_KERNEL);			   \
+	rblock = ehca_alloc_fw_ctrlblock();				   \
 	if (!rblock) {						           \
 		dev_err(dev, "Can't allocate rblock memory.");		   \
 		return 0;						   \
@@ -439,12 +476,12 @@ static ssize_t  ehca_show_##name(struct 
 									   \
 	if (hipz_h_query_hca(shca->ipz_hca_handle, rblock) != H_SUCCESS) { \
 		dev_err(dev, "Can't query device properties");	   	   \
-		kfree(rblock);					   	   \
+		ehca_free_fw_ctrlblock(rblock);			   	   \
 		return 0;					   	   \
 	}								   \
 									   \
 	data = rblock->name;                                               \
-	kfree(rblock);                                                     \
+	ehca_free_fw_ctrlblock(rblock);                                    \
 									   \
 	if ((strcmp(#name, "num_ports") == 0) && (ehca_nr_ports == 1))	   \
 		return snprintf(buf, 256, "1\n");			   \
--- linux-2.6.18.noarch/drivers/infiniband/hw/ehca/Kconfig.64k	2006-11-13 15:12:06.000000000 -0500
+++ linux-2.6.18.noarch/drivers/infiniband/hw/ehca/Kconfig	2006-11-13 15:16:41.000000000 -0500
@@ -10,6 +10,7 @@ config INFINIBAND_EHCA
 config INFINIBAND_EHCA_SCALING
 	bool "Scaling support (EXPERIMENTAL)"
 	depends on IBMEBUS && INFINIBAND_EHCA && HOTPLUG_CPU && EXPERIMENTAL
+	default y
 	---help---
 	eHCA scaling support schedules the CQ callbacks to different CPUs.
 
--- linux-2.6.18.noarch/drivers/infiniband/hw/ehca/ehca_irq.c.64k	2006-11-13 15:12:06.000000000 -0500
+++ linux-2.6.18.noarch/drivers/infiniband/hw/ehca/ehca_irq.c	2006-11-13 15:16:41.000000000 -0500
@@ -137,7 +137,7 @@ int ehca_error_data(struct ehca_shca *sh
 	u64 *rblock;
 	unsigned long block_count;
 
-	rblock = kzalloc(H_CB_ALIGNMENT, GFP_KERNEL);
+	rblock = ehca_alloc_fw_ctrlblock();
 	if (!rblock) {
 		ehca_err(&shca->ib_device, "Cannot allocate rblock memory.");
 		ret = -ENOMEM;
@@ -158,8 +158,8 @@ int ehca_error_data(struct ehca_shca *sh
 
 		length = EHCA_BMASK_GET(ERROR_DATA_LENGTH, rblock[0]);
 
-		if (length > PAGE_SIZE)
-			length = PAGE_SIZE;
+		if (length > EHCA_PAGESIZE)
+			length = EHCA_PAGESIZE;
 
 		print_error_data(shca, data, rblock, length);
 	}
@@ -168,7 +168,7 @@ int ehca_error_data(struct ehca_shca *sh
 			 "Error data could not be fetched: %lx", resource);
 	}
 
-	kfree(rblock);
+	ehca_free_fw_ctrlblock(rblock);
 
 error_data1:
 	return ret;
--- linux-2.6.18.noarch/drivers/infiniband/hw/ehca/ehca_hca.c.64k	2006-11-13 15:12:06.000000000 -0500
+++ linux-2.6.18.noarch/drivers/infiniband/hw/ehca/ehca_hca.c	2006-11-13 15:16:41.000000000 -0500
@@ -40,6 +40,7 @@
  */
 
 #include "ehca_tools.h"
+#include "ehca_iverbs.h"
 #include "hcp_if.h"
 
 int ehca_query_device(struct ib_device *ibdev, struct ib_device_attr *props)
@@ -49,7 +50,7 @@ int ehca_query_device(struct ib_device *
 					      ib_device);
 	struct hipz_query_hca *rblock;
 
-	rblock = kzalloc(H_CB_ALIGNMENT, GFP_KERNEL);
+	rblock = ehca_alloc_fw_ctrlblock();
 	if (!rblock) {
 		ehca_err(&shca->ib_device, "Can't allocate rblock memory.");
 		return -ENOMEM;
@@ -96,7 +97,7 @@ int ehca_query_device(struct ib_device *
 		= min_t(int, rblock->max_total_mcast_qp_attach, INT_MAX);
 
 query_device1:
-	kfree(rblock);
+	ehca_free_fw_ctrlblock(rblock);
 
 	return ret;
 }
@@ -109,7 +110,7 @@ int ehca_query_port(struct ib_device *ib
 					      ib_device);
 	struct hipz_query_port *rblock;
 
-	rblock = kzalloc(H_CB_ALIGNMENT, GFP_KERNEL);
+	rblock = ehca_alloc_fw_ctrlblock();
 	if (!rblock) {
 		ehca_err(&shca->ib_device, "Can't allocate rblock memory.");
 		return -ENOMEM;
@@ -162,7 +163,7 @@ int ehca_query_port(struct ib_device *ib
 	props->active_speed    = 0x1;
 
 query_port1:
-	kfree(rblock);
+	ehca_free_fw_ctrlblock(rblock);
 
 	return ret;
 }
@@ -178,7 +179,7 @@ int ehca_query_pkey(struct ib_device *ib
 		return -EINVAL;
 	}
 
-	rblock = kzalloc(H_CB_ALIGNMENT, GFP_KERNEL);
+	rblock = ehca_alloc_fw_ctrlblock();
 	if (!rblock) {
 		ehca_err(&shca->ib_device,  "Can't allocate rblock memory.");
 		return -ENOMEM;
@@ -193,7 +194,7 @@ int ehca_query_pkey(struct ib_device *ib
 	memcpy(pkey, &rblock->pkey_entries + index, sizeof(u16));
 
 query_pkey1:
-	kfree(rblock);
+	ehca_free_fw_ctrlblock(rblock);
 
 	return ret;
 }
@@ -211,7 +212,7 @@ int ehca_query_gid(struct ib_device *ibd
 		return -EINVAL;
 	}
 
-	rblock = kzalloc(H_CB_ALIGNMENT, GFP_KERNEL);
+	rblock = ehca_alloc_fw_ctrlblock();
 	if (!rblock) {
 		ehca_err(&shca->ib_device, "Can't allocate rblock memory.");
 		return -ENOMEM;
@@ -227,7 +228,7 @@ int ehca_query_gid(struct ib_device *ibd
 	memcpy(&gid->raw[8], &rblock->guid_entries[index], sizeof(u64));
 
 query_gid1:
-	kfree(rblock);
+	ehca_free_fw_ctrlblock(rblock);
 
 	return ret;
 }
--- linux-2.6.18.noarch/drivers/infiniband/hw/ehca/ehca_qp.c.64k	2006-11-13 15:12:06.000000000 -0500
+++ linux-2.6.18.noarch/drivers/infiniband/hw/ehca/ehca_qp.c	2006-11-13 15:16:41.000000000 -0500
@@ -811,7 +811,7 @@ static int internal_modify_qp(struct ib_
 	unsigned long spl_flags = 0;
 
 	/* do query_qp to obtain current attr values */
-	mqpcb = kzalloc(H_CB_ALIGNMENT, GFP_KERNEL);
+	mqpcb = ehca_alloc_fw_ctrlblock();
 	if (mqpcb == NULL) {
 		ehca_err(ibqp->device, "Could not get zeroed page for mqpcb "
 			 "ehca_qp=%p qp_num=%x ", my_qp, ibqp->qp_num);
@@ -1225,7 +1225,7 @@ modify_qp_exit2:
 	}
 
 modify_qp_exit1:
-	kfree(mqpcb);
+	ehca_free_fw_ctrlblock(mqpcb);
 
 	return ret;
 }
@@ -1276,7 +1276,7 @@ int ehca_query_qp(struct ib_qp *qp,
 		return -EINVAL;
 	}
 
-	qpcb = kzalloc(H_CB_ALIGNMENT, GFP_KERNEL );
+	qpcb = ehca_alloc_fw_ctrlblock();
 	if (!qpcb) {
 		ehca_err(qp->device,"Out of memory for qpcb "
 			 "ehca_qp=%p qp_num=%x", my_qp, qp->qp_num);
@@ -1400,7 +1400,7 @@ int ehca_query_qp(struct ib_qp *qp,
 		ehca_dmp(qpcb, 4*70, "qp_num=%x", qp->qp_num);
 
 query_qp_exit1:
-	kfree(qpcb);
+	ehca_free_fw_ctrlblock(qpcb);
 
 	return ret;
 }
--- linux-2.6.18.noarch/drivers/infiniband/hw/ehca/hcp_phyp.c.64k	2006-11-13 15:12:06.000000000 -0500
+++ linux-2.6.18.noarch/drivers/infiniband/hw/ehca/hcp_phyp.c	2006-11-13 15:16:41.000000000 -0500
@@ -44,13 +44,14 @@
 
 int hcall_map_page(u64 physaddr, u64 *mapaddr)
 {
-	*mapaddr = (u64)(ioremap(physaddr, EHCA_PAGESIZE));
+	*mapaddr = (u64)ioremap((physaddr & PAGE_MASK), PAGE_SIZE) +
+		(physaddr & (~PAGE_MASK));
 	return 0;
 }
 
 int hcall_unmap_page(u64 mapaddr)
 {
-	iounmap((volatile void __iomem*)mapaddr);
+	iounmap((void __iomem*)(mapaddr & PAGE_MASK));
 	return 0;
 }