(6/6) bcm57xx: package

This is the bcm57xx package.  I have tested default vlan functions,
but I dont have the equipment to test more advanced setups.  The default
vlan setup seems to be working fine.  I also added the activate_gpio
parameter which will make the driver activate the switch via gpio before
probing for it.

I'm not sure which method is best for autoload.  For the wrt350n, I
need the activate_gpio parameter.  But its probably not a good idea
to add that to the autoload file.  On a system without a bcm57xx switch,
isn't it a bad idea to mess with the gpios looking for the switch? Ideally,
wouldn't it be best to load the bcm57xx module from broadcom-diag, after
it has determined which router its on?  I tried using 'request_module' from
there, but had no success.  For now, I am relying on preinit to load
the bcm57xx module with activate_gpio param, after it has failed to load
switch_robo and switch_adm.

Signed-off-by: Ben Pfountz <netprince (at) vt (dot) edu>


git-svn-id: svn://svn.openwrt.org/openwrt/trunk@11471 3c298f89-4303-0410-b956-a3cf2f4a3e73

package/broadcom-57xx/src/hndgige.c

@@ -0,0 +1,146 @@
+/*
+ * HND SiliconBackplane Gigabit Ethernet core software interface
+ *
+ * Copyright 2007, Broadcom Corporation
+ * All Rights Reserved.
+ * 
+ * THIS SOFTWARE IS OFFERED "AS IS", AND BROADCOM GRANTS NO WARRANTIES OF ANY
+ * KIND, EXPRESS OR IMPLIED, BY STATUTE, COMMUNICATION OR OTHERWISE. BROADCOM
+ * SPECIFICALLY DISCLAIMS ANY IMPLIED WARRANTIES OF MERCHANTABILITY, FITNESS
+ * FOR A SPECIFIC PURPOSE OR NONINFRINGEMENT CONCERNING THIS SOFTWARE.
+ *
+ * $Id: hndgige.c,v 1.6 2007/06/01 05:59:06 michael Exp $
+ */
+
+#include <typedefs.h>
+#include <osl.h>
+#include <pcicfg.h>
+#include <sbconfig.h>
+#include <sbutils.h>
+#include "sbgige.h"
+#include <hndpci.h>
+#include "hndgige.h"
+
+uint32
+sb_base(uint32 admatch)
+{
+	uint32 base;
+	uint type;
+
+	type = admatch & SBAM_TYPE_MASK;
+	ASSERT(type < 3);
+
+	base = 0;
+
+	if (type == 0) {
+		base = admatch & SBAM_BASE0_MASK;
+	} else if (type == 1) {
+		ASSERT(!(admatch & SBAM_ADNEG));	/* neg not supported */
+		base = admatch & SBAM_BASE1_MASK;
+	} else if (type == 2) {
+		ASSERT(!(admatch & SBAM_ADNEG));	/* neg not supported */
+		base = admatch & SBAM_BASE2_MASK;
+	}
+
+	return (base);
+}
+
+/*
+ * Setup the gige core.
+ * Resetting the core will lose all settings.
+ */
+void
+sb_gige_init(sb_t *sbh, uint32 unit, bool *rgmii)
+{
+	volatile pci_config_regs *pci;
+	sbgige_pcishim_t *ocp;
+	sbconfig_t *sb;
+	osl_t *osh;
+	uint32 statelow;
+	uint32 statehigh;
+	uint32 base;
+	uint32 idx;
+	void *regs;
+
+	/* Sanity checks */
+	ASSERT(sbh);
+	ASSERT(rgmii);
+
+	idx = sb_coreidx(sbh);
+
+	/* point to the gige core registers */
+	regs = sb_setcore(sbh, SB_GIGETH, unit);
+	ASSERT(regs);
+
+	osh = sb_osh(sbh);
+
+	pci = &((sbgige_t *)regs)->pcicfg;
+	ocp = &((sbgige_t *)regs)->pcishim;
+	sb = &((sbgige_t *)regs)->sbconfig;
+
+	/* Enable the core clock and memory access */
+	if (!sb_iscoreup(sbh))
+		sb_core_reset(sbh, 0, 0);
+
+	/*
+	 * Setup the 64K memory-mapped region base address through BAR0.
+	 * Leave the other BAR values alone.
+	 */
+	base = sb_base(R_REG(osh, &sb->sbadmatch1));
+	W_REG(osh, &pci->base[0], base);
+	W_REG(osh, &pci->base[1], 0);
+
+	/*
+	 * Enable the PCI memory access anyway. Any PCI config commands
+	 * issued before the core is enabled will go to the emulation
+	 * only and will not go to the real PCI config registers.
+	 */
+	OR_REG(osh, &pci->command, 2);
+
+	/*
+	 * Enable the posted write flush scheme as follows:
+	 *
+	 * - Enable flush on any core register read
+	 * - Enable timeout on the flush
+	 * - Disable the interrupt mask when flushing
+	 *
+	 * This differs from the default setting only in that interrupts are
+	 * not masked.  Since posted writes are not flushed on interrupt, the
+	 * driver must explicitly request a flush in its interrupt handling
+	 * by reading a core register.
+	 */
+	W_REG(osh, &ocp->FlushStatusControl, 0x68);
+
+	/*
+	 * Determine whether the GbE is in GMII or RGMII mode.  This is
+	 * indicated in bit 16 of the SBTMStateHigh register, which is
+	 * part of the core-specific flags field.
+	 *
+	 * For GMII, bypass the Rx/Tx DLLs, i.e. add no delay to RXC/GTXC
+	 * within the core.  For RGMII, do not bypass the DLLs, resulting
+	 * in added delay for RXC/GTXC.  The SBTMStateLow register contains
+	 * the controls for doing this in the core-specific flags field:
+	 *
+	 *   bit 24 - Enable DLL controls
+	 *   bit 20 - Bypass Rx DLL
+	 *   bit 19 - Bypass Tx DLL
+	 */
+	statelow = R_REG(osh, &sb->sbtmstatelow);	/* DLL controls */
+	statehigh = R_REG(osh, &sb->sbtmstatehigh);	/* GMII/RGMII mode */
+	if ((statehigh & (1 << 16)) != 0)	/* RGMII */
+	{
+		statelow &= ~(1 << 20);		/* no Rx bypass (delay) */
+		statelow &= ~(1 << 19);		/* no Tx bypass (delay) */
+		*rgmii = TRUE;
+	}
+	else					/* GMII */
+	{
+		statelow |= (1 << 20);		/* Rx bypass (no delay) */
+		statelow |= (1 << 19);		/* Tx bypass (no delay) */
+		*rgmii = FALSE;
+	}
+	statelow |= (1 << 24);			/* enable DLL controls */
+	W_REG(osh, &sb->sbtmstatelow, statelow);
+
+	sb_setcoreidx(sbh, idx);
+}