> -----Original Message----- > From: linux-mmc-owner@xxxxxxxxxxxxxxx [mailto:linux-mmc- > owner@xxxxxxxxxxxxxxx] On Behalf Of Jaehoon Chung > Sent: Friday, December 16, 2011 12:06 PM > To: Huang Changming-R66093 > Cc: Jaehoon Chung; Philip Rakity; linux-mmc@xxxxxxxxxxxxxxx; Chris Ball > Subject: Re: [PATCH 3/4 v5] SDHCI: add sdhci_get_cd callback to detect > the card > > On 12/16/2011 12:25 PM, Huang Changming-R66093 wrote: > > >>> I am very confused, why do we read the present state register on > every > >> request? > >> > >> How long time read the present state register? > > Even if one line code is performed, I think it need time to complete. > > > >>> My codes are added to the function mmc_sd_detect in file core/sd.c > >>> Function mmc_rescan has detect the card present state repeatedly, so > I > >> think we don't need to detect the card state on every request. > >>> So I think the codes to detect the card present state in > sdhci_request > >> should be removed to improve the performance. > >> > >> How did you get the performance benefit? > > Do you know there are many tools to run the performance? > > IPForward/IPSEC for network, iozone for SATA/SDHC... > > > > We have the performance result from our platform, the result will > decrease 2%~5%! > > > > If the driver has the option 'MMC_CAP_NEEDS_POLL' (many driver enable > this option) and the card is always present, the driver will send the > command "mmc_send_status" to ask the card state repeatedly. > > Do you know what it means? Many interrupts will be generated even if > the system is reading from/writing to the SD/MMC card. > > > >> And mmc_rescan is repeatedly for SD-card? > >> If mmc_rescan is repeatedly, that reason is the below code. (just my > >> thinking) > >> > >> if (host->caps & MMC_CAP_NEEDS_POLL) > >> mmc_schedule_delayed_work(&host->detect, HZ); > >> > >> what is your point related with mmc_rescan? > >> I didn't understand yours.. > > > > There are some controllers don't support the card detect, the > SDHCI_CARD_PRESENT is forever one. And some controller must enable the > MMC_CAP_NEEDS_POLL option (about this, you can search the folder > mmc/host/), that means mmc_rescan will be run repeatedly. > > In these cases, if we detect the card state though command > mmc_send_status, a lot of interrupts will be generated. > > I know that...Some controller didn't support the card detect...But I > think you can control this problem with SDHCI_QUIRK_BROKEN_CARD_DETECTION > and some flags. > If controller is always polling for card status, that's inefficient. Maybe. But, many controller will polling the card state as controller designed. > I use the SDHCI_QUIRK_BROKEN_CARD_DETECTION, but didn't generate the > interrupt at every time(didn't polling) > I think you can solve this problem in sdhci-xxx.c, not in sdhci.c > Because there is one interface to detect card state .get_cd in structure mmc_host_ops, I just implement it in sdhci.c and just run the callback if the controller support the card detection, and think about some controller don't support card detection, I add one new interface to detect the card state in special platform. For FSL eSDHC controller, I do it in sdhci-of-esdhc.c -- To unsubscribe from this list: send the line "unsubscribe linux-mmc" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
